Decision Making in Criminal Justice Revisited: Toward a ... · Criminal Justice Dependent Variable 3 Criminal ... innocent individual from being processed through the system and punished.

Decision Making in Criminal Justice Revisited: Toward a General

Theory of Criminal Justice

A dissertation submitted to the

Graduate School

of the University of Cincinnati

in partial fulfillment of the

requirements for the degree of

Doctor of Philosophy

in the School of Criminal Justice

of the College of Education, Criminal Justice, and Human Services

by

Daniel J Lytle

B.A. Marquette University, 2005 M.S. University of Cincinnati, 2007

Committee Members Lawrence F. Travis, III, Ph.D. (Chair)

James Frank, Ph.D. Paula Smith, Ph.D.

William King, Ph.D.

ii

ABSTRACT The study of correlates of decision-making has been an important and integral part of

criminal justice research. While this important research has been studied largely since the

conception of criminal justice research, the attempts to take stock and determine what the body

of research collectively “knows” has been limited. These limitations have included both the

method of synthesis and also the focus of the synthesis. Traditionally, criminal justice scholars

have taken stock of correlate research by using narrative literature review techniques. While this

research does make some contribution to the field, it is flawed because of the inherent problems

in all narrative reviews, namely, double counting studies, and the subjective nature of narrative

reviews. The flaw has been that reviews have focused on a few important variables, race and

gender, and not on a more comprehensive view.

This dissertation seeks to expand upon the current state of correlates of decision-making

research by conducting a meta-analysis to examine decision making across the system,

specifically, arrest, sentencing, and parole revocation. Results indicated that several variables

are important at multiple stages of the criminal justice system. These predictors are both legal

and extra-legal variables, legal predictors of decision-making included seriousness of the

offense, and the offender’s prior criminal record, while extra-legal predictors of decision-making

included race, gender, and ethnicity. These findings were not mitigated by moderating factors,

but instead persisted across moderator categories. In addition to system-wide correlates, there

were several factors, which were unique to a specific decision point, these included, at arrest,

suspect demeanor, and, at sentencing, mode of conviction.

iii

iv

ACKNOWLEDGEMENTS There are many people that I would like to thank that made this process possible. First, I

would like to thank my dissertation chair and mentor Dr. Lawrence F. Travis, III. You have

mentored me since I walked through the doors at the University of Cincinnati and I am eternally

grateful for your guidance, mentorship, and the occasional figurative swift kick in the pants.

You have helped to mold me into a scholar. Second, I would like to thank Dr. James Frank;

thank you for your guidance on this project and your mentorship as I progressed through the

doctoral program at the University of Cincinnati. Third, I would like to thank Dr. Paula Smith

for your guidance and assistance with the technical aspects of conducting this meta-analysis.

Fourth, I would like to thank Dr. William King for serving as my outside reader on this project.

Thank you all for your service, helpful comments, and assistance in this dissertation process.

I would also like to thank the entire criminal justice faculty at the University of

Cincinnati. I would especially like to thank Dr. Robin Engel for first exposing me to criminal

justice theory and for writing the piece that served as an inspiration for this project. There are

two other professors that I have had over the course of my academic career that deserve special

mention, Dr. Carol Archbold and Dr. Amy Stichman. Thank you to you both for setting me on

the path of graduate school at the University of Cincinnati.

Finally, I would like to thank my friends and family. First and foremost, I would like to

thank my parents Joseph and Juanita Lytle. Without your encouragement and support this

process would never have been possible. Second, there are several friends that deserve special

mention. I would like to thank Cody Stoddard, Ryan Randa, Rich Lemke and Jackee Rohrbach.

I would also like to thank Jim McCafferty for assisting with my reliability coding, and Kevan

Galyean and Cheryl Jonson for answering those random questions.

v

TABLE OF CONTENTS ABSTRACT ii

ACKNOWLEDGEMENTS iv LIST OF TABLES viii LIST OF FIGURES ix CHAPTER 1: PROBLEM STATMENT 1 What is Criminal Justice? 1 Criminal Justice Dependent Variable 3 Criminal Justice Theory 4 Duality of criminal justice 6 Meta-analysis, criminal justice and criminal justice theory 7 Criminal Justice Knowledge Synthesis 9 Chapter Summary 10 A Look Ahead: The Plan for this Dissertation 11 CHAPTER 2: LITERATURE REVIEW 13 Situational Theories 14 Gottfredson and Gottfredson 14 Focal concerns theory 15 Actor Theories 16 Workgroup/Organizational Theories 16 Courtroom workgroup 17 Police workgroup 19 Community Theories 21 Black 21 Duffee 23

Wilson 26 Conflict Theory 29 Summary 30 Research Synthesis in Criminal Justice 31 Narrative Reviews 32 Arrest 33 Summary 37 Sentencing 37 Parole revocation 38 Meta-Analysis 39 Gender 40 Race 40 Victim characteristics 42

vi

Chapter Summary 42 CHAPTER 3: META-ANALYSIS: BENEFITS AND CRITIQUES 46 Traditional Narrative Literature Reviews 47 Vote-counting 48 Meta-Analysis as an Alternative 49 Advantages of Meta-Analysis 50 Effect size 50 Methodological adjustments 51 Replication 51 Dynamic 52 Critiques of Meta-Analysis 52 File drawer problem 52 Apples-and-oranges problem 54 Independence of effect sizes 55 Chapter Summary 56 CHAPTER 4: METHODS 59 Research Questions 59 Sample of Studies 60 Inclusion Criteria 62 Selection Criteria 62 Dependent Variable 63 Independent Variables 65 System-wide variables 65 Arrest 66 Sentencing 66 Parole revocation 66 Moderating Variables 67 Study characteristics 67 Methodological quality 68 Reliability of Coding 69 Analytic Strategy 70 Effect size measure 70 Heterogeneity of measures 72 Fail-safe N statistic 74 Conclusions 74 Limitations 76 CHAPTER 5: RESULTS 77 Publication Characteristics 77 Arrest 81 Sample characteristics 81

vii

Effect sizes 84 Fail-safe N analysis 86 Moderating variable analysis 87 Sentencing 113

Sample characteristics 113 Effect sizes 116 Fail-safe N analysis 118 Moderating variable analysis 119 Parole Revocation 138 Sample characteristics 138 Effect sizes 141 Fail-safe N analysis 142 Moderating variable analysis 143 Chapter Summary 149 CHAPTER 6: DISCUSSION 151 Summary of Findings 152 Missing variables 156 Criminal Justice Theory Organization 157 Toward a General Theory of Criminal Justice 159 Future Research 160 Limitations 161 REFERENCES 164 APPENDIX A 180 Arrest Coding Guide 180 Sentencing Coding Guide 193 Parole Revocation Coding Guide 205

viii

LIST OF TABLES Table 5.1: Publication Characteristics 80 Table 5.2: Arrest Sample Characteristics 83 Table 5.3: Mean Effect Sizes Arrest 85 Table 5.4: Suspect Race Moderator Analysis 91 Table 5.5: Suspect Gender Moderator Analysis 95 Table 5.6: Suspect Ethnicity Moderator Analysis 99 Table 5.7: Suspect Intoxication Moderator Analysis 102 Table 5.8: Suspect Weapon Use Moderator Analysis 105 Table 5.9: Suspect Demeanor Moderator Analysis 109 Table 5.10: Sentencing Sample Characteristics 115 Table 5.11: Mean Effect Sizes Sentencing 117 Table 5.12: Defendant Race Moderator Analysis 122 Table 5.13: Defendant Gender Moderator Analysis 125 Table 5.14: Defendant Ethnicity Moderator Analysis 128 Table 5.15: Prior Record Moderator Analysis 131 Table 5.16: Attorney Type Moderator Analysis 134 Table 5.17: Mode of Conviction Moderator Analysis 136 Table 5.18: Parole Revocation Sample Characteristics 140 Table 5.19: Mean Effect Sizes Parole Revocation 141 Table 5.20: Offender Race Moderator Analysis 145 Table 5.21: Risk Moderator Analysis 147

ix

LIST OF FIGURES Figure 2.1: Diagram of Gottfredson and Gottfredson’s Theory of Criminal Justice 14 Figure 2.2: Diagram of Focal Concerns Theory 15 Figure 2.3: Diagram of Eisenstein and Jacob’s Courtroom Workgroup Theory 19 Figure 2.4: Diagram of Klinger’s Ecological Theory of Police Behavior 20 Figure 2.5: Diagram of Black’s Theory of the Behavior of Law 22 Figure 2.6: Diagram of Duffee’s Theory of Criminal Justice 25 Figure 2.7: Diagram of Wilson’s Theory of Police Behavior 27 Figure 2.8: Diagram of Scheingold’s Theory of Policy Creation 29

1

CHAPTER 1

PROBLEM STATEMENT

According to Marenin and Worrall (1998), “criminal justice is an academic discipline in

practice but not yet in theory” (p. 465). The quote by Marenin and Worrall sums up the major

problem with the field of criminal justice science research. There is a perception that criminal

justice science as a field of study is atheoretical. However, in order for any scientific field to

become a full-fledged discipline, it must create theory. The field of criminal justice is

developing toward a strong scientific discipline. In order to achieve this status and be seen as a

scientific discipline in its own right, criminal justice must continue the trend of recent

scholarship that has put a direct emphasis on examining this critical issue (see Bernard and

Engel, 2001; Crank and Bowman, 2008 and Duffee and Maguire, 2007). The goal of this

dissertation is to continue this recent trend of criminal justice theory development by performing

a comprehensive review of correlates of decision-making. Through this analysis I will seek to

determine the most appropriate way to categorize and organize the field of criminal justice

theory.

What is Criminal Justice?

The concept of criminal justice is complex. Packer (1968) argues that there are two ideal

types of criminal justice systems, the due process and crime control models. The due process

model is described as an obstacle course, where blockades are constantly put up to prevent any

innocent individual from being processed through the system and punished. Conversely, the

crime control model is described as an assembly line. Justice must be meted out with

consistency to have the impact on criminal conduct. Duffee and Allan (2007) attempt to clarify

the study of criminal justice by defining criminal justice science as “the study of governmental

2

social control premised on punishment or blameworthiness” (p. 8). In contrast, Duffee and Allan

(2007) define criminology as “the study of crime and criminal behavior” (p. 8). Using these

definitions, a study that examines the factors that influence arrest would be a study of criminal

justice and a study that looks at the effective crime reduction of a policing intervention would be

a study of criminology. The critical difference between the two fields is that criminal justice

science seeks to understand the process of the criminal justice system and criminology is

concerned with the effectiveness of criminal justice interventions. Thus, criminology is the

study of crime and criminal justice is the study of discretion. Discretion is the ability to choose,

and within the context of the criminal justice process the decision of interest is whether to

process an individual through the criminal justice system (Davis, 1969).

Since discretion is the ability to make a choice, it is necessary to consider the definition

of a decision. According to Gottfredson and Gottfredson (1989), decisions have 3 components:

goals, alternatives, and information. By goals, they mean that a decision must have some

objective. For example, the decision to arrest may have a goal of taking a suspect away from the

person he or she is victimizing. The second component to a decision is alternatives, which

means that there must be multiple options. If there was only one option in a “decision,” one

could not make a choice. Instead, that person would be mandated to do some act. According to

the authors, the last component of a decision is information. Information is the data that are used

to make the best possible decision. The best decision is one in which there is little “uncertainty

about the consequences of the decision” (Gottfredson and Gottfredson, 1989, p. 3).

3

Criminal Justice Dependent Variable

Criminological theories center around one dependent variable, crime. While crime may

be operationalized in a variety of ways, criminological research centers on the study of crime and

delinquency. The logical organization of criminological theory, since it has one dependent

variable, is to organize theories around primary predictive factors. Since criminal justice is a

related discipline it would appear to be appropriate to categorize and organize criminal justice

theory in a similar manner.

However, scholars do not agree that organizing criminal justice theory by key

independent variables is possible for criminal justice theory. The disagreement revolves around

the number of dependent variables that criminal justice studies. Due to the relatively broad

definition above, criminal justice covers a greater number of discrete topics than criminology.

Bernard and Engel (2001) subsequently argue that since there are so many dependent variables,

criminal justice theories should be categorized around each these dependent variables.

This viewpoint creates a problem for criminal justice research. That problem is

confusion on the number of dependent variables. Since criminal justice is the study of the

process by which formal social control is executed, criminal justice research is chiefly concerned

with discretionary application of that social control. Contrary to Bernard and Engel, it is

possible to think of all the discrete variables studied in criminal justice as different ways to

operationalize discretion. For example, Lundman and Kaufman (2003) and Ulmer, Kurlychek

and Kramer (2007) both conducted studies of criminal justice decision making and both used a

conflict theory framework. Lundman and Kaufman are studying traffic citations and Ulmer and

colleagues are studying judicial discretion in sentencing. Both studies are arguing essentially the

same ideas, yet, they operationalize the dependent variable differently. Since they are examining

4

the same idea, these studies should be grouped together. Thus, there are two competing

organizational frameworks. One recommends that theories be grouped based on dependent

variable then by primary independent variables. The other argues that there is one dependent

variable, discretion, and as a result theories should be organized around primary independent

variables.

Criminal Justice Theory

According to Kuhn (1996), for a discipline to be scientific, it must have theory. More

importantly, a true scientific discipline requires a paradigm. Kuhn (1996) defines a paradigm as,

1) “an achievement [that] was sufficiently unprecedented to attract an enduring group of

adherents away from competing modes of scientific activity” and 2) “it was sufficiently open-

ended to leave all sorts of problems for the redefined group of practitioners to resolve” (p. 10).

This means that a paradigm frames a discipline and dictates the important questions. When new

information creates a conflict with the current paradigm a process of scientific revolution occurs

(Kuhn, 1996). This state of flux is the current state of criminal justice theory. There are many

different perspectives operating at multiple levels and the field is waiting for a dominant

paradigm to emerge. It may be that criminal justice has created theories but no one theory has

supplanted the others to emerge as the paradigm of the field.

An equally important and related issue to the paradigm of criminal justice is how criminal

justice theory should be organized. However, because of the issue with dependent variables,

there appears to be a divide in how to categorize and organize criminal justice theory. There

have been two previous significant attempts to categorize, and organize the various perspectives

of criminal justice theory. First, Hagan (1989) attempted to categorize criminal justice theory

breaking down criminal justice theories into two perspectives: consensus and conflict. He

5

argued that all theories that place the greatest emphasis on legal variables are consensus theories,

and those theories that place the greatest emphasis on extra-legal variables would be conflict

theories. Some scholars rejected this method as too simple.

Second, Bernard and Engel (2001) argued that criminal justice theories should be

organized by the dependent variable and then by key independent variables. They argued for

grouping theories around three broad categories of dependent variables: individual criminal

justice agents, criminal justice organizations, and overall system and its components. Under this

scheme, conflict theory could be split across all three variable types when the theory is applying

the same concepts to explaining the process of criminal justice, just on different dependent

variables. These two perspectives create a fundamental problem, should criminal justice theories

focus only on the system as a whole or should criminal justice theories be decision-specific?

If criminal justice theories are grander theoretical approaches focusing on the system as a

whole, then predictors in criminal justice theory will be consistent across the system as a whole.

Under this grand theory perspective, predictors of decision-making will be the same at each stage

because the same theoretical concepts are impacting decision makers in the same way. For

example, Gottfredson and Gottfredson (1989) argue that three factors: seriousness of the offense,

victim-offender relationship and criminal history predict the actions of decision makers across

the system. From this perspective, these three factors are the primary predictors of criminal

justice actor behavior for all criminal justice actors.

Duffee and Allan (2007) argue that criminal justice theories need not address all parts of

the system to be a theory of criminal justice. Under this perspective, criminal justice theories

may be decision-specific and factors may vary across decision points. For example, Eisenstein

and Jacob (1997) argue that dynamics of the courtroom workgroup are the key predictors of

6

court-case decision making. Conversely, Wilson (1968) argues that the local political culture

and the effectiveness of the chief of police in controlling his or her officers are the key to

predicting the type of police department, which in turn influences the behavior of officers.

Under the grand theory perspective, correlates of decision-making will be consistent across the

system and various decision points within each component. If the decision-specific perspective

is correct, then correlates of decision-making will be different across the system and each

decision point.

Duality of criminal justice

Another issue to consider, as part of the conception of criminal justice decision-making

research, is the issue of the duality of criminal justice. Several scholars have commented on the

duality of the American criminal justice system, for example, Packer (see above). More recently,

Gottfredson and Gottfredson addressed this issue of the duality of criminal justice as a function

of seriousness of the offense

Gottfredson and Gottfredson (1989) argue that there are two criminal justice systems, one

for serious offenses and a second for less serious offenses. They noticed that decision-making

research indicated that criminal justice actors exhibited less variation in the exercise of their

discretion in certain situations. Low-variation discretionary situations were characterized by

serious criminal incidents, those committed by offenders with long criminal histories, and/or

those cases where the victim and offender are strangers to each other. In these cases, there was

“very little discretion exhibited at any of the major criminal justice decisions” (Gottfredson and

Gottfredson, 1989, p. 260). It is near certainty that these serious situations will be processed

through the criminal justice system, provided there is sufficient evidence.

7

The second criminal justice system is one that involves the opposite type of incidents

from the first system. Cases in this second criminal justice system involve less serious crimes,

those committed by individuals with short or no criminal history, and/or crimes between those

with a closer relational distance. Gottfredson and Gottfredson (1989) argue that this second

system is where extralegal factors become more prominent because of greater amounts of

discretion afforded to the criminal justice actor. For example, in this second criminal justice

system demeanor of the suspect may play a larger role in the decision to make an arrest.

Gottfredson and Gottfredson (1989) assert that this second system is created because 1) “the

desire by some to pursue alternatives to full processing”, 2) “the inconsistency arising from a

lack of explicit guidelines”, and/or 3) “the result from the use of invidious decision criteria” (p.

261). It is in this second system that there is less certainty that a case will be processed fully

through the criminal justice system and greater variation in the exercise of discretion.

It appears that criminal justice decision making may be a function of offense seriousness.

This means that to fully examine the effect of predictors, other than offense seriousness, offense

seriousness must first be controlled. After offense seriousness is controlled, the effect of other

predictors, such as suspect characteristics, criminal justice agent characteristics, organizational

characteristics or community characteristics may explain the remaining variation in decision

making.

Meta-analysis, criminal justice and criminal justice theory

In order to evaluate these organization methods, and this idea of the duality of criminal

justice, it is necessary to take stock of existing research and evaluate the research on correlates of

decision-making. One approach is to use narrative literature reviews. While this approach has

been a good first step, a newer and potentially superior technique for research synthesis is meta-

8

analysis. Meta-analysis statistically examines the effect of independent variables on the

dependent variable across studies. According to Hunt (1997), meta-analysis combines the effects

of variables across studies and creates a single measure of strength for that predictor on the

dependent variable. This technique has been used to evaluate criminological theories and other

anti-crime interventions (see Hubbard and Pratt, 2002; Makarios and Pratt, 2012; Pratt, Cullen,

Blevins, Daigle, and Madensen 2006; and Pratt and Cullen, 2000, 2005). This technique can be

applied to evaluate the issue of a grand theory approach, or a decision-specific approach.

Meta-analysis can be used to evaluate the correlates of decision-making across the

system. If a majority of the same correlates are significant predictors across the system, then the

grand theory approach would appear to be the most appropriate organization method. However,

if each decision yields a number of unique correlates, then criminal justice theories should be

organized around specific decisions. Moreover, for the decision-specific model to be the

preferred model, unique factors would also have to play a larger role than system-wide factor.

This requirement demonstrates that there is some quality about each decision that is a stronger

factor than system-wide factors.

It is likely not possible to fully explore the idea of the duality of criminal justice using

meta-analysis. None of the studies separated their samples based on the seriousness of their

cases. However, the idea may be partially supported if several conditions are met. First,

seriousness of the offense must be paramount. The duality idea proposed by Gottfredson and

Gottfredson hinges on the idea that there is low variation in the exercise of discretion for serious

incidents and higher variation in the exercise of discretion in less serious incidents. This means

that the mean effect sizes for measures of seriousness must be stronger than the mean effect sizes

for other predictors. Second, the mean effect size for seriousness measures must indicate that as

9

seriousness increases, the likelihood of processing also increases. There must be a positive

relationship between the dependent variable, arrest, sentencing or parole revocation, and the

independent variable. Third, mean effect sizes of seriousness measures must be significant

across decisions. Gottfredson and Gottfredson (1989) indicated that their finding regarding the

effect of seriousness was consistent at any decision point.

Criminal Justice Knowledge Synthesis

A second issue with criminal justice research is the lack of a comprehensive synthesis of

criminal justice correlates across the system. Over thirty years ago, Gottfredson and Gottfredson

undertook an attempt to synthesize knowledge of correlates of criminal justice decision-making.

No study since has attempted to replicate their work on a similar scale. Instead, research

synthesis studies have focused on one branch of the criminal justice system or on one decision

point in the criminal justice system. Synthesis research has also at times focused too myopically

on one particular variable, e.g., race, rather than take stock of the entire field of criminal justice

decision making. When the results from these individual branch/decision points are combined a

picture begins to emerge. That picture indicates that similar correlates of decision-making have

been examined at multiple stages of criminal justice processing. For example, race, gender, and

age are constantly examined in numerous studies. Yet, often the effects of these variables are

conflicting from study to study. This conflict creates problems of interpretation in research

synthesis.

The relationship of suspect demeanor to arrest is a more specific example of research

conflict. Generally, the effect of suspect demeanor as a predictor of arrest was that disrespectful

suspects were more likely to be arrested than those that deferred to the officer’s authority (see

Smith, Visher, and Davidson, 1984; Worden, 1989; and Worden and Shepard, 1996). However,

10

Klinger (1994) argued that suspects that were disrespectful were not more likely to be arrested.

Instead these supposedly disrespectful suspects were actually committing new crimes during the

interaction phase of the officer-suspect encounter. This interaction phase crime was responsible

for the arrest, not demeanor. Klinger (1994) found that demeanor was not a statistically

significant predictor when controlling for interaction-phase crime. Since Klinger’s study,

Mastrofski, Snipes, Parks and Maxwell (2000) found that police were not significantly

influenced by a suspect’s disrespect.

Meta-analysis can provide a solution to the problem of conflicting information. If the

effect size for a particular variable is significant then that variable would be seen as critical to the

understanding of that decision. For example, if measures of suspect demeanor have an overall

significant effect, then Klinger (1994) is incorrect and suspect demeanor plays a role in arrest

decisions. If demeanor does not produce a significant effect size, then Klinger is correct and

demeanor does not influence officer decision making and instead it is some other mechanism.

Meta-analysis has had limited implementation in criminal justice research, but is an analytical

technique that is expanding in popularity. The potential benefits of the technique have had even

less use in criminal justice decision-making research outside of sentencing.

Chapter Summary

There are several problems with the current state of criminal justice research. First, there

is a perception that the discipline of criminal justice is atheoretical. This problem leads to the

discipline of criminal justice being seen as a secondary or applied science. This lack of theory

development leads to disagreement as to appropriate scope of theories within the discipline,

namely, a system-wide approach versus a decision-specific approach. One goal of this

dissertation is to examine this issue and see if correlates of decision-making are similar across

11

the system or if they are unique to particular decision points. If correlates are consistent across

the system, a system-wide approach would appear to be more applicable. However, if correlates

vary from decision point to decision point, then decision-specific explanations of criminal justice

behavior would be more appropriate. In other words, the goal is to determine if there is a

possibility to create a general theory criminal justice or should the field continue on its current

trajectory which focuses more on decision-specific theorizing.

Second, there is a lack of updated knowledge as to correlates of criminal justice

processing. There are a number of predictors that may influence decision making across the

system. However, past efforts at synthesis research were simple literature reviews. These

literature reviews led researchers to treat variables with mixed results as non-significant. One

way to combat this problem is through meta-analysis, which allows for synthesis of statistical

information across studies. The potential solution to both of these issues involves examination

of the correlates of criminal justice decision making using meta-analysis to make more definitive

rejections or acceptance of various influences on decision making.

A Look Ahead: The Plan for this Dissertation

The primary goal of this dissertation is to continue what Gottfredson and Gottfredson

began over thirty years ago. They examined the correlates of criminal justice decision-making

across the system. This dissertation will also examine correlates of decision-making across the

system. Specifically I examine correlates of arrest, sentencing, and parole revocation. Each of

these decisions was chosen because 1) each of these decision points are completely within a

particular branch of the criminal justice system and, 2) each decision point represents one of the

main functions of that branch of the criminal justice system. Unlike Gottfredson and

Gottfredson, this dissertation will use meta-analysis to fully examine the effects of various

12

predictors of arrest, sentencing, and parole revocation. The purpose of this examination is to

attempt to determine the best way to organize criminal justice theory and to determine what

correlates should be included in any theory of criminal justice.

In the chapters ahead, Chapter 2 presents a review of current criminal justice theories and

current knowledge of correlates of decision-making based, primarily on the narrative literature

reviews. Chapter 3 presents a critique of narrative literature review methods and the benefits of

meta-analysis. Chapter 4 outlines the research questions, sample generation procedures, and

statistical procedures of the analyses of this dissertation. Chapter 5 discusses the results of the

meta-analysis. Finally, Chapter 6 discusses conclusions and provides a general summary of this

dissertation.

13

CHAPTER 2

LITERATURE REVIEW

While criminal justice theory is still developing and organizing, there are distinct and

established theoretical perspectives. These theoretical perspectives vary on two key criteria.

First, variation occurs because of the theory’s primary predictive variables. Many scholars have

suggested that certain factors may influence the behavior of criminal justice agents. Each

scholar that has theorized a perspective has stipulated that there are certain key predictors. For

example community theories emphasize certain aspects of the macro-social community. The

theoretical categories are: situational theories, actor theories, workgroup/organizational theories,

community theories, and conflict theories. The second point of variation is the scope of the

theory, total-system versus decision-specific theories. Total-system theories seek to predict the

use of discretion across the criminal justice system, while decision-specific theories are directed

at one particular decision.

First, there is a discussion of each theoretical perspective in criminal justice. Following

each theory explanation, there is a discussion of the current evidence for each theory. Second,

this chapter explores the current state of knowledge regarding each decision point examined in

this dissertation. There is a discussion of arrest, sentencing, and parole revocation. The section

reviewing decision making is divided into two subsections. The first subsection contains

information based on narrative literature reviews. The second subsection contains information

based on meta-analyses.

14

Situational Theories

Situational theories are those theories of criminal justice where the main predictive

elements are aspects of the context in which discretion is exercised. Situational characteristics

are often studied in decision-making research and as such it is natural that theories emerge

emphasizing these factors. These theories promote the idea that the criminal justice system does

not operate in a vacuum and aspects of the criminal event can and will influence, at least in part,

the decisions of the criminal justice actors and the system. Two current situational theories are

Gottfredson and Gottfredson’s theory of criminal justice and focal concerns theory.

Gottfredson and Gottfredson

Gottfredson and Gottfredson’s theory states that situational factors about the criminal

event dictate the decisions made by criminal justice actors (see the discussion in Chapter 1).

These situational factors influence the amount of the exercise discretion for a particular decision,

which ultimately leads to the particular processing decision. Through their analysis, they find

that these factors continually inform all decision points in the criminal justice system, from the

decision of the victim to contact the police through the parole release decision. Numerous

empirical assessments since Gottfredson and Gottfredson discussed their theory have confirmed

the impact of seriousness of the offense, prior record, and victim-offender relationship. See the

discussion below on narrative reviews of decision making for more information.

Figure 2.1: Diagram of Gottfredson and Gottfredson’s theory of criminal justice

Prior record

Seriousness of the offense Discretion Processing decision

Victim-offender relationship

15

Focal concerns theory

Focal concerns theory is a theory of the exercise of discretion. The theory stipulates that

criminal justice decision making is based on three focal concerns (Steffensmeier, Kramer, and

Ulmer, 1998). According to Steffensmeier et al. (1998), these focal concerns are

blameworthiness of the offender, protection of the community, and practical constraints and

consequences. Blameworthiness deals with the culpability of the offender in the criminal act and

the degree of injury. Protection of the community pertains to the potential for future harm from a

particular offender, and an element of deterrence of other offenders. Finally, Steffensmeier and

his colleagues explain that practical constraints and consequences are other miscellaneous

considerations like the offender’s ability to survive the punishment, and organizational issues

such as the availability of resources and space. These focal concerns shape the exercise of a

criminal justice actor’s discretion and determine their response in a particular criminal event.

Figure 2.2: Diagram of focal concerns theory

Blameworthiness

Protection of the community Sentence

Practical constraints

Empirical assessments of focal concerns theory have produced findings consistent with

the theory. In addition to Steffensmeier’s and his colleague’s initial assessment of focal

concerns theory, many other studies have found results consistent with the theory (Johnson,

2003; Johnson, 2005; Johnson 2006; Kramer and Ulmer, 2002; Steffensmeier and Britt, 2001;

Ulmer and Johnson, 2004; Ulmer, Kurlychek, and Kramer, 2007). These studies have

consistently shown that male individuals, convicted of more serious offenses, with longer

criminal histories are more likely to not only be incarcerated but also receive longer prison

16

sentences. These offenders are considered more blameworthy and able to survive the prison

environment. Additionally, these studies found that those that committed more serious violent

crime were more likely to receive longer prison terms.

Actor Theories

The critical component that sets actor theories of criminal justice apart from other

theories is that actor theories place a greater emphasis on the individual actors in the criminal

justice system. While the individual is part of the situation, situational theories place a greater

emphasis on the totality of circumstances. Actor theories place a greater emphasis on one aspect

of the situation, the individual criminal justice actor. While there is not a specific theory of

criminal justice based on actor characteristics, it is a studied area of decision-making research

and is a plausible explanation of criminal justice behavior. If criminal justice actor

characteristics emerge as a significant correlate of decision-making, then a theory should be

developed focusing on this particular set of predictors.

Workgroup/Organizational Theories

Criminal justice actors act within the context of a situation, but there are also collective

aspects of their working environment and organization that influence their behavior and the

exercise of their discretion. This idea is the central theme of workgroup and organizational

theories. Workgroup theories argue that the collective norms that develop between coworkers

within an organization are the key factors in the exercise of discretion. Two examples of these

types of theories are Eisenstein and Jacob’s courtroom workgroup theory and Klinger’s

ecological theory of police behavior.

17

Courtroom workgroup

Eisenstein and Jacob (1977) postulate that it is the courtroom workgroup that heavily

influences discretionary decisions made in the criminal justice system and the criminal courts

specifically. By workgroup, Eisenstein and Jacob are referring to the main three courtroom

actors, the judge, prosecutor, and defense attorney. In their original conceptualization of their

workgroup theory, they argue that the political environment influences both the sponsoring

organization characteristics and the workgroup characteristics. These sponsoring organization

characteristics and workgroup characteristics then directly influence the outcomes of a case. The

political environment also directly impacts the outcomes of a case, and the sponsoring

organizations also impact the workgroup characteristics.

Each of the above variables includes multiple components. The political environment

incorporates the larger contextual environment. The police are seen as gatekeepers of the

system; if they do not make arrests then the courts have no cases to process. Therefore,

Eisenstein and Jacob (1977) argue that police are part of the political environment and context.

Specifically, they argue that police influence cases via the quality of evidence they collect. The

next component of the political context is the legislature and appellate courts. Each of these

entities creates rules that the workgroup must follow; the legislature via creation of the court

system and establishment of the law and appellate courts by creating procedural rules that the

workgroup must follow. According to Eisenstein and Jacob (1977), the next component is prison

authorities. This entity influences case processing because they regulate sentencing, judges may

only sentence defendants to the available space. The final influential member of the

environment is the media. The authors explain that the sponsoring agencies and the media have

18

a reciprocal relationship. The sponsoring agencies want to be portrayed positively so they use

the media for that end and the media uses the sponsoring agencies for news content.

Workgroups are complex entities. There is more to the workgroup than simply who is a

member. Many factors influence workgroups. Eisenstein and Jacob (1977) explain that the

degree of workgroup stability and the degree of familiarity that members have with one another

influence the workgroup. They argue that workgroups that are stable and familiar with one

another are able to process cases more informally and rely less on the adversarial process and

more on a negotiation style of case disposition. Additionally, each workgroup works toward

certain goals. The authors explain that goals are shaped by two factors. The first is the origin of

the goal. Origin refers to where the goal originates, either external or internal to the workgroup.

For example, a goal that results from pressures from one of the entities of the political

environment is an external goal. The second goal-shaping component is the function of the goal.

The authors explain that a goal may be expressive or symbolic in nature, or a goal may be

instrumental, or focused on improving efficiency. Each combination of these two categories

produces a different goal that in turn influences the workgroup.

The final independent variable in the workgroup theory is the supporting agency

characteristics. The supporting agencies are those agencies that have control over the personnel

of the workgroup. Eisenstein and Jacob (1977) argue that since these agencies have control over

the workgroup personnel, recruitment and assignment procedures impact the workgroup because

of their effects on stability and familiarity. Assignment procedures that constantly change the

workgroup create instability in the workgroup and the workgroup begins to rely more on the

adversarial process versus negotiations. Eisenstein and Jacob (1977) also argue that the amount

of control a sponsoring agency has over it’s members influences the workgroup. For example, if

19

a prosecutor’s office creates a policy that restricts plea-bargaining in felony cases, then

prosecutors must turn toward a more adversarial case processing because their ability to

negotiate has been reduced by the supporting agency

Figure 2.3: Diagram of Eisenstein and Jacob’s courtroom workgroup theory

Sponsoring Organization Characteristics Informal Norms Outcomes

Political Environment Workgroup Characteristics

Police workgroup

Klinger provides another workgroup theory that is aimed at explaining the exercise of

police discretion. Klinger (1997) argues that the amount of deviance in a district leads to an

overall district workload, but also impacts officers’ perceptions of deviance in that area and the

seriousness of the immediate offense. As the district workload also increases, resource

constraints increase along with perceptions of deviance among officers. He argues that an

officer’s understanding of district deviance impacts what crimes they perceive as normal, the

deservedness of victims and levels of police cynicism. As perceptions of deviance increase,

more serious crimes become “normal”, victims become less deserving, and police officers

become more cynical. According to Klinger (1997), resource constraints, police cynicism,

deservedness of victims, and perceptions of normal crime all impact workgroup rules about

discretion. These workgroup rules about police discretion then directly impact the exercise of

discretion. Finally, the seriousness of the immediate offense also directly impacts police action.

20

Figure 2.4: Diagram of Klinger’s ecological theory of police behavior

Level of district deviance Seriousness of the immediate offense

Normal crimes

Officer’s understanding Deserving victims Group rules Vigor of district deviance Police cynicism

District workload Resource constraints

Recently scholars have begun to test Klinger’s theory. Sobol (2010) attempted a partial

test of Klinger’s theory examining measures of cynicism and district workload on police vigor.

He found that neither of these two variables were significantly associated with vigor in the full

model. Workload was significantly related to vigor at the bivariate level but the relationship was

rendered non-significant in the full model, which lead Sobol (2010) to conclude that the

relationship was spurious. While not a complete test, Philips and Sobol (2011) were able to

support the workload hypothesis of Klinger’s theory. They found that officers that worked in

districts with higher workloads were less likely to view force as unnecessary. None of the above

studies were able to completely test Klinger’s theory. Both studies were unable to tap into

victim deservedness. Another limitation of Philips and Sobol’s (2011) study is that they used

vignettes and were not able to examine police cynicism.

21

Community Theories

Community theories argue that the larger community is the driving factor in criminal

justice decision making. In other words, the collective norms and attitudes of the community

dictate the type of criminal justice organizations and thus, the actions of the individual criminal

justice actors. Three examples of these types of theories are Black’s theory of the behavior of

law, Duffee’s community theory of criminal justice, and Wilson’s community theory of police.

Black (1976) argues that the law, which is defined as “governmental social control” (p.2) and

can be interpreted as the discretionary action of a criminal justice system, is the product of five

macro-social constructs: stratification, morphology, culture, organization, and social-control.

Duffee (1990) argues that criminal justice action is the result of a type of community. That

community is the product of two characteristics vertical relations and horizontal articulation.

Each of these constructs leads to a particular type of community, which leads to a particular type

of criminal justice organization. These organizations will have differing approaches to criminal

justice. Finally, Wilson (1968) argues that the local political culture of the community is critical

in determining the type of police agency in the community. The critical component in all three

of these theories is the role of the community in shaping the type of criminal justice exercised

within that community.

Black

Black argues that five predictor variables: stratification, morphology, culture,

organization, and social control, influence the behavior of law. According to Black (1976),

stratification is a measure of the wealth of a society, and law varies directly with stratification.

Black (1976) also states that stratification is “the vertical distance between the people of a social

setting” (p. 13). In other words, stratification refers to a measure of social status. Morphology is

22

a person’s relational distance to another individual; it is a person’s level of interconnectedness.

He explains that if stratification is the vertical distance of society, then morphology is the

horizontal distance of society. This concept is similar to collective efficacy. According to

Sampson, Raudenbush and Earls (1997), collective efficacy is a community’s level of

integration. When communities have a greater degree of interconnection they are able to police

themselves and have less reliance on formal social control. Black (1976) explains that the

relationship between law and morphology is curvilinear.

Black’s final three variables are culture, organization, social control. According to Black

(1976), “culture is the symbolic aspect of social life, including expressions of what is true, good,

and beautiful” (p. 61). Culture incorporates a society’s level of education, religion, and

technology. Black (1976) states that societies with more culture will have more law. The fourth

variable is organization. Organization is a society’s ability to use collective action. He argues

that societies with greater amounts of organization will have greater amounts of law. The final

variable is social control. Black (1976) explains that social control is society’s informal social

control. Communities that have greater amounts of social control will have less law.

Figure 2.5: Diagram of Black’s theory of the behavior of law

Stratification Morphology Culture Law

Organization

Social Control

Empirical assessments of Black’s theory have been partially supportive. Gottfredson and

Hindelang (1979) and Braithwaite and Biles (1980) tested the theory and found that some of

23

Black’s constructs did not hold up empirically. Gottfredson and Hindelang (1979) found support

for organization; they found that businesses were more likely to report their victimization to the

police. There was also support for culture, measured as education, but the effect of culture was

small. The finding for the culture construct was not as large as Black indicated it should have

been. Braithwaite and Biles (1980) also found that there was limited evidence in support of

Black’s theory.

More recent examinations of Black’s theory have found similar results to previous

examinations. Kuo, Longmire, Cuvelier, and Chang (2010) found that measures of culture and

organization significantly impacted prosecutorial decision making in Taiwan in the directions

predicted by Black. Clay-Warner and McMahon-Howard (2009) found that one measure of

stratification, household income, significantly predicted rape reporting. Additionally, they

found that as relational distance increased odds of reporting increased. There were no significant

results for the other constructs: organization, culture, or social control.

Duffee

Duffee (1980) argues that two factors, a community’s vertical relations and a

community’s horizontal articulation, shape the community’s criminal justice system. Vertical

relations refer to a community’s reliance on other non-local organizations. Communities that

have a high level of vertical relations are reliant on private or the state and Federal Government

to provide goods and services. Duffee (1980) describes horizontal articulation in similar terms to

Black’s concept of morphology; it is the degree of cohesiveness of the community.

Communities with strong horizontal articulation “interface and cooperate with each other”

(Duffee, 1980, p. 153). Populations that live in communities with high levels of horizontal

articulation are homogeneous and share similar norms and values.

24

Duffee (1980) describes each of the four types of communities based on the level amount

of vertical relations and horizontal articulation. The first community, disorganized, has both low

vertical relations and low horizontal articulation. Norms in these communities are not well

defined. Duffee (1980) explains that “there will be no group boundaries against which to define

deviant action, and on the other hand there will be no processes by which groups inculcate

mutual standards in individuals” (p. 155). Attempts at formal control have little legitimacy.

When criminal justice action does occur it is limited to major issues and crises. Criminal justice

agents are seen as outsiders and any criminal justice action is seen as coercive.

The second type of community, solidary, has low vertical relations and high horizontal

articulation. Duffee (1980) argues that the norms of this community are so well established that

there is little need for formal sanctions. He states that enforcement is often based on peer

pressure. This community is self-contained and is able to control actions informally rather than

formally. Formal sanctions that take place will need to conform to local goals and norms.

The third type of community has high vertical relations but low horizontal articulation.

This community is described as a fragmented community (Duffee, 1980). According to Duffee,

(1980) this community uses organizations that focus on special individual clients “without regard

for the community” (p. 157). This community type has unclear norms. They also have a high

reliance on external entities. The combination of these two situations creates an environment

where multiple organizations have very specific functions. These specifically focused

organizations often have turf wars. The focus of these agencies tends to be client centered. In

these communities criminal justice is concerned with “reduction of future crime[…]rather than a

concern for maintaining an existing normative pattern” (Duffee, 1980, p. 157).

25

The final community has high vertical relations and high horizontal articulation.

According to Duffee (1980), this community is interdependent; it wants to rely on outsiders to

provide services, but at the same time it is interconnected to be self-sufficient. This community

is somewhat odd and not likely to exist. Like the solidary community type they can police

themselves because they have good integration amongst members, but yet they rely on an outside

authority due to the high vertical relations. Duffee explains that “administration of formal

organizations involved in social control would seem to be a difficult activity here” (Duffee,

1980, p. 157). There is a reliance on complex outside funded entities, but community members

will likely be dissatisfied with the services provided when services are not directed toward the

community as a whole.

Figure 2.6: Diagram of Duffee’s theory of criminal justice

Vertical relations

Community type Execution of criminal justice

Horizontal articulation

While not explicitly testing Duffee’s theory, Liederbach and Travis (2008) found that

police agency style, a proxy for criminal justice system type, was dependent upon community

size. Specifically, they found that larger communities tended toward legalistic type department.

Smaller communities tended toward order maintenance policing. Legalistic departments tend

toward the law enforcement function and rely on the formal sanction to deal with problems

(Wilson, 1968). As the community grows larger it cannot maintain that high level of horizontal

articulation and begins to rely on outsiders for criminal justice services. They are simply too big

to maintain close interpersonal relationships within the community as a whole. Smaller

26

communities can move away from formal law enforcement function because they may be more

likely to have higher levels of horizontal articulation and thus, can resolve matters informally.

Wilson

Criminal justice systems and specifically policing systems in a community may be

influenced by the local political culture of the community. Wilson (1968) argues that local

political culture plays an important role in the creation of a certain type of police department.

Political culture has little direct impact on the actual decision making of police officers, but it

does impact organizational and policy issues, and choosing a chief. Because the chief plays an

important role in controlling officers’ behavior via policy, the community gains control over

officers through the local political culture. Wilson (1968) states that city managers and mayors

were concerned about illegal business and those political leaders placed pressure on the chief to

intervene, and the chief emphasized illegal business issues to his officers.

The local political culture and chief led to the creation of a certain type of police

department. According to Wilson (1968), there are three types of departments: legalistic,

watchmen, and service. Each one of the departments emphasizes a different function of police.

Legalistic departments focus on exercise of formal authority. They emphasize the law

enforcement function. Service departments emphasize the service function. Officers in these

departments tend to resolve problems informally. Finally, watchmen departments emphasize the

order maintenance function. These officers use a combination of formal and informal sanctions

to execute the order maintenance function. The type of department then influences how officers

interact with citizens.

Wilson (1968) explains that there are four types of discretionary situations and responses

are based on whether or not the police have frequent or infrequent contact with citizens and

27

whether or not the situation is formal or informal. Liederbach and Travis (2008), argue that

frequent, formal intervention is common in legalistic departments. These police departments

stress law enforcement and these officers issue many citations and make more arrests. Service

departments also intervene regularly but the solutions and issues tend to be less formal. The

emphasis on informality is consistent with officers in service agencies. Finally, officers in

watchmen agencies do not intervene often with citizens, and when interventions do occur they

tend to be a mix of formal and informal. Thus, officers in these agencies emphasize the order

maintenance function.

Figure 2.7: Diagram of Wilson’s theory of police behavior

Community

Local political culture

Dept. type Primary police function

Chief

Wilson’s theory has been assessed, at least in part, several times. The results have been

mixed. Langworthy (1985) assessed the relationship between arrest rates and government type,

i.e., what Wilson identified as a good government run by a city manager as opposed to a more

traditional mayoral system of government. He found that it was indeed the case that arrest rates

for marginal crimes varied based on the government type. While not a perfect test of Wilson’s

theory, Langworthy’s test did partially support Wilson’s conception of police behavior. Crank

(1990) also found support for the government-arrest hypothesis. He found that rural

municipalities were significantly more likely to have higher arrest rates for trespassing,

disorderly conduct, motor vehicle theft, and cannabis control. In urban environments, police in

areas that were run by city managers were significantly more likely to make arrests for motor

28

vehicle theft, but not for the other crime categories. Additionally, Smith (1984) found that

correlates of arrest varied depending on the police style of the department. He concluded that

any explanation of legal force would be incomplete without a discussion of police organizational

characteristics.

While some scholars have found limited support for Wilson’s theory, other scholars have

not found the same results when testing Wilson’s theory. Several studies have failed to find a

relationship between political culture and police organization (Hassell, Zhao, and Maguire, 2003;

Liederbach and Travis, 2008; Zhao and Hassell, 2005; and Zhao, Ren, and Lovrich, 2010). Zhao

and Hassell (2005) went on to claim that this consistent lack of a finding of a correlation between

political structure and police style was evidence of falsification of Wilson’s theory. However,

Liederbach and Travis (2008) were not as quick to signal the end of Wilson’s theory. Instead,

they argued that no assessment had fully replicated Wilson’s original test and that each test,

including theirs, was subject to issues of model misspecification. The primary model

misspecification is that no study to date had adequately examined the correct combination of

communities. Often communities were fairly constant in assessments of Wilson’s theory.

In sum, three scholars created a theory that places an emphasis on the impact of the

community. Black outlines the social dynamics of communities that lead to the amount of

formal criminal justice processing. Duffee argued that qualities of a community, vertical

relations and horizontal articulation, led to the creation of a certain type of criminal justice

system, which dictated how offenders are processed. Finally, Wilson stated that local political

culture has an impact on the type of police department and the chief; who sets policy that dictates

how officers handle situations.

29

Conflict Theory

The final theoretical perspective is the conflict perspective. Conflict theory argues that

macro-social group dynamics influence the behavior of criminal justice actors. According to

Hagan (1989), conflict theory argues that extra-legal factors like race or gender will play a larger

role in criminal justice decision making. Scheingold’s (1984) conception of conflict theory

argues that increases in victimization, both direct and vicarious victimization via media, increase

fear, which leads to increased punitive decision making. Under Scheingold’s (1984) model

extra-legal factors play a large role in vicarious victimization; vicarious victimization occurs due

to increased media attention, which is the result of increased levels of crime. Additionally,

Scheingold (1984) argues that punitive policy responses create yet more increased media

attention on crime. Thus, culturally driven fear creates a picture of a typical offender which

leads to punitive punishments and punitive decision making toward that group.

Figure 2.8: Diagram of Scheingold’s theory of policy creation

Increased crime victimization non-punitive response non-punitive policy FEAR Increased media attention vicarious victimization punitive response punitive policy

More recently, conflict theory has emerged as a form of the minority group threat

hypothesis/perspective. Under this perspective, as certain minority groups increase in size they

become more threatening and laws are created to keep that minority population in check (Kane,

2002). Kane (2002) argues that as threatening groups increase in size police are deployed to

those areas and use stricter enforcement practices. While he specifically applies the minority

30

group threat hypothesis to police decision making, the same idea can be applied to the other

branches of the criminal justice system.

According to Hagan (1989), conflict theory is examined when scholars look at extra-legal

explanations of criminal justice outcomes. Many scholars have examined the impact of these

extra-legal variables such as age, race, ethnicity and gender on criminal justice decision making.

Empirical assessments of these relationships have been mixed (see below).

Summary

There are a variety of criminal justice theory perspectives. These theories are broken

down into 5 classifications based on the primary predictive variables. These five theory

categories are: situational, actor, organizational/workgroup, community and conflict

perspectives. Gottfredson and Gottfredson and Steffensmeier and colleagues created situational

theories that stress the importance of the context in which the criminal event occurs in order to

explain why individuals are processed through the system. Eisenstein and Jacob and Klinger

both created theories that emphasize the role of the workgroup on actor decision making. Black,

Duffee, and Wilson all created theories that emphasize the role of community as a predictive

factor of the criminal justice process. Black emphasized the dynamics of societal interaction,

and Duffee emphasized the two elements of the community, vertical relations and horizontal

articulation. Wilson argued that local political culture influence police departments and through

departments the actions of officers. Lastly, several conflict perspectives have emerged each

emphasizing the role of societal conflict on criminal justice decision making.

Some perspectives apply a system-wide approach. For example, Gottfredson and

Gottfredson, Duffee, and Black all created theories to explain the process of criminal justice

generally. Other scholars created theories that focus on a particular branch of the criminal justice

31

system, for example, Steffensmeier and colleagues, Eisenstein and Jacob, Klinger, and Wilson.

Some of the decision-specific theories can be applied to other components of the criminal justice

system. Focal concerns theory was created as a way to explain judicial discretion. However, the

concepts of blameworthiness, protection of the community, and other practical concerns are

applicable to other criminal justice decision points and decision makers. For example, an officer

making an arrest could very easily be deciding who to arrest based on all three of the factors

discussed in focal concerns theory. Additionally, when some decision-specific theories are

considered together the key components appear to be applied at multiple stages of the criminal

justice system. For example, both workgroup theories are decision-specific but examining the

theories in concert shows how the concept of a workgroup can apply across the system. The

workgroup may change slightly, but the concept is consistent across the system.

Research Synthesis in Criminal Justice

Now that each of the theoretical perspectives has been discussed, it is important to take

stock of the current empirical knowledge on criminal justice decision making. The research

synthesis takes two primary forms, narrative literature reviews and meta-analysis.1 Criminal

justice has traditionally used narrative reviews to assess correlates of decision-making (see

Gottfredson and Gottfredson, 1989; National Research Council, 2004, 2008; Riksheim and

Chermak, 1993; Sherman, 1980; and Spohn, 2000). With the exception of Gottfredson and

Gottfredson, each of the aforementioned reviews focuses on one decision point and at times on

the effects of one particular variable (see Spohn, 2000).2 The use of meta-analysis has been

much more limited. Meta-analysis has been primarily used to examine correlates of sentencing,

but recently meta-analysis was used to examine the effect of suspect race on arrest (see Daly and

1 The merits and difficulties of each form of synthesis are discussed in detail in the next chapter. 2 Since Gottfredson and Gottfredson (1989) has been discussed in detail above there is no need to include another discussion of their work.

32

Bordt, 1995; Kochel et al., 2011; Mitchell, 2005; Pratt, 1998; and Whatley, 1996). Each of these

meta-analyses, discussed in detail below, focuses both on only one decision point and on one

particular relationship.

Correlates of decision-making have been a long-standing research interest of criminal

justice science for some time. The vast majority of synthesis literature has been decision-

specific; emphasis has been given to one particular decision point. The discussion that follows

will first discuss narrative literature reviews. Finally, this section concludes with a discussion of

the current meta-analytic studies on criminal justice decision making.

Narrative Reviews

Arrest

Several studies have summarized the correlates of arrest. The first study to

systematically evaluate the correlates of arrest was Sherman (1980). Sherman’s study was

followed by Riksheim and Chermak (1993). Finally, the National Research Council (2004)

reviewed the correlates of arrest as part of a larger examination of all aspects of policing

research. Each of these studies is a narrative review and each will be reviewed in detail below.

Sherman

Sherman’s (1980) study was the first narrative review of the correlates of police officer

decision making. Sherman (1980) separates explanations of officer behavior into five

classifications: individual, situations, organizational, community and legal. With regard to

individual explanations, Sherman (1980) found that less experienced officers, male officers,

black officers, better educated officers, officers with greater job satisfaction, and officers that had

negative views of black individuals were more likely to make arrests. Situational factors that

increased the odds of arrest included: when officers proactively enter into a police-citizen

33

encounter, when more officers are present at the police-citizen encounter, when suspects are

disrespectful, when the victim does not know the offender, when suspects are younger, when

suspects are male, and when suspects are members of the lower class. Additionally, when the

complainant prefers arrest, and when the complainant is respectful, an officer is more likely to

make an arrest. He also notes that legal variables play a significant role in predicting arrest.

Offense seriousness, and the amount of evidence increase the odds of arrest.

There was a large amount of research on individual and situational effects on arrest.

Sherman (1980) states that there has been little examination of the organizational effects on

arrest behavior. Finally, Community variables have also been neglected in terms of examination

regarding their impact on arrest behavior.

Riksheim and Chermak

Riksheim and Chermak (1993) separate arrest predictors into four categories: individual-

level, situational-level, organizational-level, and community-level. They depart from Sherman

slightly by collapsing legal-based explanations into situational-level variables. They organize

their discussion of correlates as pre- and post-1980.

The increased amount of research changed the impact of individual factors. Riksheim

and Chermak (1993) found that an officer’s education level had a mixed impact. They found

that some research showed that an individual officer’s education had no impact on arrest, but

some departments with better-educated officers had fewer arrests. They cited other research,

which reported that the impact of education was conditional on situational factors. Officer

experience as a predictor of arrest also became more muddled. Some studies found evidence that

was consistent with Sherman, and others found that there was no effect of officer experience on

arrest. Post-1980 research appears to contradict earlier findings with regard to the effect of

34

officer race, gender and attitudes. Riksheim and Chermak (1993) found that these variables had

no effect on arrest in studies conducted post-1980. However, this finding may be a

methodological artifact due to the majority of pre-1980 studies focusing on bivariate

relationships and post-1980 there was an increase in the use of multivariate statistics.

Situational characteristics are likely the most commonly studied correlate of officer

decision making and there is little consistency in findings from before 1980 to those findings

from studies after 1980. Riksheim and Chermak (1993) found that the research was consistent

with respect to the effect of lower class individuals; they were more likely to be arrested. They

found that results were mixed on the impact of victim initiated police-citizen encounters in terms

of arrest. Some studies found police were less likely to arrest when a victim calls the police in

domestic disturbance calls; other studies found no effect. Research conducted in the 1980s

indicates many mixed findings. Gender, race, and age have had mixed findings. A largely equal

number of studies found that males were more likely to be arrested while other studies found no

effect. Race was one of the most commonly studied variables in arrest research, yet the effect of

race was inconclusive. A large number of studies found no effect of race and only a handful

found that minorities were more likely to be arrested, and one study found that whites were more

likely to be arrested. According to Riksheim and Chermak (1993), age has generally been a non-

significant predictor of arrest. But, some studies have found that age played a role with certain

groups such as females, and adult misdemeanants.

Other situational variables have shown mixed effects. The overwhelming majority of

research indicates that complainant preference influences arrest, but this factor had no effect in

some situations such as department type. Overall, research cited by Riksheim and Chermak

(1993) demonstrates that disrespectful suspects are more likely to be arrested. Other research

35

has indicated that suspect demeanor has no effect in certain situations. Victim and offender

relationship has also produced mixed findings. In many studies cited by Riksheim and Chermak

(1993) when the victim and offender do not know each other the suspect is more likely to be

arrested. Studies of police-citizen encounters involving alleged domestic violence found that

victim-offender relationship did not have a significant effect. This finding may be the result of

changing societal attitudes or department policies. The location of the incident has shown to

have mixed results as well. Some research indicated that there was no difference between

incidents that were in public areas and private areas. Other research showed that males were

arrested more often in public locations than females. Police were more likely to arrest for more

serious incidents. This effect held even after creating more detailed scales than felony vs.

misdemeanor. However, some research cited by Riksheim and Chermak (1993) found that the

presence of an injury did not significantly increase the likelihood of arrest. A new variable,

examining the influence of police supervisors, had mixed results. For some department types

this increased arrest, while for others it decreased arrest odds or had no effect.

Few organizational-level variables have shown an effect on arrest behavior. The other

issue with organizational-level variables is that they have not been extensively studied and some

of the measures are crude at best. Research on department deployment strategies has found

mixed results. According to research cited by Riksheim and Chermak (1993), one study found

no effect of deployment strategy on arrest. Another study found that deploying officers to

smaller assignment areas was associated with higher odds of arrest in areas with high violent

crime. Also, target specific deployments, and more aggressive police tactics appear to increase

arrest rates. The number and length of shift durations do not appear to influence arrest

productivity. Research on department type has produced mixed results; some research has found

36

that certain department types increase arrest. Other studies have found no effect of department

type on arrest. Finally, the effect of department size is also unclear; an equal number of studies

have found that officers in larger departments were less likely to arrest as studies that have found

no effect of department size on arrest.

Like organizational-level variables, research on community-level variables has been

limited. There has been some study of community-level economic variables; research cited by

Riksheim and Chermak (1993) indicates that police have higher arrest rates in poorer

communities. Many different measures of poverty appear to indicate the same findings. City

manager style government appears to increase arrests rates for specific crimes, which include:

property crime, DWI, trespassing, disorderly conduct, and motor vehicle crimes. Research cited

by Riksheim and Chermak (1993) was inconclusive on percent non-white. Some research found

that this variable was correlated with increased arrest rates. While other research showed that

when this variable was measured as percent Hispanic the results varied based on the location of

the police department, urban versus rural. Finally, Riksheim and Chermak (1993) cited

contradictory research regarding the effect of neighborhood crime rates and victimization rates

on arrest.

National Research Council

The National Research Council (2004) evaluated the majority of the currently available

information regarding correlates of police behavior. They argued that police action is most

heavily influenced by legal variables, particularly seriousness of the offense, evidence of

wrongdoing, and willingness to file a complaint. The committee also stated that there is

consistent evidence to indicate that high crime areas or disadvantaged areas will increase the

odds of arrest. Finally, the committee found that there is some evidence that police agencies are

37

influenced by court rulings but the committee also argues that there is evidence that agencies try

and circumvent rulings as well. These three factors were the only influences on police behavior

that showed consistent effects across studies using the narrative review approach. For all other

explanations of police behavior there was either mixed findings or not enough research to draw a

conclusion with confidence.

Summary

Taken together these narrative reviews begin to create a picture that emphasizes

situational-level factors such as suspect demeanor and legal variables. What is consistent across

these literature reviews is inconsistency. At some points in time scholars were willing to assert

with more conviction that certain factors are significant and as time progresses confusion appears

to increase. In this case, more data seems to be cumbersome for traditional narrative reviews.

Another issue with these narrative reviews is that they are reporting on multiple studies that use

the same data over and over. This fact may be contributing to the confusion rather than helping

to eliminate it. Simply because even within the same data set, findings appear to disagree with

each other, for example, research cited by Riksheim and Chermak (1993) showed that some

research from the Police Services Study found that race was a significant factor in arrest

behavior while other research did not find an effect for race.

Sentencing

Spohn (2000) reviewed the correlates of sentencing, but her review focuses on how race

influences sentencing. All other correlates are always put in the context of the effect of race on

sentencing. Meta-analytic reviews of sentencing have discussed other issues with sentencing.

They are discussed after the narrative literature review section.

38

Spohn (2000) reviews sentencing literature to examine the impact of race and ethnicity

on sentencing. Of the studies reviewed, Spohn (2000) found that 43.2 percent of state-level

studies found an effect for black vs. white defendants and 27.6 percent of state-level studies

found an effect for Hispanic vs. white defendants. Additionally, she found that 55.5 percent of

studies found a significant effect of race on the in/out decision and 23.1 percent of studies found

an effect of race on the length of sentencing decision. Finally, 41.7 percent of state-level studies

found an impact for ethnicity on the in/out decision and one study found an effect of ethnicity on

the length of the sentence. It is also important to note that many studies that did not find race

and/or ethnicity had main effects did find interaction effects of race and/or ethnicity combining

with other factors.

Spohn (2000) also reviewed the impact of race and ethnicity on federal sentencing

studies. Research cited by Spohn (2000) showed that 68.2 percent of studies of federal-level

sentencing data found an effect of race. Also, 47.6 percent of studies of federal-level sentencing

data found an effect of ethnicity. Three of seven studies that examined the in/out decision found

significant effects for race and three of seven studies found a significant effect for ethnicity.

Finally, six of nine studies on race found significant effect of race on the length of sentence and

two of eight studies on ethnicity found a significant effect of ethnicity on the length of sentence.

According to Spohn (2000), based on the above picture, race appears to play a role in sentencing.

Parole revocation

The status of review research on parole revocation is very limited. Summary research on

parole often barely mentions the topic of parole revocation and instead focuses more on racial

disparities and parole. For example, Latessa and Smith (2007), Travis and Lawrence (2002) and

the National Research Council’s (2008) discussions of parole fail to explain the correlates of

39

parole revocation. Caplan (2007) reviewed literature surrounding parole but focused entirely on

release and not revocation. Instead each of these research endeavors has focused on other issues

of parole such as release or the elimination of parole as a release mechanism. Additionally,

Gottfredson and Gottfredson (1989) offer limited information on the correlates of revocation.

Research cited by Gottfredson and Gottfredson (1989) states that prison performance predicts

parole success or failure. This measure is similar to the idea of criminal history and is consistent

with one of Gottfredson and Gottfredson’s (1989) three main correlates of criminal justice

decision making.

In sum, narrative literature reviews have shown limited consistencies across the system.

For example, Gottfredson and Gottfredson (1989) found victim-offender relationship,

seriousness of the offense and criminal history were the primary explanatory factors across the

system. Aside from the influence of these variables, narrative literature reviews have been

inconsistent across the system. At the arrest decision, literature reviews are rather exhaustive.

Yet, at the sentencing decision, narrative reviews have focused almost exclusively on the effects

of race and have virtually ignored the effects of other theoretically relevant variables. Finally,

the status of synthesis of research on parole revocation is very limited to the point of non-

existence. Reviews of literature on parole generally have focused on general issues with parole

or the release decision not the revocation decision.

Meta-Analysis

Meta-analysis has had some use in criminal justice decision-making research. Meta-

analysis has primarily been used to examine the correlates of sentencing. Daly and Bordt,

(1995) examine the effects of gender on sentencing. Mitchell (2005) and Pratt (1998) both

examine the effects of race on sentencing. Finally, Whatley (1996) examines the effects of

40

victim blameworthiness. More recently, Kochel, Wilson, and Mastrofski (2011) analyzed the

effects of race on the decision to arrest.

Gender

It had long been argued that gender played a role in sentencing decisions. Daly and

Bordt (1995) examined the effects of gender on sentencing meta-analytically. They found that

even after controlling for legally relevant variables, such as prior record, gender remained a

significant correlate of decision-making. The vast majority of studies that showed gender effects

showed effects that favored women. Additionally, gender effects were a more common result

than race effects. Finally, Daly and Bordt (1995) found that gender effects were more likely to

be found when the study had a high proportion of non-white defendants. The authors do

conclude that gender effects may be contextual depending on the context of the criminal justice

decision.

Race

When examining the effect sizes of race, prior record and seriousness of the offense, Pratt

(1998) found that only measures of offense seriousness had a significant impact on measures of

sentence length. Given the consistent finding across studies that prior record plays a large role,

Pratt’s findings are somewhat surprising. Pratt argues that prior record may not have been a

significant correlate due to a measurement issue. He states that the measures of prior record

were often interval measures which implies a linear relationship and the relationship between

prior record and sentence length may be non-linear, hence no relationship. Pratt (1998) argues

that his non-significant race finding may also be due to measurement issues. There are two

majors ways that race was operationalized across studies: white/black, and white/nonwhite. Pratt

also created a third, other category, which acted as a catch all for any other race

41

conceptualizations. Pratt (1998) found that there was a significant difference between the

operationalization of race and that variables impact on sentence length. Specifically, he found

that white/non-white operationalization had the greatest likelihood of finding a significant effect

of race on sentence length. Thus, it may be possible that measures of race were masking a real

effect (Pratt, 1998).

Unlike Pratt’s meta-analysis, Mitchell’s (2005) meta-analytic examination of sentencing

correlates found that race was a significant factor in sentencing. Mitchell (2005) found a small

but statistically significant effect of race even after controlling for sentence severity.3 African

Americans were punished more severely than white Americans. Additionally, he found that the

magnitude of the effect size varied depending on several measurement issues. For example,

larger effect sizes were found with less precise measures of criminal history. Race appeared to

have a stronger influence on sentencing when worse measures of criminal history were used in a

particular study. Even after controlling for other variables such as use of a weapon, trial type,

type of defense counsel, and socio-economic, a small but statistically significant race effect

persisted (Mitchell, 2005).

Kochel et al. (2011) examined the effect of race on officer decision to arrest. They found

that effect sizes indicated that race was a significant predictor of officer arrest decisions. They

found that African Americans were significantly more likely to be arrested. This effect persisted

even after controlling for a variety of relevant moderating variables. These control variables

included whether the study examined domestic violence, other legally relevant variables such as

seriousness of the offense, victim request for arrest, presence of a witness, quality of evidence,

use of alcohol or drugs, and prior record. In addition, they controlled for extra-legal variables

3 Mitchell (2005) analyzed studies that used multiple dependent variables unlike Pratt (1998), who examined only studies that used sentence length as a metric outcome.

42

such as demeanor of the suspect and they examined whether or not the effect of race on arrest

changed over time. In all of these instances the effect of race remained.

Victim characteristics

Another consideration in sentencing is victim responsibility. To examine this issue,

Whatley (1996) meta-analyzed studies investigating the effects of victim characteristics on

victim blameworthiness. He found that victims were more likely to be blamed when: they wore

revealing clothing, had questionable character, and had a close relationship with their attacker.

He did not support the hypothesis that unattractive victims would be more likely to be blamed.

In sum, meta-analysis has been used to examine issues with sentencing and, very

recently, arrest. Through the use of meta-analysis, scholars were able to make sense of the

mixed effects of several variables and make definitive conclusions regarding those variables.

Daly and Bordt (19955) found that there was a gender effect on sentencing. While Pratt’s (1998)

study did not find significant race effects on sentencing, Mitchell’s (2005) meta-analysis did find

a small but significant effect of race on sentencing decisions. Finally, Whatley (1996) found that

victim characteristics played a role in victim blameworthiness.

Chapter Summary

Meta-analysis is a new technique that could be used to garner new information regarding

the effects of predictors of criminal justice decision making on various decisions. The majority

of synthesis research has taken the form of narrative literature reviews. With the exception of a

few relationships, the results of these reviews have often ended with the call for more research to

assess conflicting information. Meta-analysis has had limited use in the area of sentencing but

has not been used extensively in any other area of criminal justice decision making. Meta-

analysis may be able to shed new light on correlates of decision-making across the system and

43

assess the validity of criminal justice theory. If correlates of decision-making are consistent

across the criminal justice system a grand theory approach may be more appropriate. However,

if there are few consistent correlates across decisions and the strongest relationships are

predictors that are unique to a particular decision, then a decision-specific approach may be most

appropriate.

Given the above review, several factors emerge for the entire system under various

domains or groups of variables. Legal variables appear to consistently influence decision

making across the system. These variables include measures of seriousness of the offense, prior

record and victim-offender relationship. When crimes were considered more serious, suspects

had longer records and the victim and offender did not know one another the system was more

punitive. Suspect characteristics, such as race, ethnicity, age and gender, have also been studied

repeatedly but have had mixed results. Due to the conflicted nature of these results, they would

be ideal to include in a meta-analysis. Other variables that have produced conflicted results

include criminal justice actor characteristics, such as actor race, age/experience, gender,

ethnicity, education level4, and cynicism. The aforementioned variables all tend to focus on the

incident level of case processing. However, research has produced mixed results with more

macro level constructs. These variables that have produced inconsistent results at the macro

level include: neighborhood poverty levels, family structure, and racial heterogeneity. These

variables are system-wide and are often measured at each stage of the criminal justice system.

There are a number of factors that are unique to a particular decision for a variety of

reasons.5 At the arrest decision a number of factors have been examined that are unique to the

arrest decision. The first unique set of factors is organizational characteristics. Measures that

4 While, judges have a similar education level, this does vary for police officers and probation officers. 5 Some factors are considered unique simply because they are not measured at multiple stages of the criminal justice system. For example, demeanor may be applicable at multiple stages but is only measured at the arrest stage.

44

are organizational characteristics are measures of organizational hierarchy and police agency

size. The arrest decision point also officers several unique situational factors that are not

necessarily found in other decision points. These factors include measures of poor citizen

demeanor, citizen intoxication, citizen non-compliance,6 interaction-phase criminal conduct and

lay perceptions of mental disorder.7 While some of these factors may not seem unique to the

arrest decision, for example suspect demeanor, organizational hierarchy, or police agency size,

they are not measured in studies of other decision points. It is possible that court defendants or

parolees may have a poor demeanor or parole officers may operate within certain organizational

hierarchies or within agencies of varying size, however, these variables are not measured in

studies of sentencing or parole revocation, and thus, these variables included as unique.

Like arrest, sentencing offers a number of unique factors due to the nature of the

sentencing decision. Like arrest, some “unique” factors are unique because they are not

measured at the other stages. These unique characteristics are broken down into court

characteristics and case characteristics. Items included under court characteristics include

measures such as: jurisdiction of the court, the geographic trial location, and the court size.

Jurisdiction examines Federal versus state, county, or local jurisdiction. Geographic trial

location focuses on regional differences in case processing. Finally, court size refers to the

amount of cases processed through a particular court. All three are applicable at any decision

point, but are not measured at other decision points.8 Items included under case characteristics

6 Some research has separated citizen compliance from demeanor. It may be possible that these measures will be correlated (see Novak et al. 2002). If that is the case, a measure of compliance will be treated as a measure for demeanor. 7 Engel and Silver (2001) established that officer perceptions of mental illness were not clinical definitions of mental illness but may influence officer decision making in arrest situations. Lay perceptions of mental illness are considered unique to arrest because at the sentencing and parole revocation stage the actor will likely have an official report and would not need to rely on lay opinions. 8 Court size would need to be reoperationalized to the volume of police citizen encounters or case loads of parolees revocated to be applied to arrest or parole revocation decisions.

45

include measures such as: mode of conviction, education level of the defendant, employment

status and type of counsel. Finally, parole revocation also presents a number of unique

measures. These factors are divided into offender characteristics and parole characteristics.

Items included under offender characteristics include measures such as: education level and

employment status.9 The unique parole characteristic is program completion.

9 Education level and employment status are considered unique factors for sentencing and parole revocation because these considerations are not measured at the arrest decision so they are not system-wide and thus are unique to these two particular decision points.

46

CHAPTER 3

META-ANALYSIS: BENEFITS AND CRITIQUES

There have been many attempts to examine correlates of criminal justice decision

making. The purpose of this dissertation is to examine these correlates in the context of the

criminal justice system as a whole and determine if theoretical perspectives should focus on the

entire system or specific decisions. While there have been many examinations of criminal

justice decision making there have been limited attempts to synthesize the research and what has

been done has had several problems. First, other than Gottfredson and Gottfredson, all attempts

at research synthesis have been specific to a particular decision. This creates a myopic approach

that prevents scholars from seeing the larger picture, i.e. that factors that influence decision

making may be consistent across the system. The second limitation, as seen in the previous

chapter, is that the majority of these reviews have been narrative, or vote counting-based

reviews, which in and of themselves are problematic.

In the pages that follow, issues of synthesis literature are discussed. First, there is a

discussion of the problems of narrative literature and vote-counting reviews. Second there is a

discussion of the benefits of meta-analysis. Third, there is a discussion the problems with meta-

analysis and the solutions to these problems.

Scholars have always strived to make sense, and take stock of their fields of research.

Cullen (2011) issued a call for scholars in the fields of criminology and criminal justice to

advance the field by taking stock of research and creating a better understanding of these fields

through research synthesis. Many issues in research, such as sampling and measurement

methods, create inconsistencies in findings and these inconsistencies lead to a state of confusion

or “scientific chaos” (Hunt, 1997, p. 1). In an attempt to make order from the chaos, scholars

47

would conduct a massive narrative literature review or a vote-count review in order to make

sense of inconsistent findings (Hunter and Schmidt, 1990). However, these traditional methods

of research synthesis are flawed.

Traditional Narrative Literature Reviews

Traditionally, research on criminal justice decision making was reviewed via the

narrative literature review format. These reviews permeate research on arrest and sentencing

(see National Research Council, 2004; Riksheim and Chermak, 1993; Sherman, 1980; Spohn,

2000). Under these narrative reviews a scholar gathers all available research on a particular

topic, evaluates the research, and makes a determination as to the importance of certain

relationships. While these types of reviews provide valuable insights, the method is a flawed and

antiquated approach to synthesis research.

The first problem is the subjective nature of narrative reviews (Pratt, 2002; Wolf, 1986).

Since the review is based solely on the researcher’s evaluation of the literature, any conclusions

will potentially be biased based on the researcher’s examination of the literature. Essentially,

two scholars could read the same inconsistent evidence and come to opposite conclusions. This

qualitative judgment has no standardized process and this can possibly bias the evaluations.

A second and related issue is there is no set process to conduct a narrative literature

review (Jonson, 2010). This means that there is no procedure for finding, collecting and

evaluating the studies for a narrative review. As a result, bias may be introduced as a result of a

scholar consciously or unconsciously omitting studies from the review. This omitted study bias

creates a subjective review, which may not necessarily be able to be replicated. For example, the

National Research Council’s discussion of correlates of arrest fails to mention any articles using

48

data from the Cincinnati Police Observation study. The omission of these studies is a blatant

example of omitted study bias in narrative literature reviews.

A third issue of narrative reviews is that a large number of studies makes the technique

cumbersome and potentially decreases the accuracy of the findings (Hunter and Schmidt, 1990;

Jonson, 2010). When the literature for a particular topic becomes too voluminous, a researcher

may fail to include studies. The researcher may choose to use a sub-sample of studies in order to

make the review more manageable (Jonson, 2010). There are several problems with this

approach. First, the narrative review is no longer a complete review. Second, omitting studies

creates a selection bias and possibly leads to inaccurate results because critical studies may have

been missed or omitted entirely.

Vote-counting

Another type of traditional research synthesis is the vote-count approach. Under the

vote-count approach, a researcher tallies the results of studies (Hunt, 1997; Hunter and Schmidt,

1990). A researcher divides the findings of studies into three groups: positive relationships,

negative relationships and no relationships. Whichever group is the largest becomes the

interpretation of the relationship for the review. Unlike narrative reviews, vote-counting does

have some objective means of evaluating relationships. However, like narrative reviews, vote-

counting has the same potential for selection bias because of omitted studies. A second and

equally large problem is that vote-counts treat all studies equally (Jonson, 2010). A study with a

sample size of 40 is weighted the same as a study with a sample size of 1,000. Since sampling

error decrease with larger samples, more weight should be placed on the study with a larger

sample size since there is more confidence in those results (Hanushek and Jackson, 1977).

Another related problem is that vote-counting does not account for varying magnitudes of

49

findings (Hunt, 1997; Jonson, 2010). For example, suppose a study found a weak positive effect

and another found a strong negative effect for the same relationship. Under the vote-counting

scheme each study carries the same weight. Essentially the two findings would cancel each other

out rather than the stronger negative effect taking precedence. Thus, while vote-counting is a

slightly more objective approach than narrative literature reviews, it still suffers from flaws and

there is a need for a better way to synthesize research.

Meta-Analysis as an Alternative

Meta-analysis is an analytical technique that can overcome the inherent problems in

narrative and vote-counting reviews. Meta-analysis is a technique designed to make sense of and

evaluate a large amount of information. Meta-analysis is still a relatively young technique; it

was developed in the 1970s (Hunt, 1997). According to Glass (1976), meta-analysis is a

statistical method for the purpose of integrating and evaluating the findings of multiple studies.

Jonson (2010) comments that “[meta-analysis involves a] five-step process: 1) formulating the

problem, 2) collecting the data, 3) evaluating the data, 4) synthesizing the data, and 5) presenting

the findings” (p. 73). This technique has become increasingly more common in the fields of

criminology and criminal justice (Pratt, 2010; Wells, 2009). Wells (2009) found that the

majority of meta-analyses in criminology and criminal justice have been conducted since the

1990s. Furthermore, Jonson (2010) states that meta-analysis is a common technique of the

broader social science research community. However, not all scholars have been advocates of

meta-analysis. Eysenck (1976) referred to the technique as an “exercise in mega-silliness” (p.

517). With this disagreement in mind it is important to discuss the strengths and weaknesses of

the technique.

50

Advantages of Meta-Analysis

Meta-analysis has several advantages over traditional synthesis techniques. According to

Pratt (2001) there are four advantages of meta-analysis; first, meta-analysis can derive exact

measures of strength for a particular relationship called an effect size. Second, meta-analysis can

use multivariate analysis to examine if relationships change under different methodological

conditions. Third, other scholars may replicate meta-analyses. Fourth, meta-analysis is

dynamic, as opposed to static, and when new studies are conducted they can easily be added to

existing meta-analyses and the results may be updated. Additionally, Jonson (2010) stated that

meta-analysis is a better approach for policy analysis and consequently meta-analysis can assess

the implications of policy much better than traditional literature reviews.

Effect size

First, meta-analysis provides an exact measure of strength of the relationship between

two variables. The effect size measure is superior to narrative literature reviews because

traditional reviews can at best vote count (Pratt, 2001). Vote counting may be biased by multiple

studies using the same single data set. The effect size measure gets around this problem by

creating an effect size per data set. The effect size provides a precise measure of strength and

direction for the relationship between the two variables (Hunt, 1997; Lipsey and Wilson, 2001;

Pratt, 2001). Additionally, Jonson (2010), states that effect sizes account for variation in sample

sizes. Thus the studies with larger, more representative samples are granted more weight than

smaller samples. Under vote-counting each sample is given the same weight regardless of

sample size. The effect size essentially creates a better understanding of the true relationship

between the predictors and the outcome variable (Hunt, 1997; Hunter and Schmidt, 1990;

Jonson, 2010).

51

Methodological adjustments

Second, meta-analysis can adjust magnitude measures to account for varying research

methodologies. According to Pratt (2001), Meta-analysis allows for the separation of studies

into groups based on methodological differences and calculates effect sizes for each group.

Meta-analysis allows the researcher to examine if certain pertinent methodological factors

influence the relationship of interest. These methodological considerations include:

demographic information (age, race and gender), study characteristics (longitudinal versus cross

sectional, study location, or year published), and measurement of the outcome variable (Durlak

and Lipsey, 1991; Jonson, 2010; Lipsey and Wilson, 2001). For example, Pratt (1998) found

that the effect of race on arrest was contingent on how race was measured (see discussion of

Pratt, 1998 in Chapter 2).

Replication

Third, according to Pratt (2001), “coding decisions in a meta-analysis are ‘public’” (p.

31). Coding procedures are described in detail with the study and coding sheets are often

published with the study (Jonson, 2010). Because coding decisions are public, the results may be

replicated and research integrity is preserved. In addition to a public coding process, the

eligibility criteria for the meta-analysis are also included and thus another aspect of the meta-

analysis may be replicated (Jonson, 2010). The transparent nature of meta-analysis allows for

any other scholar to replicate the findings and thus ensure research integrity. This transparency

is not found with traditional literature reviews where inclusion criteria are not provided.

Consequently, the sample in a traditional literature review is potentially subject to selection bias.

52

Dynamic

Finally, meta-analysis is a dynamic syntheses method. As such, when other studies are

written they can be added to the meta-analysis and the information can continually be

reevaluated (Pratt, 2001). Weisburd, Telep, Hinkle and Eck (2010) recently reviewed the status

of problem oriented policing as part of the Campbell Collaborative’s systematic review process.

Their review specifically mentioned that as new studies emerged they would be added to the

review and the meta-analysis on the effectiveness of problem oriented policing would

continually be updated. This example demonstrates the dynamic nature of meta-analysis. As

new studies emerge on a particular topic, in this case problem oriented policing, new information

can be added to the meta-analysis and new effect sizes may be calculated with relative ease.

Thus, meta-analysis allows for the most up-to-date and accurate reviews and synthesis of

information possible.

Critiques of Meta-Analysis

Meta-analysis is not a solution to research synthesis that is free from flaws. However,

scholars have created ways to account for these potential problems and adjust the analyses.

There are three problems for this meta-analysis: 1) the file drawer problem, 2) the apples-and-

oranges problem, and 3) the issue of independence of studies. The following section will discuss

each problem and potential solutions.

File drawer problem

According to Rosenthal (1991), the process of publishing research biases meta-analysis.

There is an emphasis on publishing findings that support research questions and thus, academic

journals may not want to publish null findings. These null studies are therefore sitting in the

53

bottom of a scholar’s file drawer. Consequently, meta-analyses may be biased in the direction of

finding an effect. If these unpublished studies included, effect size calculations may be different.

While this is a valid criticism, according to Jonson (2010) this problem is equally valid

for narrative literature reviews and vote-counting studies. Pratt (2010) contends that this

problem may not be as serious of an issue because of the nature of tenure and promotion

amongst scholars in academe. The “publish or perish” phenomenon does not create an incentive

for research to sit in a file drawer. A scholar will not continue to be employed if he/she sits on

null findings and does not make an attempt to get them published. Additionally, given the

research surrounding certain variables in criminal justice research, for example, suspect

demeanor, null findings may question the validity of past research and are certainly worth

publishing.

With these considerations in mind, it appears that the file drawer problem may not be as

large of a problem as originally perceived. There are two approaches to examine this problem.

The first is to try and retrieve unpublished studies (Lipsey and Wilson, 2001). These

unpublished reports may take the form of government reports, dissertations, conference

presentations, and directly contacting well-known scholars in the field for any additional studies.

By using these methods, there is potentially less bias in meta-analytic calculations. The second

method is to statistically examine for bias using the fail-safe N statistic (Rosenthal, 1991). This

statistic calculates how many null findings would be necessary to render the effect size

statistically zero (Lipsey and Wilson, 2001). Higher values on a fail-safe N statistic indicate that

a large number of studies would be necessary to nullify the effect size. As a result, meta-

analyses with higher fail-safe N values can be interpreted as more likely to be valid because it

would take an unlikely number of null findings to invalidate the findings.

54

Apples-and-oranges problem

The second problem with meta-analysis is the apples-and-oranges problem. This

problem is related to varying measures of independent and dependent variables (Lipsey and

Wilson, 2001; Pratt, 2001, 2002). If analyzed studies are not measuring constructs in the same

manner then it may be inappropriate to meta-analyze them together. Meta-analysis assumes that

each study is measuring the same variables in the same way (Hunter and Schmidt, 1990). If this

is not the case then the measures are not necessarily tapping the same idea. Consequently, the

effect size is meaningless (Hunter and Schmidt, 1990). For example, Mitchell (2005) used

multiple measures of sentence severity that included: in/out decisions, sentence length, ordinal

scales of severity, discretionary release (reversed coded so that positive values indicated more

punitive punishments) and discretionary punitiveness. Critics would argue that meta-analyzing

these measures would be inappropriate because each is different. However, proponents would

argue that on all of these scales increasing values indicated more punitive sentencing. Thus,

while individual measures of punishment are slightly different, the overall indication of each

measure was in the same direction and each measure is tapping into the same construct.

A related issue to the apples-and-oranges problem is the quality of the research being

analyzed. Critics have argued that more rigorous studies, for example longitudinal and classical

experiment studies, should not be meta analyzed with less rigorous studies such as quasi-

experimental and non-experimental designs (Logan and Gaes, 1993). Varying quality of

methodological rigor means that studies will vary in internal and external validity and this can

bias the effect size. Critics argue that these meta-analysis attempts statistically improve poor

research (Jonson, 2010). This idea is commonly referred to as “garbage in, garbage out,” which

means that poor data will produce poor statistics. There are two solutions to this issue. The first

55

is to include only studies that measure all variables the same way and/or have greater

methodological rigor. However, this solution would severely limit the sample. The alternative

solution is to statistically control this issue. By statistically controlling the issue, methodological

rigor may be a factor influencing the relationships between independent variables and the

dependent variable.

Independence of Effect Sizes

A third problem for this particular meta-analysis is independence of effect sizes.

According to Pratt (2000), current studies in criminal justice and criminology use multiple

models that can each produce multiple effect sizes. This is especially true in arrest research

given that a large amount of knowledge of arrest is produced via large systematic social

observations like the Project on Policing Neighborhoods, the Cincinnati Study of Police

Behavior and the Police Services Study. The lack of independence creates errors in estimates,

which decrease the variance and increase the likelihood that an effect size will be significantly

different from zero (Pratt, 2000). In other words, the likelihood of making a type II error

increases dramatically. In order to deal with this issue, scholars suggest using a system of

selection criteria to select the best possible representation. This process has been used in

previous meta-analyses of criminal justice decision making (see Kochel et al., 2011, and

Mitchell, 2005). While some scholars have suggested the use of independence adjustment or

hierarchical linear modeling to deal with these issues, Lipsey and Wilson (2001) suggest that if

possible using one representative per data set is the preferred method.

56

Chapter Summary

Meta-analysis has been used for a variety of research questions. Most recently, meta-

analysis has been used to evaluate criminological theory. Some of the meta-analytically

evaluated theories include: Gottfredson and Hirschi’s general theory of crime, social learning

theory, deterrence theory, and bio-social predictors of crime (Paternoster, 1987; Pratt and Cullen,

2000; Pratt, Cullen, Blevins, et al., 2002; Pratt, Cullen, Blevins, et al. 2006; Pratt, Cullen, Sellers

et al., 2010; Pratt, McGloin and Fearn, 2006; Walters, 1992). Each of these evaluations used

meta-analysis to determine the overall worth of the theory and the efficacy of each of the

components of the theory. For example, Pratt and Cullen (2000) found that low self-control had

an effect size over .20 and stated that self-control is “one of the strongest known correlates of

crime” (p. 952). Meta-analysis serves to create an alternative to evaluate theories other than

purely narrative literature reviews.

Meta-analysis can also be used to evaluate the relative strength of known predictors of a

relationship. For example, Pratt and Cullen (2005) used meta-analysis to evaluate a number of

macro-level predictors of crime, such as unemployment, racial heterogeneity, regional effects,

and residential mobility among others. Pratt and Cullen (2005) examined the effects of 23

different variable types on crime. This same principle can be applied to the determinants of

criminal justice decision making.

Meta-analysis provides a method for synthesizing a large number of studies. It is

superior to traditional narrative literature reviews and vote-counting reviews in a number of

ways. First, meta-analysis allows for precise measures of strength of relationships via the effect

size. This statistic indicates whether the relationship between an independent variable exists and

the strength of that relationship. Second, meta-analysis allows for methodological adjustment.

57

Under traditional research synthesis, each study was given the same weight regardless of the

quality of the study. Meta-analysis can correct for this and adjust the effect size based on

methodological considerations. Third, meta-analysis can be replicated. Coding decisions are

public because coding protocols are often included with the presentation of results. This allows

another scholar to replicate the process or modify the process used in the original meta-analysis.

For example, Mitchell replicated Pratt’s meta-analysis of the sentencing literature. The two

scholars came to different conclusions (see Chapter 2). Finally, meta-analysis is dynamic. With

meta-analysis, when more studies are generated new effect sizes for those studies can be

calculated and a new average effect size can be calculated and the relationship can be

reevaluated. Mitchell was able to add additional studies that had been published between his

study and Pratt’s. With narrative literature reviews an entirely new narrative literature review

would have to be conducted and that process would be long and involved. Meta-analysis offers a

faster and more efficient alternative for updating a particular body of knowledge.

While there are many benefits to meta-analysis there are several problems, but scholars

have worked to correct these known issues. The first problem is commonly known as the “file-

drawer” problem. This problem involves the bias of results in favor of a relationship because of

the nature of the publication process. This problem is corrected in two ways, first the meta-

analyst should do his/her best to find as many unpublished studies as possible, and second they

should calculate a fail-safe N statistic (see Chapter 4). The second problem is the “apples-and-

oranges” problem. This problem refers to a bias that is created when a meta-analyst is not

comparing similar constructs. Because of the nature of research not every study is identical. As

a result, there is a possibility that the measures being meta-analyzed may not be the same for a

variety of reasons both on a measurement level and a study design level (i.e. longitudinal versus

58

cross sectional etc.). To combat this problem a meta-analyst can either remove studies of low

methodological rigor or they can include moderator variables and treat methodological rigor as a

factor in the relationship between the independent and dependent variables. The final problem is

the potential lack of independence of effect sizes. This issue is the result of massive data

gathering projects. When large amounts of money are used to generate data, scholars often do

not want to publish just one paper from the data, but instead publish multiple papers from one

data set. This creates a problem for meta-analysis because it requires that each effect size be

generated from an independent source. To combat this problem Kochel et al. (2011) and

Mitchell (2005) suggest that selection criteria be created to select the best possible representation

of a particular data set (see chapter 4 for the specific selection criteria). With these benefits and

problems in mind, the next chapter outlines the statistical procedures for this dissertation.

59

CHAPTER 4

METHODS

With the benefits, and potential problems, of meta-analysis in mind, this chapter outlines

the statistical procedures used in this dissertation. First, there is a discussion of the research

questions. Second, there is an examination of the sampling procedures and a discussion of the

ways to acquire unpublished studies as an attempt to counter the file-drawer problem (see

Chapter 3 for a discussion of the file-drawer problem). Third, there is a discussion of the

dependent, independent and moderating variables that will be used in the statistical analysis.

Fourth, analytical procedures are discussed. Fifth there is a discussion of how the analysis will

indicate which correlates are significant, how it was determined whether criminal justice

theoretical perspectives should be system-wide or decision-specific. This chapter concludes with

a brief discussion of the limitations of this study.

Research Questions

This dissertation is examining the current status of criminal justice theory via an

examination of correlates of criminal justice decision-making. In an effort to examine this idea,

this dissertation proposed to examine several decision points in the criminal justice system in

order to determine the answers to the following questions:

1. What factors influence decision making at arrest, sentencing, and parole revocation?

2. Given these factors, are they consistent across the system as a whole?

3. Given the presence or lack of consistency, should theories of criminal justice focus on the

system as a whole or a particular branch/decision point?

These questions are designed to assess the correlates at each stage and then the entire system. As

a result of this process it is possible to determine what factors influence each decision point and

60

if those are the same influential factors across the system or if factors vary based on the decision

of interest.

Sample of Studies

The first step in the process of meta-analysis is to collect all relevant studies. Several

search strategies were employed to gather all the relevant studies. First, the bibliographies of the

current synthesis studies were scoured to gather all possible studies. However, for any particular

decision, the most recent narrative review is between six and ten years old. Consequently, other

search strategies were utilized.

The second search strategy was to search various criminal justice and social science

research databases. Separate keyword searches were performed on the following databases:

Criminal Justice Abstracts, Criminal Justice Periodical Index, Dissertation Abstracts Online,

ERIC, National Criminal Justice Reference Service, PsycINFO, Social Sciences Index,

Sociological Abstracts, and SocINDEX. These are the most common databases searched in

criminology and criminal justice meta-analyses. The search terms for these searches were kept

intentionally broad due to the generalist nature of the research questions discussed. For arrest,

the search terms used “police” “arrest NOT cardiac” “decision or decisions, or ‘decision

making’”. For sentencing, the search terms used were “sentencing” and “decision, decisions or

‘decision making’”. For parole revocation, the search terms used were “revocation” and

“decision, decisions or ‘decision making’. When compared to existing meta-analyses and

narrative reviews these search procedures produced virtually identical study lists.

The third search strategy was to search all relevant peer-reviewed academic journals issue

by issue to look for more recent information that may not yet be in the databases.10 Jonson

10 Due to the thoroughness of the searches themselves, journal search was limited to online-first articles as those already in print were in the databases and covered by second search strategy.

61

(2010) performed a Google search to find unpublished data and the tactic did glean a few state

and national reports. Thus, the fourth search strategy performed was a Google search to look for

unpublished reports. Because of the broad nature of the searches used Google searches were

impractical. While they were performed, only the first ten pages of results were searched

because search result pages numbered in the thousands and the results went too far off topic to

have any utility past this point. A negligible number of potential studies were actually gleaned

from this method.

In total, the raw number of results from these various methods was over 3000 for arrest,

over 1000 for sentencing and approximately 500 hits for parole revocation (all three estimates

are inclusive of duplicates). After each search was performed, abstracts and titles were inspected

for topic appropriateness. For arrest, 236 articles were determined to have a potential to be

coded, and of those, 62 were deemed codeable and subsequently coded. A total of 54 separate

data sets or studies were found. The reason for the difference is that some articles used the same

data set to produce the results and could not all be included in the meta-analysis due to

dependency issues (see below).

For sentencing, 396 articles were determined to have potential, and of those, 65 were

deemed codeable, and a total of 46 separate data sets or studies were collected. It is important to

note that 20 of the 396 studies were removed because they dealt with juveniles. Since the

juvenile adjudication process is substantively different from adults and are two separate systems,

studies focusing on juveniles were not included in analyses of sentencing and parole revocation.

However juvenile focused studies were included in arrest because the arrest process is not

substantively different for juveniles or adults. Juveniles are arrested by the same officers that

arrest adults. Finally, for parole revocation, a total of 82 studies were thought to have potential,

62

and of those 17 were coded, which yielded 19 separate data sets/studies. Surprisingly, parole

revocation research did not utilize massive data sets to produce multiple studies as often as arrest

or sentencing articles.

Inclusion Criteria

For studies to be included in this meta-analysis they must have met the following criteria.

First, the study must have examined one of the three decisions of interest i.e. arrest, sentencing,

or parole revocation, and the measurement of those decision points must be consistent with the

discussion in the dependent variable section (see below). Second, the study must include at least

one measure per system-wide domain for the independent variable. These independent variable

domains include measures for offense severity11, and suspect/defendant/offender

characteristics.12 Third, the study must provide the necessary information to calculate an effect

size. These inclusion criteria are relatively lax in an effort to include more studies as opposed to

limiting the number of studies. Including multiple studies allows for better examination of

moderating variables such as methodological rigor.

Selection Criteria

Meta-analysis requires every effect size to be independent. If effect sizes are not

independent then the mean effect size may be inflated and produce biased results in favor of

finding an effect (Lipsey and Wilson, 2001). Because of the nature of the studies that were

analyzed, this was not always possible if every possible. Every article did not pertain to one data

set. For example, many sentencing research articles were generated from sentencing data from

11 Domain refers to a grouping of predictors for a particular outcome. 12 Criminal justice actor characteristics were not included as very few studies include these measures and including this requirement would needlessly shrink the sample size. Community characteristics were not included here because they are often aggregate measures and the nature of decision-making research tends to be more micro level analysis and as a result valid studies may be needlessly eliminated if community characteristic measures were made mandatory.

63

Pennsylvania and the years of data often overlap. The same data cannot be continually used as it

would bias the results. As a result, some criteria are necessary to select an article that best

represents that particular data set. This dissertation used criteria modified from Mitchell (2005)

and Kochel et al. (2011). The criteria for selecting the ideal representation of a particular data

set were: 1) codeability- the particular article that provides the most information for coding

purposes was included, 2) number of control variables- the article that has the greatest number of

control variables was included, and 3) sample size- the article with the largest sample size was

included because that represents the largest number of cases and larger samples are generally

seen as more representative. These selection criteria are consistent with current meta-analyses of

criminal justice decision making.

Dependent Variable

The overall dependent variable in this study is criminal justice decision processing. The

definition in Chapter 1 established that criminal justice examines “governmental social control

premised on punishment or blameworthiness” (Duffee and Allan, 2007, p. 8). As such, criminal

justice is the study of processing. More simply, criminal justice is the study of discretion. It

seeks to examine what causes people to be pushed further through the system and others are

released or not pursued through the system. Operationalization of this concept varies depending

upon the decision point of interest. Because of the nature of this study, each decision point

presents a different operationalization of the dependent variable. For example, a study of

sentencing would not operationalize criminal justice processing in the same manner as a study of

arrest. They are looking at different steps in the criminal justice process. However, many

studies measure process outcomes similarly in that they measure the dependent variable

dichotomously as yes or no.

64

The arrest decision, which is defined as a police officer taking a person into custody in

order to be processed further into the criminal justice system, measures processing as a yes/no

dichotomy. In this dissertation, studies used for the arrest analysis, measured the arrest

dichotomously. The decision to only use studies that measure arrest dichotomously is to

eliminate studies that examine use of force or a continuum of law enforcement sanctions as they

are outside the scope of the arrest decision, and could lead to a problem with heterogeneity (see

below for a discussion of heterogeneity).

The sentencing decision point is more complex than the arrest decision point from an

operationalization perspective. Sentencing is defined here as a judge’s determination of the

appropriate punishment for an individual. Many studies of sentencing focus on the incarceration

decision as an in/out dichotomy. That is, studies focus on the decision to incarcerate or not. Yet,

these same studies often also collect information on the length of the incarceration term.

Mitchell (2005) used multiple sentencing measures, this strategy was not used in this

dissertation. This decision was made to avoid both dependency and the apples-and-oranges

problem. By limiting the sample in this way, the issue of dependency is avoided. There is no

overlap between metric sentence length measures and dichotomous in/out scales that apply to the

same set of defendants. This limitation still allows for a sufficiently large sample, but avoids any

chance that the dependent variables are not compatible with one another. Additionally, limiting

to the in/out decision required fewer artificial transformation procedures of effect sizes.

Like sentencing, parole revocation is more complex, in terms of measurement, than

arrest. For this study, parole revocation is defined as a parole officer's or parole board’s pursuit

of re-incarceration of an individual serving the remainder of their sentence under parole

supervision. This decision point has been measured dichotomously using the yes/no for a

65

measure of parole revocation. Since the previous two decisions used a dichotomous dependent

variable measure, it is appropriate that the same standard is applied to the parole revocation

decision. The reasons for using only a revoke/no revoke are the same as in sentencing, to avoid

dependency issues, and the apples-and-oranges problem.

Independent Variables

With the research questions in mind, there should be some variables that are common to

each branch of the system, for example, race of the criminal justice actor and other variables that

are decision-specific, for example attorney type. The following discussion will explain what

variables should function system-wide and which variables are unique to a particular decision

point.

System-wide variables

Several factors have consistently been predicted to explain decision making across the

system. Each of these factors constitutes a separate domain of independent variables in this

meta-analysis. The four principle domains cover offense seriousness, criminal justice actor

characteristics, suspect/defendant characteristics, and community characteristics. Items under

the seriousness domain include measures of seriousness of the offense (e.g. felony versus

misdemeanor), and prior record. Items included under the criminal justice actor domain would

include measures of race, ethnicity, gender, cynicism, professionalism/education,13 and

experience of the criminal justice actor. Items included under the suspect/defendant

characteristic domain would include race, ethnicity, age, gender, and social class. Finally, items

included under the community conditions domain include poverty, family structure, community

racial composition and other socio-economic predictors.

13 Education and professionalism are combined because they are generally measured interchangeably see Smith (1984).

66

Arrest

While the above domains do apply to the arrest decision, there are several domains that

appear to be uniquely applied to the arrest decision. These two arrest specific domains are

organizational characteristics, and other situation characteristics. Items included under the

organizational characteristics domain include measures of organizational hierarchy and police

agency size. Items included under other situational characteristics include measures of suspect

demeanor, intoxication, suspect non-compliance, interaction-phase crime, and lay perceptions of

mental illness. It should be noted that items included under these domains are “unique” because

they are only used in arrest research but are applicable to other decision points. Prior arrest

research has examined these factors, but studies of sentencing or parole revocation have not

included these measures. As a result, they are “unique” to arrest for analytical purposes.

Sentencing

As with arrest, the sentencing decision adds other domains that are unique to the

sentencing decision. These new domains include court characteristics, and case characteristics.

Items under the court characteristics domain include measures of jurisdiction of the trial (e.g.

federal versus local court), Trial location (e.g. region of the United States in which the trial took

place, or a particular region of the state examined for the study), and court size (e.g. the number

of judges in a particular jurisdiction). Items under the case characteristics domain include

measures of mode of conviction (e.g. plea versus guilt at trial) and type of counsel (e.g. as public

versus private counsel). While case characteristics are unique to the sentencing decision, the

court characteristics are similar theoretically speaking to organizational characteristics described

in the previous section. Like arrest, these court characteristics are unique because they are not

measured at other stages.

67

Parole revocation

The parole revocation decision presents another set of variables that are unique to the

particular decision. These new factors are divided into two domains: offender characteristics,

and parole conditions. Factors included under the offender characteristics domain include:

education level of the offender (e.g. ordinal scale or a series of dummy variables), employment

status (e.g. ordinal scale or as a series of dummy variables). The new factor under the parole

condition domain is program completion. Again, like the previous two decision-specific domain

discussions, offender characteristics can be applied to other domains, but they are not measured

in studies of the other decision points.

Moderating Variables

Moderating variables are similar to control variables but in this case they are variables

that pertain to the studies themselves and not necessarily the theories within the study. This is

one of the advantages of meta-analysis because it allows for considerations of items such as

whether the study utilized a particular data collection method, or if the data were collected in a

particular time period. Each variable is discussed below.

Study characteristics:

The present meta-analysis coded data on several study characteristics. First, the

publication year and decade were recorded. Additionally, the decade and year of data collection

were also recorded. The type of report (e.g. book, book chapter, journal article, or conference

paper), and the location of the study (what region of the United States: Northeast, South,

Midwest, Southwest, and Pacific/West) were recorded. A variable was created for whether or

not the study was multi-jurisdictional/national.

68

Methodological quality:

The second domain of moderating variables was related to the methodological quality.

Since this meta-analysis did not want to eliminate all studies of poor quality, the alternative

solution is to examine the impact of methodological quality on outcome variables.

Methodological quality was assessed via the type of data collected. The duration of the study

was recorded. In order to account for the possibility of seasonality bias, a dummy variable was

created to examine whether or not data collection was continuous or had periodic breaks. For

example, the Cincinnati Police Observation Study was conducted year round as opposed to the

Project on Policing Neighborhoods that collected data over two consecutive summers.

Interaction effects are an important consideration in decision-making research. For

example, many studies have found that it is not necessarily young offenders, racial/ethnic

minorities or males that alone are significantly more likely to be incarcerated, but rather it is the

interaction of all three so that young minority males are most likely to be severely punished

(Spohn, 2000). Wooldredge (1998) argues that studies of sentencing need to improve

measurement and include measures of interaction effects. Therefore, a dummy variable was

created to indicate whether or not the effect size was generated from a study that measured

interaction effects.

Pratt (1998) found that the manner in which race was measured had an impact on the

effect of race. A moderating variable that notes the way in which race is measured was recorded

(coded as a three-category variable, where 1 = Black/White, 2 = White/non-White, and 3 = some

other racial classification scheme). Because race and ethnicity are closely related and at times

used interchangeably, a moderating variable noting the measurement of ethnicity was also

69

created (coded as a three-category variable, where 1 = Hispanic/non-Hispanic, 2 =

Black/White/Hispanic, and 3 = some other ethnicity classification scheme).

Arrest research has two additional moderating variables that apply to the arrest analysis.

First, systematic social observation is a well-respected form of data collection and studies not

using this data collection method may not be as methodologically sound. However, not all

police research has been conducted via systematic social observation. A dummy variable was

created to measure whether or not the effect size was generated from data recorded via

systematic social observation. This variable applies only to the arrest assessment.

Second, an issue that has been raised in research on arrest is the operationalization of

disrespect or demeanor. According to Klinger (1994), suspect demeanor was not operationalized

properly and as a result what was actually interaction-phase crime that should have accounted for

the arrest was measured as suspect disrespect. Due to this critique, another moderating variable

was created to account for this operationalization issue. This variable was coded as 0 when there

was no interaction-phase crime measure and 1 when an interaction-phase crime measure was

present.

Reliability of Coding

It is important to produce coded results that are reliable. One of the benefits of meta-

analysis is that it can be reproduced because meta-analysts are proponents of transparency and

replication. This dissertation evaluated coding reliability using the Yeaton-Wortman method.

Yeaton and Wortman (1993) suggest that agreement is equal to the total number of agreements

divided by the total number of comparisons. Thus, if reliability is checked and the two coders

agree 50 times over 100 total points of comparison then the amount of agreement would be equal

to 0.50. Values of 1.00 indicate total agreement and values of 0.00 indicate a complete lack of

70

agreement. The principle coder selected thirteen studies at random and had those studies coded

by a reliability coder. After the thirteen studies were double coded, Yeaton and Wortman’s

method was applied. The level of agreement between the reliability coder and the principle

coder was .92. This value is well above any threshold where reliability is questioned.

Analytic Strategy

In order to address the research questions listed above, a series of analyses were

conducted. First, a series of univariate statistics were calculated. These statistics are designed to

generate a general description of the sample. Second, effect size measures were calculated

between the independent variables of interest and the appropriate dependent variables. Next, a

mean effect size was calculated for each relationship. Third, heterogeneity of effect sizes and the

fail-safe N were evaluated in order to address common criticisms of meta-analysis, namely the

apples-and-oranges problem and the file-drawer problem. Each of the techniques is described in

detail below.

Effect size measure

Consistent with the current meta-analysis literature (see for example: Berlin, and Colditz,

1990; Brind, Chinchilli, Severs, et al., 1996; Cosgrove, Sakoulas, Perencevich, et al., 2003;

Fleiss 1993; Fleiss and Berlin, 2009; Gaugler, Duval, Anderson, et al., 2007; Geddes, and

Lawrie, 1995; Greenland, 1987; Greenland, 1993; Haddock, Rindskopf, and Shadish, 1998;

Kochel et al., 2011; Lasky-Su, Faraone, Glatt, et al., 2005; Lipsey and Wilson, 2001; Lӧsel and

Schmucker, 2005; Mitchell, 2005; Tenback, van Harten, van Os, 2009), the measure of effect

size estimate used here was a logged odds ratio. This effect size estimate was chosen for several

reasons. First, logged odds ratios were the most common effect size available for all three

groups of studies. Second, logged odds ratios are symmetrical around zero making them easily

71

interpreted regarding significance and directionality (Hanushek and Jackson, 1977). Third,

formulas exist for transforming other test statistics (such as t, F, and χ2) into a logged odds ratio

value (Lipsey and Wilson, 2001). Additionally, Hedges and Olkin (1985) provide conversion

formulas for probit coefficients and variances. Fourth, logged odds ratios can be used even with

the presence of covariates and “inferences about the [odds ratios] may be drawn in the usual

ways” (Fleiss and Berlin, 2009, p. 245). Sixth, logged odds ratios can be easily transformed into

their more interpretable odds ratio counterpart by simply taking the natural anti-log of the logged

odds ratio, giving them easy interpretation. The natural anti-log, also known as exponent B, is

easily interpreted as an increase in the dependent variable given an increase in the independent

variable.

After each effect size was calculated, it was then weighted by the inverse of the variance.

This step was taken to ensure that studies with larger samples are given more influence. Effect

sizes generated from larger samples are given more weight because larger samples are assumed

to be more representative of the population and thus, more accurate. The weighting procedure

follows those outlined by Lipsey and Wilson (2001) where weights are equal to the product of

the effect size, in this case the natural log of the odds ratio, and the inverse of the variance. The

inverse of the variance is equal to:

wLOR = ______abcd______ ab(c+d) + cd(a+b)

where wLOR = the inverse variance weight for the effect size

a = when both variables have a value of zero b = when the independent variable has a value of zero and the dependent variable has a value of one c = when the independent variable has a value of one and the dependent variable has a value of zero d = when both variables have a value of zero

72

Once each effect size has been weighted using the inverse variance method, a mean effect

size was calculated. The mean effect size is equal to the sum of the weighted effect sizes divided

by the sum of the inverse of the variance for each weighted effect size (Lipsey and Wilson,

2001). In order to determine that the mean effect size was significantly different from zero, a

95% confidence interval was calculated. If the confidence interval contains zero, then that

particular mean effect size was considered non-significant. If zero was not contained within the

confidence interval, then the mean effect size was considered significant at the .05 level and that

relationship was included in the grand theory versus decision-specific discussion.14

Independence of effect size measures

There is a large chance that effect sizes drawn may not be independent of one another.

As a result, the effect sizes generated from these data sets may not be independent. Inclusion of

dependent effect sizes will bias the mean effect size. This issue was combated by selecting only

one measure per variable per data set. This method is consistent with current criminal justice

decision-making meta-analyses.

Heterogeneity of measures

A criticism of meta-analysis is that meta-analysts combine measures that are too

dissimilar to be meta-analyzed properly. If measures generating effect sizes are not similar or

tapping into the same constructs, then the effect sizes estimates are effectively meaningless.

However, meta-analysts are aware of this issue and they created Q statistics to combat this

specific problem (Pratt, 2001). The Q-statistic uses a chi-square distribution. According to

Lipsey and Wilson (2001) the formula used to assess heterogeneity of effect size estimates is as

follows:

14 All effect size calculations, Q, and moderating variable analyses were conducted using STATA macros available on David Wilson’s website http://mason.gmu.edu/~dwilsonb/ma.html

73

Qes = ∑(wi(ESi – ES+) 2)

where wi is the inverse variance weight for effect size i.

ESi is the individual effect size

and ES+ is the weighted mean effect size.

Once Qes is calculated for each effect size, all values are summed. The summed value is

designated Qobt. Values of Qobt were compared to critical values of the chi-square distribution

with k-1 degrees of freedom where k is the number of effect size estimates. Values found to be

significant were considered heterogeneous. According to Pratt (2001), values found to be

significant may potentially be too dissimilar to be combined. However, Lipsey and Wilson

(2001) suggest that significant Q values simply represent data gathered from varying

populations. Effect sizes generated from heterogeneous populations can be adjusted and

reanalyzed once outliers are removed (Lipsey and Wilson, 2001). When Q values were

significant, several steps were taken. First, Bonta et al. (1998) suggest that any value greater

than three standard deviations from the mean should be considered an outlier and removed.

Second, estimates that are discontinuous from the rest of the distribution were also removed

(Durlak and Lipsey, 1991). Once the outliers were removed, a new mean effect size, weighted

mean effect size, and confidence interval were calculated. Both results, one with outliers and

one without outliers, are presented. Additionally, random effects modeling was used for effect

size analyses. Random effects modeling generates larger confidence intervals than fixed effects

modeling, and is a more conservative approach (Lipsey and Wilson, 2001).15

15 For mean effect sizes both the fixed and random effects results are presented in the next chapter.

74

Fail-safe N statistic

As discussed above, a critique of meta-analysis is that studies with null findings are less

likely to be published and rather will sit in a scholars file drawer collecting dust. Given the

current environment of scholars at universities, it seems unlikely that someone would hold a

paper in a file drawer. Moreover, null findings that question the validity of previous research are

equally worthy of publication. Nonetheless, meta-analysts have developed statistics to combat

this file-drawer problem critique, the fail-safe N statistic. This statistic estimates the number of

studies with alternative findings necessary to render the current effect size to the point of

insignificance (Wolf, 1986). The larger the value of the fail-safe N statistic, the more studies

with alternative findings that would be necessary to find a non-significant effect size (Hunter and

Schmidt, 1990; Lipsey and Wilson, 2001; Pratt, 2001). According to Lipsey and Wilson (2001)

the fail-safe N formula is as follows:

k0 = k[(ESk/ESc)-1]

where: k = number of studies used to calculate the mean effect size

ESk = the mean effect size

ESc = the alternative effect size value of non-significance

The values used for ESc varied depending on the particular calculation. The value used was the

minimum value that would incorporate the null value, zero, within a 95% confidence interval.

Conclusions

The analyses discussed above were designed to assess the average strength of correlates

of decision-making at three distinct decision points across the criminal justice system. The

ultimate goal of this assessment is to examine the support for various criminal justice theories

and to work toward the plausibility of a general theory of criminal justice. In order to achieve

75

these goals, each relationship was assessed via the mean effect size process. Correlates that

generate significant effect sizes were retained for each particular decision point. Those that did

not generate a significant mean effect size were eliminated from consideration toward a general

theory of criminal justice.

After determining which correlates of decision-making were significant, an assessment

was made to examine the existence of a grand theory approach versus decision-specific approach

to theories of criminal justice. When mean effect sizes for various predictors are significant at

multiple decision points and have sufficiently large fail-safe N values, then a grand theory of

criminal justice is the recommended pathway. It is not necessary that correlates function in the

same manner at each decision point. It is conceivable that the same variable may act differently

depending on the decision point. For example, it is possible that officers are more likely to push

women through the formal criminal justice process and judges are more likely to be lenient at

sentencing because their action removes women from their familial responsibilities for a longer

period of time than an arrest. In this case, the same variable, gender, is functioning differently at

two different decision points. Yet the end result is that gender matters at both decision points.

The alternative approach to criminal justice theory organization is a series of decision-

specific theories. These theories would emphasize decision-specific variables or each decision

point would emphasize different predictor variables. If decision-specific variables produce

larger effect sizes or each decision point has a unique set of significant effect sizes, with

sufficiently large fail-safe N values, then decision-specific theories will be most appropriate for

theories of criminal justice.

76

Limitations

There have been some concerns raised about meta-analysis. These concerns about meta-

analysis are three main problems: 1) the publication bias, 2) the incompatibility problem and 3)

the lack of independence of effect sizes. Each of these problems has been discussed in detail

above and solutions were proposed.

Another issue is that the data are limited to mostly published materials and as a result this

study will be biased by the same biases in those other studies when assessing the correlates of

decision-making. This problem is commonly known as “garbage in, garbage out”. A flaw with

this meta-analysis, and any meta-analysis, is that the effect sizes are only as good as the data

used to generate them. There is no cure for this issue other than going through the moderating

variable analysis. Missing variables are another related issue, and they happen for a multitude of

reasons. Since this study is a meta-analysis there is even less control over missing variables.

Omitted variable bias is even more of an issue. To combat this issue, random effects modeling

was used to generate larger confidence intervals and thus, more conservative estimates.

Finally, these analyses cannot examine interaction effects. The present analysis cannot

include interaction effects because those measures are potentially dependent with the non-

interactive variables. For example, the inclusion of race_X_gender interaction effect may cause

dependency issues with the separate measures of race and gender. The best this meta-analysis

can do is to assess potential differences in studies that used interaction effects and those that did

not via moderating variable analysis.

77

CHAPTER 5

RESULTS

This dissertation set out to answer three questions. First, what factors influence three

decisions across the criminal justice system? Each of these decisions, arrest, sentencing, and

parole revocation, are assessed and discussed. The second question examined in this dissertation

is, given significant factors, are they consistent across the system or are significant factors unique

to a particular decision point? The third question examined is, should criminal justice theory

take a whole-system or decision-specific approach when formulating theories of criminal justice?

The second and third questions will be discussed in the next chapter.

This chapter will discuss the results from the meta-analysis described in the previous

chapter. The first part of this chapter is a discussion of the general publication characteristics of

each sample. The second part is a discussion of the results of the analysis of each of the

decisions. The discussion of each decision is subdivided into sample characteristics, effect sizes

and moderating variable analysis. It is important to note that effect sizes and confidence

intervals in tables were left as logged odds ratios. This was done for ease of interpretation of the

confidence intervals as logged odds ratios are symmetrical around zero and thus, any confidence

interval containing zero is not significant at 95 percent confidence.

Publication Characteristics

As shown in Table 5.1, the analysis of arrest was based on 54 studies. The majority of

studies were published recently. Approximately 96 percent of studies were published since

1990. However, only 70.3 percent of studies began data collection during that same time frame.

This is likely due to large data collection efforts like POPN, PSS, or the Cincinnati study, which

are still being used to generate publications today. The vast majority of data were generated

78

from studies published in journals. Geographically, studies were scattered. The Midwest region

was where the largest number of studies were conducted, 27.80 percent. The next most common

geographic categories were the South and multi-site studies (18.50 percent of studies each). The

least common study location was the Pacific/Northwest category. Finally, the majority of

authors were university/college professors and the most common discipline was

criminology/criminal justice.

As shown in Table 5.1, the sentencing analysis was based on 46 studies. The majority of

studies were published recently. Specifically, 66.70% of studies were published since 2000.

However, some studies were based on older data. Approximately one quarter of studies were

based on data gathered in the 1960s or 1970s. Sentencing studies tended to be scattered across

the United States. The most common study region was the Midwest, 34.80 percent, followed by

the South, 28.30 percent. Eight studies were based on data collected in multiple states or federal

court. The most common publication type was a journal, with about 70 percent of studies

published in journals. The vast majority, of authors (95.30%) were affiliated with universities or

colleges. Finally, approximately 70 percent of authors worked in the criminology/criminal

justice discipline.

Table 5.1 also presents the general publication information for the parole revocation

analysis. It should be noted that studies of parole revocation were the least consistent. The

parole revocation analysis presented the smallest sample size, with only 19 accepted studies.

Additionally, the parole revocation publication analysis had a fair amount of missing data given

the small sample size. The results indicate that the majority of studies were published recently.

63.10 percent of studies had been published since 2000. The majority of data collection has

taken place since 1990. As with previous decisions, studies are scattered throughout the United

79

States. Parole revocation studies were most commonly conducted on the West Coast, 33.30

percent, followed by the Midwest, 22.20 percent. The most common publication type was an

academic journal, 73.70 percent. Finally, authors were primarily affiliated with universities and

colleges and the most common discipline was criminology/criminal justice or mixed, 36.80

percent each.

80

Table 5.1: Publication Characteristics Arrest Sentencing

Parole Revocation

Publication Characteristic N % N % N % Publication Decade 1970 1 1.90 0 0.00 1 5.30 1980 1 1.90 6 13.30 3 15.80 1990 20 37.00 9 20.00 3 15.80 2000 32 59.30 27 60.00 5 26.30 2010 0 0.00 3 6.70 7 36.80 Publication Type Journal 37 68.50 32 69.60 14 73.70 Dissertation/Thesis 11 20.40 12 26.10 2 10.50 Agency Report 6 11.10 1 2.20 3 15.80 Other 0 0.00 1 2.20 0 0.00 Data Gathered 1960 2 3.70 2 4.30 0 0.00 1970 6 11.10 10 21.70 0 0.00 1980 8 14.80 4 8.70 3 18.80 1990 20 37.00 24 52.20 5 31.80 2000 18 33.30 6 13.00 8 50.00 Geographic Region of the Study Northeast 9 16.70 4 8.70 1 5.60 South 10 18.50 13 28.30 3 16.70 Midwest 15 27.80 16 34.80 4 22.20 Southwest 7 13.00 4 8.70 3 16.70 Pacific/Northwest 3 5.60 1 2.20 6 33.30 Mixed/Nationwide 10 18.50 8 17.40 1 5.60 Author Affiliation University/College 51 94.40 41 95.30 15 78.90 State Agency 0 0.00 0 0.00 1 5.30 Mixed 1 1.90 2 4.70 3 15.80 Other 2 3.70 0 0.00 0 0.00

81

Table 5.1 continued Arrest Sentencing Parole

Revocation Publication Characteristic N % N % N % Author Discipline Criminology/Criminal Justice 34 63.00 29 69.00 7 36.80 Psychology 1 1.90 0 0.00 1 5.30 Sociology 9 16.70 10 21.30 2 10.50 Social Work 1 1.90 0 0.00 1 5.30 Mixed 3 5.60 2 4.80 7 36.80 Other 6 11.10 1 2.40 1 5.30 N 54 46 19

Arrest

Sample characteristics

Table 5.2 describes the sample characteristics for the arrest study analysis. As shown in

Table 5.2, the average sample size was 60,423.61 cases and across all 54 studies, 3,262,875 cases

were studied. The mean number of agencies involved in each study was 4.311 agencies. The

mean study length was 21.67 months. Approximately, 20 percent of studies reported at least one

analysis with interaction effects, a quarter of studies used systematic social observation, and

nearly all studies presented information on control variables and used continuous data collection,

92.60 and 98.10 respectively. Finally, approximately 70 percent of studies collected information

on offense specific arrests.

Table 5.2 also describes what variables were collected in various studies. These variables

are separated into 4 groups: suspect characteristics, officer characteristics, seriousness

characteristics, and other characteristics. For suspects, the majority of studies collected

information on race, gender and age of suspects, 83.30, 72.20, and 66.70 respectively.

Approximately 30 to 25 percent of studies reported information on suspect ethnicity, demeanor,

82

or the presence of interaction phase crime. Finally, few studies collected information on suspect

compliance, or whether the suspect was mentally ill, 16.70 and 5.60 percent, respectively.

Officer characteristics were not commonly recorded. Officer race and experience

measures were most common. Each was recorded in 20.40 percent of studies. Officer gender

was slightly less common 18.50 percent of studies. Officer education was recorded in 16.70

percent studies.

As shown in Table 5.2, the mean number of seriousness indicators was 1.407 per study.

Weapon use and a general seriousness measure, most commonly measured as felony versus

misdemeanor, were recorded in 33.30 and 31.50 percent studies respectively. An intoxication

measure was recorded in 29.60 percent of the studies. Evidence and victim preference for arrest

were recorded in about one quarter of studies, 25.90 and 24.10 percent respectively. Finally,

other characteristics were relatively uncommon. Less than 10 percent of studies recorded

organization measures and slightly more than 10 percent recorded information on community

variables. However, it should be noted that there was very little consistency in these measures

and therefore, they are not included in the effect size analysis.

83

Table 5.2: Arrest Sample Characteristics Variable Mean/% S Total General Study Characteristics Agencies 4.31 7.25 Offense Specific (Yes=1) 68.50 Control Variables 92.60 Systematic Social Observation 25.90 Interaction Effects 18.50 Length 21.67 29.19 Continuous Data Collection 98.10 Size 60423.61 149133.61 3262875 Suspect Characteristics Suspect Race Measure 83.30 Suspect Gender Measure 72.20 Suspect Age Measure 66.70 Suspect Ethnicity Measure 27.80 Demeanor Measure 29.60 Interaction Phase Crime Measure 24.10 Compliance Measure 16.70 Mental Illness Measure 5.60 Officer Characteristics Officer Race Measure 20.40 Officer Gender Measure 18.50 Officer Education Measure 16.70 Officer Experience Measure 20.40 Seriousness Characteristics Seriousness Measure 31.50 Evidence Measure 25.90 Intoxication Measure 29.60 Weapon Use Measure 33.30 Victim Preference Measure 24.10 Total Serious Measures 1.41 1.27 Other Characteristics Organization Variables Measured 9.40 Community Variables Measured 11.10 N 54

84

Effect sizes

Table 5.3 displays results for the arrest effect size analysis. Suspect race, gender and

ethnicity significantly influenced arrest. Specifically, black suspects were approximately 1.40

times more likely to be arrested than white suspects, male suspects were approximately 1.50

times more likely to be arrested than female suspects and Hispanic suspects were approximately

1.25 times more likely to be arrested than non-Hispanic suspects across studies. Age and mental

illness were not significantly related to arrest. Additionally, all three of the characteristics were

all significant at p < .001. No officer characteristics were significantly related to arrest.

Seriousness measures are all significant and in the expected directions. The strongest

seriousness indicator was strength of evidence. As amounts of evidence increased, arrest

chances were almost 7 times higher. When the victim preferred arrest, the odds of arrest

increased 4.10 times. Felonies were approximately 2.50 times more likely to end in arrest than

misdemeanors. Finally, suspect intoxication and the presence or use of a weapon increased the

odds of arrest by 1.53 and 1.58 times respectively.

Suspect demeanor measures were also influential in arrest outcomes. Demeanor was not

stronger than the effect of interaction phase crime. Demeanor increased the odds of arrest by

almost 200 percent, but when a suspect committed an interaction phase crime the odds of arrest

increased 3.17 times across studies. Additionally, when suspects were considered non-

compliant, their odds of arrest increased 1.79 times.

85

Table 5.3: Mean Effect Sizes Arrest Variable k N Mean

ES W Mean

ES 95% CI

Min 95% CI

Max

Suspect Variables Suspect Black*** 42 3253123 0.66 0.33 0.221 0.438

(0.01) (0.02) (0.03)

Without Outliers*** 41 3239557 0.52 0.33 0.222 0.438

(0.01 (-0.02) (0.03)

Suspect Male*** 38 3213071 0.83 0.40 0.29 0.52

(0.16 (0.12) (0.19)

Without Outliers *** 37 3199505 0.38 0.40 0.29 0.52

(0.16) (0.12) (0.19)

Suspect Age 35 2120709 0.06 -0.001 -0.004 0.003

(0.001) (0.001) (0.001)

Suspect Ethnicity*** 15 2858798 0.26 0.22 0.104 0.343

(0.02) (0.02) (0.03)

Suspect Mental Illness 4 4364 -0.51 -0.47 -1.43 0.49

(-0.34) (-0.77) (0.09)

Officer Variables Officer Black 11 1822723 -0.07 -0.035 -0.097 0.027

(0.01) (-0.01) (0.02)

Officer Male 10 1787322 0.32 0.095 -0.069 0.26

(-0.020) (-0.05) (0.01)

Officer Experience 8 1733917 0.12 0.009 -0.018 0.037

(0.003) (0.002) (0.01)

Seriousness Variables Seriousness*** 14 19782 1.91 0.925 0.585 1.265 (0.34) (0.27) (0.40)

86

Table 5.3 continued

Variable k N Mean ES

W Mean ES

95% CI Min

95% CI Max

Evidence*** 12 1728460 2.63 1.93 1.38 2.49

(0.67) (0.59) (0.74)

Intoxication*** 17 64711 0.60 0.43 0.27 0.59

(0.38) (0.30) (0.45)

Weapon Use*** 15 73113 0.59 0.46 0.30 0.62

(0.34) (0.25) (0.42)

Arrest Preference*** 13 12811 1.45 1.41 0.87 1.94

(1.57) (1.36) (1.77)

Demeanor Variables Demeanor*** 16 150640 0.03 0.69 0.36 1.02

(0.02) (0.002) (0.04)

Without Outliers*** 13 150050 0.68 0.64 0.31 0.97

(0.02) (0.001) (0.04)

Interaction Phase Crime** 9 43810 1.81 1.15 0.35 1.96

(0.19) (0.01) (0.36)

Compliance** 9 6794 0.66 0.58 0.15 1.02 (0.45) (0.31) (0.58) (Fixed effects results in parentheses) * sig p< .05 ** sig p<.01 *** sig p< .001

Fail-safe N analysis

The fail-safe N statistic determines the number of studies necessary to render the effect

size to a value of non-significance. Higher values indicate that more studies with null values are

required to nullify the mean effect size. The value chosen for the negligible value varied

depending on the particular mean effect size. A fail-safe N value was calculated for each

87

significant weighted mean effect size value. Overall, there is confidence in these significant

effect sizes. The average fail-safe N value was 36. There was some variation in fail-safe N

values amongst the various significant mean effect sizes. The “suspect black” finding would

require another 86 studies to render the finding presented above to an insignificant value.16 The

effect of suspect gender was also a stable finding; the fail-safe N value for gender was 93.

Ethnicity was the least stable significant suspect characteristic with a fail-safe N value of 15.

While there is some variation amongst these significant suspect characteristics, they are

relatively stable.

All of the seriousness measures are stable as well. The relationship between seriousness

of the offense and arrest would require another 25 null effect sizes, and the evidence effect size

would require an additional 30 null effect sizes to render these findings non-significant.

Additionally, the intoxication effect size’s fail-safe N value is 28, and the weapon use effect size

fail-safe N value is 31. The arrest preference effect size would require an additional 21 effect

sizes to render that finding non-significant. Finally, the fail-safe N values for various demeanor

measures were the least stable. The fail-safe N value for demeanor is 13. The mean effect sizes

for interaction-phase crime and compliance are much less stable than the fail-safe N for

demeanor. The values are 4 and 3 respectively.

Moderating variable analysis

An attempt was made to dig deeper into the effect size analysis above by conducting an

analysis of moderating variables. The moderator variable analysis for arrest is limited. The

primary limitation is lack of variation in moderating variables. Due to the consistency of study

variables there is little variation in the variables found in studies. As a result, effect sizes used in

the moderating variable analysis are largely lumped into one category and very few in the others. 16 Where applicable fail-safe N values are reported with outliers removed.

88

A second limitation is there were few mean effect sizes with enough k values for meaningful

moderator analysis. Using a cutoff of a k value of 15, the variables that could be further

analyzed were: suspect race, suspect gender, suspect ethnicity, suspect intoxication, suspect

weapon use, and suspect demeanor. Due to a technical issue with the STATA software, too

many cases had to be removed in order to analyze suspect age. As a result, that analysis was not

conducted.

Tables 5.4-5.9 below display the results of the moderating variable analysis. Across all

moderator analyses, variables that were significant in the mean effect size analysis remained

significant across moderator categories.17 Table 5.4 displays the full results of the suspect race

moderator analysis. Effect sizes generated by authors that were categorized as “other” discipline

did not report significant race effect sizes. Studies conducted in the South did not report

significant race effect sizes, but all other geographic regions did report significant race effect

sizes. The mean effect size for studies using data gathered in the 1980s did not report a

significant race effect. However, it should be noted that the k value for this finding is small.

Across other moderating variables, race remained significant. There were only a handful of non-

significant categories and the majority of these were based on a small number of effect sizes, one

to three, which made these results less meaningful.

The findings for the moderator analysis of gender effect sizes are similar to the findings

for race in that, across moderator categories, mean effect sizes remained significant. The full

results are displayed in Table 5.5. Like race, mean effect sizes for studies conducted by those in

the “other” discipline category were not significant. Upon closer examination of when studies

were published, only those published between 2000 and 2009 produced significant mean effect

sizes for gender. However, studies based on data gathered in the 1960s, 1970s, and 1980s did 17 The findings discuss those results that are contrary to the general finding.

89

not produce significant mean effect sizes. Only two studies were based on data from the 1960s

and 1980s, and six studies used data from the 1970s. Finally, the remaining moderator

categories either produced mean effect sizes for gender that remained significant or those that

were not significant were based on a negligible number of studies.

Table 5.6 displays the full results for the ethnicity moderating variable analysis. Relative

to race and gender there are significantly fewer effect sizes making up the suspect ethnicity

moderator analysis. The inclusion of Hispanic status as an empirical variable is still a relatively

modern addition. The majority of studies analyzed here were published between 2000 and 2009

and a few in the 1990s. Nevertheless, there were a sufficient number of effect sizes to conduct a

moderating variable analysis.

The pattern found in these results is similar to the previous moderator analyses for

individual suspect characteristics. Specifically, the majority of mean effect sizes across

categories are significant and those that are not are often limited to one or two effect sizes to

generate the mean effect size. Significant mean effect sizes were found among those that were

conducted by individuals in criminal justice or mixed disciplines. Studies conducted in the

Northeast and the Southwest produced significant mean effect sizes for ethnicity. Hispanics

were significantly more likely to be arrested according to studies conducted between 2000 and

2009. However, it should be noted that the smaller number of studies spread over more

categories creates a small number of effect sizes generating the mean effect size in this finding.

Only state/local reports reported significantly more Hispanics being arrested than non-Hispanics.

Studies that included a measure of age found that Hispanic status increased arrest.

Two seriousness measures had a sufficient number of effect sizes to perform a

meaningful moderator analysis, suspect intoxication and suspect weapon use. Table 5.7 displays

90

the full results for the moderating variable analysis for suspect intoxication. Consistent with

previous moderator analyses, mean effect sizes across moderator categories are generally

significant. Those mean effect sizes that were not significant are those that were created from

one or two effect sizes. One exception is that mean effect sizes generated from studies that did

not measure gender and those that did not measure age did not produce significant mean effect

sizes for suspect intoxication. Table 5.8 displays the full results for suspect weapon use. For this

analysis all mean effect sizes were significant across moderator categories except for a handful

of categories that had one to three effect sizes to create the mean effect size.

The final variable examined for moderator variable analysis was suspect demeanor. The

full results for this analysis are displayed in Table 5.9. The results here are similar to all of the

previous moderator results for the arrest decision. Specifically, mean effect sizes are generally

significant across moderator categories and those that are not significant are often generated

from one to three effect sizes. There were some unique findings. The mean effect size for

studies conducted by sociologists did not find demeanor significant. Studies conducted in the

Northeast and Midwest did not find demeanor significantly predicted arrest. Studies based on

data from the 1980s did not find demeanor significantly predicted arrest. Studies that did not

measure intoxication did not produce a significant mean effect size for demeanor. Finally, both

studies that did and those that did not record a measure of interaction phase crime produced

significant mean effect sizes for demeanor and arrest.

91

Table 5.4: Suspect Race Moderator Analysis Moderator Variable k ES 95% CI Min 95% CI Max SE Discipline Criminal Justice/Criminology 26 0.34 0.22 0.45 0.06 Psychology 0 0.00 0.00 0.00 0.00 Sociology 9 0.52 0.26 0.77 0.13 Social Work 0 0.00 0.00 0.00 0.00 Mixed 3 0.35 0.07 0.62 0.35 Others 6 0.12 -0.04 0.28 0.08 Publication Decade 1970 1 0.63 0.12 1.14 0.26 1980 1 0.71 -0.59 2.02 0.66 1990 15 0.34 0.09 0.59 0.13 2000 28 0.30 0.19 0.42 0.06 Geographic Region Northeast 6 0.27 0.05 0.48 0.11 South 7 -0.06 -0.47 0.35 0.21 Midwest 14 0.49 0.27 0.71 0.11 Southwest 7 0.43 0.25 0.61 0.09 Pacific/Northwest 3 0.51 0.22 0.79 0.15 Mixed 8 0.15 0.00 0.29 0.07 Data Gathered 1960 2 0.41 0.07 0.75 0.17 1970 6 0.52 0.09 0.96 0.22 1980 4 0.31 -0.13 0.75 0.23 1990 16 0.43 0.21 0.64 0.11 2000 17 0.25 0.12 0.39 0.07 Publication Type Journal 30 0.23 0.11 0.36 0.06 Graduate Dissertation/Thesis 9 0.33 0.15 0.50 0.09 State or Local Report 6 0.46 0.27 0.64 0.10

92

Table 5.4 continued Moderator Variable k ES 95% CI Min 95% CI Max SE Agency Number 1 28 0.39 0.26 0.51 0.06 2 1 0.21 -0.26 0.68 0.24 3 1 0.63 0.19 1.08 0.23 4 1 0.53 -0.29 1.35 0.44 8 1 0.31 -0.31 0.92 0.32 9 3 1.33 0.51 2.14 0.42 20 1 0.94 0.27 2.16 0.62 24 1 0.45 -0.08 0.98 0.27 28 1 0.71 -0.57 1.99 0.65 Offense Specific No 16 0.45 0.26 0.64 0.10 Yes 29 0.16 0.16 0.39 0.06 Control Variables No 4 0.52 0.08 0.97 0.23 Yes 41 0.32 0.21 0.42 0.05 SSO Use No 32 0.29 0.18 0.40 0.06 Yes 13 0.49 0.24 0.74 0.13 Interaction Effect Use No 37 0.35 0.23 0.46 0.06 Yes 8 0.25 0.00 0.51 0.13 Continuous Data Collection No 1 0.21 -0.35 0.77 0.29 Yes 44 0.34 0.22 0.45 0.06 Offender Gender Recorded No 7 0.42 0.05 0.79 0.19 Yes 38 0.32 0.21 0.43 0.06 Offender Age Recorded No 11 0.26 0.07 0.45 0.10 Yes 34 0.36 0.23 0.49 0.07

93

Table 5.4 continued Moderator Variable k ES 95% CI Min 95% CI Max SE Offender Ethnicity Recorded No 30 0.24 0.11 0.36 0.06 Yes 15 0.42 0.27 0.56 0.07 Officer Race Recorded No 34 0.38 0.25 0.51 0.07 Yes 11 0.23 0.58 0.40 0.09 Officer Gender Recorded No 36 0.35 0.21 0.49 0.07 Yes 9 0.30 0.10 0.51 0.11 Seriousness Recorded No 30 0.32 0.20 0.44 0.06 Yes 15 0.33 0.15 0.52 0.09 Evidence Recorded No 31 0.27 0.15 0.38 0.06 Yes 14 0.44 0.26 0.61 0.09 Intoxication Recorded No 32 0.35 0.21 0.48 0.07 Yes 13 0.31 0.06 0.56 0.13 Weapon Use Recorded No 35 0.36 0.23 0.50 0.07 Yes 10 0.26 0.01 0.50 0.13 Victim Preference Recorded No 35 0.30 0.19 0.42 0.06 Yes 10 0.43 0.18 0.69 0.13 Demeanor Recorded No 30 0.27 0.16 0.39 0.06 Yes 15 0.49 0.27 0.70 0.11

94

Table 5.4 Continued Moderator Variable k ES 95% CI Min 95% CI Max SE Interaction Phase Crime Recorded No 35 0.34 0.22 0.45 0.06 Yes 10 0.29 0.03 0.55 0.13 Compliance Recorded No 37 0.31 0.20 0.42 0.06 Yes 8 0.50 0.17 0.83 0.17 Organization Recorded No 41 0.35 0.28 0.43 0.04 Yes 3 -0.10 -0.23 0.04 0.07 Community Recorded No 41 0.33 0.22 0.44 0.06 Yes 4 0.32 0.02 0.62 0.15

95

Table 5.5: Suspect Gender Moderator Analysis Moderator Variable k ES 95% CI Min 95% CI Max SE Discipline Criminal Justice/Criminology 23 0.42 0.28 0.55 0.07 Psychology 0 0.00 0.00 0.00 0.00 Sociology 7 0.38 0.10 0.66 0.14 Social Work 0 0.00 0.00 0.00 0.00 Mixed 3 0.80 0.46 1.13 0.80 Others 5 0.15 -0.08 0.37 0.15 Publication Decade 1970 1 0.53 -0.09 1.15 0.32 1980 1 0.53 -0.78 1.85 0.67 1990 13 0.22 -0.03 0.47 0.13 2000 23 0.45 0.31 0.59 0.07 Geographic Region Northeast 3 0.53 0.20 0.87 0.17 South 5 0.65 0.17 1.14 0.25 Midwest 13 0.29 0.04 0.55 0.13 Southwest 7 0.56 0.32 0.80 0.12 Pacific/Northwest 2 0.50 0.13 0.86 0.19 Mixed 8 0.24 0.04 0.44 0.10 Data Gathered 1960 2 0.38 -0.05 0.80 0.22 1970 6 0.29 -0.17 0.75 0.24 1980 2 -0.03 -0.54 0.48 0.26 1990 13 0.42 0.19 0.66 0.12 2000 15 0.46 0.29 0.64 0.09 Publication Type Journal 26 0.28 0.13 0.42 0.08 Graduate Dissertation/Thesis 6 0.56 0.32 0.80 0.12 State or Local Report 6 0.54 0.31 0.77 0.12

96

Table 5.5 continued Moderator Variable k ES 95% CI Min 95% CI Max SE Agency Number 1 22 0.56 0.42 0.71 0.07 2 1 0.23 -0.27 0.73 0.25 3 1 0.53 0.01 1.04 0.26 8 1 0.93 0.33 1.52 0.30 9 3 -0.19 -0.74 0.37 0.28 20 1 0.23 -1.36 1.83 0.81 24 1 0.40 -0.24 1.04 0.22 28 1 0.53 -0.74 1.80 0.65 Offense Specific No 14 0.36 0.15 0.57 0.11 Yes 24 0.42 0.28 0.57 0.07 Control Variables No 3 0.50 -0.02 1.01 0.26 Yes 35 0.40 0.28 0.52 0.06 SSO Use No 27 0.40 0.28 0.53 0.06 Yes 11 0.39 0.12 0.67 0.14 Interaction Effect Use No 31 0.43 0.30 0.57 0.07 Yes 7 0.29 0.01 0.56 0.14 Continuous Data Collection No 1 0.23 -0.36 0.82 0.30 Yes 37 0.41 0.29 0.53 0.06 Offender Age Recorded No 8 0.28 0.08 0.48 0.10 Yes 30 0.45 0.32 0.57 0.07 Offender Ethnicity Recorded No 26 0.30 0.15 0.44 0.07 Yes 12 0.53 0.37 0.70 0.08

97

Table 5.5 continued Moderator Variable k ES 95% CI Min 95% CI Max SE Officer Race Recorded No 27 0.41 0.27 0.56 0.07 Yes 11 0.38 0.20 0.56 0.09 Officer Gender Recorded No 30 0.41 0.27 0.56 0.07 Yes 8 0.38 0.17 0.60 0.11 Seriousness Recorded No 26 0.39 0.25 0.53 0.07 Yes 12 0.44 0.21 0.66 0.11 Evidence Recorded No 26 0.29 0.16 0.42 0.07 Yes 12 0.60 0.42 0.79 0.09 Intoxication Recorded No 27 0.40 0.27 0.54 0.07 Yes 11 0.41 0.16 0.65 0.12 Weapon Use Recorded No 31 0.43 0.30 0.56 0.07 Yes 7 0.31 0.06 0.56 0.13 Victim Preference Recorded No 31 0.38 0.26 0.51 0.06 Yes 7 0.50 0.21 0.80 0.15 Demeanor Recorded No 24 0.41 0.27 0.54 0.07 Yes 14 0.39 0.17 0.61 0.11 Interaction Phase Crime Recorded No 29 0.40 0.28 0.53 0.06 Yes 9 0.41 0.11 0.71 0.15

98

Table 5.5 continued Moderator Variable k ES 95% CI Min 95% CI Max SE Compliance Recorded No 30 0.41 0.29 0.54 0.06 Yes 8 0.33 0.02 0.64 0.16 Organization Recorded No 34 0.43 0.33 0.54 0.05 Yes 3 0.00 -0.22 0.22 0.11 Community Recorded No 35 0.39 0.27 0.51 0.06 Yes 3 0.54 0.17 0.91 0.19

99

Table 5.6: Suspect Ethnicity Moderator Analysis Moderator Variable k ES 95% CI Min 95% CI Max SE Discipline Criminal Justice/Criminology 8 0.22 0.05 0.40 0.09 Psychology 0 0.00 0.00 0.00 0.00 Sociology 4 0.18 -0.20 0.56 0.19 Social Work 0 0.00 0.00 0.00 0.00 Mixed 2 0.49 0.10 0.87 0.19 Others 1 0.25 -0.19 0.69 0.27 Publication Decade 1990 3 0.30 -0.84 1.44 0.58 2000 12 0.25 0.12 0.39 0.07 Geographic Region Northeast 3 0.49 0.20 0.79 0.15 South 2 -0.10 -0.37 0.17 0.14 Midwest 2 0.21 -0.01 0.44 0.11 Southwest 4 0.35 0.17 0.54 0.05 Pacific/Northwest 3 0.21 -0.01 0.44 0.11 Mixed 1 0.11 -0.43 0.65 0.28 Data Gathered 1980 3 0.30 -0.84 1.43 0.58 1990 2 0.28 -0.16 0.72 0.22 2000 10 0.25 0.10 0.40 0.08 Publication Type Journal 4 0.07 -0.15 0.28 0.11 Graduate Dissertation/Thesis 5 0.17 -0.07 0.41 0.12 State or Local Report 6 0.38 0.22 0.53 0.08 Agency Number 1 12 0.25 0.11 0.40 0.07 4 1 0.44 -0.18 1.05 0.31

100

Table 5.6 continued Moderator Variable k ES 95% CI Min 95% CI Max SE Offense Specific No 4 0.15 -0.39 0.70 0.28 Yes 11 0.26 0.12 0.40 0.07 SSO Use No 13 0.25 0.12 0.39 0.07 Yes 2 0.30 -0.84 1.44 0.58 Interaction Effect Use No 11 0.26 0.10 0.42 0.08 Yes 4 0.23 -0.09 0.55 0.16 Offender Gender Recorded No 3 0.31 -0.04 0.67 0.18 Yes 12 0.24 0.09 0.40 0.08 Offender Age Recorded No 5 0.23 -0.02 0.48 0.13 Yes 10 0.27 0.10 0.44 0.09 Officer Race Recorded No 10 0.22 0.05 0.39 0.09 Yes 5 0.31 0.03 0.53 0.11 Officer Gender Recorded No 10 0.21 0.03 0.38 0.09 Yes 5 0.31 0.11 0.52 0.10 Seriousness Recorded No 12 0.24 0.10 0.38 0.07 Yes 3 0.43 -0.14 1.00 0.29 Evidence Recorded No 8 0.22 0.00 0.43 0.11 Yes 7 0.28 0.09 0.47 0.10

101

Table 5.6 continued Moderator Variable k ES 95% CI Min 95% CI Max SE Weapon Use Recorded No 14 0.24 0.11 0.38 0.07 Yes 1 0.44 -0.17 1.05 0.31 Victim Preference Recorded No 14 0.24 0.11 0.38 0.07 Yes 1 0.44 -0.17 1.05 0.31 Demeanor Recorded No 12 0.27 0.12 0.42 0.08 Yes 3 0.15 -0.26 0.57 0.21 Interaction Phase Crime Recorded No 13 0.26 0.11 0.41 0.08 Yes 2 0.25 -0.19 0.68 0.22 Community Recorded No 14 0.22 0.09 0.35 0.07 Yes 1 0.57 0.13 1.01 0.23

102

Table 5.7: Suspect Intoxication Moderator Analysis Moderator Variable k ES 95% CI Min 95% CI Max SE Discipline Criminal Justice/Criminology 13 0.41 0.18 0.64 0.12 Psychology 0 0.00 0.00 0.00 0.00 Sociology 2 0.76 0.12 1.40 0.33 Social Work 0 0.00 0.00 0.00 0.00 Mixed 1 0.48 -0.21 1.16 0.35 Others 1 0.43 -0.16 1.01 0.30 Publication Decade 1980 1 0.40 -1.06 1.86 0.74 1990 4 0.68 0.26 1.10 0.21 2000 12 0.38 0.20 0.56 0.09 Geographic Region Northeast 1 -0.26 -0.62 0.10 0.18 South 6 0.52 0.19 0.85 0.17 Midwest 4 0.28 0.07 0.49 0.11 Southwest 2 0.57 0.16 0.98 0.21 Pacific/Northwest 1 0.62 0.18 1.07 0.23 Mixed 3 0.52 0.34 0.70 0.09 Data Gathered 1970 3 0.78 0.23 1.34 0.28 1980 0 0.00 0.00 0.00 0.00 1990 12 0.48 0.27 0.70 0.11 2000 2 0.12 -0.27 0.52 0.20 Publication Type Journal 15 0.48 0.31 0.65 0.08 Graduate Dissertation/Thesis 2 0.12 -0.24 0.49 0.19

103

Table 5.7 continued Moderator Variable k ES 95% CI Min 95% CI Max SE Agency Number 1 9 0.38 0.10 0.58 0.11 2 1 0.59 0.17 1.00 0.21 3 0 0.00 0.00 0.00 0.00 4 1 0.62 0.13 1.12 0.25 8 1 0.48 -0.01 0.96 0.25 9 0 0.00 0.00 0.00 0.00 20 1 1.44 0.08 2.80 0.69 21 1 -0.26 -0.68 0.16 0.22 24 1 0.74 0.24 1.24 0.25 28 1 0.40 -1.02 1.82 0.73 Offense Specific No 5 0.63 0.36 0.90 0.14 Yes 12 0.32 0.13 0.51 0.10 Control Variables No 2 1.08 0.13 2.02 0.48 Yes 15 0.41 0.24 0.57 0.08 SSO Use No 11 0.30 0.11 0.50 0.10 Yes 6 0.65 0.38 0.91 0.13 Interaction Effect Use No 15 0.39 0.23 0.56 0.09 Yes 2 0.74 0.22 1.27 0.27 Continuous Data Collection No 1 0.59 0.06 1.12 0.27 Yes 16 0.41 0.24 0.59 0.09 Offender Race Recorded No 3 0.01 -0.35 0.37 0.18 Yes 17 0.48 0.33 0.62 0.08

104

Table 5.7 continued Moderator Variable k ES 95% CI Min 95% CI Max SE Offender Gender Recorded No 6 0.24 -0.05 0.53 0.15 Yes 11 0.50 0.31 0.69 0.09 Offender Age Recorded No 6 0.25 -0.05 0.55 0.15 Yes 11 0.49 0.30 0.67 0.09 Offender Ethnicity Recorded No 16 0.41 0.24 0.58 0.09 Yes 1 0.62 0.03 1.22 0.30 Officer Race Recorded No 14 0.47 0.26 0.68 0.11 Yes 3 0.37 0.05 0.68 0.16 Officer Gender Recorded No 12 0.61 0.38 0.84 0.12 Yes 5 0.27 0.07 0.48 0.10 Seriousness Recorded No 9 0.52 0.25 0.79 0.14 Yes 8 0.37 0.16 0.59 0.11 Evidence Recorded No 10 0.40 0.16 0.63 0.12 Yes 7 0.49 0.22 0.76 0.14 Weapon Use Recorded No 8 0.27 0.04 0.50 0.12 Yes 9 0.52 0.33 0.71 0.10 Victim Preference Recorded No 6 0.23 0.01 0.46 0.12 Yes 11 0.58 0.37 0.78 0.11

105

Table 5.7 continued Moderator Variable k ES 95% CI Min 95% CI Max SE Demeanor Recorded No 9 0.33 0.09 0.58 0.13 Yes 8 0.55 0.24 0.81 0.13 Interaction Phase Crime Recorded No 12 0.32 0.14 0.51 0.09 Yes 5 0.64 0.36 0.91 0.14 Compliance Recorded No 13 0.45 0.24 0.66 0.11 Yes 4 0.41 0.10 0.72 0.16 Organization Recorded No 15 0.52 0.32 0.73 0.10 Yes 2 0.12 -0.28 0.52 0.20 Community Recorded No 15 0.42 0.23 0.61 0.10 Yes 2 0.49 0.11 0.87 0.19

106

Table 5.8: Suspect Weapon Use Moderator Analysis Moderator Variable k ES 95% CI Min 95% CI Max SE Discipline Criminal Justice/Criminology 11 0.56 0.34 0.78 0.11 Psychology 0 0.00 0.00 0.00 0.00 Sociology 0 0.00 0.00 0.00 0.00 Social Work 1 0.27 -0.33 0.88 0.31 Mixed 1 0.34 -0.99 1.66 0.68 Others 2 0.35 -0.09 0.78 0.22 Publication Decade 1990 5 0.55 0.27 0.83 0.14 2000 10 0.44 0.20 0.67 0.12 Geographic Region Northeast 2 0.41 0.07 0.75 0.17 South 5 0.73 0.23 1.23 0.26 Midwest 2 0.71 0.32 1.10 0.20 Southwest 1 0.34 -0.96 1.63 0.66 Pacific/Northwest 0 0.00 0.00 0.00 0.00 Mixed 6 0.35 0.11 0.59 0.12 Data Gathered 1970 1 1.16 0.39 1.93 0.39 1980 3 0.59 0.24 0.95 0.18 1990 9 0.42 0.21 0.63 0.11 2000 2 0.30 -0.04 0.65 0.18 Publication Type Journal 13 0.47 0.30 0.65 0.09 Graduate Dissertation/Thesis 2 0.44 -0.04 0.92 0.24 Agency Number 1 7 0.80 0.46 1.15 0.18 2 1 0.02 -0.91 0.95 0.47 8 1 0.34 -0.95 1.67 0.66 9 1 0.64 0.19 1.09 0.23 24 2 0.43 0.10 0.77 0.17

107

Table 5.8 continued Moderator Variable k ES 95% CI Min 95% CI Max SE Offense Specific No 3 0.66 0.33 1.00 0.17 Yes 12 0.39 0.23 0.55 0.08 SSO Use No 13 0.44 0.28 0.61 0.08 Yes 2 0.70 0.11 1.29 0.30 Interaction Effect Use No 14 0.41 0.27 0.56 0.07 Yes 1 1.16 0.41 1.91 0.38 Continuous Data Collection No 1 0.02 -0.92 0.04 0.48 Yes 14 0.48 0.31 0.64 0.08 Offender Race Recorded No 6 0.48 0.20 0.77 0.15 Yes 9 0.49 0.25 0.73 0.12 Offender Gender Recorded No 8 0.50 0.23 0.77 0.14 Yes 7 0.47 0.22 0.72 0.13 Offender Age Recorded No 5 0.55 0.23 0.88 0.17 Yes 10 0.46 0.23 0.68 0.11 Officer Race Recorded No 14 0.48 0.31 0.64 0.08 Yes 1 0.02 -0.92 0.96 0.48 Officer Gender Recorded No 13 0.54 0.36 0.72 0.09 Yes 2 0.18 -0.18 0.54 0.19

108

Table 5.8 continued Moderator Variable k ES 95% CI Min 95% CI Max SE Seriousness Recorded No 12 0.40 0.25 0.55 0.07 Yes 3 0.96 0.43 1.49 0.27 Evidence Recorded No 12 0.48 0.31 0.65 0.09 Yes 3 0.35 -0.27 0.97 0.32 Intoxication Recorded No 7 0.50 0.28 0.72 0.11 Yes 8 0.44 0.14 0.74 0.15 Victim Preference Recorded No 8 0.38 0.23 0.54 0.08 Yes 7 0.73 0.38 1.07 0.18 Demeanor Recorded No 13 0.44 0.28 0.61 0.09 Yes 2 0.70 0.11 1.29 0.30 Interaction Phase Crime Recorded No 11 0.41 0.25 0.57 0.08 Yes 4 0.73 0.31 1.14 0.21 Compliance Recorded No 14 0.48 0.31 0.64 0.08 Yes 1 0.02 -0.92 0.96 0.48 Organization Recorded No 14 0.52 0.34 0.71 0.09 Yes 1 0.21 -0.20 0.61 0.21 Community Recorded No 13 0.45 0.29 0.62 0.09 Yes 2 0.57 0.02 1.12 0.28

109

Table 5.9: Suspect Demeanor Moderator Analysis Moderator Variable k ES 95% CI Min 95% CI Max SE Discipline Criminal Justice/Criminology 10 0.70 0.35 1.06 0.18 Psychology 0 0.00 0.00 0.00 0.00 Sociology 5 0.37 -0.34 1.08 0.36 Social Work 0 0.00 0.00 0.00 0.00 Mixed 0 0.00 0.00 0.00 0.00 Others 1 1.03 0.11 1.95 0.47 Publication Decade 1970 1 1.03 0.08 0.20 0.49 1980 0 0.00 0.00 0.00 0.00 1990 9 0.52 0.03 1.01 0.25 2000 6 0.73 0.30 1.17 0.22 Geographic Region Northeast 1 0.01 -0.73 0.74 0.38 South 4 0.62 0.02 0.12 0.31 Midwest 6 0.52 -0.01 1.06 0.27 Southwest 2 1.00 0.28 1.72 0.37 Pacific/Northwest 0 0.00 0.00 0.00 0.00 Mixed 3 0.90 0.43 1.37 0.24 Data Gathered 1960 1 1.03 0.32 1.74 0.36 1970 5 1.09 0.45 1.73 0.33 1980 3 0.07 -0.43 0.56 0.25 1990 6 0.67 0.28 1.06 0.20 2000 1 0.89 0.08 1.70 0.41 Publication Type Journal 13 0.74 0.30 1.19 0.23 Graduate Dissertation/Thesis 2 0.51 -0.35 1.36 0.44 State or Local Report 1 0.89 -0.39 2.17 0.65

110

Table 5.9 continued Moderator Variable k ES 95% CI Min 95% CI Max SE Agency Number 1 12 0.48 0.18 0.79 0.16 2 1 0.49 -0.27 1.25 0.39 3 1 1.03 0.32 1.74 0.36 20 1 1.26 -0.02 2.55 0.65 24 1 1.17 0.39 1.95 0.40 Offense Specific No 10 0.70 0.24 1.16 0.24 Yes 6 0.74 0.05 1.44 0.36 Control Variables No 3 1.09 0.44 1.75 0.33 Yes 13 0.57 0.26 0.88 0.16 SSO Use No 3 0.61 -0.21 1.43 0.42 Yes 13 0.74 0.30 1.18 0.22 Interaction Effect Use No 11 0.77 0.37 1.18 0.21 Yes 5 0.52 -0.09 1.12 0.31 Continuous Data Collection No 1 0.49 -0.63 1.61 0.57 Yes 15 0.72 0.36 1.08 0.18 Offender Race Recorded No 1 0.20 -1.00 1.40 0.61 Yes 15 0.74 0.39 1.09 0.18 Offender Gender Recorded No 2 0.08 -0.57 0.74 0.34 Yes 14 0.81 0.49 1.14 0.16

111

Table 5.9 continued Moderator Variable k ES 95% CI Min 95% CI Max SE Offender Age Recorded No 3 0.71 -0.11 1.54 0.42 Yes 13 0.69 0.32 1.06 0.19 Offender Ethnicity Recorded No 13 0.80 0.44 1.17 0.19 Yes 3 0.33 -0.31 0.96 0.33 Officer Race Recorded No 13 0.80 0.35 1.26 0.23 Yes 3 0.46 -0.31 1.23 0.39 Officer Gender Recorded No 13 0.80 0.35 1.26 0.23 Yes 3 0.46 -0.31 1.23 0.39 Seriousness Recorded No 7 0.70 0.17 1.24 0.27 Yes 9 0.69 0.26 1.11 0.22 Evidence Recorded No 10 0.65 0.13 1.17 0.26 Yes 6 0.78 0.22 1.34 0.29 Intoxication Recorded No 8 0.47 -0.11 0.96 0.27 Yes 8 0.96 0.45 1.47 0.26 Weapon Use Recorded No 13 0.71 0.33 1.09 0.19 Yes 3 0.64 0.00 1.27 0.32 Victim Preference Recorded No 9 0.28 0.01 0.56 0.14 Yes 7 0.97 0.68 1.26 0.15

112

Table 5.9 continued Moderator Variable k ES 95% CI Min 95% CI Max SE Interaction Phase Crime Recorded No 7 0.71 0.27 1.14 0.22 Yes 9 0.66 0.18 1.14 0.24 Compliance Recorded No 9 0.77 0.29 1.25 0.25 Yes 7 0.62 -0.05 1.28 0.34 Community Recorded No 14 0.71 0.34 1.08 0.19 Yes 2 0.64 -0.19 1.46 0.42

113

Sentencing

Sample characteristics

Table 5.10 describes the sample characteristics for the sentencing study analysis. As

shown in Table 5.10, the average sample size contained 49,382.72 cases and across all 46

studies, 2,271,605 cases were studied. The mean number of court systems involved in each

study was 1.83. The mean length of the study period was 37.89 months. Approximately 30

percent of studies reported at least one analysis with interaction effects, and nearly all studies

used presented information on control variables and used continuous data collection, 95.70 and

87.00 percent respectively. Finally, approximately 33 percent of studies collected information

on offense specific cases.

Table 5.10 also describes what variables were collected in various studies. These

variables were separated into 4 groups: suspect characteristics, judge characteristics, seriousness

characteristics, and other characteristics. For suspects, the majority of studies collected

information on race, gender, and age of suspects, 84.80, 69.60, and 87.00 percent respectively.

Approximately 35 and 20 percent of studies reported information on suspect ethnicity, and

suspect employment.

Judge characteristics were not commonly recorded. Judge race was most common. Judge

race was recorded in 10.90 percent of studies. Judge gender was less common with only 6.40

percent of studies reporting information on the gender of the judge. Judicial experience was also

not a common variable. Only 6.5 percent of studies recorded judge experience measures.

Measures of seriousness generally took the form of two variables, seriousness of the

offense and prior record. Each was recorded in 91.30 and 95.70 percent of studies respectively.

Weapon use information was recorded in approximately 40 percent of studies, and drug use was

114

rarely recorded, with only 6.5 percent of studies measuring drug use. The average number of

seriousness measures was 2.00 per study. Overall, other court characteristics were also not

common. Federal court study, court organization measures, and community measures were each

recorded in approximately 6.5 percent of cases. Mode of conviction and attorney type measures

were more common, and were reported in 60.90 and 32.60 percent respectively of all studies.

115

Table 5.10: Sentencing Sample Characteristics Variable Mean S Total General Study Characteristics Number of Court Systems 1.83 3.97 Offense Specific 32.60 Control Variables 95.70 Interaction Effects 28.30 Length 37.89 32.65 Continuous Data Collection 87.00 Size 49,382.72 136,070.28 2,271,605 Defendant Characteristics Defendant Race Measure 84.80 Defendant Gender Measure 69.60 Defendant Age Measure 87.00 Defendant Ethnicity Measure 34.80 Defendant Mentally Ill Measure 0.00 Defendant Employment Measure 21.70 Judge Characteristics Judge Race Measure* 10.90 Judge Gender Measure 6.40 Experience 6.50 Seriousness Characteristics Seriousness Measure 91.30 Weapon Use Measure 41.30 Drug Use Measure 06.50 Prior Record Measure 95.70 Total Seriousness Measures 2.00 0.87 Other Court Characteristics Federal vs. Local 6.40 Mode of Conviction Measure 60.90 Attorney Type Measure 32.60 Court Organization Measures 6.50 Community Measures 6.50 N 46 *one study included information on ethnicity of the judge but it was included with race

116

Effect sizes

Table 5.11 presents the findings from the effect size analysis. Defendant race, gender

and ethnicity are each significant. Specifically, black defendants were 1.23 times more likely to

be incarcerated over white defendants, males were 1.63 times more likely to be incarcerated than

females and Hispanics were 1.29 times more likely to be incarcerated than non-Hispanics.

However, age and employment status were not significant predictors of incarceration.

Additionally, like arrest, judge characteristics were not significantly related to the outcome of

interest. It is noteworthy that the effect size for judge race approached significance, .073, and

that black judges were more lenient than white judges. This finding is the only criminal justice

actor variable that approaches significance.

Prior record had a stronger overall effect than seriousness of the offense. Specifically,

those with longer prior records were almost 2 times more likely to be incarcerated. Seriousness

of the offense increased the odds of incarceration by approximately 1.5 times. Drug use also

increased incarceration (1.45 times). With regard to other court characteristics, only mode of

conviction was significant. Those that opted for a plea were 1.29 times less likely to be

incarcerated. Attorney type and court size correlations were not significant but were in the

expected directions.

117

Table 5.11: Sentencing Mean Effect Sizes

Variable k N Mean ES W

Mean ES

95% CI Min

95% CI Max

Defendant Characteristics Suspect Black*** 39 2247871 0.17 0.21 0.11 0.30

(0.28) (0.27) (0.29)

Suspect Male*** 32 2243915 0.48 0.489 0.334 0.645

(0.29) (0.29) (0.31)

Suspect Age 39 2214935 -0.02 0.001 -0.004 0.002

(-0.002) (-0.002) (-0.001)

Suspect Ethnicity*** 16 1481682 0.22 0.25 0.16 0.34

(0.16) (0.13) (0.18)

Suspect Employment 10 20694 0.45 0.45 -0.17 1.07

(0.45 (-0.17) (1.07)

Judge Characteristics Judge Race 7 268988 -0.17 -0.16 -0.33 0.01

(-0.10) (-0.15) (-0.06)

Judge Gender 5 252702 0.09 0.14 -0.10 0.39

(0.002) (0.00) (0.004)

Judge Experience 4 200920 0.00 -0.001 -0.02 0.02

(0.01) (0.01) (0.01)

Seriousness Characteristics Seriousness*** 24 1991280 0.58 0.39 0.34 0.43

(0.02) (0.02) (0.02)

Without outliers*** 22 1438972 0.35 0.25 0.22 0.29 (0.02) (0.02) (0.02)

118

Table 5.11 continued

Variable k N Mean ES W

Mean ES

95% CI Min

95% CI Max

Prior Record*** 44 2214373 0.73 0.67 0.54 0.81

(0.23) (0.22) (0.24)

Without outliers*** 42 2193655 0.66 0.61 0.48 0.74

(0.22) (0.21) (0.23)

Drug Use* 4 17164 0.55 0.37 0.04 0.71

(0.01) (-0.001) (0.01)

Other Court/Case Characteristics Court Size 5 974747 -0.82 -0.83 -2.05 0.40

(-0.69) (-0.74) (-0.65)

Counsel Type 15 326882 -0.04 -0.02 -0.19 0.15

(-0.03) (-0.08) (0.01)

Mode of Conviction* 28 1292317 -0.08 -0.26 -0.46 -0.06

(-0.02) (-0.04) (0.01)

Without outliers* 26 905324 -0.11 -0.20 -0.38 -0.02 0.004 (-0.02) (0.03) (Fixed effects results in parentheses) * sig p< .05 ** sig p<.01

Fail-safe N analysis

Consistent with the fail-safe N results for the arrest analysis, the fail-safe N values for

sentencing are generally stable. Fail-safe N values for suspect race and gender were 47 and 70

respectively. Ethnicity was the least stable finding in the suspect characteristics domain with 29

additional studies needed to nullify the ethnicity effect size. Seriousness of the offense and prior

record were both very stable findings. Their fail-safe N values were 129 and 160. The effect

119

size for drug use produced the smallest fail-safe N value of any decision point; the value is less

than 1. Clearly, the results for drug use should be interpreted with caution. Finally, the fail-safe

N value for the mode of conviction effect size was 5. Like arrest, the decision-specific variables

produced the smallest fail-safe N values. Overall, these findings, like arrest, are stable and can

be viewed with confidence, except where noted. The mean fail-safe N value was 90 additional

effect sizes.

Moderating variable analysis

A series of moderating variable analyses were also performed for the sentencing decision.

Like the arrest analyses, these analyses were limited to a handful of variables that had a

sufficient number of effect sizes for meaningful analysis. The results for each analysis are

displayed in Tables 5.12-5.17.

Table 5.12 displays the results of the moderating variable analysis for defendant race.

There is a similar pattern for the moderator analysis of sentencing outcomes as the arrest

outcome. For defendant race, the majority of mean effect sizes were significant across

moderator categories and the analysis follows the same pattern that African Americans were

more severely punished. However, there are several other noteworthy findings. Race was not a

significant predictor of sentencing for studies that used multiple study locations. These multiple

location studies were often studies of sentencing at the federal level. Race was also not

significantly related to sentencing for studies that used data gathered in the 1970s or the 1980s.

Race was also not related to sentencing in offense specific studies or studies that did not use

continuous data. Race was not significantly related to sentencing in studies that measured

suspect employment or counsel type.

120

Table 5.13 displays the results of the moderating variable analysis for defendant gender.

Results are similar to previous moderator analyses. The majority of mean effect sizes are

significant across moderator categories. Those that are not significant are generally the result of

a small number of effect sizes used to generate the mean effect size. Males were not

significantly more likely to be sentenced to prison in studies conducted in the Northeast, South,

Pacific, or those using multiple locations or the federal level. Gender was not a significant

predictor of sentencing in studies based on data gathered in the 1970s or the 1980s. Like the

suspect race analysis, gender was not a significant predictor of sentencing in studies that

measured suspect employment.

Table 5.14 displays the results for defendant ethnicity moderator analysis. Unlike race

and gender, the ethnicity analysis is based on a much smaller number of effect sizes. The

majority of mean effect sizes were significant across moderator categories. Those that were not

were generally limited to only a few effect sizes to create the mean effect sizes. However, unlike

the previous two analyses, Hispanics were significantly more likely to be incarcerated only in

studies that had multiple locations or Federal court. There was little variation in the other

variable categories.

Table 5.15 displays the results for the moderator analysis for prior record. Generally,

patterns here were similar to previous moderator discussions. Prior record was not a significant

predictor of sentencing for studies published since 2010 and studies conducted in the Southwest.

It should be noted that these findings are based on a small number of effect sizes. All other mean

effect sizes were significant across moderator categories. Those that were not were based on a

negligible number of effect sizes.

121

The final two moderator analyses were conducted on attorney type and mode of

conviction. Table 5.16 displays the results for attorney type and Table 5.17 displays the results

for mode of conviction. Mean effect sizes for attorney type are generally not significant across

moderator categories. However the one exception was in the data gathered moderator. The

results show that studies conducted in the 1960s were significantly more likely to find that those

represented by private attorneys were more likely to be incarcerated. In the 1970s defendants

were significantly more likely to be incarcerated when represented by public defenders or court

appointed counsel. Mean effect sizes for mode of conviction were generally significant. Those

that were not significant were often based on a small number of effect sizes. Studies published

between 2000 and 2009, and studies based on data collected in the South, found that people that

pled guilty were significantly less likely to be sentenced to prison. Significant mean effect sizes

were found for studies based on data from the 1990s indicating those that took a plea were less

likely to be incarcerated. Pleas also produced significantly less prison sentences in studies that

were offense specific.

122

Table 5.12: Defendant Race Moderator Analysis Moderator Variable k ES 95% CI Min 95% CI Max SE Discipline Criminology/Criminal Justice 21 0.20 0.12 0.28 0.04 Psychology 0 0.00 0.00 0.00 0.00 Sociology 11 0.19 0.08 0.30 0.06 Social Work 0 0.00 0.00 0.00 0.00 Mixed 2 0.34 -0.04 0.54 0.15 Other 1 0.34 -0.01 0.68 0.17 Publication Decade 1980 6 0.28 0.07 0.48 0.10 1990 8 0.18 0.03 0.33 0.08 2000 21 0.16 0.08 0.25 0.04 2010 3 0.44 0.25 0.64 0.10 Geographic Region Northeast 4 0.36 0.20 0.52 0.07 South 7 0.19 0.06 0.32 0.07 Midwest 16 0.17 0.06 0.29 0.06 Southwest 3 0.34 0.17 0.51 0.09 Pacific/Northwest 1 0.48 0.17 0.79 0.16 Mixed 8 0.11 -0.01 0.23 0.06 Data Gathered 1960 2 0.54 0.08 0.99 0.23 1970 10 0.12 -0.09 0.34 0.11 1980 3 0.75 -0.24 0.39 0.16 1990 18 0.20 0.07 0.32 0.06 2000 6 0.31 0.11 0.51 0.10 Publication Type Journal 30 0.20 0.10 0.30 0.05 Graduate Dissertation/Thesis 7 0.22 0.03 0.41 0.10 State or Local Report 1 0.17 -0.28 0.62 0.23 Other 1 0.35 -0.10 0.80 0.23

123

Table 5.12: continued Moderator Variable k ES 95% CI Min 95% CI Max SE Agency Number 1 37 0.21 0.11 0.31 0.05 17 1 0.25 -0.27 0.77 0.27 23 1 0.18 -0.35 0.71 0.27 Offense Specific No 30 0.21 0.10 0.31 0.05 Yes 9 0.21 -0.03 0.46 0.13 Control Variables No 2 0.26 -0.31 0.66 0.20 Yes 36 0.21 0.12 0.31 0.05 Continuous Data Collection No 6 0.10 -0.14 0.35 0.13 Yes 33 0.23 0.12 0.33 0.05 Defendant Gender Recorded No 8 0.30 0.02 0.57 0.14 Yes 31 0.20 0.10 0.30 0.05 Defendant Age Recorded No 7 0.28 0.05 0.50 0.11 Yes 32 0.19 0.09 0.30 0.05 Defendant Ethnicity Recorded No 25 0.17 0.07 0.28 0.05 Yes 14 0.26 0.14 0.39 0.06 Defendant Employment Status Recorded No 32 0.23 0.13 0.33 0.05 Yes 7 0.06 -0.21 0.34 0.14 Seriousness Recorded No 4 0.37 0.06 0.68 0.16 Yes 35 0.19 0.08 0.29 0.05

124

Table 5.12: continued Moderator Variable k ES 95% CI Min 95% CI Max SE Prior Record Recorded No 2 0.35 -0.03 0.73 0.19 Yes 37 0.20 0.10 0.30 0.05 Mode of Conviction Recorded No 20 0.22 0.07 0.36 0.07 Yes 18 0.18 0.04 0.32 0.07 Counsel Type Recorded No 31 0.21 0.11 0.32 0.05 Yes 7 0.17 -0.03 0.37 0.10 Organization Recorded No 37 0.23 0.13 0.33 0.05

Yes 2 -0.13 -0.51 0.24 0.19

Community Recorded No 36 0.21 0.11 0.31 0.05 Yes 3 0.19 -0.11 0.50 0.15

125

Table 5.13: Defendant Gender Moderator Analysis Moderator Variable k ES 95% CI Min 95% CI Max SE Discipline Criminology/Criminal Justice 20 0.51 0.30 0.72 0.11 Psychology 0 0.00 0.00 0.00 0.00 Sociology 8 0.45 0.14 0.76 0.16 Social Work 0 0.00 0.00 0.00 0.00 Mixed 2 0.18 0.37 -0.54 0.90 Other 1 0.74 -0.19 1.68 0.48 Publication Decade 1980 1 0.13 -0.48 0.74 0.31 1990 6 0.46 0.17 0.75 0.15 2000 21 0.49 0.34 0.64 0.08 2010 3 0.66 0.29 1.02 0.19 Geographic Region Northeast 3 0.45 -0.07 0.98 0.27 South 6 0.37 -0.01 0.75 0.19 Midwest 12 0.66 0.38 0.94 0.14 Southwest 4 0.68 0.21 1.16 0.24 Pacific/Northwest 1 0.45 -0.43 1.33 0.45 Mixed 6 0.21 -0.16 0.58 0.19 Data Gathered 1970 4 0.30 -0.18 0.79 0.25 1980 4 0.46 -0.02 0.95 0.25 1990 18 0.52 0.30 0.73 0.11 2000 6 0.52 0.16 0.88 0.18 Publication Type Journal 23 0.50 0.33 0.67 0.09 Graduate Dissertation/Thesis 7 0.41 0.12 0.71 0.15 State or Local Report 1 0.71 -0.02 1.44 0.37 Other 1 0.47 -0.26 1.20 0.37

126

Table 5.13: continued Moderator Variable k ES 95% CI Min 95% CI Max SE Agency Number 1 30 0.49 0.32 0.65 0.08 17 1 0.53 -0.29 1.36 0.42 23 1 0.51 -0.32 1.34 0.42 Offense Specific No 29 0.19 0.33 0.66 0.08 Yes 3 0.45 -0.10 0.99 0.28 Continuous Data Collection No 6 0.43 0.07 0.79 0.19 Yes 26 0.50 0.33 0.68 0.09 Defendant Race Recorded No 1 0.77 -0.29 0.18 0.54 Yes 31 0.48 0.33 0.64 0.08 Defendant Age Recorded No 4 0.66 0.21 1.11 0.23 Yes 28 0.47 0.30 0.63 0.08 Defendant Ethnicity Recorded No 18 0.48 0.26 0.70 0.11 Yes 14 0.50 0.27 0.73 0.12 Defendant Employment Status Recorded No 27 0.50 0.33 0.66 0.08 Yes 5 0.43 -0.01 0.87 0.22 Seriousness Recorded No 3 0.51 -0.01 1.03 0.27 Yes 29 0.49 0.33 0.65 0.08 Prior Record Recorded No 2 0.19 -0.40 0.78 0.30 Yes 30 0.51 0.35 0.67 0.08

127

Table 5.13: continued Moderator Variable k ES 95% CI Min 95% CI Max SE Mode of Conviction Recorded No 14 0.50 0.25 0.75 0.13 Yes 16 0.46 0.24 0.68 0.11 Organization Recorded No 30 0.50 0.34 0.67 0.08 Yes 2 0.32 -0.25 0.90 0.30 Community Recorded No 29 0.50 0.34 0.67 0.09 Yes 3 0.37 -0.11 0.85 0.25

128

Table 5.14: Defendant Ethnicity Moderator Analysis Moderator Variable k ES 95% CI Min 95% CI Max SE Discipline Criminology/Criminal Justice 11 0.33 0.23 0.43 0.05 Psychology 0 0.00 0.00 0.00 0.00 Sociology 4 0.15 -0.01 0.30 0.08 Social Work 0 0.00 0.00 0.00 0.00 Mixed 0 0.00 0.00 0.00 0.00 Other 0 0.00 0.00 0.00 0.00 Publication Decade 1990 2 -0.14 -0.47 0.20 0.17 2000 11 0.32 0.21 0.43 0.06 2010 3 0.15 -0.07 0.37 0.17 Geographic Region Northeast 3 0.22 -0.02 0.46 0.12 South 2 0.26 -0.04 0.56 0.15 Midwest 3 0.28 .-.00946 0.57 0.15 Southwest 3 0.21 -0.03 0.44 0.12 Pacific/Northwest 0 0.00 0.00 0.00 0.00 Mixed 5 0.27 0.09 0.46 0.09 Data Gathered 1980 2 0.19 -0.15 0.53 0.17 1990 11 0.27 0.15 0.39 0.06 2000 3 0.19 -0.04 0.42 0.12 Publication Type Journal 14 0.26 0.15 0.37 0.06 Graduate Dissertation/Thesis 2 0.21 -0.03 0.46 0.13 Agency Number 1 14 0.24 0.14 0.35 0.05 17 1 0.27 -0.05 0.60 0.17 23 1 0.30 -0.04 0.64 0.17

129

Table 5.14: continued Moderator Variable k ES 95% CI Min 95% CI Max SE Continuous Data Collection No 2 0.29 0.06 0.52 0.12 Yes 14 0.24 0.14 0.34 0.05 Defendant Race Recorded No 1 0.22 -0.21 0.64 0.22 Yes 15 0.25 0.16 0.53 0.05 Defendant Gender Recorded No 1 0.09 -0.27 0.45 0.18 Yes 15 0.26 0.15 0.37 0.06 Defendant Age Recorded No 2 0.24 -0.02 0.50 0.13 Yes 14 0.25 0.14 0.37 0.06 Defendant Employment Status Recorded No 15 0.25 0.16 0.35 0.05 Yes 1 0.22 -0.21 63799.00 0.22 Seriousness Recorded No 1 0.27 -0.06 0.60 0.17 Yes 15 0.25 0.15 0.34 0.05 Prior Record Recorded No 2 0.32 0.05 0.60 0.14 Yes 14 0.24 0.15 0.34 0.05 Mode of Conviction Recorded No 5 0.28 0.12 0.45 0.08 Yes 11 0.23 0.10 0.36 0.07 Counsel Type Recorded No 13 0.23 0.13 0.32 0.05 Yes 3 0.41 0.16 0.65 0.13

130

Table 5.14: continued Moderator Variable k ES 95% CI Min 95% CI Max SE Organization Recorded No 14 0.27 0.17 0.36 0.05 Yes 2 0.11 -0.16 0.38 0.14 Community Recorded No 15 0.25 0.15 0.34 0.05 Yes 1 0.30 -0.03 0.63 0.17

131

Table 5.15: Prior Record Moderator Analysis Moderator Variable k ES 95% CI Min 95% CI Max SE Discipline Criminology/Criminal Justice 27 0.70 0.53 0.88 0.09 Psychology 0 0.00 0.00 0.00 0.00 Sociology 10 0.74 0.44 1.05 0.15 Social Work 0 0.00 0.00 0.00 0.00 Mixed 2 0.57 -0.17 1.30 0.38 Other 1 0.20 -0.69 1.08 0.45 Publication Decade 1980 5 0.72 0.24 1.21 0.25 1990 9 0.46 0.18 0.74 0.14 2000 26 0.75 0.59 0.91 0.08 2010 3 0.48 -0.06 1.02 0.28 Geographic Region Northeast 4 0.61 0.25 0.97 0.18 South 12 0.77 0.55 0.99 0.11 Midwest 15 0.75 0.54 0.96 0.11 Southwest 4 0.15 -0.26 0.56 0.21 Pacific/Northwest 1 2.14 1.43 2.85 0.36 Mixed 8 0.49 0.22 0.77 0.14 Data Gathered 1960 2 0.57 -0.15 1.28 0.37 1970 9 0.84 0.50 1.17 0.17 1980 4 0.30 -0.11 0.72 0.21 1990 24 0.72 0.55 0.90 0.09 2000 5 0.73 0.32 1.14 0.21 Publication Type Journal 30 0.63 0.50 0.76 0.07 Graduate Dissertation/Thesis 12 0.60 0.39 0.81 0.11 State or Local Report 1 1.90 .1.258 2.53 0.32 Other 1 1.32 0.68 1.96 0.33

132

Table 5.15: continued Moderator Variable k ES 95% CI Min 95% CI Max SE Agency Number 1 42 0.71 0.56 0.85 0.08 17 1 0.35 -0.53 1.23 0.45 23 1 0.64 -0.24 1.52 0.45 Offense Specific No 29 0.66 0.50 0.82 0.08 Yes 15 0.76 0.51 1.02 0.13 Control Variables No 2 0.55 -0.04 1.14 0.30 Yes 41 0.69 0.55 0.82 0.07 Continuous Data Collection No 6 0.84 0.46 1.21 0.19 Yes 38 0.67 0.52 0.82 0.08 Defendant Race Recorded No 7 0.54 0.22 0.86 0.16 Yes 37 0.72 0.58 0.87 0.07 Defendant Gender Recorded No 14 0.77 0.51 1.02 0.13 Yes 30 0.66 0.51 0.82 0.08 Defendant Age Recorded No 7 0.57 0.23 0.90 0.17 Yes 37 0.71 0.57 0.86 0.07 Defendant Ethnicity Recorded No 30 0.82 0.66 0.98 0.08 Yes 14 0.44 0.22 0.66 0.11 Defendant Employment Status Recorded No 36 0.69 0.55 0.83 0.07 Yes 8 0.68 0.37 1.00 0.16

133

Table 5.15: continued Moderator Variable k ES 95% CI Min 95% CI Max SE Seriousness Recorded No 4 0.54 0.07 1.01 0.24 Yes 40 0.71 0.56 0.85 0.07 Mode of Conviction Recorded No 14 0.46 0.25 0.66 0.10 Yes 30 0.77 0.62 0.91 0.07 Counsel Type Recorded No 30 0.65 0.49 0.81 0.08 Yes 14 0.74 0.49 1.00 0.13 Organization Recorded No 42 0.79 0.57 0.87 0.08 Yes 2 0.30 -0.34 0.93 0.32 Community Recorded No 42 0.67 0.54 0.81 0.07 Yes 2 0.98 0.41 1.54 0.29

134

Table 5.16: Attorney Type Moderator Analysis Moderator Variable k ES 95% CI Min 95% CI Max SE Discipline Criminology/Criminal Justice 7 0.10 -0.22 0.42 0.16 Psychology 0 0.00 0.00 0.00 0.00 Sociology 3 -0.52 -1.05 0.01 0.27 Social Work 0 0.00 0.00 0.00 0.00 Mixed 5 0.15 -0.44 0.73 0.30 Other 0 0.00 0.00 0.00 0.00 Publication Decade 1980 5 0.05 -0.36 0.45 0.21 1990 2 -0.23 -0.65 0.20 0.22 2000 8 0.01 -0.23 0.25 0.12 Geographic Region Northeast 1 0.48 0.12 0.83 0.18 South 5 0.15 -0.08 0.37 0.11 Midwest 6 -0.17 -0.41 0.07 0.12 Southwest 1 -1.18 -1.86 -0.50 0.34 Pacific/Northwest 0 0.00 0.00 0.00 0.00 Mixed 2 -0.08 -0.32 0.16 0.12 Data Gathered 1960 2 0.57 0.09 1.06 0.25 1970 5 -0.33 -0.64 -0.02 0.16 1980 0 0.00 0.00 0.00 0.00 1990 8 0.02 -0.20 0.23 0.11 Publication Type Journal 14 -0.03 -0.26 0.20 0.12 Graduate Dissertation/Thesis 1 -0.08 -0.79 0.63 0.36 Offense Specific No 9 -0.01 -0.25 0.23 0.12 Yes 6 -0.07 -0.41 0.28 0.18

135

Table 5.16: continued Moderator Variable k ES 95% CI Min 95% CI Max SE Control Variables No 1 0.48 0.06 0.89 0.21 Yes 14 -0.07 -0.22 0.08 0.08 Defendant Gender Recorded No 6 0.13 -0.17 0.43 0.15 Yes 9 -0.06 -0.24 0.11 0.09 Defendant Age Recorded No 4 0.19 -0.07 0.46 0.14 Yes 11 -0.10 -0.27 0.07 0.09 Defendant Ethnicity Recorded No 10 0.11 -0.14 0.36 0.13 Yes 5 -0.28 -0.62 0.05 0.17 Defendant Employment Status Recorded No 14 0.00 -0.17 0.18 0.09 Yes 1 -0.53 -1.36 0.30 0.42 Seriousness Recorded No 3 -0.01 -0.44 0.42 0.22 Yes 12 -0.04 -0.27 0.19 0.12 Mode of Conviction Recorded No 5 0.02 -0.29 0.34 0.16 Yes 10 -0.04 -0.25 0.17 0.11 Organization Recorded No 14 -0.03 -0.26 0.20 0.12 Yes 1 -0.08 -0.79 0.63 0.36

136

Table 5.17: Mode of Conviction Moderator Analysis Moderator Variable k ES 95% CI Min 95% CI Max SE Discipline Criminology/Criminal Justice 17 -0.49 -0.80 -0.17 0.16 Psychology 0 0.00 0.00 0.00 0.00 Sociology 7 0.22 -0.33 0.77 0.28 Social Work 0 0.00 0.00 0.00 0.00 Mixed 1 0.31 -1.50 2.11 0.92 Other 0 0.00 0.00 0.00 0.00 Publication Decade 1980 4 0.20 -0.43 0.83 0.32 1990 7 -0.24 -0.71 0.23 0.24 2000 15 -0.32 -0.58 -0.06 0.13 2010 2 -0.39 -1.12 0.33 0.37 Geographic Region Northeast 5 -0.07 -0.69 0.56 0.32 South 4 -1.23 -1.96 -0.51 0.37 Midwest 10 -0.28 -0.77 0.22 0.25 Southwest 1 0.11 -1.23 1.50 0.71 Pacific/Northwest 1 -0.86 -2.26 0.54 0.71 Mixed 7 0.61 -0.03 1.26 0.33 Data Gathered 1960 2 -0.13 -1.22 0.96 0.56 1970 6 0.07 -0.66 0.79 0.37 1980 3 -0.18 -0.97 0.62 0.41 1990 13 -0.44 -0.85 -0.03 0.21 2000 4 0.12 -0.58 0.83 0.36 Publication Type Journal 22 -0.19 -0.52 0.14 0.17 Graduate Dissertation/Thesis 5 -0.34 -0.99 0.31 0.33 State or Local Report 1 -0.04 -1.43 1.35 0.71 Agency Number 1 27 -0.24 -0.45 -0.04 0.11 23 1 -0.62 -1.61 0.37 0.51

137

Table 5.17: continued Moderator Variable k ES 95% CI Min 95% CI Max SE Offense Specific No 21 -0.28 -0.48 -0.08 0.10 Yes 7 -0.17 -0.67 0.34 0.26 Control Variables No 1 -0.44 -1.37 0.49 0.48 Yes 26 -0.24 -0.45 -0.03 0.11 Continuous Data Collection No 3 -0.51 -1.12 0.10 0.31 Yes 25 -0.23 -0.44 -0.01 0.11 Defendant Race Recorded No 1 0.11 -0.84 1.05 0.48 Yes 27 -0.27 -0.49 -0.06 0.11 Defendant Gender Recorded No 8 0.16 -0.39 0.71 0.28 Yes 20 -0.34 -0.65 -0.03 0.16 Defendant Age Recorded No 4 -0.14 -0.80 0.52 0.34 Yes 24 -0.24 -0.53 0.05 0.15 Defendant Ethnicity Recorded No 15 -0.25 -0.65 0.08 0.19 Yes 13 -0.18 -0.53 0.17 0.18 Defendant Employment Status Recorded No 25 -0.31 -0.53 -0.10 0.11 Yes 3 0.44 -0.32 1.20 0.39 Seriousness Recorded No 1 0.31 -1.30 1.92 0.82 Yes 27 -0.27 -0.47 -0.06 0.10

138

Table 5.17: continued Moderator Variable k ES 95% CI Min 95% CI Max SE Prior Record Recorded No 2 0.30 -0.40 0.99 0.36 Yes 26 -0.31 -0.52 -0.10 0.11 Counsel Type Recorded No 18 -0.21 -0.46 0.04 0.13 Yes 10 -0.35 -0.72 0.02 0.19 Organization Recorded No 24 -0.29 -0.52 -0.07 0.11 Yes 4 -0.05 -0.59 0.48 0.27 Community Recorded No 27 -0.24 -0.45 -0.04 0.11 Yes 1 -0.62 -1.61 0.37 0.51

Parole Revocation

Sample characteristics

Table 5.18 describes the sample characteristics for the parole revocation study analysis.

As shown in Table 5.18, the average sample size was 8,979.16 cases and across all 19 studies,

170,604 cases were studied. The mean number of parole revocation systems involved in each

study was 1.11 agencies. The mean length of study period was 15.33 months. Approximately 15

percent of studies reported at least one analysis with interaction effects, and nearly all studies

used continuous data collection (88.90%). Approximately six percent of studies collected

information on offense specific cases (5.60%).

Table 5.18 also describes what variables were collected in various studies. These

variables are separated into 2 groups: offender characteristics, and seriousness characteristics.

For suspects, the majority of studies collected information on race, gender, and employment

status of offenders, 84.20, 52.60, and 52.60 percent respectively. About 37 percent of studies

139

collected information on offender age. One quarter of studies recorded information on ethnicity

of offenders and 21 percent of studies recorded information on offender education. No studies

included in this analysis recorded measures of parole officer characteristics.

The most common seriousness measure was a risk/prior record measure. Eighty percent

of studies recorded a measure of risk (78.90%). Slightly more than half of studies recorded

information on the seriousness of the offense and roughly one quarter recorded information on a

need measure, 52.60 and 26.30 percent respectively. Very few studies (5.30%) recorded

information on weapon use. The average number of seriousness measures was 1.26 per study.

140

Table 5.18: Parole Revocation Sample Characteristics Variable Mean/% S Total General Study Characteristics Number of PR Systems 1.11 0.74 Offense Specific 5.60 Control Variables 63.20 Interaction Effects 15.80 Length 15.33 7.554 Continuous Data Collection 88.90 Size 8,979.16 20,856.20 170,604 Suspect Characteristics Suspect Race Measure 84.20 Suspect Gender Measure 52.60 Suspect Age Measure 36.80 Suspect Ethnicity Measure 26.00 Suspect Education Measure 21.10 Suspect Employment Measure 52.60 Seriousness Characteristics Seriousness Measure 52.60 Evidence Measure 0.00 Weapon Use Measure 05.30 Risk/Prior Record Measure 78.90 Need Measure 26.30 Total Serious Measures 1.26 0.45 N 19

141

Effect sizes

Table 5.19 presents the effect sizes for parole revocation. As shown in table 5.19, only

two variables are significantly related to parole revocation, seriousness and risk. Seriousness

increased the odds of revocation 1.18 times and those parolees with greater risk were 1.15 times

more likely to be revoked. No other measures were significantly related to parole revocation. It

is noteworthy that the effect size for offender race approached significance (p<0.051). This

finding that race was nearly significant is consistent with findings from the other two decisions

that indicated that race was a significant predictor of decision-making. Further, this race finding

may be an artifact of the small number of studies assessed.

Table 5.19: Mean Effect Sizes Parole Revocation

Variable k N Mean ES

W Mean

ES

95% CI Min

95% CI Max

Offender Characteristics Offender Black 16 157177 0.76 0.22 0.00 0.44

(0.19) (0.13) (0.25)

Without outliers 15 126541 0.15 0.15 0.00 0.31

(0.19) (0.13) (0.25)

Offender Male 10 149928 -0.39 -0.04 -0.36 0.29

(0.11) (0.06) (0.16)

Offender Age 7 21385 -0.16 -0.02 -0.05 0.01

(-0.02) (-0.03) (-0.01)

Offender Ethnicity 5 115738 -0.01 0.04 -0.13 0.21

(0.13) (0.09) (0.17)

Offender Employment 10 19689 -0.29 -0.29 -0.75 0.17 (-0.24) (-0.38) (-0.11)

142

Table 5.19 Continued

Variable k N Mean ES

W Mean

ES

95% CI 95% CI

Min Max

Offender Education 6 45652 -0.49 -0.28 -0.83 0.26 (-0.18) (-0.37) (0.01)

  Seriousness Characteristics Seriousness** 10 154527 -0.12 0.17 0.05 0.28

(0.20) (0.16) (0.25)

Risk/Prior Record*** 16 167528 0.63 0.14 0.09 0.20

(0.05) (0.05) (0.05)

Without Outliers*** 15 136892 0.38 0.13 0.08 0.19

(0.05) (0.05) (0.05)

Need** 4 1095 0.14 0.47 0.16 0.79 (0.49) (0.25) (0.73) (Fixed effects results in parentheses) * sig p< .05 ** sig p<.01 *** sig p< .001

Fail-safe N analysis

The results of the parole revocation fail-safe N analysis indicate that the results were less

stable than results of the other decisions. This is likely due to the smaller k values. Risk had the

largest fail-safe N value; 25 additional studies would be necessary to nullify the risk mean effect

size. There was no substantive difference between the risk fail-safe N analysis with or without

outliers. Seriousness and need had similar fail-safe N values, 4 and 2 respectively. The results

regarding these two variables should be used with caution given the small fail-safe N value. The

value for risk appears to be stable but the values for seriousness and need cannot be considered

stable. The mean fail-safe N value was 14 additional studies.

143

Moderating variable analysis

The moderator analysis for parole revocation is limited to offender race and risk effect

sizes. The limited analysis is due to the small number of studies in parole revocation generally

and offender race and risk were the only two variables that had a sufficient number of effect

sizes for meaningful moderator analysis. As with previous moderator variable analyses

discussed above, the general trends of significance mirror those at the mean effect size level.

Thus, across moderator categories, mean effect sizes for offender race are generally not

significant. However, there were some exceptions. The full results are found in Table 5.20.

Studies conducted by those working in either “sociology” or “social work” categories produced a

significant mean effect size. Those two groups were found that black individuals were more

likely to be revocated. Significant mean effect sizes were also found in studies that were

conducted in the South. Black individuals in the South were more likely to be revocated.

Studies that had a continuous data collection generated a significant mean effect size. Those

studies that coded both race and a measure of risk had a significant mean effect size. Finally,

studies that measured race as white/non-white produced a significant mean effect size. Like

Pratt (1998), the effects of race on parole revocation may be masked by measurement issues,

while those that measured race as white/black did not produce a significant mean effect size.

Risk was also measured in a sufficient number of studies to perform further analysis. As

with previous moderator analyses, the general pattern of significance follows the mean effect

size significance. In other words, the majority of mean effect sizes across moderator categories

were significant for the risk variable. Table 5.21 displays the full results for this analysis.

Studies based on data gathered between 2000 and 2009 did not produce a significant mean effect

size, but studies based on data collected in the 1980s and 1990s did produce a significant mean

144

effect size. All other categories of moderator variables produced significant effect sizes or were

based on one to two effect size values.

145

Table 5.20: Offender Race Moderator Analysis Moderator Variable k ES 95% CI Min 95% CI Max SE Discipline Criminal Justice/Criminology 6 -0.15 -0.56 0.26 0.21 Psychology 1 0.48 -0.27 1.24 0.38 Sociology 2 1.34 0.60 2.08 0.38 Social Work 1 0.99 0.05 1.92 0.48 Mixed 6 0.11 -0.22 0.44 0.17 Other 1 0.00 -1.53 1.53 0.78 Publication Decade 1970 1 0.48 -0.38 1.35 0.44 1980 1 0.99 -0.04 2.01 0.52 1990 3 0.07 -0.48 0.62 0.28 2000 5 0.15 -0.40 0.69 0.28 2010 7 0.21 -0.16 0.58 0.19 Geographic Region Northwest 0 0.00 0.00 0.00 0.00 South 3 0.43 0.12 0.74 0.16 Midwest 4 0.02 -0.25 0.30 0.14 Southwest 2 -0.66 -1.40 0.08 0.28 Pacific/Northwest 6 0.10 -0.10 0.32 0.10 Mixed 1 9.91 7.41 12.40 1.27 Data Gathered 1980 2 0.06 -0.62 0.74 0.35 1990 5 0.22 -0.25 0.68 0.24 2000 8 0.16 -0.22 0.53 0.19 Publication Type Journal 13 0.20 -0.05 0.45 0.13 Graduate Dissertation/Thesis 2 0.72 -0.08 1.52 0.41 State or local report 2 0.03 -0.64 0.70 0.34 Agency Number 1 15 0.12 -0.03 0.27 0.08 4 1 9.91 7.41 12.40 1.27

146

Table 5.20: continued Moderator Variable k ES 95% CI Min 95% CI Max SE Control Variables No 5 0.37 -0.07 0.80 0.22 Yes 12 0.17 -0.10 0.44 0.14 Interaction Effects No 14 0.28 0.00 0.57 0.15 Yes 3 0.10 -0.40 0.59 0.25 Continuous Data Collection No 2 -0.31 -0.84 0.23 0.27 Yes 14 0.26 0.03 0.50 0.12 Risk Recorded No 4 -0.14 -0.59 0.30 0.23 Yes 13 0.38 0.10 0.67 0.14 Need Recorded No 13 0.16 -0.09 0.41 0.13 Yes 4 0.45 -0.03 0.94 0.25 Race Measure White/Non-white 8 0.43 0.03 0.83 0.20 White/Black 9 0.13 -0.15 0.40 0.14 Offender Gender Exclusively Male 4 0.48 0.06 0.89 0.21 Exclusively Female 1 0.21 -0.68 1.10 0.46 Mainly Male 10 0.10 -0.19 0.39 0.15

147

Table 5.21: Risk Moderator Analysis Moderator Variable k ES 95% CI Min 95% CI Max SE Discipline Criminal Justice/Criminology 5 0.32 0.05 0.60 0.14 Psychology 0 0.00 0.00 0.00 0.00 Sociology 2 0.17 -0.07 0.41 0.12 Social Work 1 1.62 0.69 2.28 0.34 Mixed 7 0.21 0.09 0.33 0.06 Other 1 -0.38 -1.75 0.99 0.70 Publication Decade 1980 3 1.46 1.00 6.27 0.23 1990 2 0.22 0.05 2.56 0.09 2000 5 0.24 0.06 2.61 0.09 2010 5 0.09 -0.02 1.64 0.05 Geographic Region Northwest 1 1.37 0.51 2.23 0.44 South 2 0.23 0.06 0.41 0.09 Midwest 3 0.21 0.03 0.39 0.09 Southwest 2 1.11 0.19 2.03 0.47 Pacific/Northwest 6 0.12 0.01 0.22 0.05 Mixed 1 4.39 1.61 7.16 1.42 Data Gathered 1980 2 0.21 0.00 0.42 0.11 1990 5 0.28 0.13 0.44 0.08 2000 7 0.12 -0.01 0.25 0.07 Publication Type Journal 10 0.12 0.03 0.21 0.04 Graduate Dissertation/Thesis 2 1.26 0.67 1.84 0.30 State or local report 3 0.74 0.42 0.99 0.15 Other 1 0.80 0.28 1.32 0.26 Agency Number 1 14 0.20 0.10 0.29 0.05 4 1 4.39 1.61 7.16 1.42

148

Table 5.21: continued Moderator Variable k ES 95% CI Min 95% CI Max SE Control Variables No 6 0.84 0.59 1.09 0.13 Yes 10 0.14 0.04 0.23 0.05 Interaction Effects No 13 0.37 0.24 0.50 0.07 Yes 3 0.07 -0.09 0.23 0.08 Continuous Data Collection No 1 0.80 0.27 1.33 0.27 Yes 14 0.18 0.09 0.28 0.05 Need Recorded No 11 0.21 0.10 0.32 0.06 Yes 5 0.45 0.21 0.69 0.12 Race Measure White/Non-white 7 0.49 0.27 0.71 0.11 White/Black 8 0.17 0.06 0.28 0.06 Offender Race Mainly Black 2 0.62 0.32 0.93 0.16 Mixed 12 0.15 0.06 0.23 0.05 Offender Gender Exclusively Male 3 0.19 0.05 0.33 0.07 Exclusively Female 1 -1.42 -2.32 -0.52 0.46 Mainly Male 10 0.20 0.09 0.32 0.06

149

Chapter Summary

This chapter presented the results of the meta-analysis that was conducted as a part of this

dissertation. Across decision points, the majority of studies were published since 1990 and were

published in journals. The majority of studies used data that were gathered relatively recently.

Finally, studies were conducted across the United States and were not limited to a single

jurisdiction.

There was less consistency with independent variables across studies. Race, gender, and

measures of seriousness of the offense were measured often across decisions. Measures of other

variables were more limited. Age was measured in a majority of studies of arrest and sentencing

but not parole revocation. Suspect ethnicity has been a more recent addition to studies of

decision-making and is measured in a few studies across decision points. Criminal justice actor

characteristics were not measured in parole revocation studies, but were measured in a few of the

studies arrest and sentencing. Community characteristics were not measured often at any

decision point.

The effect size analysis indicates that measures of seriousness are significant across

decision points. The correlations between seriousness variables and the three dependent

variables estimated some of the largest mean effect sizes across decision points. Suspect

characteristics were also significant across decision points. Race was significantly related to

arrest and sentencing, but approached significance in the parole revocation analysis. The

findings indicated that black individuals were significantly more likely to be processed further

through the system. Gender was also significant at two decision points, arrest and sentencing.

Males were more likely to be arrested and incarcerated. Ethnicity was also a significant factor.

Hispanic individuals were more likely to be arrested and sentenced to prison.

150

Each decision point also had significant decision-specific results. At arrest, significant

mean effect sizes were found for all measures of suspect demeanor. At sentencing, the mode of

conviction was significant. More specifically, those that chose to take a plea were significantly

less likely to be sentenced to prison than those that went to trial. Finally, the analysis of parole

revocation indicated that those with a greater level of need were more likely to be revoked.

Several system-wide and decision-specific variables were not significant. Criminal

justice actor characteristics and community characteristics were not significant across decision

points. Suspect/defendant age was also not significant across decision points. The decision-

specific variable counsel type was not significant at sentencing. Finally, the moderator analysis

does not appear to nullify the effect size findings.

151

CHAPTER 6

DISCUSSION

This dissertation seeks to address two major limitations within the field of criminal

justice: 1) the lack of extensive reviews regarding decision-making correlates throughout

criminal justice system, and 2) the general perception that criminal justice is an atheoretical field.

Reviews of decision-making correlates are limited in both scope and number; additionally, the

majority of these reviews focus on specific decisions within the system or on a particular

variable’s (such as race and gender) impact on an outcome within the system (see Chapter 2 for a

detailed discussion). While these research endeavors have been valuable contributions to the

field of criminal justice research, they are limited. The only comprehensive system-wide review,

which was a massive and time-consuming undertaking, was conducted by Gottfredson and

Gottfredson in the late 1980s and has yet to be sufficiently updated. System-wide reviews of

decision-making correlates are valuable in that they help us understand what factors influence or

structure discretion. The system-wide focus also presents a fuller picture because the entire

system is examined, as opposed to examining each piece independently.

A second limitation of criminal justice is the general perception that criminal justice is an

atheoretical field. Marenin and Worrall (1998) argue that criminal justice “has not yet achieved

theoretical integrity and coherence” and that the only general consensus within the field is that

theories and concepts are drawn from a variety of other disciplines—such as sociology, biology,

political science—and that these derived theories coalesce and contribute to the “consolidation of

the new discipline” (p. 465). An atheoretical perception reduces legitimacy of criminal justice

theory and suggests a lack of development as a field. The lack of legitimacy may also reduce

interest in the topic amongst scholars, which also leads to the lack of development.

152

This dissertation seeks to address both of these limitations by recreating Gottfredson and

Gottfredson’s work on a smaller scale. First, correlates of decision-making are examined at

multiple decision points in an attempt to examine the “bigger picture” of criminal justice

decision making. Second, the correlates of decision-making are considered in the context of

system-wide versus decision-specific in order to determine if criminal justice theories should be

focused on the criminal justice system as a whole or by looking at each decision point separately.

The purpose of this dissertation has been to examine, in the ten plus years since Marenin and

Worrall’s comments, whether the field of criminal justice has begun to create a theoretical

foundation.

This section of the dissertation—which is broken up into five different sections—will

discuss these findings and comment on their relevance to these two limitations. First, there is a

summary of the findings. Second, there is a discussion of how to organize criminal justice

theory given the results. Third, there is a discussion of a theory that can be gleaned from the

findings of this dissertation. Fourth, there suggestions for future research in criminal justice

theory are presented. Finally, there is a brief discussion of the limitations of this study.

Summary of Findings

The overall findings suggest a lack of consistency in decision-making research across the

system, but there was some degree of similarities with respect to certain variables, such as

measures of seriousness and race. Measures of seriousness were significant across the system.

These measures were generally the strongest predictors of criminal justice processing. These

results indicated that those who committed more serious acts were more likely to be processed

through that decision point. Seriousness measures increased the likelihood of arrest between

1.53 and 6.91 times. The weakest seriousness relationship was between weapon use and arrest.

153

The results indicated that those that used a weapon were 1.53 times more likely to be arrested.

The strongest relationship was between evidence and arrest. Cases with more evidence were

nearly 7 times more likely to end in arrest. For sentencing, the increase in likelihood of

sentencing was between 1.45 and 1.96 times across seriousness measures.18 The weakest

relationship was between drug use and sentencing, which indicated that suspects that used illicit

substances were 1.45 times more likely to be incarcerated. The findings indicated that prior

record was the strongest predictor of incarceration; those with longer prior records were nearly 2

times more likely to be incarcerated. Results regarding seriousness measures for parole

revocation were similar to the other two decision points. Risk, which is essentially prior record,

was the weakest measure of seriousness. The results indicated that those with greater risk were

1.15 times to be revoked. The strongest measure of seriousness was need. Those with greater

need were 1.60 times more likely to be revoked.

These findings support the idea that criminal justice decision making is largely a function

of seriousness of the offense. Furthermore, the idea of two criminal justice systems is supported

by the finding from arrest analysis, which dichotomizes seriousness of offense as felony versus

misdemeanor. This indicates that those suspected of felony crimes are significantly more likely

(2.50 times) to be arrested. Additionally, measures of seriousness were generally the strongest

predictors across decision points and the direction of the relationship was positive. This

information indicates that it may be the case that the duality of criminal justice idea discussed by

the Gottfredsons may be correct. However, the data presented here cannot provide a definitive

answer.

18 The reduced effect of seriousness measures for sentencing may have been due to drug charging as a result of the “War on Drugs”. Mandatory drug penalties may have caused some defendants to be sentenced to prison when they would otherwise would not have been sentenced as harshly

154

Race appears to be a significant correlate of criminal justice decision making at multiple

stages of the criminal justice system. Black individuals are significantly more likely to be

processed than white individuals, 1.39 times more likely to be arrested and 1.23 times more

likely to be sentenced to prison. Offender race was nearly significant at parole revocation and

would likely have been significant with the addition of a small number of studies. This finding

may be due to the small number of studies on parole revocation. Additionally, the moderator

analysis indicates that African Americans are more likely to be revoked in the South (1.54 times

more likely). This evidence indicates that race appears to be an important predictor of criminal

justice decision making. Gender and ethnicity were significant at two decision points. Results

indicated that males were more likely to be arrested or sentenced than females, 1.49 and 1.63

times respectively, and Hispanics were more likely to be arrested or sentenced than non-

Hispanics, 1.25 and 1.29 times respectively. Another important aspect of the finding that black

individuals and Hispanics are more likely to be processed through the system is that it is not

limited temporally. This means, potentially, disparate treatment or discriminatory treatment is

still on going.

Several multi-decision relationships were not significant at any decision point. Age was

not significant at any decision point. It could be that the effect of age was absorbed by measures

of race, gender, and ethnicity. Measures of criminal justice agent characteristics, race and

gender, were also not significant.19 There are many possible explanations for this finding. First,

the lack of influence for criminal justice actor characteristics could be an artifact of the small

number of studies that actually measured these concepts, but the results of available studies

19 Criminal justice actor ethnicity was only measured in a few studies in either arrest or sentencing and was not included in parole revocation studies. Therefore, criminal justice actor ethnicity was included in with race for meta-analytic purposes. Hispanic was treated as black or non-white and non-Hispanic was treated as white. In the event that a study measured both officer race and officer ethnicity, officer ethnicity was dropped in favor of the officer race measure.

155

indicated that agent characteristics were not significantly related to the processing decisions. A

second and related issue is that some studies examined actor race, gender or ethnicity in a way

that was not conducive to meta-analysis. For example, Brown and Frank (2006) examined the

effect of officer race on arrest but did so by creating a multivariate model for white, and a second

model for black officers. The model containing arrests made by both black and white officers

did not include officer race. Thus, their finding that black officers were more likely to arrest

black suspects could not be included in this meta-analysis; there was no data available to

generate an effect size for the relationship between officer race and arrest.

Third, the lack of significance for criminal justice actor predictors is limited evidence of

discrimination in the system. The finding that race, gender, and ethnicity of suspects/defendants

is significant, but those same variables are not significant for criminal justice actors, could

indicate that potential discrimination is structural and not based on individual actors. In other

words, discrimination is potentially institutionalized and not just limited to a few “bad apples”.

The training process may be very thorough and could cause any individual differences to be

reduced or eliminated. Instead the individual criminal justice agent functions as a part of the

workgroup for that particular branch of the criminal justice system (Klinger, 1996).

A final interpretation of the significance of suspect demographic characteristics but not

criminal justice agent demographics is that suspect race, gender and ethnicity may be proxies for

other theoretically relevant but unmeasured variables. None of the studies that measured

criminal justice agent factors also measured factors pertaining to social class or education, and

only a handful of studies measured employment status. However, these variables have been

theorized to be explanations of criminal justice actor behavior. For example, Black (1976)

argues that social status, education (a measure of culture), and employment status (a measure of

156

organization) are critical factors to explaining the behavior of governmental social control.

These other demographic characteristics may be correlated with race, gender, and ethnicity and

act as proxy measures.

Both sentencing and arrest also had unique, decision-specific measures that were

significant. Both measures of demeanor were significant. Specifically, suspects that were

disrespectful or non-compliant were significantly more likely to be arrested. The third demeanor

related variable—interaction-phase crime—was also significant. Based on the findings

presented in this study, it appears that Klinger may have been correct: interaction-phase crime

does significantly increase the odds of arrest. The mean effect size of interaction-phase crime

was almost twice as strong as the mean effect size of demeanor. However, the moderator

analysis showed that regardless of whether a study measured interaction-phase crime, demeanor

was still significant. Suspect mental illness was not significantly related to arrest, but this could

be an artifact of the small number of studies that actually measured suspect mental state. The

direction of the effect size indicated that officers were less likely to arrest mentally ill suspects.

Sentencing also produced a unique finding. Mode of conviction was significantly related

to sentencing. Those that pled guilty were 1.29 times less likely to go to prison. However,

suspect employment, court size, and the type of attorney were not significantly related to

sentencing. Finally, moderator analysis for all three decision points did not really create any

significant modifications to the general mean effect size findings.

Missing variables

Another finding, when looking at the bigger picture of the results, is that there is a clear

lack of consistency in decision-making research. The results generally indicate that seriousness

of the offense and prior record are important at all stages examined, and race, gender, and

157

ethnicity are important at multiple stages. However, when it comes to other correlates, there is

little consistency across research. Potentially relevant variables are simply not measured. This

finding has implications for the “atheoretical” perception of criminal justice research. Without a

clear theory or theories to provide guidance on not only what to measure, but how to measure

concepts, this lack of consistency will continue, which will cause the “atheoretical” perception of

criminal justice research to thrive.

Criminal Justice Theory Organization

Given the above results, and the general lack of effects from moderators, it appears that

correlates are generally consistent across the system. Some measure of seriousness was

significant at all three stages of the criminal justice system. Additionally, several variables were

significant predictors (race, gender, and ethnicity) of at least two decision points. These

variables also performed in the same directions at all stages examined. The effects of factors that

are unique to a particular decision were not as strong as those that are generally applicable.

Factors that were unique to a decision may have only been unique because they were not

measured at other stages of the criminal justice system. Arrest specific factors, demeanor and

mentally ill status, are not exclusive to arrest. Defendants could potentially have a less than

cooperative disposition and judge may enforce a harsher punishment. Interaction-phase crime is

really a measure of seriousness and the strong effect is more applicable in the context of a

measure of seriousness than as a unique demeanor measure.

The effects of unique sentencing factors are virtually non-existent and the one decision-

specific relationship that produced a significant effect size, mode of conviction, was weaker than

any effect size in suspect characteristics, or seriousness measures. While the effects of need

158

were a strong unique factor in parole revocation, it was also a measure of seriousness. This

means that need functions in a similar manner as interaction-phase crime did at arrest.

Given this information, it seems most appropriate to approach criminal justice theory

from a whole-system perspective versus a decision-specific approach. Given the leaning toward

the whole-system approach for theory organization, the idea that criminal justice theory has one

dependent variable, discretionary processing, is also supported. However, due to limited data on

some unique factors, it is not prudent to dismiss them entirely, and it is probably the case that the

majority of the decision making of criminal justice agents is based largely on these whole-system

variables with some influence by decision-specific characteristics.

The findings here do not appear to be consistent with Bernard and Engel’s

conceptualization of criminal justice theory organization. Bernard and Engel’s (2001) attempt

was useful and an important contribution to this area of study, but their organizational approach

may miss higher order, system-wide influences on processing. A new organizational method

based on primary predictor variables would prevent theories with similar theoretical explanations

from being separated simply because they operationalize a dependent variable differently. For

example, Klinger’s ecological theory of policing focuses on police workgroups and the influence

of those workgroup norms on police decisions. Eisenstein and Jacob’s workplace theory of court

processing argues that the courtroom workgroup is a critical component to sentencing. If the

focus is on the dependent variable, then these two theoretically similar theories would not be

grouped together. Focusing on primary correlates as an organizational approach would also

bring criminal justice in line with other social science disciplines, such as criminology and

sociology, and increase the scientific legitimacy of criminal justice as a discipline.

159

The theoretical categories, discussed in chapter two, represent the new organizational

model proposed in this study. Theories should be grouped into the following categories:

situational theories, actor theories, workgroup/organizational theories, community theories, and

conflict theories. This newly proposed model breaks each type of theory into their key predictive

components. Additionally, while actor characteristics, organizational, and community

characteristics were not supported in the effect size analysis, they should not be omitted. These

categories of predictors have not been sufficiently studied to warrant their removal as theory

categories.

Toward a General Theory of Criminal Justice

A general explanation for the behavior of criminal justice actors would have to include

indicators of seriousness and suspect characteristics—race, ethnicity, and gender. There would

also seem to be some effect of unique characteristics that function as part of the general

explanation. Unfortunately, the literature examined here did not provide consistent measures of

community influences or organizational factors. The handful of studies that actually recorded

this information did not measure with any degree of consistency. Court size, the one

organizational factor that was measured somewhat frequently was not significant. However, it

would be inappropriate to dismiss both organizational and community explanations and the

theories based on this level of explanation such as Wilson’s theory of police behavior, and

Duffee’s community theory of criminal justice, based on this one finding. Nevertheless, it

appears that a theory of criminal justice should first focus on the entire criminal justice system.

Second, theories of criminal justice should include measures of seriousness and suspect

characteristics.

160

Future Research

First, there is a general lack of research on certain variables; specifically, there is a lack

of research on the effect of organizational and community factors on decision-making behavior.

The examination of characteristics of criminal justice actors is very limited. Eleven studies of

arrest and seven studies of sentencing examined here also measured the effects of criminal

justice actors on decision making. Suspect employment was only examined in ten sentencing

studies and ten parole revocation studies; this is a relatively small portion when compared to

other factors like race and gender.

A related issue is that other potentially relevant measures, like education and social class,

were virtually unmeasured. Some studies examined a quasi-social class characteristic; however,

any examination was based more on where the arrest or sentencing took place and not an

individual’s social class. Future research may want to improve upon the current state of

decision-making research by striving to incorporate the effect of these relevant but unmeasured

variables. This would aide in potentially disentangling the effects of race, gender, and ethnicity,

and how they may be serving as a proxy measure of other variables like education and social

class.

Second, it appears that prior record/risk is a key factor in criminal justice decision

making. However, prior record/risk has not been measured at the arrest stage. If possible, it

could be useful to incorporate some measure of prior record into examinations of the arrest

decision. This concept could be incorporated into systematic social observation studies by

asking officers if they have had repeated contact with the suspect or by checking arrest records of

suspects to obtain the identity of the arresting officer. Another potential measure could be the

extent of the record check when an officer performs a record check during a stop. Finally, there

161

is a very limited knowledge of ethnicity relative to other suspect characteristics. While the

results are significant, perhaps as time passes and more studies measuring ethnicity are

accumulated the results may change.

Third, while specific variables are limited, there is a general lack of studies on parole

revocation. This important final decision has very little research relative to arrest and

sentencing. Given that this decision potentially pulls someone back into prison, it would be

important to study the factors that influence this decision. However, based on the small number

of studies analyzed here, it would appear that parole revocation is the most neutral and objective

decision. Research in this area can be improved by creating a more uniform measure of

revocation and consistent operational definitions of concepts. Parole revocation as a field of

study was the least developed of any of the decisions examined.

Finally, there is a need of greater research that uses a system-wide approach to decision

making. No studies analyzed here actually took cases from arrest to parole revocation. It would

be helpful, and a rich source of data, to be able to analyze cases from arrest to parole revocation

to see if the same factors are influencing decision making. A limitation of the meta-analysis

presented in this dissertation is that the cases examined in arrest are not necessarily the same

cases examined in sentencing or parole revocation. Having data that actually examines

processing from the beginning to the end of the system might be able to put an end to the whole-

system versus decision-specific approach debate.

Limitations

While every effort was made to minimize the limitations of this study, there were several

limitations. First, it is likely that some studies may have been missed. Efforts were made to be

as inclusive as possible, but some of the results should be used with caution. However, the

162

majority of significant mean effect sizes produced stable results. Second, this project was that

only direct effects could be measured. It may be that some of the non-significant factors were

functioning as part of an interaction effect or a non-direct effect. It may be the case that some

effects were masked by unexamined interaction effects. Third, the results are limited by the

quality of the studies included. However, this limitation was mitigated somewhat by the

moderator analysis, which indicated a great deal of uniformity across studies. Finally, as

mentioned above there were some predictors of decision-making that were simply insufficiently

measured. This reduced the ability to explore all facets of the research question and truly

develop a general theory of criminal justice. Finally, two large groups of factors, organizational

and community, were simply not able to be examined.

The third and fourth limitations can be corrected with future research. The solution to

these limitations is for future research to strive to create more uniform measures and models.

More complete models would correct the second limitation because it would improve the overall

quality of the study by having more complete data. Further, standardized measures and models

would provide effect size information that is more directly comparable and less likely to

experience problems with heterogeneity. More compatible information would reduce the risk of

heterogeneity in the study effect sizes. The fourth limitation can also be corrected with this

proposed idea of a standardized model of decision making. By standardizing the measures and

models currently omitted variables would be more likely to be included and therefore a more

definitive decision could be made as to the effect of actor, organizational, and community

characteristics. Variables that could be included in this standardized model would include

system-wide variables: offender race, offender gender, offender ethnicity, seriousness of the

offense, risk/prior record. Additionally, the model would also need to include significant

163

decision-specific variables, where applicable, such as demeanor, and mode of conviction.

Finally, the standardized model should also include agent characteristics, race, gender, ethnicity,

and experience, organizational characteristics, and community characteristics in order to fully

determine if these factors are important to criminal justice processing. Standardized models may

also help to conceptualize new criminal justice theories.

Despite these limitations this dissertation has contributed to the field of decision-making

research. This study examined decision making at multiple decision points, which has not been

done since Gottfredson and Gottfredson’s study. Further, this dissertation improved upon the

Gottfredson’s work by adding meta-analysis to the examination of correlates of decision-making.

The results indicate that seriousness is the chief decision-making component of decision makers

of the criminal justice system. Further, the results indicate, race, gender, and ethnicity are

important factors in decision making. Black individuals, males, and Hispanics appear to be more

likely to be processed through the system.

Finally, this dissertation has also presented results that appear to indicate that criminal

justice theory should focus on a system-wide approach versus creating decision-specific theories.

Correlates that are significant predictors of decision-making are consistent across decision

points. Independent variables that were significant and unique to a particular decision point were

generally unique in application, not in theory. In other words, there were no theoretical reasons

why, for example, a suspect/defendants demeanor would not impact a judge’s decision in the

same way it appears to impact an officer’s exercise of discretion. Finally, moderator analyses do

not seem to negate any of the major findings.

164

REFERENCES20 *Alpert, G.P., Becker, E., Gustafson, M.A., Meister, A.P., Smith, M.R., and Strombom, B.A. (2006). Pedestrian and Motor Vehicle Post-Stop Data Analysis. Los Angeles, CA: City of Los Angeles. *Arazan, C.L. (2007). Contextual Effects and Punishment Decisions: A Multilevel Analysis. Ph.D. Dissertation, College of Criminology and Criminal Justice, Florida State University, Tallahassee, FL. *Atherton, M.C. (2005). The Study of Hispanic Defendants in United States federal court: A Multi-Level Analysis of Racial Sentencing Outcomes. Ph.D. Dissertation, Department of Sociology, Pennsylvania State University, University Park, PA. Berlin, J.A., and Colditz, G.A. (1990). “A meta-analysis of physical activity in the prevention of coronary heart disease”. American Journal of Epidemiology. 132: 612-628. Bernard, T. and Engel, R. (2001). “Conceptualizing criminal justice theory” Justice Quarterly. 18: 1-30. *Bickle, G.S. and Peterson, R.D. (1991). “The impact of gender-based family roles on criminal sentencing”. Social Problems. 38: 372-394. Black, D. (1976). The behavior of law. New York: Academic Press. Bonta, J., Law, M., & Hanson, K. (1998). “The prediction of criminal and violent recidivism among mentally disordered offenders: A meta-analysis”. Psychological Bulletin, 123: 123-142. *Bourassa, S.C. and Andreescu, V. (2009). “Decomposition of racial differences in sentencing: Application of an econometric technique to cocaine possession cases”. Journal of Ethnicity in Criminal Justice. 7: 186-204. Braithwaite, J. and Biles, D. (1980). “Empirical verification and Black’s ‘behavior of law’”. American Sociological Review. 45: 334-338. Brind, J., Chinchilli, V.M., Severs, W.B., and Summy-Long, J. (1996). “Induced abortion as an independent risk factor for breast cancer: A comprehensive review and meta-analysis”. Journal of Epidemiology and Community Health. 50: 481-496. *Brown, M.A. (2003). Similarities and Differences in the Sentencing Decisions of Men and Women Judges in Cook County Circuit Court. Ph.D. Dissertation, University of Nebraska, Omaha, NE. Brown, R. and Frank, J. (2006). “Race and officer decision making: Examining differences in arrest outcomes between black and white officers”. Justice Quarterly, 23: 96-126. 20 * indicates a reference that was used for the meta-analysis

165

*Brown, R.A., Novak, K.J., Frank, J. (2009). “Identifying variation in police officer behavior between juveniles and adults”. Journal of Criminal Justice. 37: 200-208. *Bushway, S.D. and Piehl, A.M. (2001). “Judging judicial discretion: Legal factors and racial discrimination in sentencing”. Law & Society Review. 35: 733-764. *Buzawa, E., Austin, T.L., and Buzawa, C.G. (1995). “Responding to crimes of violence against women: Gender differences versus organizational imperatives”. Crime & Delinquency. 41: 443-466. Caplan, J. (2007). “What factors effect parole: A review of empirical research”. Federal Probation. 71: 16-19. *Cho, H. (2006). Effects of Arrest on Intimate Partner Violence Incidence and Revictimization: Logistic Regression and Regression Time Series Analysis of the National Crime Victimization Survey from 1987-2003. Ph.D. Dissertation, College of Social Work, Florida State University. Clay-Warner, J. and McMahon-Howard, J. (2009). “Rape reporting: ‘Classic rape’ and the behavior of law”. Violence and Victims. 24: 723-743. *Cohen, T. (2003). The New Penology: How Court Sentencing Practices Have Been Influenced by the Growing Emphasis on Actuarialism and Managerialism. Ph.D. Dissertation. Rutgers, The State University of New Jersey. Crank, J. (1990). “The influence of environmental and organizational factors on police style in urban and rural environments”. Journal of Research in Crime and Delinquency. 27: 166-189. Crank, J. and Bowman, B. (2008). “What is good criminal justice theory?”. Journal of Criminal Justice. 36: 563-572. *Crawford, C. (2000). “Race and pretextual stops: Noise enforcement in Midwest City”. Social Pathology. 6: 213-227. *Crow, M.S. and Bales, W. (2006). “Sentencing guidelines and focal concerns: The effect of sentencing policy as a practical constraint on sentencing decisions”. American Journal of Criminal Justice. 30: 285-304. Cullen, F. (2011). “Beyond adolescence-limited criminology: Choosing our future – The American Society of Criminology 2010 Sutherland address”. Criminology. 49: 287-330. Daly, K. and Bordt, R. (1995). “Sex effects and sentencing: An analysis of the statistical literature”. Justice Quarterly. 12: 141-175. Davis, K.C. (1969). Discretionary justice. Baton Rouge, LA: Louisiana State University

166

*Doran, D.E. (2007). Racial Profiling in Las Vegas: A Reexamination of Police Stop Data in Las Vegas. Master’s Thesis, Department of Criminal Justice, University of Nevada Las Vegas. Duffee, D. (1980). Explaining criminal justice: Community theory and criminal justice reform. Cambridge, MA: Oelgeschlager, Gunn & Hain. Duffee, D. and Allan, E. (2007). “Criminal Justice, criminology and criminal justice theory”. In Duffee, D. and Maguire, E. (eds). Criminal justice theory: Explaining the nature and behavior of criminal justice. New York: Routledge. Duffee, D. and Maguire E. (2007). Criminal justice theory: Explaining the nature and behavior of criminal justice. New York: Routledge. Durlak, J. A., & Lipsey, M. W. (1991). “A practitioner’s guide to meta-analysis”. American Journal of Community Psychology, 19: 291-332. Eisenstein, J. and Jacob, H. (1977). Felony justice: An organizational analysis of criminal courts. Boston: Little, Brown. *Eitle, D. (2005). “The influence of mandatory arrest policies, police organizational characteristics, and situational variables on the probability of arrest in domestic violence cases”. Crime & Delinquency. 51: 573-597. *Eitle, D., Stolzenberg, L., and D’Alessio, S.L. (2005). “Police organizational factors, the racial composition of the police, and the probability of arrest”. Justice Quarterly. 22: 30-57. *Ekpunoboi, A.E. (1999). Judicial Decision-Making Under Michigan Sentencing Guidelines. Ph.D. Dissertation, School of Public Affairs, Western Michigan University, Kalamazoo, MI. *Engel, R.S., and Calnon, J.M. (2004). “Examining the influence of drivers’ characteristics during traffic stops with police: Results from a national survey”. Justice Quarterly. 21: 49-90. *Engel, R.S., Calnon, J.M., Liu, L., and Johnson, R. (2004). Project on Police-Citizen Contacts: Year 1 Final Report. Harrisburg, PA: Pennsylvania State Police. *Engel, R.S., Calnon, J.M., Tillyer, R., Johnson, R., Liu, L., and Wang, X. (2005). Project on Police-Citizen Contacts: Year 2 Final Report. Harrisburg, PA: Pennsylvania State Police. *Engel, R.S., Chekauskas, J.C., and Smith, M.R. (2008). Traffic Stop Data Analysis Study: Year 2 Final Report. Phoenix, AZ: Department of Public Safety. *Engel, R.S., Cherkauskas, J.C., Smith, M.R., Lytle, D., and Moore, K. (2009). Traffic Stop Data Analysis Study: Year 3 Final Report. Phoenix, AZ: Department of Public Safety. *Engel, R.S. and Silver, E. (2001). “Policing mentally disordered suspects: A reexamination of the criminalization hypothesis”. Criminology. 39: 225-252.

167

*Engel, R.S., Sobol, J.J., and Worden, R.E. (2000). “Further exploration of the demeanor hypothesis: The interaction effects of suspects’ characteristics and demeanor on police behavior”. Justice Quarterly. 17: 235-258. *Engel, R.S., Tillyer, R., Cherkauskas, J.C., and Frank, J. (2007). Traffic Stop Data Analysis Study: Year 1 Final Report. Phoenix, AZ: Department of Public Safety. *Erez, E. and Tontodonato, P. (1989). “Patterns of reported parent-child abuse and police response”. Journal of Family Violence. 4: 143-159. *Everett, R.S. and Nienstedt, B.C. (1999). “Race, remorse, and sentence reduction: Is saying you’re sorry enough?” Justice Quarterly. 16: 99-122. Eysenck, H. J. (1978). “An exercise in mega-silliness”. American Psychologist. 33: 517. *Farrell, A., Ward, G., and Rousseau, D. (2009). “Race effects of representation among federal court workers: Does black workforce representation reduce sentencing disparities?”. The ANNALS of the American Academy of Political and Social Science. 623: 121-133 *Fearn, N. (2003). Community Context and Sentencing Decisions: A multilevel Analysis. Ph.D. Dissertation, Department of Criminology and Criminal Justice, University of Missouri-St. Louis, St. Louis, MO. *Feder, L. (1996). “Police handling of domestic calls: The importance of offender’s presence in the arrest decision”. Journal of Criminal Justice. 24: 481-490. *Felson, R.B. and Ackerman, J. (2001). “Arrests for domestic and other assaults”. Criminology. 39: 655-676. Fleiss, J.L. (1993). "The statistical basis of meta-analysis". Statistical Methods in Medical Research, 2: 121-145. Fleiss, J.L. and Berlin, J.A. (2009). "Effect sizes for dichotomous data". In Cooper, H., Hedges, L. and Valentine, J. The Handbook of Research Synthesis and Meta-Analysis (2nd Ed). New York: Russell Sage Foundation *Franklin, T.W. (in press). “Sentencing native Americans in US Federal courts: An examination of disparity”. Justice Quarterly 1-30. *Freeman, J.R. (1992). The Social Ecology of Police Discretion. Ph.D. Dissertation. Northwestern University, Evanston, IL. *Friedrich, R.J. (1977). The Impact of Organizational, Individual, and Situational Factors on Police Behavior. Ph.D. Dissertation, Department of Political Science, University of Michigan.

168

*Fyfe, J.J., Klinger, D.A., and Flavin, J.M. (1997). “Differential police treatment of male-on-female spousal violence”. Criminology. 35: 455-473. Gaugler, J.E., Duval, S., Anderson, K.A., and Kane, R.L. (2007). "Predicting nursing home admissions in the U.S: A meta-analysis". BMC Geriatrics. 7: 1-14. *Gibbs, J. (2003). Factors Affecting Police Officers’ Decisions to Arrest in Domestic Violence Situations: A Look at the Niagara Falls Police Department. Master’s Thesis, University of Niagra. Glass, G. V. (1976). "Primary, secondary, and meta-analysis". Educational Researcher, 5: 3-8. Geddes, J.R., and Lawrie, S.M. (1995). “Obstetric complications and schizophrenia: A meta-analysis”. The British Journal of Psychiatry. 167: 786-793. Gottfredson, M. and Gottfredson, D. (1989). Decision making in criminal justice: Toward the rational exercise of discretion. (2nd Ed.). New York: Plenum Press. Gottfredson, M. and Hindelang, M. (1979). “A study of the behavior of law”. American Sociological Review. 44: 3-18. Greenland, S. (1987). “Quantitative methods in the review of epidemiologic literature”. Epidemiological Reviews. 9: 1-30 Greenland, S. (1993). “A meta-analysis of coffee, myocardial infarction, and coronary death”. Epidemiology. 4: 366-374. *Griffin, T.W. (2002). The Impact of Presumptive Sentencing Guidelines on Disparity in Sentencing in Ohio. Ph.D. Dissertation, Division of Criminal Justice, University of Cincinnati. *Gumbhir, V.K. (2005). Racial Profiling in Eugene, Oregon: A Case Study in Race, Community, and Law Enforcement. Ph.D. Dissertation. Department of Sociology. University of Oregon. Haddock, C.K., Rindskopf, D., and Shadish, W.R. (1998). “Using odds ratios as effect sizes for meta-analysis of dichotomous data: A primer on methods and issues”. Psychological Methods. 3: 339-353. *Harrington, M.P. and Spohn, C. (2007). “Defining sentence type: Further evidence against use of the total incarceration variable”. Journal of Research in Crime and Delinquency. 44: 36-63. *Hadwiger, J. (1972). The Significance of Legal and Extra Legal Variables in Predicting Sentencing Outcomes Under Oklahoma’s Community Sentencing Act. Ph.D. Dissertation, Oklahoma State University.

169

Hagan, J. (1989). “Why is there so little criminal justice theory? Neglected macro- and micro-level links between organization and power.” Journal of Research in Crime and Delinquency. 26: 116-135 Hanushek, E.A., and Jackson, J.E. (1977). Statistical methods for social scientists. Orlando, FL: Academic Press Inc. Hassell, K., Zhao, J., and Maguire, E. (2003). “Structural arrangements in large municipal police organizations: Revisiting Wilson’s theory of local political culture”. Policing: An International Journal of Police Strategies and Management. 26: 231-250. *Hawkins, H.C. (2008). “Race and sentencing outcomes in Michigan”. Journal of Ethnicity in Criminal Justice. 3: 91-109. *Haynes, S.H., Ruback, R.B., and Cusick, G.R. Courtroom Workgroups and Sentencing: The effects of Similarity, Proximity, and Stability. Unpublished manuscript. Hedges, L.V., and Olkin, I. (1985). Statistical methods for meta-analysis. London: Academic Press Inc. *Heilburn, A.B. (1978). “Race, criminal violence, and length of parole: A new look at parole outcome”. British Journal of Criminology. 18: 53-61. *Ho, T. (2000). “Domestic violence in a southern city: The effects of a mandatory arrest policy on male-versus-female aggravated assault incidents”. American Journal of Criminal Justice. 25: 107-118. *Ho, T. (2003). “The influence of suspects gender in domestic violence arrests”. American Journal of Criminal Justice. 27: 183-195. Hubbard, D. and Pratt, T. (2002). “A meta-analysis of the predictors of delinquency among girls”. Journal of Offender Therapy. 34(3): 1-13. Hunt, J. (1997). How science takes stock: The story of meta-analysis. New York: Russell Sage Foundation. Hunter, J. E., & Schmidt, F. L. (1990). Methods of meta-analysis. Newburk Park, CA: Sage Publications. *Jhi, K.Y., and Joo, H.J. (2009). “Predictors of recidivism across major age groups of parolees in Texas”. Justice Policy Journal. 6: 1-28. Johnson, B.D. (2003). “Racial and ethnic disparities in sentencing departures across modes of conviction”. Criminology. 41: 449-488.

170

Johnson, B.D. (2005). “Contextual disparities in guideline departures: Courtroom social contexts, guideline compliance and extralegal disparities in criminal sentencing”. Criminology. 43: 761-796. *Johnson, B.D. (2006). “The multilevel context of criminal sentencing: Integrating judge- and county-level influences”. Criminology. 44: 259-298. *Johnson, Y.W. (2010). Racial Disparity in Sentencing Outcomes: A Study of Sentencing Decisions Under Arizona’s Presumptive Sentencing Structure. Ph.D. Dissertation, Capella University, Minneapolis, MN. *Johnson, B.D. and Betsinger, S. (2009). “Punishing the ‘model minority’: Asian-American criminal sentencing outcomes in federal district courts”. Criminology. 47: 1045-1090. Jonson, C.L. (2010). The Impact of Imprisonment on Reoffending: A Meta-Analysis. PhD. Dissertation, Department of Criminal Justice, University of Cincinnati, Cincinnati, OH. *Jones, D.A. and Belknap, J. (1999). “Police responses to battering in a progressive pro-arrest jurisdiction”. Justice Quarterly. 16: 249-273. Kane, R. (2002). “The social ecology of police misconduct”. Criminology. 40: 867-896. Klinger, D. (1994). “Demeanor or crime? Why ‘hostile’ citizens are more likely to be arrested”. Criminology. 32: 475-493. *Kassenbaum, G., Davidson-Coronado, J., Perrone, P., and Allen, J. (2001). Parole Decision Making in Hawaii. Hawaii Correctional Industries. *Kassenbaum, G., Davidson-Coronado, J., Silverio, M. and Marker, N. (1999). Survival on Parole: A Study of Post-Prison Adjustment and the Risk of Returning to Prison in the State of Hawaii. Hawaii Correctional Industries. *Klinger, D. (1994). “Demeanor or crime? Why ‘hostile’ citizens are more likely to be arrested”. Criminology. 32: 475-493. *Klinger, D. (1996). “More on demeanor and arrest in Dade County”. Criminology. 34: 61-82. Klinger, D. (1997). “Negotiating order in patrol work: An ecological theory of police response to deviance”. Criminology. 35: 277-306. Kochel, T.R., Wilson, D.B., and Mastrofski, S.D. (2011). “Effect of suspect race on officers’ arrest decisions”. Criminology. 49: 473-512. *Koons-Witt, B.A. (2002). “The effects of gender on the decision to incarcerate before and after the introduction of sentencing guidelines”. Criminology. 40: 297-328.

171

*Kramer, J. and Steffensmeier, D. (1993). “Race and imprisonment decisions”. The Sociological Quarterly. 34: 357-376. Kramer, J.H. and Ulmer, J.T. (2002). “Downward departures for serious violent offenders: Local court ‘corrections’ to Pennsylvania’s sentencing guidelines”. Criminology. 40: 897-931. Kraska, P. (2006). “Criminal justice theory: Toward legitimacy and an infrastructure”. Justice Quarterly. 23:167-185. *Krauss, M.W. (1981). The Mentally Disabled Parolee: Predictors of Parole Outcome. Ph.D. Dissertation, The Florence Heller Graduate School for Advanced Studies in Social Welfare, Brandeis University, Waltham, MA. Kuhn, T. (1996). The structure of scientific revolutions. (3rd Ed.). Chicago: University of Chicago Press. Kuo, S., Longmire, D., Cuvelier, S., and Chang, S. (2010). “Prosecutorial decision making in Taiwan: A partial test of Black’s behavior of law”. International Journal of Offender Therapy and Comparative Criminology vol. 54: 1023-1046 Lasky-Su, J.A., Faraone, S.V., Glatt, S.J., and Tsuang, M.T. (2005). “Meta-analysis of the association between two polymorphisms in the serotonin transporter gene and affective disorders” American Journal of Medical Genetics. 133: 110-115. Langworthy, R. (1985). “Wilson’s theory of police behavior: A replication of the constraint theory”. Justice Quarterly. 2: 89-98. Latessa, E. and Smith, P. (2007). Corrections in the community (4th ed.) Cincinnati, OH: Anderson Publishing. *Leiber, M.J. and Blowers, A.N. (2003). “Race and misdemeanor sentencing”. Criminal Justice Policy Review. 14: 464-485. *Leinfelt, F.H. (2006). “Police officer discretion and race in an American Midwestern city: Predicting arrests, citations, warnings: A multivariate examination”. The Police Journal. 79: 3-27. *Liederbach, J. (2007). “Controlling suburban and small-town hoods: An examination of police encounters with juveniles”. Youth Violence and Juvenile Justice. 5: 107-124. Liederbach, J. and Travis, L. (2008). “Wilson redux: Another look at varieties of police behavior”. Police Quarterly. 11: 447-467. Lipsey, M.W. and Wilson, D.B. (2001). Practical meta-analysis. Thousand Oaks, CA: Sage Publications Inc.

172

*Lin, J., Grattet, R., and Petersilia, J. (2010). "'Back-end sentencing' and reimprisonment: Individual, organizational, and community predictors of parole sanctioning decisions". Criminology. 48: 759-795. Logan, C. H., & Gaes, G. G. (1993). Meta-analysis and the rehabilitation of punishment. Justice Quarterly, 10, 245-263. Lӧsel, F. and Schmucker, M. (2005). “The effectiveness of treatment for sexual offenders: A comprehensive meta-analysis”. Journal of Experimental Criminology. 1: 117-146. *Lunden, R. (1987). Risk and Recidivism Among Massachusetts Parolees: An Update. Boston, MA: Massachusetts Parole Board, Research and Planning Unit. *Lundman, R.J. (1994). “Demeanor or crime? The Midwest city police-citizen encounters study”. Criminology. 32: 631-656. *Lundman, R.J. (1996). “Demeanor and arrest: Additional evidence from previously unpublished data”. Journal of Research in Crime and Delinquency. 33: 306-323. Lundman, R. J., & Kaufman, R. L. (2003). Driving while black: Effects of race, ethnicity, and gender on citizen self-reports of traffic stops and police actions. Criminology, 41: 601–626. Makrarios, M. and Pratt, T. (2012), “The effectiveness of policies and programs that attempt to reduce firearm violence: A meta-analysis”. Crime & Delinquency. 58: 222-244. Marenin, O., & Worrall, J. (1998). “Criminal justice: Portrait of a discipline in progress”. Journal of Criminal Justice, 26: 465–480. Mastrofski, S., Snipes, J., Parks, R. and Maxwell, C. (2000). “The helping hand of the law: Police control of citizen’s request”. Criminology. 38: 307-342 *Mastrofski, S.D., Worden, R.E., and Snipes, J.B. (1995). “Law enforcement in a time of community policing”. Criminology. 33: 539-563. *Mignon, S.I., and Holmes, W.M. (1995). “Police response to mandatory arrest laws”. Crime & Delinquency. 41: 430-442. *Miller, J.M., and Miller, H.V. (2011). “Considering the effectiveness of drug treatment behind bars: Findings from the South Carolina RSAT evaluation”. Justice Quarterly. 28: 70-86. Mitchell, O. (2005). “A meta-analysis of race and sentencing research: Explaining the inconsistencies”. Journal of Quantitative Criminology. 21: 439-466. *Morris, P.W. (2009). Dual Arrest in Intimate Partner Violence Incidents: The Influence of Police Officer, Incident, and Organizational Characteristics. Ph.D. Dissertation, City University of New York, New York, NY.

173

*Myers, M.A. and Talarico, S.M. (1986). “Urban justice, rural injustice? Urbanization and its effects on sentencing”. Criminology. 24: 367-391. *Nobiling, T.L. (2005). A Tale of Two Cities: The Effects of Offender’s Employment Status on Sentence Severity in Chicago and Kansas City. Ph.D. Dissertation, University of Nebraska, Omaha, NE. *Novak, K.J. and Engel, R.S. (2005). “Disentangling the influence of suspects’ demeanor and mental disorder on arrest”. Policing: An International Journal of Police Strategies and Management. 28: 493-512. Novak, K., Frank, J., Smith, B., and Engel, R. (2002). “Revisiting the decision to arrest: comparing beat and community officers”. Crime & Delinquency. 48:70-98. National Research Council (2004). Fairness and effectiveness in policing: The evidence. Committee to Review Research on Police Policy and Practices. W. Skogan and K. Frydl, Eds. Committee on Law and Justice, Division of Behavioral and Social Sciences and Education. Washington, DC: National Academies Press. National Research Council (2008). Parole, desistance from crime, and community integration. Committee on Community Supervision and Desistance from Crime. Committee on Law and Justice, Division of Behavioral and Social Sciences and Education. Washington, DC: The National Academies Press. *Ozgenturk, Ilyas (2009). An Analysis of the Characteristics Related to the Victims and Suspects of Domestic Violence. D.Ed. Dissertation, College of Education, Spalding University, Louisville, KY. Packer, H. (1968). The limits of the criminal sanction. Stanford, CA: Stanford University Press. Paternoster, R. (1987). “The deterrent effect of the perceived certainty and severity of punishment: A review of the evidence and issues”. Justice Quarterly. 4: 173-217. *Patnoe, J.L. (1990). The Demographic and Ecological Distribution of Police Discretion in an Urban Area. Ph.D. Dissertation, Department of Sociology, University of Arizona, Tuscon, AZ. Philips, S.W. and Sobol, J.J. (2011). “Police attitudes about the use of unnecessary force: An ecological explanation”. Journal of Police and Criminal Psychology. 26: 47-57. *Pogrebin, M.R., Poole, E.D., and Regoli, R. (1986). “Parole decision making in Colorado”. Journal of Criminal Justice. 14: 147-155. Pratt, T. C. (1998). “Race and sentencing: A meta-analysis of conflicting empirical research results” Journal of Criminal Justice. 26: 513-523.

174

Pratt, T. C. (2001). Assessing the relative effects of macro-level predictors of crime: A meta-analysis. Ph.D. dissertation, Department of Criminal Justice, University of Cincinnati, Cincinnati, OH. Pratt, T. C. (2002). "Meta-analysis and its discontents". Journal of Offender Rehabilitation, 35: 23-40. Pratt, T. (2010). “Meta-analysis in criminal justice and criminology: What is it, when it’s useful, and what to watch out for”. Journal of Criminal Justice Education. 21:152-168. Pratt, T., and Cullen, F. (2000). “The empirical status of Gottfredson and Hirschi’s general theory of crime: A meta-analysis”. Criminology. 38: 921-964. Pratt, T., and Cullen, F. (2005). “Assessing macro-level predictors and theories of crime: A meta-analysis”. In Tonry, M. (Ed.) Crime and justice: A review of research (Vol. 32 pp. 373-450). Chicago: University of Chicago Press. Pratt, T., Cullen, F., Blevins, K., Daigle, L., and Madensen, T. (2006). “The empirical status of deterrence theory: A meta-analysis”. In Cullen, F., Wright, J., and Blevins, K. (Eds.) Taking Stock: The Status of Criminological Theory: Advances in Criminological Theory (Vol. 15 pp. 367-396). New Brunswick, NJ: Transition. Pratt, T., Cullen, F., Blevins, K., Daigle, L., and Unnever, J.D. (2002). “The relationship between attention deficit hyperactivity disorder to crime and delinquency: a meta-analysis”. International Journal of Police Science and Management. 4: 344-360. Pratt, T.C., McGloin, J.M., and Fearn, N.E. (2006). “Maternal cigarette smoking during pregnancy and criminal/deviant behavior: A meta-analysis”. International Journal of Offender Therapy and Comparative Criminology. 50: 672-690. *Prozesky, L. (2009). Parole Bureaucracy: The Decision Making Process of Paroling Authorities. Ph.D. dissertation, Southern Illinois University Carbondale, Carbondale, IL. *Pruitt, C.R. and Wilson, J.Q. (1983). “A longitudinal study of the effect of race on sentencing”. Law & Society Review. 17: 613-636. Riksheim, E. and Chermak, S. (1993). “Causes of police behavior revisited”. Journal of Criminal Justice. 21: 353-382. *Robinson, A.L. and Chandek, M.S. (2000). “The domestic violence arrest decision: Examining demographic, attitudinal, and situational variables”. Crime & Delinquency. 46: 18-37. *Rodriquez, S.F., Curry, T.R., and Lee, G. (2006). “Gender differences in criminal sentencing: Do effects vary across violent, property, and drug offenses?”. Social Science Quarterly. 87: 318-339.

175

Rosenthal, R. (1991). Meta-analytic procedures for social research. Newbury Park, CA: Sage Productions Inc. *Rydberg, J., and Terrill, W. (2010). “The effect of higher education on police behavior”. Police Quarterly. 13: 92-120. Sampson, R.J., Raudenbush, S.W., and Earls, F. (1997). “Neighborhoods and violent crime: A multileveled study of collective efficacy”. Science. 277: 918-924. Scheingold, S. (1984) The politics of law and order: Street crime and public policy. New York: Longman. *Schram, P.J., Koons-Witt, B.A., Williams, F.P., and McShane, M.D. (2006). “Supervision strategies and approaches for female parolees: Examining the link between unmet needs and parolee outcome”. Crime & Deliquency. 52: 450-471. *Schwaner, S. (1998). “Patterns of violent specialization: Predictors of recidivism for a cohort of parolees”. American Journal of Criminal Justice. 23: 1-17. Sealock, M.D. and Simpson, S.S. (1998). “Unraveling bias in arrest decisions: The role of juvenile offender type-scripts”. Justice Quarterly. 15: 427-457. Shaffer, D. K. (2006). Reconsidering drug court effectiveness: A meta-analysis. Ph.D. dissertation, Department of Criminal Justice, University of Cincinnati, Cincinnati, OH. Sherman, L. (1980). “Causes of police behavior: The current state of quantitative research”. Journal of Research in Crime and Delinquency. 17: 69-100. Smith, D. (1984). “The organizational context of legal control”. Criminology. 22: 19-38. Smith, D., Visher, C., and Davidson, L. (1984). “Equity and discretionary justice: The influence of race on arrest decisions”. Journal of Criminal Law and Criminology. 75: 234-249. *Smith, M.R., Makarios, M., and Alpert, G.P. (2006). “Differential suspicion: Theory specification and gender effects in the traffic stop context”. Justice Quarterly. 23: 271-295. Sobol, J.J. (2010). “Social ecology and police discretion: The influence of district crime, cynicism, and workload on the vigor of police response”. Journal of Criminal Justice. 38: 481-488. *Spohn, C. (1990A). “Decision making in sexual assault cases: Do black and female judges make a difference?”. Women & Criminal Justice. 2: 83-105. *Spohn, C. (1990B). “The sentencing decisions of black and white judges: Expected and unexpected similarities”. Law & Society Review. 24: 1197-1216.

176

Spohn, C. (2000). “Thirty years of sentencing reform: the quest for a racially neutral sentencing process”. Criminal Justice: The National Institute of Justice Journal. 3: 427-501. *Spohn, C. and DeLone, M. (2000). “When does race matter? An analysis of the conditions under which race affects sentence severity”. Sociology of Crime, Law, and Deviance. 2: 3-37 *Spohn, C., Gruhl, J., and Welch, S. (1981-1982). “The effect of race on sentencing: A re-examination of an unsettled question”. Law & Society Review. 16: 71-88. *Spohn, C. and Holleran, D. (2000). “The imprisonment penalty paid by young, unemployed black and Hispanic male offenders”. Criminology, 38: 281-306. *Spohn, C. and Spears. J. (1997). “Gender and case processing”. Women & Criminal Justice. 8: 29-59. *Steen, S., Engen, R.L., and Gainey, R.R. (2005). “Images of danger and culpability: Racial stereotyping, case processing, and criminal sentencing”. Criminology. 43: 435-468. *Steffensmeier, D. and Britt, C. (2001). “Judges race and judicial decision making: Do black judges sentence differently?”. Social Science Quarterly. 82: 749-764 *Steffensmeier, D. and Demuth, S. (2001). “Ethnicity and judges’ sentencing decisions: Hispanic-black-white comparisons”. Criminology. 39: 145-178. *Steffensmeier, D. and Demuth, S. (2006). “Does gender modify the effects of race—ethnicity on criminal sanctioning? Sentences for male and female and white, black, and Hispanic defendants”. Journal of Quantitative Criminology. 22: 241-261. *Steffensmeier, D. and Herbert, C. (1999). “Women and men policymakers: Does the judge’s gender affect the sentencing of criminal defendants?”. Social Forces. 77: 1163-1196. *Steffensmeier, D., Ulmer, J. and Kramer, J. (1998). “The interaction of race, age, and gender in criminal sentencing: The punishment cost of being young, black, and male”. Criminology. 36: 763-798. *Steen, S. and Opsal, T. (2007). “‘Punishment on the installment plan’: Individual-level predictors of parole revocation in four states”. The Prison Journal. 87: 344-366. *Steiner, B., Travis, L., Makarios, M., and Meade, B. “Examing the effect of sanctions on offender recidivism”. Unpublished manuscript. *Steiner, B., Travis, L.F., Makarios, M.D., and Meade, B. (2011). “Short-term effects of sanctioning reform on parole officers’ revocation decisions”. Law and Society Review. 45: 371-400.

177

*Tartaro, C. and Sedemaier, C.M. (2009). “A tale of two counties: the impact of pretrial release, race, and ethnicity upon sentencing decisions”. Criminal Justice Studies. 22: 203-221. Tenback, D.E., van Harten, P.N., van Os, J. (2009). “Non-therapeutic risk factors for onset of tardive dyskinesia in schizophrenia: A meta-analysis”. Movement Disorders. 24: 2309-2315. *Teplin, L.A. (2000). “Keeping the peace: Police discretion and mentally ill persons”. National Institute of Justice Journal. 8-15. *Tillyer, R. (2008). Social Conditioning of Police Officers: Exploring the Interactive Effects of Driver Demographics on Traffic Stop Outcomes. Ph.D. Dissertation, Division of Criminal Justice, University of Cincinnati. Travis, J. and Lawrence, S. (2002). Beyond Prison Gates: The State of Parole in America. Urban Institute Justice Policy Center: Washington, DC. Ulmer, J.T. and Johnson, B.D. (2004). “Sentencing in context: A multilevel analysis”. Criminology. 42: 137-177. Ulmer, J.T., Kurlychek, M.C., and Kramer, J.H. (2007). “Prosecutorial discretion and the imposition of mandatory minimum sentences”. Journal of Research in Crime and Delinquency. 44: 427-458. *Unnever, J.D. (1982). “Direct and organizational discrimination in the sentencing of drug offenders”. Social Problems. 30: 212-225. *Van Voorhis, P., Spruance, L.M., Ritchey, P.N., Listwan, S.J., and Seabrook, R. (2004). “The Georgia cognitive skills experiment: A replication of reasoning and rehabilitation”. Criminal Justice and Behavior. 31: 282-305. *Viehe, M.E. (2003). Criminal Justice Response to Domestic Violence: A Closer Look at Los Angeles County, 1995-1998. Ph.D. Dissertation, Department of Sociology, University of Southern California *Walker, J.T., Hartley, R.D., Maddan, S., VanHouten, A.C., and Ervin-McLarty, G. (2004). Sentencing Practices Under the Arkansas Sentencing Guidelines. Little Rock, AR: Arkansas Crime Information Center. Walker, S. (1992). “Origins of the contemporary criminal justice paradigm: The American Bar Foundation survey, 1953-1969”. Justice Quarterly. 9: 47-75 Walters, G.D. (1992). “A meta-analysis of the gene-crime relationship”. Criminology. 30: 595-613. *Walsh, A. (1990). “Twice labeled: The effect of psychiatric labeling on the sentencing of sex offenders”. Social Problems. 37: 375-389.

178

*Wang, X. and Mears, D.P. (2010). “A multilevel test of minority threat effects on sentencing”. Journal of Quantitative Criminology. 26: 191-215. Weisburd, D., Telep, C.W., Hinkle, J.C., and Eck, J.E. (2010). “Is problem-oriented policing effective in reducing crime and disorder?” Criminology & Public Policy. 9: 139-172 *Weisburd, D., Waring, E., and Wheeler, S. (1990). “Class, status, and the punishment of white-collar criminals”. Law & Social Inquiry. 15: 223-243. Wells, E. (2009) “Uses of meta-analysis in criminal justice research: A quantitative review”. Justice Quarterly. 26: 268-294. Whatley, M.A. (1996). “Victim characteristics influencing attributions of responsibility of rape victims: A meta-analysis”. Aggression and Violent Behavior. 1: 81-95. *Wilmot, K.A. and DeLone, M.A. (2010). “Sentencing of native Americans: A multistage analysis under the Minnesota sentencing guidelines”. Journal of Ethnicity in Criminal Justice. 8: 151-180. Wilson, J.Q. (1968). Varieties of police behavior. Cambridge, MA: Harvard University Press. Wolf, F. M. (1986). Meta-analysis: Quantitative methods for research synthesis. Beverly Hills, CA: Sage Publications. *Winfree, L.T., Wooldredge, J., Sellers, C.S., and Ballard, V.S. (1990). “Parole survival and legislated change: A before/after study of parole revocation decision making”. Justice Quarterly. 7: 151-173. *Withrow, B.L. (2007). “Race-based policing: A descriptive analysis of the Wichita stop study”. Police Practice and Research: An International Journal. 5: 223-240. *Wooldredge, J. (1998). “Analytical rigor in studies of disparities in case processing”. Journal of Quantitative Criminology. 14: 155-179. Worden, R. (1989). “Situational and attitudinal explanations of police behavior: A theoretical reappraisal and empirical assessment”. Law and Society Review. 23: 667-711. Worden, R. and Shepard, R. (1996). “Demeanor crime and police behavior: A reexamination of Police Services Study data”. Criminology. 34: 83-105. *Wordes, M., and Bynum, T.S. (1995). “Policing juveniles: Is there bias against youths of color?” In Kempf-Leonard, K., Pope, C.E., and Feyerherm, W. (eds). Minorities in Juvenile Justice. Thousand Oaks, CA: Sage Publications. *Wu. J. (2009). Immigration, Race/Ethnicity, and Differential Sentencing: An Examination of the Effects of Citizenship Status on Sentencing Outcomes in Federal Court. Ph.D. Dissertation, University of Nebraska, Omaha, NE.

179

Yeaton, W. and Wortman, P. (1993). “On the reliability of meta-analytic reviews: The role of intercoder agreement”. Evaluation Review. 17: 292-309. Zhao, J. and Hassell, K. (2005). “Policing styles and organizational priorities: Retesting Wilson’s theory of local political culture”. Police Quarterly. 8: 411-430. Zhao, J., Ren, L., and Lovrich, N. (2010). “Wilson’s theory of local political culture revisited in today’s police organizations: Findings from longitudinal panel study”. Policing: An International Journal of Police Strategies and Management. 33: 287-304.

180

APPENDIX A

Arrest Coding Guide: (All missing data coded as 99) I. STUDY INFORMATION: A. Identification Number: [STUDID] ________ B. Study accepted or rejected? (yes=1; no=0) [INCLUD] ________ C. Eligibility criteria 1. Study is examining arrest measured dichotomously? 1=YES 0=NO 2. Includes at least one measure of offense severity? 1=YES 0=NO 3. Includes at least one measure of offender characteristics? 1=YES 0=NO 4. Includes at least one measure of officer characteristics? 1=YES 0=NO Notes [DATASET]: _____________________________________________________________ ____________________________________________________________________________________________________________________________________________________________   D. Article author(s): _____________________________________________________________ E. Title: ______________________________________________________________________ F. Author affiliation: ____________________________________________________________ [AFFIL] __________ 1 = University 2 = State Agency 3 = Federal Agency 4 = Mixed 5 = Other _________________ 99 = Missing G. Author discipline: 1 = Criminal Justice/Criminology [DISCIP] _________ 2 = Psychology 3 = Sociology 4 = Social Work 5 = Mixed 6 = Other _____________

99 = Missing H. Coder: 1 = Daniel Lytle [CODER] _________

181

0 = Other 99 = Missing I. Funding Agency: _____________________________________________________________ 1 = Unfunded [FUND] __________ 2 = Independent agency/foundation funded 3 = State funded 4 = Federal funded 5 = Other ______________ 99 = Missing J. Publication year: [PUBYR] _________ Decade: [PUBDEC] ________ 1 = 1950 2 = 1960 3 = 1970 4 = 1980 5 = 1990 6 = 2000 7 = 2010 99 = Missing K. Geographic region of the study: [GEOREG] ________ 1 = Northeast 2 = South 3 = Midwest 4 = Southwest 5 = Pacific/Northwest 6 = Mixed 99 = Missing L. Data collection year: [START] _________ [FINISH] _________ Data gathered decade: [DATAGA] _______ 1 = 1950 2 = 1960 3 = 1970 4 = 1980 5 = 1990 6 = 2000 7 = 2010 99 = Missing

182

M. Publication type: [PUBTYP] ________ 1 = Journal 2 = Doctoral dissertation 3 = Book 4 = Book chapter 5 = Federal report 6 = State or local report 7 = Conference paper 8 = Other ________________ 99 = Missing O. Number of police agencies surveyed: [AGENUM] _______ II. SAMPLE DEMOGRAPHICS: A. Offense specific:           Is  the  current  study  limited  to  a  type  of  offense  (1=YES,  0=NO)    [TYPOFF0]  __________     Specific  type  of  offense:             [TYPOFF1]  __________     1 = Drug offenses 2 = Property offenses 3 = Violent offenses 4 = DUI 5 = Mixed 6 = Other _______________ 99 = Missing/na Percent arrested [PERARR] ________ B. Age: Suspects are: [SUSAGE] ________ 1 = Exclusively adults 2 = Exclusively juveniles 3 = Mainly adults (over 80%) 4 = Mainly juveniles (over 80%) 5 = Mixed 99 = Missing Mean age of suspects [MAGESU] ______

183

C. Gender Suspects are: [SUSGEN] ________ 1 = Exclusively male 2 = Exclusively female 3 = Mainly male (over 80%) 4 = Mainly female (over 80%) 5 = Mixed 99 = Missing Percent Female suspect: [PRFSUS] _______ Officers are: [OFFGEN] ________ 1 = Exclusively male 2 = Exclusively female 3 = Mainly male (over 80%) 4 = Mainly female (over 80%) 5 = Mixed 99 = Missing Percent Female officer: [PRFOFF] _______ D. Race Suspects are: [SUSRAC] ________ 1 = Exclusively White 2 = Exclusively Black 3 = Mainly White (over 80%) 4 = Mainly Black (over 80%) 5 = Mixed 99 = Missing Percent Suspect Non-white: [PRNWS] _________ Officers are: [OFFRAC] ________ 1 = Exclusively White 2 = Exclusively Black 3 = Mainly White (over 80%) 4 = Mainly Black (over 80%) 5 = Mixed 99 = Missing Percent Officer Non-white [PRNWO] ________

184

Race is measured as: 1 = White/Non-white [RACMEA] _______ 2 = White/Black 3 = Other _________________ 99 = Missing/not measured E. Ethnicity Suspects are: [ETHSUS] ________ 1 = Exclusively Hispanic 2 = Exclusively Non-Hispanic 3 = Mainly Hispanic (Over 80%) 4 = Mainly Non-Hispanic (Over 80%) 5 = Mixed 99 = Missing Percent Hispanic: [PRHISS] _________ Officers are: [ETHOFF]_________ 1 = Exclusively Hispanic 2 = Exclusively Non-Hispanic 3 = Mainly Hispanic (Over 80%) 4 = Mainly Non-Hispanic (Over 80%) 5 = Mixed 99 = Missing Percent Hispanic: [PRHISO] _________ Ethnicity is measured as: [ETHMEA] _______ 1 = Hispanic/Non-Hispanic 2 = Other ________________ 99 = Missing F. Offense Severity Sample is: [SERFEL] ________ 1 = Exclusively felonies 2 = Exclusively misdemeanors 3 = Mainly felonies (Over 80%) 4 = Mainly misdemeanors (Over 80%) 5 = Mixed 99 = Missing

Percent felony: [PRFEL] __________

Evidence:

185

Was the amount of evidence measured? (1=YES, 0=NO) [EVIREC] ________ Citizen intoxication: Did the study measure citizen intoxication? (1=YES, 0=NO) [TOXREC] ________ Percent of citizens intoxicated: [PRTOX] _________ Weapon Use: Was the weapon use in the encounter recorded? (1=YES, 0=NO) [WEAPREC] ______ Percent of encounters involving a weapon: [PRWEAP] ________ Victim request arrest: Was victim preference recorded? (1=YES, 0=NO) [VPREFREC] ______ Percent of encounters where victim requested arrest: [PRVPREF] _______ Total number of offense severity measures: [SERTOT] ________ G. Demeanor Suspect demeanor:

Was suspect demeanor recorded? (1=YES, 0=NO) [DEMREC] _______

Percent non-deferential: [PRDEM] ________ Interaction-phase crime: Was IPC recorded? (1=YES, 0=NO) [IPCREC] _________ Percent of encounters with IPC: [PRIPC] __________ Suspect compliance: Was suspect compliance recorded? (1=YES, 0=NO) [COMREC] _______ Percent of encounters where suspect was non-compliant [PRCOM] _________ F. Mental illness Mental illness: Was suspect mental illness recorded? (1=YES, 0=NO) [MIREC] _________ Percent mentally ill: [PRMENIL] _______ G. Other officer characteristics Education: Is officer education recorded? (1=YES, 0=NO) [EDREC] _________

186

Officers are: [COLED] _________ 1 = Exclusively college educated 2 = Exclusively high school/GED educated 3 = Mainly college educated (Over 80%) 4 = Mainly high school/GED educated (Over 80%) 5 = Mixed 99 = Missing Percent college educated: [PRCOLED] _______ Experience: Was a measure of officer experience recorded? (1=YES, 0=NO) [XPREC] _________ H. Organizational characteristics: Were organizational level variables recorded? (1=YES, 0=NO) [ORGREC] ________ Mean number of organizational levels: [MNLEV] _________ Mean agency size: [MNSIZE] ________ I. Community characteristics:   Were  community  level  variables  recorded?  (1=YES, 0=NO) [UNITREC] _______ III. METHODOLOGICAL RIGOR ASSESSMENT A. Control Variables included: (1=YES, 0=NO) [CONVAR] _______ B. Systematic social observation used: (1=YES, 0=NO) [SSOUSE] ________ C.  Includes  interaction-­‐effects:  (1=YES, 0=NO) [INEFUSE] _______ D.  How  long  was  the  study  conducted  in  months:         [LENGTH]  ___________    E.  Was  data  collection  continuous:   (1=YES, 0=NO)       [CONTIN]  ____________    IV.  SAMPLE  SIZE    A.  Sample  size:                 [SIZE]  ________________              

187

   V.  EFFECT  SIZE  DATA    A.  Suspect  Race:    

Odds  ratio  (logistic  regression)  of  race  and  arrest       [ORRACE]  ___________  Inverse  variance:               [IVRACE]  ____________  

    Type  of  statistical  test  for  race  and  arrest         [RACETEST]  _________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Race and arrest effect size: [ESRACE] ________  B.  Suspect  Gender:  

Odds  ratio  (logistic  regression)  of  gender  and  arrest     [ORGEN]  ___________  Inverse  variance:               [IVGEN]  ____________  

    Type  of  statistical  test  for  gender  and  arrest       [GENTEST]  _________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Gender and arrest effect size: [ESGEN] ________ C. Suspect Age:

Odds  ratio  (logistic  regression)  of  age  and  arrest       [ORAGE]  ___________  Inverse  variance:               [IVAGE]  ____________  

    Type  of  statistical  test  for  age  and  arrest         [AGETEST]  _________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Race and arrest effect size: [ESAGE] ________ D. Suspect Ethnicity:

Odds  ratio  (logistic  regression)  of  ethnicity  and  arrest     [ORHISP]  ___________  Inverse  variance:               [IVHISP]  ____________  

188

      Type  of  statistical  test  for  ethnicity  and  arrest       [HISPTEST]  _________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Ethnicity and arrest effect size: [ESHISP] ________ E.  Officer  Race:    

Odds  ratio  (logistic  regression)  of  officer  race  and  arrest     [ORORAC]  ___________  Inverse  variance:               [IVORAC]  ____________  

    Type  of  statistical  test  for  officer  race  and  arrest       [ORACTEST]  _________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Officer race and arrest effect size: [ESORAC] ________  F.  Officer  Gender:  

Odds  ratio  (logistic  regression)  of  officer  gender  and  arrest   [OROGEN]  ___________  Inverse  variance:               [IVOGEN]  ____________  

    Type  of  statistical  test  for  officer  gender  and  arrest     [OGENTEST]  ________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Officer gender and arrest effect size: [ESOGEN] ________ G. Officer Ethnicity:

Odds  ratio  (logistic  regression)  of  officer  ethnicity  and  arrest   [OROHIS]  ___________  Inverse  variance:               [IVOHIS]  ____________  

    Type  of  statistical  test  for  officer  ethnicity  and  arrest     [OHISTEST]  _________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Officer ethnicity and arrest effect size: [ESOHIS] ________

189

H. Offense Severity (felony v. misdemeanor):

Odds  ratio  (logistic  regression)  of  felony  and  arrest     [ORFELO]  ___________  Inverse  variance:               [IVFELO]  ____________  

    Type  of  statistical  test  for  felony  and  arrest       [FELOTEST]  _________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Felony and arrest effect size: [ESFELO] ________ I. Evidence:

Odds  ratio  (logistic  regression)  of  evidence  and  arrest     [OREVID]  ___________  Inverse  variance:               [IVEVID]  ____________  

    Type  of  statistical  test  for  evidence  and  arrest       [EVIDTEST]  _________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Evidence and arrest effect size: [ESEVID] ________ J. Citizen Intoxication

Odds  ratio  (logistic  regression)  of  intoxication  and  arrest   [ORTOX]  ___________  Inverse  variance:               [IVTOX]  ____________  

    Type  of  statistical  test  for  intoxication  and  arrest       [TOXTEST]  _________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Intoxication and arrest effect size: [ESTOX] ________ K. Weapon Use

Odds  ratio  (logistic  regression)  of  weapon  use  and  arrest   [ORWU]  ___________  Inverse  variance:               [IVWU]  ____________  

    Type  of  statistical  test  for  weapon  use  and  arrest       [WUTEST]  _________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated

190

4: descriptive statistic 99: MISSING Weapon use and arrest effect size: [ESWU] ________ L. Victim preference for arrest

Odds  ratio  (logistic  regression)  of  victim  preference  and  arrest   [ORVPREF]___________  Inverse  variance:               [IVVPREF]____________  

    Type  of  statistical  test  for  victim  preference  and  arrest     [VPREFTEST]________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Victim preference and arrest effect size: [ESVPREF]________ M. Demeanor

Odds  ratio  (logistic  regression)  of  demeanor  and  arrest     [ORSDEM]  ___________  Inverse  variance:               [IVSDEM]  ____________  

    Type  of  statistical  test  for  demeanor  and  arrest       [SDEMTEST]  ________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Demeanor and arrest effect size: [ESSDEM] ________ N. Interaction Phase Crime

Odds  ratio  (logistic  regression)  of  IPC  and  arrest       [ORIPC]  ___________  Inverse  variance:               [IVIPC]  ____________  

    Type  of  statistical  test  for  IPC  and  arrest         [IPCTEST]  _________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING IPC and arrest effect size: [ESIPC] ________ O. Suspect Compliance

Odds  ratio  (logistic  regression)  of  compliance  and  arrest   [ORSCOM]  ___________  Inverse  variance:               [IVSCOM]  ____________  

191

      Type  of  statistical  test  for  compliance  and  arrest       [SCOMTEST]_________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Compliance and arrest effect size: [ESSCOM] ________ P. Suspect Mentally Ill

Odds  ratio  (logistic  regression)  of  mentally  ill  and  arrest     [ORSMI]  ___________  Inverse  variance:               [IVSMI]  ____________  

    Type  of  statistical  test  for  mentally  ill  and  arrest       [SMITEST]  _________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Mentally ill and arrest effect size: [ESSMI] ________ Q. Officer Education

Odds  ratio  (logistic  regression)  of  officer  education  and  arrest   [OROED]  ___________  Inverse  variance:               [IVOED]  ____________  

    Type  of  statistical  test  for  officer  education  and  arrest     [OEDTEST]  _________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Officer education and arrest effect size: [ESOED] ________ R. Officer Experience

Odds  ratio  (logistic  regression)  of  experience  and  arrest     [ORXP]  ___________  Inverse  variance:               [IVXP]  ____________  

    Type  of  statistical  test  for  experience  and  arrest       [XPTEST]  _________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Experience and arrest effect size: [ESXP] ________

192

S. Rank Structure (number of organizational levels)

Odds  ratio  (logistic  regression)  of  rank  structure  and  arrest   [ORLEVE]  ___________  Inverse  variance:               [IVLEVE]  ____________  

    Type  of  statistical  test  for  rank  structure  and  arrest     [LEVETEST]  _________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Rank structure and arrest effect size: [ESLEVE] ________ T. Agency Size

Odds  ratio  (logistic  regression)  of  agency  size  and  arrest     [ORSIZE]  ___________  Inverse  variance:               [IVSIZE]  ____________  

    Type  of  statistical  test  for  agency  size  and  arrest       [SIZETEST]  _________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Agency size and arrest effect size: [ESSIZE] ________ U. Community Characteristics (crime rate)

Odds  ratio  (logistic  regression)  of  community  and  arrest   [ORCOMM]___________  Inverse  variance:               [IVCOMM]____________  

    Type  of  statistical  test  for  community  and  arrest       [COMMTEST]  ________  

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Community and arrest effect size: [ESCOMM] _______

193

Sentencing Coding Guide: (All missing data coded as 99) I. STUDY INFORMATION: A. Identification Number: [STUDID] ________ B. Study accepted or rejected? (YES=1; NO=0) [INCLUD] ________ C. Eligibility criteria 1. Study is examining sentencing measured dichotomously? 1=YES 0=NO 2. Includes at least one measure of offense severity? 1=YES 0=NO 3. Includes at least one measure of offender characteristics? 1=YES 0=NO 4. Includes at least one measure of officer characteristics? 1=YES 0=NO [DATASET]: __________________________________________________________________ ____________________________________________________________________________________________________________________________________________________________ D. Article author(s): _____________________________________________________________ E. Title: ______________________________________________________________________ F. Author affiliation: ____________________________________________________________ [AFFIL] __________ 1 = University 2 = State Agency 3 = Federal Agency 4 = Mixed 5 = Other _________________ 99 = Missing G. Author discipline: 1 = Criminal Justice/Criminology [DISCIP] _________ 2 = Psychology 3 = Sociology 4 = Social Work 5 = Mixed 6 = Other _____________

99 = Missing H. Coder: 1 = Daniel Lytle [CODER] _________ 0 = Other Fill in name __________________ 99 = Missing

194

I. Funding Agency: _____________________________________________________________ 1 = Unfunded [FUND] __________ 2 = Independent agency/foundation funded 3 = State funded 4 = Federal funded 5 = Other ______________ 99 = Missing J. Publication year: [PUBYR] _________ Decade: [PUBDEC] ________ 1 = 1950 2 = 1960 3 = 1970 4 = 1980 5 = 1990 6 = 2000 7 = 2010 99 = Missing K. Geographic region of the study: [GEOREG] ________ 1 = Northeast 2 = South 3 = Midwest 4 = Southwest 5 = Pacific/Northwest 6 = Mixed 99 = Missing L. Data collection year: [START] _________ [FINISH] _________ Data gathered decade: [DATAGA] _______ 1 = 1950 2 = 1960 3 = 1970 4 = 1980 5 = 1990 6 = 2000 7 = 2010 99 = Missing

195

M. Publication type: [PUBTYP] ________ 1 = Journal 2 = Doctoral dissertation 3 = Book 4 = Book chapter 5 = Federal report 6 = State or local report 7 = Conference paper 8 = Other ________________ 99 = Missing O. Number of courts surveyed: [AGENUM] _______ II. SAMPLE DEMOGRAPHICS: A. Offense specific: Is the current study limited to a type of offense (1=YES, 0=NO) [TYPOFF0] _______ Specific type of offense: [TYPOFF1] _______ 1 = Drug offenses 2 = Property offenses 3 = Violent offenses 4 = DUI 5 = Mixed 6 = Other _______________ 99 = Missing/na Percent sentenced to prison [PERARR] ________ B. Age: Defendants are: [DEFAGE] ________ 1 = Exclusively adults 2 = Exclusively juveniles 3 = Mainly adults (over 80%) 4 = Mainly juveniles (over 80%) 5 = Mixed 99 = Missing Mean age of defendants [MAGEDF] ______

196

C. Gender Defendants are: [DEFGEN] ________ 1 = Exclusively male 2 = Exclusively female 3 = Mainly male (over 80%) 4 = Mainly female (over 80%) 5 = Mixed 99 = Missing Percent Female defendants: [PRFDEF] _______ Judges are: [JUDGEN] ________ 1 = Exclusively male 2 = Exclusively female 3 = Mainly male (over 80%) 4 = Mainly female (over 80%) 5 = Mixed 99 = Missing Percent Female officer: [PRFJUD] _______ D. Race Defendants are: [DEFRAC] ________ 1 = Exclusively White 2 = Exclusively Black 3 = Mainly White (over 80%) 4 = Mainly Black (over 80%) 5 = Mixed 99 = Missing Percent Defendants Non-white: [PRNWD] ________

Judges are: [JUDRAC] ________ 1 = Exclusively White 2 = Exclusively Black 3 = Mainly White (over 80%) 4 = Mainly Black (over 80%) 5 = Mixed 99 = Missing Percent judges Non-white [PRNWJ] ________

197

Race is measured as: 1 = White/Non-white [RACMEA] _______ 2 = White/Black 3 = Other _________________ 99 = Missing/not measured E. Ethnicity Defendants are: [ETHDEF] ________ 1 = Exclusively Hispanic 2 = Exclusively Non-Hispanic 3 = Mainly Hispanic (Over 80%) 4 = Mainly Non-Hispanic (Over 80%) 5 = Mixed 99 = Missing Percent Hispanic defendants: [PRHISD] _________ Judges are: [JUDHIS] _________ 1 = Exclusively Hispanic 2 = Exclusively Non-Hispanic 3 = Mainly Hispanic (Over 80%) 4 = Mainly Non-Hispanic (Over 80%) 5 = Mixed 99 = Missing Percent Hispanic judge: [PRHISJ] _________ Ethnicity is measured as: [ETHMEA] _______ 1 = Hispanic/Non-Hispanic 2 = Other ________________ 99 = Missing F. Offense Severity Sample is: [SERFEL] ________ 1 = Exclusively felonies 2 = Exclusively misdemeanors 3 = Mainly felonies (Over 80%) 4 = Mainly misdemeanors (Over 80%) 5 = Mixed 99 = Missing

Percent felony: [PRFEL] __________

Evidence: Was the amount of evidence measured? (1=YES, 0=NO) [EVIREC] ________

198

Weapon Use: Was the weapon use in the case recorded? (1=YES, 0=NO) [WEAPREC] ______ Percent of cases involving a weapon: [PRWEAP] ________ Drug and Alchohol use: Was drug and/or alcohol use recorded? (1=YES, 0=NO) [DRUGREC] ______ Percent of cases involving drugs or alcohol [PRDRUG] _______ Prior record/criminal history

Criminal history/prior record recorded? (1=YES, 0=NO) [PRIORREC]______ How was prior record measured? 1= 1 or more or none [PRIORMEA] _____ 2= discrete scale 3= count 4= other ________________

Total number of offense severity measures: [SERTOT] ________ F. Mental illness Mental illness: Was defendant mental illness recorded? (1=YES, 0=NO) [MIREC] _________ Percent mentally ill: [PRMENIL] _______ G. Other court characteristics Experience: Was a measure of judicial experience recorded? (1=YES, 0=NO) [XPREC] _________ Mean experience [XPMEAN] _______ Federal vs. local Cases are: [FVLREC] ________ 1 = Exclusively Federal 2 = Exclusively local 3 = Mostly Federal (over 80%) 4 = Mostly local (over 80%) 5 = Mixed 99 = Missing Percent Federal [PRFVL] _________

199

Mode of conviction (plea v. trail conviction) Cases are: [PVTCREC] _______ 1 = Exclusively pleas 2 = Exclusively trial convictions 3 = Mostly pleas (80%) 4 = Mostly trial convictions (80%) 5 = Mixed 99 = Missing Percent plea [PRPVTC] ________ Counsel type Defendants had: [PVPREC] _______ 1 = Exclusively public 2 = Exclusively private 3 = Mostly public (over 80%) 4 = Mostly private (over 80%) 5 = Mixed 99 = Missing Percent public defenders [PRPVP] __________ H. Organizational characteristics: Were organizational level variables recorded? (1=YES, 0=NO) [ORGREC] ________ Mean court size: [MNSIZE] ________ I. Community characteristics: Were community level variables recorded? (1=YES, 0=NO) [UNITREC] _______ III. METHODOLOGICAL RIGOR ASSESSMENT A. Control Variables included: (1=YES, 0=NO) [CONVAR] _______ B. Includes interaction-effects: (1=YES, 0=NO) [INEFUSE] _______ C. How long was the study conducted in months: [LENGTH] _______ D. Was data collection continuous: (1=YES, 0=NO) [CONTIN] ________ IV. SAMPLE SIZE A. Sample size: [SIZE] ___________ V. EFFECT SIZE DATA

200

A. Defendant Race: Odds ratio (logistic regression) of race and sentence [ORRACE] ________ Inverse variance: [IVRACE] ________

Type of statistical test for race and sentence [RACETEST]______

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Race and sentence effect size: [ESRACE] ________ B. Defendant Gender:

Odds ratio (logistic regression) of gender and sentence [ORGEN]_________ Inverse variance: [IVGEN] _________

Type of statistical test for gender and sentence [GENTEST] _______

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Gender and sentence effect size: [ESGEN] ________ C. Defendant Age:

Odds ratio (logistic regression) of age and sentence [ORAGE] _______ Inverse variance: [IVAGE] _________

Type of statistical test for age and sentence [AGETEST] ______

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Race and sentence effect size: [ESAGE] ________ D. Defendant Ethnicity:

Odds ratio (logistic regression) of ethnicity and sentence [ORHISP] ________ Inverse variance: [IVHISP]__________

Type of statistical test for ethnicity and sentence [HISPTEST] _______

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING

201

Ethnicity and sentence effect size: [ESHISP] ________ E. Judge Race:

Odds ratio (logistic regression) of judge’s race and sentence [ORJRAC] ________ Inverse variance: [IVJRAC] _________

Type of statistical test for judge’s race and sentence [JRACTEST] ______

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Judge’s race and sentence effect size: [ESJRAC] ________ F. Judge Gender:

Odds ratio (logistic regression) of judge’s gender and sentence [ORJGEN]_________ Inverse variance: [IVJGEN] _________

Type of statistical test for judge’s gender and sentence [JGENTEST]_______

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Judge’s gender and sentence effect size: [ESJGEN] ________ G. Judge’s Ethnicity:

Odds ratio (logistic regression) of judge’s ethnicity and [ORJHIS]_________ sentence Inverse variance: [IVJHIS] _________

Type of statistical test for judge’s ethnicity and sentence [JHISTEST] ______

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Judge’s ethnicity and sentence effect size: [ESJHIS] ________ H. Offense Severity (felony v. misdemeanor):

Odds ratio (logistic regression) of felony and sentence [ORFELO] ________ Inverse variance: [IVFELO] _________

Type of statistical test for felony and sentence [FELOTEST] ______

1: r 5: t test

202

2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Felony and sentence effect size: [ESFELO] ________ I. Evidence:

Odds ratio (logistic regression) of evidence and sentence [OREVID] ________ Inverse variance: [IVEVID] _________

Type of statistical test for evidence and sentence [EVIDTEST] ______

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Evidence and sentence effect size: [ESEVID] ________ J. Weapon Use

Odds ratio (logistic regression) of weapon use and sentence [ORWU] ________ Inverse variance: [IVWU]___________

Type of statistical test for weapon use and sentence [WUTEST] ______

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Weapon use and sentence effect size: [ESWU] ________ K. Experience

Odds ratio (logistic regression) of experience and sentence [ORXP] _________ Inverse variance: [IVXP] __________

Type of statistical test for experience and sentence [XPTEST] ______

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Experience and sentence effect size: [ESXP] ________ L. Court size

Odds ratio (logistic regression) of court size and sentence [ORSIZE] _________ Inverse variance: [IVSIZE]__________

203

Type of statistical test for court size and sentence [SIZETEST]_______ 1: r 5: t test

2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Court size and sentence effect size: [ESSIZE] ________ M. Community Characteristics

Odds ratio (logistic regression) of community and sentence [ORCOMM]_______ Inverse variance: [IVCOMM]________

Type of statistical test for community and sentence [COMMTEST] ________

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Community and sentence effect size: [ESCOMM] _______ N. Federal v. Local

Odds ratio (logistic regression) of jurisdiction and sentence [ORJUR]________ Inverse variance: [IVJUR]_________

Type of statistical test for jurisdiction type and sentence [JURTEST]________

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Jurisdiction type and sentence effect size: [ESJUR] ________ O. Counsel Type (public v. private)

Odds ratio (logistic regression) of counsel type and sentence [ORPVP]________ Inverse variance: [IVPVP]_________

Type of statistical test for counsel type and sentence [PVPTEST] ________

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Counsel type and sentence effect size: [ESPVP] _______ P. Plea type (plea v. trial conviction)

204

Odds ratio (logistic regression) of plea type and sentence [ORPVTC] ________ Inverse variance: [IVPVTC] ________

Type of statistical test for plea type and sentence [PVTCTEST] ______

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Plea type and sentence effect size: [ESPVTC] _______ Q. Drug and/or Alcohol use

Odds ratio (logistic regression) of Drug/alcohol use [ORDRUG] ______ Inverse variance: [IVPVTC]________

Type of statistical test for plea type and sentence [DRUGTEST] _____ 1: r 5: t test

2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Plea type and sentence effect size: [ESDRUG] _______ R. Prior record/Criminal History

Odds ratio (logistic regression) of Drug/alcohol use [ORPRIOR] _______ Inverse variance: [IVPRIOR] _______

Type of statistical test for plea type and sentence [PRIORTEST] _____ 1: r 5: t test

2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Plea type and sentence effect size: [ESPRIOR] _______

205

Parole Revocation Coding Guide: (All missing data coded as 99) I. STUDY INFORMATION: A. Identification Number: [STUDID] ________ B. Study accepted or rejected? (yes=1; no=0) [INCLUD] ________ C. Eligibility criteria 1. Study is examining parole revocation measured dichotomously? 1=YES 0=NO 2. Includes at least one measure of offense severity? 1=YES 0=NO 3. Includes at least one measure of offender characteristics? 1=YES 0=NO 4. Includes at least one measure of officer characteristics? 1=YES 0=NO Data Set [DATASET]: ___________________________________________________________ ____________________________________________________________________________________________________________________________________________________________ D. Article author(s): _____________________________________________________________ E. Title: ______________________________________________________________________ F. Author affiliation: ____________________________________________________________ [AFFIL] __________ 1 = University 2 = State Agency 3 = Federal Agency 4 = Mixed 5 = Other _________________ 99 = Missing G. Author discipline: 1 = Criminal Justice/Criminology [DISCIP] _________ 2 = Psychology 3 = Sociology 4 = Social Work 5 = Mixed 6 = Other _____________

99 = Missing H. Coder: 1 = Daniel Lytle [CODER] _________ 0 = Other If zero, Identify coder __________________ 99 = Missing

206

I. Funding Agency: _____________________________________________________________ 1 = Unfunded [FUND] __________ 2 = Independent agency/foundation funded 3 = State funded 4 = Federal funded 5 = Other ______________ 99 = Missing J. Publication year: [PUBYR] _________ Decade: [PUBDEC] ________ 1 = 1950 2 = 1960 3 = 1970 4 = 1980 5 = 1990 6 = 2000 7 = 2010 99 = Missing K. Geographic region of the study: [GEOREG] ________ 1 = Northeast 2 = South 3 = Midwest 4 = Southwest 5 = Pacific/Northwest 6 = Mixed 99 = Missing L. Data collection year: [START] _________ [FINISH] _________ Data gathered decade: [DATAGA] _______ 1 = 1950 2 = 1960 3 = 1970 4 = 1980 5 = 1990 6 = 2000 7 = 2010 99 = Missing (If data collection crosses decade use starting decade for DATAGA)

207

M. Publication type: [PUBTYP] ________ 1 = Journal 2 = Doctoral dissertation 3 = Book 4 = Book chapter 5 = Federal report 6 = State or local report 7 = Conference paper 8 = Other ________________ 99 = Missing O. Number of police agencies surveyed: [AGENUM] _______ II. SAMPLE DEMOGRAPHICS: A. Offense specific: Is the current study limited to a type of offense (1=YES, 0=NO) [TYPOFF0] _______ Specific type of offense: [TYPOFF1] _______ 1 = Drug offenses 2 = Property offenses 3 = Violent offenses 4 = DUI 5 = Mixed 6 = Other _______________ 99 = Missing/na Percent revocated [PERARR] ________ B. Age: Offenders are: [OFFAGE] ________ 1 = Exclusively adults 2 = Exclusively juveniles 3 = Mainly adults (over 80%) 4 = Mainly juveniles (over 80%) 5 = Mixed 99 = Missing Mean age of offenders [MAGEOF] ______

208

C. Gender Parole officers are: [POGEN] ________ 1 = Exclusively male 2 = Exclusively female 3 = Mainly male (over 80%) 4 = Mainly female (over 80%) 5 = Mixed 99 = Missing Percent Female suspect: [PRFPO] __________ Offenders are: [OFFGEN] ________ 1 = Exclusively male 2 = Exclusively female 3 = Mainly male (over 80%) 4 = Mainly female (over 80%) 5 = Mixed 99 = Missing Percent Female offenders: [PRFOFF] _______ D. Race Parole officers are: [PORAC] ________ 1 = Exclusively White 2 = Exclusively Black 3 = Mainly White (over 80%) 4 = Mainly Black (over 80%) 5 = Mixed 99 = Missing Percent parole officers Non-white: [PRNWPO] _______ Offenders are: [OFFRAC] ________ 1 = Exclusively White 2 = Exclusively Black 3 = Mainly White (over 80%) 4 = Mainly Black (over 80%) 5 = Mixed 99 = Missing Percent Offender Non-white [PRNWO] ________

209

Race is measured as: 1 = White/Non-white [RACMEA] _______ 2 = White/Black 3 = Other _________________ 99 = Missing/not measured E. Ethnicity Parole officers are: [ETHPO] ________ 1 = Exclusively Hispanic 2 = Exclusively Non-Hispanic 3 = Mainly Hispanic (Over 80%) 4 = Mainly Non-Hispanic (Over 80%) 5 = Mixed 99 = Missing Percent Hispanic: [PRHISPO] _______ Offenders are: [ETHOFF] ________ 1 = Exclusively Hispanic 2 = Exclusively Non-Hispanic 3 = Mainly Hispanic (Over 80%) 4 = Mainly Non-Hispanic (Over 80%) 5 = Mixed 99 = Missing Percent Hispanic: [PRHISOF]________ Ethnicity is measured as: [ETHMEA] _______ 1 = Hispanic/Non-Hispanic 2 = Other ________________ 99 = Missing F. Offense Severity Sample is: [SERFEL] ________ 1 = Exclusively felonies 2 = Exclusively misdemeanors 3 = Mainly felonies (Over 80%) 4 = Mainly misdemeanors (Over 80%) 5 = Mixed 99 = Missing

Percent felony: [PRFEL] __________

Evidence: Was the amount of evidence measured? (1=YES, 0=NO) [EVIREC] ________

210

Weapon Use: Was the weapon use in the encounter recorded? (1=YES, 0=NO) [WEAPREC] ______ Percent of encounters involving a weapon: [PRWEAP] ________ Risk:

Criminal History/Risk Recorded? (1=YES, 0=NO) [RISKREC] _______ Need:

Criminogenic Need Recorded ? (1=YES, 0=NO) [NEEDREC] ______ Total number of offense severity measures: [SERTOT] ________ G. Education Education: Is offender’s education recorded? (1=YES, 0=NO) [EDREC] _________ Offenders are: [COLED] _________ 1 = Exclusively college educated 2 = Exclusively high school/GED educated 3 = Mainly college educated (Over 80%) 4 = Mainly high school/GED educated (Over 80%) 5 = Mixed 99 = Missing Percent college educated offenders: [PRCOLED] _______ Parole officer’s education recorded? (1=YES, 0=NO) [POEDREC] _______

Officers are: [COLEDPO] ______ 1 = Exclusively college educated 2 = Exclusively high school/GED educated 3 = Mainly college educated (Over 80%) 4 = Mainly high school/GED educated (Over 80%) 5 = Mixed 99 = Missing Percent college educated parole officers: [PRCOLEDPO] ____ Experience: Was a measure of parole officer experience recorded? [XPREC] _________ (1=YES, 0=NO) Mean officer experience [MNXP] __________

211

H. Employment status: Offenders are: [EMPOREC] ______ 1 = Exclusively employed full time or part time 2 = Exclusively not employed 3 = Mostly employed full or part time (over 80%) 4 = Mostly not employed (over 80%) 5 = Mixed 99 = Missing Percent employed full or part time [PREMP] _________ I. Community characteristics: Were community level variables recorded? (1=YES, 0=NO) [UNITREC] _______ J. Program completion: Were program completion variables recorded? (1=YES, 0=NO) [PCREC] _________ Percent that have completed programming requirements [PRPC] ___________ III. METHODOLOGICAL RIGOR ASSESSMENT A. Control Variables included: (1=YES, 0=NO) [CONVAR] _______ B. Includes interaction-effects: (1=YES, 0=NO) [INEFUSE] _______ C. How long was the study conducted in months: [LENGTH] ________ D. Was data collection continuous: (1=YES, 0=NO) [CONTIN] ________ IV. SAMPLE SIZE A. Sample size: [SIZE] ___________ V. EFFECT SIZE DATA A. Offender Race:

Odds ratio (logistic regression) of race and parole revocation [ORRACE] ________ Inverse variance: [IVRACE] ________

Type of statistical test for race and parole revocation [RACETEST] ______

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Race and parole revocation effect size: [ESRACE] ________

212

B. Offender Gender: Odds ratio (logistic regression) of gender and parole [ORGEN]_________ revocation Inverse variance: [IVGEN] _________

Type of statistical test for gender and revocation [GENTEST] _______

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Gender and parole revocation effect size: [ESGEN] ________ C. Offender Age:

Odds ratio (logistic regression) of age and parole revocation [ORAGE] _________ Inverse variance: [IVAGE] _________

Type of statistical test for age and parole revocation [AGETEST] _______

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Race and parole revocation effect size: [ESAGE] ________ D. Offender Ethnicity:

Odds ratio (logistic regression) of ethnicity and parole [ORHISP]_________ revocation Inverse variance: [IVHISP] _________

Type of statistical test for ethnicity and parole revocation [HISPTEST] ______

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Race and parole revocation effect size: [ESHISP] ________ E. Offense Severity (felony v. misdemeanor):

Odds ratio (logistic regression) of felony and parole revocation [ORFELO] ________ Inverse variance: [IVFELO] _________

213

Type of statistical test for felony and parole revocation [FELOTEST] ______ 1: r 5: t test

2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Felony and parole revocation effect size: [ESFELO] ________ F. Evidence:

Odds ratio (logistic regression) of evidence and parole [OREVID] ________ revocation Inverse variance: [IVEVID] _________

Type of statistical test for evidence and parole revocation [EVIDTEST] ______

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Evidence and parole revocation effect size: [ESEVID] ________ G. Weapon Use

Odds ratio (logistic regression) of weapon use and parole [ORWU] ________ revocation

Inverse variance: [IVWU] __________ Type of statistical test for weapon use and parole revocation [WUTEST] ________

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Weapon use and arrest effect size: [ESWU] ________ H. Parole Officer Education

Odds ratio (logistic regression) of officer education and parole [ORPOED] ________ revocation Inverse variance: [IVPOED] ________

Type of statistical test for officer education and parole [POEDTEST] ______ revocation

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING

214

Officer education and parole revocation effect size: [ESPOED] ________ I. Parole Officer Race

Odds ratio (logistic regression) of officer race and parole [ORPORAC] ______ revocation Inverse variance: [IVPORAC] _______

Type of statistical test for officer race and parole revocation [PORACTEST]_____

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Officer race and parole revocation effect size: [ESPORAC] _______ J. Parole Officer Gender:

Odds ratio (logistic regression) of officer gender and parole [ORPOGEN] ______ revocation Inverse variance: [IVPOGEN] _______

Type of statistical test for officer gender and parole revocation [POGENTEST] ____

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Officer gender and parole revocation effect size: [ESPOGEN] _______ K. Parole Officer Ethnicity:

Odds ratio (logistic regression) of officer ethnicity and parole [ORPOETH] ______ revocation Inverse variance: [IVPOETH] _______

Type of statistical test for officer ethnicity and parole [POETHTEST] ____ revocation

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Officer ethnicity and parole revocation effect size: [ESPOETH] _______ L. Offender Employment Status

Odds ratio (logistic regression) of employment and parole [OREMP] _______ revocation Inverse variance: [IVEMP] _________

215

Type of statistical test for employment and parole revocation [EMPTEST] _______

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Employment and parole revocation effect size: [ESEMP] ________ M. Offender Education

Odds ratio (logistic regression) of education and parole [OREDU] _________ revocation Inverse variance: [IVEDU] _________

Type of statistical test for education and parole revocation [EDUTEST] _______

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Education and parole revocation effect size: [ESEDU] _______ N. Program Completion

Odds ratio (logistic regression) of program completion and [ORPROG] _______ parole revocation Inverse variance: [IVPROG] ________

Type of statistical test for program completion and parole [PROGTEST] ______ revocation

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Program completion and parole revocation effect size: [ESPROG] _______ O. Parole Officer Experience

Odds ratio (logistic regression) of experience and parole [POXP] ___________ revocation Inverse variance: [IVXP] __________

Type of statistical test for experience and parole revocation [XPTEST] ________

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING

216

Experience and parole revocation effect size: [ESXP] ________ P. Community Characteristics

Odds ratio (logistic regression) of community and parole [ORCOMM] _______ revocation Inverse variance: [IVCOMM] _______

Type of statistical test for community and parole revocation [COMMTEST] _____

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Community and parole revocation effect size: [ESCOMM] _______ Q. Criminal History/Risk

Odds ratio (logistic regression) of risk and parole [ORRISK] ________ revocation Inverse variance: [IVRISK] _________

Type of statistical test for risk and parole revocation [RISKTEST] ______

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Risk and parole revocation effect size: [ESRISK] _______ R. Need

Odds ratio (logistic regression) of need and parole [ORNEED] ________ revocation Inverse variance: [IVNEED] ________

Type of statistical test for need and parole revocation [NEEDTEST] ______

1: r 5: t test 2: ANOVA 6: p value not otherwise specified 3: Chi-square 7: Φ calculated 4: descriptive statistic 99: MISSING Need and parole revocation effect size: [ESNEED] _______