A longitudinal analysis of reparative probation and recidivism

This article was downloaded by: [ ]On: 18 September 2012, At: 08:40Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954 Registeredoffice: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Criminal Justice Studies: A CriticalJournal of Crime, Law and SocietyPublication details, including instructions for authors andsubscription information:http://www.tandfonline.com/loi/gjup20

A longitudinal analysis of reparativeprobation and recidivismJohn A. Humphrey a , Gale Burford b & Meredith Huey Dye ca Criminal Justice Department, St. Anselm College, PO Box 1622,100 St. Anselm Drive, Manchester, NH, 03102, USAb Department of Social Work, University of Vermont, 443Waterman Building, 85 So. Prospect St, Burlington, VT,05405-0160, USAc Department of Sociology and Anthropology, Middle TennesseeState University, PO Box 0010, Murfreesboro, TN, 37132, USA

Version of record first published: 19 Jul 2012.

To cite this article: John A. Humphrey, Gale Burford & Meredith Huey Dye (2012): A longitudinalanalysis of reparative probation and recidivism, Criminal Justice Studies: A Critical Journal ofCrime, Law and Society, 25:2, 117-130

To link to this article: http://dx.doi.org/10.1080/1478601X.2012.699731

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions

This article may be used for research, teaching, and private study purposes. Anysubstantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,systematic supply, or distribution in any form to anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any representationthat the contents will be complete or accurate or up to date. The accuracy of anyinstructions, formulae, and drug doses should be independently verified with primarysources. The publisher shall not be liable for any loss, actions, claims, proceedings,demand, or costs or damages whatsoever or howsoever caused arising directly orindirectly in connection with or arising out of the use of this material.

A longitudinal analysis of reparative probation and recidivism

John A. Humphreya*, Gale Burfordb and Meredith Huey Dyec

aCriminal Justice Department, St. Anselm College, PO Box 1622, 100 St. Anselm Drive,Manchester, NH 03102, USA; bDepartment of Social Work, University of Vermont, 443

Waterman Building, 85 So. Prospect St, Burlington, VT 05405-0160, USA; cDepartment ofSociology and Anthropology, Middle Tennessee State University, PO Box 0010,

Murfreesboro, TN 37132, USA

Longitudinal data on 9078 probationers were used to assess the impact ofVermont’s reparative probation program on criminal recidivism. A quasi-experi-mental design was employed to compare five-year reconviction rates of 6682standard and 2396 reparative probationers sentenced during the years 1998,1999, and 2000. Propensity score methods were used to address selection biasand to generate two equivalent groups of probationers for comparison. Resultsfrom a Cox regression model indicated that over the five-year period followingthe imposition of the original probationary sentence, reparative probationersevidenced a significantly lower risk of reconviction than standard probationers.This disparity in recidivism was maintained when holding constant probationer’sprior record, type of offense, age, and gender. Implications for policy arediscussed.

Keywords: restorative justice; reparative probation; probation; recidivism;Vermont; propensity score methods

Introduction

Reliance on probationary strategies as an alternative to incarceration is increasinglyevident across the USA. A recent Pew Foundation report found that of the one in31 convicted adults in the USA, two-thirds are supervised in the community. ThePew report argued that carefully designed, evidence-based community correctionalprograms are essential for the reduction of the $68 billion burden of maintaining alargely ineffective correctional system in the USA (Pew Center on the States,2009).

Restorative justice has been advanced as a theoretically derived alternative tothe traditional retributive model of justice (Bazemore, 2004a, 2004b; Bazemore &Umbreit, 2001; Braithwaite, 1989, 2002; Umbreit, Coates, & Vos, 2004, 2007). Var-ious practical forms of restorative justice have been proposed and implemented(London, 2006; Maxwell & Morris, 2006; McCold, 2003), many with positive out-comes for restoring victims and communities and reducing criminal re-offending(Latimer, Dowden, & Muise, 2005; Maxwell, Kingi, Robertson, & Morris, 2004;Maxwell, Morris, & Anderson, 1999; Schiff, Bazemore, & Erbe, 2001). Based on

*Corresponding author. Email: [email protected]

Criminal Justice StudiesVol. 25, No. 2, June 2012, 117–130

ISSN 1478-601X print/ISSN 1478-6028 online� 2012 Taylor & Francishttp://dx.doi.org/10.1080/1478601X.2012.699731http://www.tandfonline.com

Dow

nloa

ded

by [

] a

t 08:

40 1

8 Se

ptem

ber

2012

the principles of restorative justice, Vermont’s reparative probation program isdescribed as an effective community-based alternative to traditional probation and issupported by criminal justice professionals, community members, and offendersalike (Karp, 2001, 2002, 2004; Karp & Drakulich, 2004).

In keeping with the Pew recommendations, the purpose of this evaluation is todetermine whether or not participation in Vermont’s reparative probation programsignificantly reduces the risk of criminal recidivism compared to placement on stan-dard probation. The present study provides analyses of the outcomes of reparativeand standard probation cases remanded to the Vermont Department of Correctionsfor probation supervision during the three-year period, 1998–2000.1 Longitudinaldata provide the basis for an analysis of recidivism for a five-year period from thetime the offender was placed on probation. While the reparative program has beenthe focus of several studies, the analyses presented in the current study are the firstthat compare recidivism for standard and reparative probationers in Vermont (Karp,2001, 2002; Karp & Drakulich, 2004; Karp, Sprayregen, & Drakulich, 2002; Karp& Walther, 2001).

Background

Reparative probation in Vermont

Vermont first implemented the reparative probation condition option in 1995, with aview to involving communities in the justice process while at the same timeaddressing the needs of crime victims, restoring communities, and encouragingoffenders to take responsibility for their behavior (Karp & Drakulich, 2004). Boyes-Watson (2004, p. 687) described Vermont’s program as ‘one of the earliest andmost extensive statewide restorative initiatives in the United States’. The design andmotivation for the program initially emerged in response to the escalating use ofincarceration in the state (Perry & Gorczyk, 1997) and only later was characterizedas an example of restorative justice. Indeed, the commitment to developing collabo-rative practices between professionals and partnerships with community and con-sumer groups has a long history in Vermont (Comerford & Burford, 2002; Gorczyk& Perry, 1997). Thus, the emergence of reparative boards was consistent with othercommunity-centered developments that were emerging in the state in the late 1980sand early 1990s in partnerships between the state government’s Agency of HumanServices – of which the Department of Corrections is a part – and local communi-ties (Hogan, 1999; Mitchell, 2001; Perry, 1999). The program has enjoyed biparti-san and senior government support through transitions in governors, Secretaries ofthe Agency of Human Services, and Commissioners of Corrections. The somewhatunusual role of the Department of Corrections in spearheading the changes is notedas unique in the country. The use of volunteer community boards has been widelyrecognized as an innovative example of restorative justice and has drawn consider-able attention of researchers, policy-makers, and public and community interestgroups (Boyes-Watson, 2004; De Pommereau, 1997; Dooley, 1996; Goodenough,1997; Immarigeon, 1997; Perry, 1997, 1998; Reno, 1997; Scott, 1996). Numerousboard programs serving young offenders have been set up across the USA usingcommunity boards (Schiff, Bazemore, & Erbe, 2001) and many of these cite Ver-mont’s innovative program and leadership.

The reparative probation program provides that upon conviction of a minor,nonviolent crime, offenders may be sentenced to probation with the condition that

118 J.A. Humphrey et al.

Dow

nloa

ded

by [

] a

t 08:

40 1

8 Se

ptem

ber

2012

they appear before a reparative board composed of trained citizen volunteers. Theboard, the offender, and the victim(s), should choose to attend, and other invitedpersons who have been affected by the crime meet and, when successful, negotiatean agreement in which the offender agrees to complete a number of tasks duringthe probationary period. These agreements include tasks intended to help the offen-der better understand the harmful consequences of his or her behavior, repair theharm done to the victim, and restore the community to its pre-crime state. Tasks inthe agreements frequently include letters of apology to the victim, restitution, com-munity service, and activities aimed at developing needed competencies (Karp,2001; Karp & Drakulich, 2004). In addition, the agreements are intended to bedeveloped around a strategy aimed at reducing re-offending (Karp, Sprayregen, &Drakulich, 2002).2

Probation outcomes

Whether or not placement on reparative probation relative to standard (traditional)probation reduces the likelihood of future offending is not known. Karp et al.(2002) reported that 31% of reparative probationers were rearrested within one yearof meeting with the community board. Analyses of traditional probation outcomesreport mixed results, with recidivism rates for adults and young offenders rangingfrom estimates as high as 65% (Petersilia, Turner, Kahan, & Petersen, 1985) to aslow as 18% (Vito, 1986). The most recent estimates reported by the Bureau of Jus-tice Statistics indicate that in 2008, 63% of probationers in the USA were dis-charged after completing their full probation sentence whereas 21% wereincarcerated or absconded (Glaze & Bonczar, 2009).

Prior studies of probation outcomes have consistently found that age, gender,prior criminal record, and type of offense are among the most predictive variablesfor re-offending (MacKenzie, 2000; MacKenzie, Browning, Skroban, & Smith,1999; Morgan, 1994, 1996; Petersilia, Turner, Kahan, & Petersen, 1985; Robinson,2001; Sims & Jones, 1997; Whitehead, 1991). For example, older offenders arefound to be more successful on probation than their younger counterparts (MacKen-zie, Browning, Skroban, & Smith, 1999; Sims & Jones, 1997; Whitehead, 1991).Petersilia and her colleagues’ (1985) analysis of probation recidivism found that re-offending was significantly more likely for probationers with a prior conviction.Morgan (1994) reported a similar relationship with successful probation outcomes,gender, and prior felony convictions. More recently, Minor, Wells, and Sims (2003)found that failure on federal probation was more common among the young andoffenders with prior convictions.

Data and methods

Vermont’s existing judicial policy allows judges to sentence offenders to the repara-tive probation condition provided the offender is convicted of a minor, nonviolentoffense. Given the existing sentencing policy and due process restrictions, a randomexperimental design was not feasible for this study. Instead, all court convictions(n= 34,471 dockets) for misdemeanor charges sentenced and remanded to the Ver-mont Department of Corrections for probation supervision for the years 1998–2005were analyzed. Given the available data, probationers were tracked for five yearsfrom the point of placement on probation to determine differential recidivism

Criminal Justice Studies 119

Dow

nloa

ded

by [

] a

t 08:

40 1

8 Se

ptem

ber

2012

outcomes. During the years 1998–2000, 9400 unique offenders were sentenced toprobation in Vermont. The vast majority of probationers (n= 8492) received a singleprobation sentence while approximately 10% of probationers (n= 908) were sen-tenced to probation on multiple occasions during the time frame. Of the 908 proba-tioners with multiple probation sentences, 294 were sentenced to both standard andreparative probation at some point during the years 1998, 1999, or 2000. In orderto best isolate the effects of probation type on recidivism, probationers sentenced toboth standard and reparative probation were excluded from the analyses.3 The fol-lowing analyses were based on complete pooled data for 9078 Vermont probation-ers sentenced either to reparative (n= 2396) or standard (n= 6682) probation duringthe years 1998, 1999, or 2000.

Variables of interest

The dependent variable for this analysis was a dichotomous variable that indicatedwhether or not a probationer committed at least one new conviction within the five-year period following the imposition of the probation sentence. Probationers with anew conviction were coded 1. Central to this analysis was the effect of probationtype on recidivism. Probation type was represented by a dichotomous variablewhere placement on reparative probation was coded 1. In addition to the probationtype, several important criminogenic variables, consistently identified in the researchliterature as predictive of crime and recidivism, were considered. These included adichotomous variable indicating a record of prior misdemeanor convictions, type ofcurrent offense (property [reference category], drug, motor vehicle, and other), age(in years), and sex (male = 1).4

Analytic strategy

Several analytical approaches were used to compare recidivism of Vermont repara-tive and standard probationers. Bivariate tests, including t-tests and chi-squaredtests, were used to examine potential differences in the composition of the twogroups of probationers. Next, propensity score methods were employed to addressselection bias and to better estimate the impact of probation type on reconvictionpatterns (see Rosenbaum and Rubin, 1984, 1985; as well as Friedline, Elliott, &Nam, 2011; Garrett, 2005; Guo & Fraser, 2010; Lane, Mikelson, Sharkey, & Wisso-ker, 2001; Ridgeway, 2006; Sampson, Laub, & Wimer, 2006). Lastly, recidivismoutcomes for reparative probationers and a comparison group of standard probation-ers (obtained through propensity score matching methods) were calculated usinglogistic regression and Cox proportional hazard models (Cox, 1972).

Results

Table 1 shows descriptive statistics for the 9078 probationers sentenced either to stan-dard or reparative conditions. Reparative probationers represented approximately aquarter of probationers (26.4%) sentenced during the time frame. Forty-six percent ofVermont probationers were charged with a new conviction five years following theoriginal sentence date. A comparison of reconviction patterns for reparative and stan-dard probationers indicated that reparative probationers were significantly less likelyto be charged with a new conviction during the five-year time frame (43.5 vs. 47.5%;

120 J.A. Humphrey et al.

Dow

nloa

ded

by [

] a

t 08:

40 1

8 Se

ptem

ber

2012

p< .001). Bivariate results in Table 1 also show that standard and reparative probation-ers differed significantly on each variable of interest. For example, reparative proba-tioners were significantly younger than standard probationers (27.9 vs. 31.3 years;p< .001) and were significantly less likely to have a prior record of misdemeanor con-victions (39.2 vs. 53.8%; p< .001).5 In addition, reparative probationers were overrep-resented among property offenders and those sentenced for ‘other’ offenses, andunderrepresented among drug and motor vehicle offenses.

The differences between standard and reparative probationers support the needto control these variables statistically in multivariate analyses, but they also point toselection bias. Indeed, results of a logistic regression model revealed that althoughreparative probationers were significantly less likely to recidivate than standard pro-bationers, prior record, gender, age, and current offense also predicted the likelihoodof a new conviction.6 For all of these variables, the higher percentage of crimino-genic risk factors was found for standard probationers indicating a greater a prioririsk of recidivating than reparative probationers. Thus, significant differences inrecidivism between reparative and standard probationers may be explained by pre-existing differences between the groups (e.g. prior criminal record) rather thanplacement on reparative probation. To take into account potential selection biases,

Table 1. Characteristics of Vermont Department of Corrections 1998, 1999, and 2000probation cohorts by probation type (n = 9078).

Standardprobationers

Reparativeprobationers Total

Covariate imbalance prior to matching

n = 6682 n= 239673.6% 26.4% 100.0%

New offense (% yes) 47.6% 43.5% 46.5%

Pearson χ2 = 11.893 df = 1, p< .001

Prior record (% yes) 53.8% 39.2% 49.9%

Pearson χ2 = 150.852 df = 1, p< .001

Current offenseProperty 17.0% 19.7% 17.7%Drug 5.5% 3.8% 5.1%Motor vehicle 55.6% 44.3% 52.6%Othera 21.9% 32.2% 24.7%

Pearson χ2 = 136.89 df = 3, p< .001

Age (years) 31.3 years 27.9 years 30.4 years (SD= 10.9, range16–80)

t= 13.103, p< .001

Sex (% male) 79.7% 73.1% 78.0%

Pearson χ2 = 45.955 df = 1, p< .001

aIncludes disorderly conduct, prohibited acts, disturbing the peace, violations of conditions, alcohol pos-session, negligent operation, and similar misdemeanor violations.

Criminal Justice Studies 121

Dow

nloa

ded

by [

] a

t 08:

40 1

8 Se

ptem

ber

2012

each probationer was assigned a propensity score representing the predictedprobability of being placed on reparative probation. Propensity scores were obtainedby regressing probation type on the observed explanatory variables (sex, age, priorrecord, and current offense). Initially, propensity scores were incorporated into alogistic regression model to attenuate any selection bias. In these models, reparativeprobationers continued to evidence lower odds of a new conviction relative to stan-dard probationers (p< .05). However, this model also indicated that probationerswith a greater probability of placement on reparative probation – regardless ofactual placement – were significantly less likely to recidivate (p< .001).

To further address the effects of selection bias on recidivism and isolate theeffects of reparative probation on the likelihood of a new offense, propensity scoreswere used to generate a comparison sample of standard probationers which reflectedthe characteristics of those sentenced to reparative probation.7 To do this, we usedpropensity scores and a stratification matching/resampling process. First, the propen-sity scores were sorted into five groups (i.e. quintiles). Within each strata, we

Table 2. Characteristics of matched sample of Vermont standard (n= 2396) and reparative(n= 2396) probationers.

Standard probationersReparativeprobationers Total

%Reductionin biasa

Covariate balance check after matching

New Offense (% yes) 46.1% 45.9% 44.8% N/A

Pearson χ2 = 3.039, df = 1, p= .081

Prior record (% yes) 40.0% 39.2% 40.0% 92.0

Pearson χ2 = .683, df = 1, p= .408

Current offenseProperty 20.5% 19.7% 20.1% 83.3Drug 4.2% 3.8% 4.0% 79.6Motor vehicle 44.0% 44.3% 44.1% 91.9Otherb 31.3% 32.2% 31.7% 87.6

Pearson χ2 = 1.520, df = 3, p= .678

Age (years) 28.3 years 27.9 years 28.1 years(SD= 10.6,range 16–80)

84.8

t= 1.130, p = .259

Sex (% male) 73.3% 73.1% 73.2% 97.3

Pearson χ2 = .027, df = 1, p= .870

aPercent reduction in bias represents the ratio of standardized differences (Formula:100ð�X2 � �X0Þ=

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiS22 þ S20=2

p) between the two groups of probationers prior to and after matching.

bIncludes disorderly conduct, prohibited acts, disturbing the peace, violations of conditions, alcohol pos-session, negligent operation, and similar misdemeanor violations.

122 J.A. Humphrey et al.

Dow

nloa

ded

by [

] a

t 08:

40 1

8 Se

ptem

ber

2012

examined the number of probationers placed on reparative vs. standard probationand resampled. This process generated a one-to-one match of standard to reparativeprobationers (n= 4792) based on propensity score. To access statistically whetherthe propensity score matching approach generated similar groups and reduced theselection bias, sensitivity analyses were run to check covariate balance. As shownin Table 2, the characteristics of probationers placed on reparative probation and thematched comparison group of standard probationers were quite similar statisticallyon the observed characteristics prior record, current offense, age, and gender. Thepercentage reduction in bias prior to relative to after matching ranged from 79.6 to9.3% reduction. Thus, the matching methods resulted in two ‘equivalent’ groups ofprobationers. The first group consisted of all 2396 reparative probationers sentencedduring the years 1998, 1999, and 2000. A group of 2396 standard probationers,who were statistically indistinguishable from the reparative group on observed char-acteristics, made up the comparison group.

Using these matched groups of standard and reparative probationers, recidivismpatterns over the five-year observation period were compared. First, at the bivariatelevel, 43.5% of reparative probationers and 46.1% of standard probationers experi-enced a new offense resulting in a new conviction within five years following theimposition of the sentence date (see Table 2). Next, a logistic regression model pre-dicting the odds of a new offense was performed. Consistent with the findingsreported above, reparative probationers evidenced lower odds of a new convictioncompared to standard probationers, net of other variables (p< .05). In addition, priorrecord, current offense, age, and gender all remained as significant predictors ofrecidivism. Lastly, a Cox regression model was used to estimate the impact of pro-bation type and other variables of interest on recidivism (Cox, 1972). In this analy-sis, the unit of time is days and the event is the date of first new conviction withinthe five-year observation period. Given that over half of the cases are right censored(i.e. probationers who did not experience a new conviction by the end of the five-year study period), Cox regression modeling is an appropriate analytical strategy(see Allison, 1984).8

Results from the Cox regression model are presented in Table 3. (Hazard ratesfor the Cox regression are shown in Appendix 1.) A positive coefficient indicates

Table 3. Results of the Cox regression modela (n= 4792).

B (SE) Odds

Probation type(Reparative probation = 1; Standard probation = 0)

�.118⁄ (.043) .889

Prior record (yes = 1; no = 0) .723⁄⁄ (.044) 2.060Current offense categoryProperty ReferenceDrug �.195 (.11.) .823Motor vehicle �.331⁄⁄ (.058) .718Other �.235⁄⁄ (.058) .790Age (years) �.031⁄⁄ (.003) .969Sex (male = 1; female = 0) .383⁄⁄ (.054) 1.467�2 Log likelihood 34650.67Model χ2 550.72, df = 7, p < .001

Notes: ⁄p < .01; ⁄⁄p < .001.aFor the Cox regression, the hazard rate is based on the days until first new offense with a conviction.

Criminal Justice Studies 123

Dow

nloa

ded

by [

] a

t 08:

40 1

8 Se

ptem

ber

2012

that a unit increase in the independent variable raises the risk of a new conviction;a negative coefficient means that a unit increase in the independent variabledecreases the risk of recidivism. As shown, placement on reparative probation low-ered the hazard of a new conviction. Net of other variables, placement on reparativeprobation decreased the risk of new conviction by 11% over the five-year period(p< .01). To illustrate the effect of probation type, Figure 1 shows the proportion ofstandard and reparative probationers without a new offense (surviving) at the end ofeach of the five-year intervals represented by days. As shown, holding all othervariables at their mean values, the cumulative probability of survival was greaterfor reparative probationers over the entire study period.

The control variables in the Cox model also had large substantial and significanteffect on the hazard of a new offense, all of which were in the expected direction.The largest effect was seen for prior record. Having a prior record more thandoubled the risk of a new conviction regardless of probation type. Males had higherhazards of a new conviction as well, while older probationers, motor vehicle viola-tors, and probationers with ‘other’ offenses (relative to property offenders) hadlower hazards of a new conviction. The negative coefficient for probationer ageindicated that a unit increase in age (one year) was associated with a three percentdecrease in the hazard of a new offense. A 10-year increase in age decreased therisk of recidivism by nearly 27% [100 ⁄ (1� ((exp) (�.031⁄10)) = 26.7%]. Addi-tional analyses probed a series of interaction models that allowed the effect of pro-bation type to vary by each of the sociodemographic and criminogenic variables aswell as over time. None of the interaction terms were statistically significant. Thus,the effect of reparative probation did not vary for certain types of probationers orover the five-year period of observation.

Figure 1. Proportion of standard and reparative probationers without a new conviction atthe end of each year.

124 J.A. Humphrey et al.

Dow

nloa

ded

by [

] a

t 08:

40 1

8 Se

ptem

ber

2012

Discussion and conclusions

The results of this study suggested that reparative probation, as implemented inVermont, significantly reduced subsequent criminal offending compared to moretraditional standard probation practices. Offenders who were assigned to thereparative probation condition were significantly less apt to receive a new convic-tion during the five-year period following their initial placement in the program thana similar group of standard probationers. The effects of gender, age, and prior con-victions on recidivism were consistent with previous research (see Morgan, 1994,1995, 1996). That is, males, younger offenders, and those previously convicted arefound to be at significantly higher risk for re-offending. However, the difference inrecidivism between reparative and standard probationers was maintained when hold-ing constant the risk factors of age, gender, prior record, and current offense.

Although the results reported here represent an important contribution to theresearch on reparative probation as implemented in the state of Vermont, severallimitations of the study must be noted. Most importantly, the findings of the presentstudy are limited to probationers who met the initial selection criteria for thereparative probation program in Vermont. Because of the non-experimental designof the present study, we were unable to eliminate the possibility of selection bias atthe outset. However, we addressed the issue of selection bias by assigning propen-sity scores, which represented the probability of being placed on reparative proba-tion, and then using these scores to generate an equivalent group of standardprobationers for comparison. Overall, the results of this approach were consideredmore reliable than approaches that do not address selection bias.

One disadvantage of the propensity score matching approach is that the findingscan only be interpreted as the effect of reparative probation on the types of proba-tioners most likely to be sentenced to the reparative condition. Conclusions are lim-ited to a comparison of reparative and standard probationers that have similarsociodemographic and criminogenic characteristics. In addition, any probationercharacteristics that are not incorporated in the creation of propensity score mayaffect the probability of placement on reparative probation. In this case, selectionbias may not be completely attenuated by the propensity score methods. We werequite limited in the observed characteristic that predicted placement on reparativeprobation. Indeed, the relatively low Nagelkerke coefficients in the logistic regres-sion models indicated how unmeasured variables might contribute to the offendingpatterns of probationers. Therefore, the findings cannot be generalized to probation-ers with different characteristics. This includes probationers who may have experi-enced a new arrest, but did not receive a new conviction. An analysis of newarrests could reveal a different relationship with probation types and recidivism.Subsequent investigations of reparative probation should include additional mea-sures of social bonds, (e.g. marital and employment status, educational levels, andliving arrangement) that have been found to influence re-offending among proba-tioners, but were not available in the data. Changes in probationer’s social bonds tocommunity should be assessed over time to determine changes in their impact oncriminal behavior. Future research should consider the applicability of reparativeprobation for probationers with felony convictions.

Vermont’s reparative probation program, based on key principles of restorativejustice, has yielded promising results. It is vital, however, to empirically investigatethe questions of ‘why’ and ‘how’ the principles of restorative justice operate in

Criminal Justice Studies 125

Dow

nloa

ded

by [

] a

t 08:

40 1

8 Se

ptem

ber

2012

restorative justice-based programs. More refined measures of the social processesthat underlie the relative effectiveness of the restorative programs should be devel-oped and assessed. In addition, a rigorously designed cost-benefit analysis wouldprovide critically needed information for criminal justice administrators, plannersand policy-makers in their consideration of restorative justice practices.

AcknowledgementsThis project was supported by Grant No. 2000IJCX0033 awarded by the National Instituteof Justice, US Department of Justice. Points of view in this document are those of theauthors and do not necessarily represent the official position or policies of the USDepartment of Justice. Special thanks to John Braithwaite and David Karp for theirinsightful comments on an earlier version of this paper.

Notes1. It is common for probationers to have been convicted of more than one charge. Nearly

all convictions are the result of plea agreements on multiple charges each of which hasa different sentence. It is possible for an offender to have a reparative sentence for abad check charge, traditional probation for a retail theft, and a split term of 60 days injail followed by a one-year term of probation for a domestic assault. In order to isolatethe effects of probation type on recidivism, this study only includes cases sentencedeither to reparative or standard probation between the years 1998–2000. Probationerswith multiple sentences are described in more detail below.

2. More detailed descriptions of program features can be found in Dembenski (2003,2004), Dooley (1996), Karp (2002), Karp and Walther (2001), Perry and Gorczyk(1997) and Walther and Perry (1997).

3. A comparison of excluded (n= 294) and included probationers (n= 9078) evidencedtwo differences. First, excluded probationers were significantly younger than thoseincluded in the analysis (mean age 24 vs. 30 years; p< .001). Second, nearly all theprobationers excluded from the analysis (99%) had a prior misdemeanor record. Incomparison, about half as many probationers included in the analysis had prior records(49.9%). Overall, these comparisons suggest that young offenders who have had multi-ple convictions represent a high risk group for community supervision. Reparative pro-bation was not originally intended for such chronic offenders, but for first time, lowrisk, misdemeanants. However, the dampening effect of reparative probation on the fur-ther criminality of certain chronic offenders is an area for further research.

4. Due to extensive missing data on offender’s race and marital status, these variableswere not included in the analysis. However, more than 97% of probationers in Vermontare white. In addition, changes in marital status, employment, and educational attain-ment over the five-year follow-up period were not available; therefore, these proba-tioner characteristics were not included.

5. It is important to note the discrepancy in proportion of reparative probationers with aprior record (approximate 40%) and the sentencing guidelines for reparative probation(e.g. ‘first-time offenders’). Further evaluation of this discrepancy is certainly needed.

6. Several logistic regression models were performed to test the robustness of the findings.Details of these analyses are available from the authors.

7. Propensity scores are relatively new to the field of criminology and criminal justiceresearch methods and statistical analysis, but are readily used in fields such as econom-ics and recently social work. In fact, several propensity score methods are noted in theliterature. For example, propensity scores weights can be used to create a set of standardprobationers that are indistinguishable on observed characteristics from reparativeprobationers. Another strategy involves propensity score matching, where probationerscan be sorted by their predicted probabilities of being placed on reparative probation.Standard probationers within a specified proportion of the distribution (e.g. quartile,quintile, etc.) are then matched with reparative probationers. Matching procedures canbe performed using PSMATCH2 command in STATA or GREEDY in SAS. Both pro-

126 J.A. Humphrey et al.

Dow

nloa

ded

by [

] a

t 08:

40 1

8 Se

ptem

ber

2012

pensity score methods were performed; however, the propensity score weights failed togenerate statistically equivalent groups of standard and reparative probationers. Thus, forthese data, the latter method was utilized. A variety of matching procedures were com-pared and the results of the matching procedures were the same regardless of the proce-dure. The analysis is based on stratification matching (see Cochrane & Chamers, 1965;Guo & Fraser, 2010). Further details of these analyses are available from the authors.

8. Half of the probationers with a new offense recidivated within the first year. Recidivismrates for both reparative and standard probationers gradually declined over the five-yearperiod (see comparative hazard rate patterns in the Appendix).

Notes on contributorsJohn A. Humphrey, is a Professor of criminal justice at St Anselm College and co-directorof the New Hampshire Institute of Politics’ Consortium on Justice and Society. He is agraduate of St Anselm College and received his PhD in sociology from the University ofNew Hampshire. His research interests include alternatives to incarceration, wrongfulconviction, interpersonal violence, and self-destructive behavior.

Gale Burford is a Professor of social work at the University of Vermont, director of theUniversity/State Child Welfare Training Partnership and principal investigator for theCommunity Justice Consortium located at the University of Vermont. He received hisundergraduate degree at St Martin’s University, MSW from the University of Washingtonand his PhD from the University of Sterling in Scotland. His research interests include theuse of family engagement in child welfare and youth justice and in restorative justice theoryand practice.

Meredith Huey Dye is an Assistant Professor of sociology at Middle Tennessee State University.She received her PhD from the University of Georgia in 2008. Her research interests includerestorative justice, effects of incarceration, prison suicide, and women in prison.

References

Allison, P. (1984). Event history analysis: Regression for longitudinal event data. Sage Uni-versity Paper Series on Quantitative Applications in the Social Sciences, 46. BeverlyHills, CA: Sage.

Bazemore, G. (2004a). Paradigm muddle or paradigm paralysis: The wide and narrowerroads to restorative justice (or, a little confusion may be a good thing). ContemporaryJustice Review, 7, 37–57.

Bazemore, G. (2004b). Whom and how do we reintegrate? Finding community in justice.Criminology and Public Policy, 4, 131–148.

Bazemore, G., & Umbreit, M. (2001). A comparison of four restorative justice conferencingmodels. Juvenile Justice Bulletin, Office of Juvenile Justice and Delinquency Prevention,Office of Justice Programs, US Department of Justice.

Boyes-Watson, C. (2004). The value of citizen participation in restorative/community justice:Lessons from Vermont. Criminology and Public Policy, 3, 687–692.

Braithwaite, J. (1989). Crime, shame, and reintegration. Cambridge: Cambridge UniversityPress.

Braithwaite, J. (2002). Restorative justice and responsive regulation. New York, NY: OxfordUniversity Press.

Cochrane, W., & Chambers, S. (1965). The planning of observational studies of human pop-ulations. Journal of the Royal Statistical Society, Series A, 128, 234–266.

Comerford, S., & Burford, G. (2002). Community by community: Strengthening Vermont’syouth justice system through collaboration. Burlington, VT: University of Vermont,Department of Social Work.

Cox, D.R. (1972). Regression models and life tables. Journal of the Royal Statistical Society,Series B, 34, 187–202.

De Pommereau, I. (1997). In the heart of a community. Vermont Life, Summer, 60–63.

Criminal Justice Studies 127

Dow

nloa

ded

by [

] a

t 08:

40 1

8 Se

ptem

ber

2012

Dembenski, J. (2003). Restorative justice: Vermont State Policy. The Vermont Bar Journal,29(4), 39–45.

Dembenski, J. (2004). Restorative justice in Vermont: Part II. The Vermont Bar Journal, 30(1), 49–53.

Dooley, M. (1996). Restorative justice in Vermont: Awork in progress. In E. Barajas (Ed.), Com-munity justice: Striving for safe, secure, and just communities (pp. 30–35). Washington, DC:National Institute of Corrections.

Friedline, T.L., Elliott, W., & Nam, I. (2011). Predicting savings from adolescence to youngadulthood: A propensity score approach. Journal of the Society for Social Work andResearch, 2(1), 1–22.

Garrett, B. (2005). Propensity score methods. The Urban Institute. Retrieved from http://www.urban.org/toolkit/data-methods/propensity.cfm

Glaze, L., & Bonczar, T.P. (2009). Probation and parole in the United States, 2008. Washington,DC: US Department of Justice, Office of Justice Programs, and Bureau of Justice Statistics.

Goodenough, O. (1997). Faculty adviser and panalist. Biology, behavior, and criminal law.Vermont Law Review Symposium.

Gorczyk, J.F., & Perry, J.G. (1997). What the public wants: Market research finds supportfor restorative justice. Corrections Today, 59(7), 78–83.

Guo, S., & Fraser, M. (2010). Propensity score analysis: Statistical methods and applica-tions. Los Angeles, CA: Sage.

Hogan, C. (1999). Vermont Communities Count: Using results to strengthen services forfamilies and children. The Annie E. Casey Foundation. Retrieved from http://www.aecf.org/publications/vermont.pdf

Immarigeon, R. (1997). Probation and domestic violence: New parameters, model practices.Community Corrections Report on Law and Corrections Practice, 4(6), 91–94.

Karp, D. (2001). Harm and repair: Observing restorative justice in Vermont. Justice Quar-terly, 18, 727–757.

Karp, D. (2002). The offender/community encounter: Stakeholder involvement in the Ver-mont community reparative boards. In D.R. Karp & T.R. Clear (Eds.), What is commu-nity justice? Case studies of restorative justice and community supervision (pp. 61–86).Thousand Oaks, CA: Sage.

Karp, D. (2004). The role and attitudes of restorative board members: A case study of vol-unteers in community justice. Crime & Delinquency, 50(4), 487–515.

Karp, D., & Drakulich, K.M. (2004). Minor crime in a quaint setting: Practices, outcomes, andlimits of Vermont reparative probation boards. Criminology and Public Policy, 3, 655–686.

Karp, D., Sprayregen, M., & Drakulich, K. (2002). Vermont reparative probation year 2000outcome evaluation final report. Saratoga Springs, NY: Skidmore College, Departmentof Sociology. Retrieved from http://www.skidmore.edu/~dkarp/Karp%20Vitae_files/VT%20Reparative%20Probation%20Year%202000%20Outcome%20Evaluation.pdf

Karp, D., & Walther, L. (2001). Community reparative boards in Vermont. In G. Bazemore& M. Schiff (Eds.), Restorative community justice: Repairing harm and transformingcommunities (pp. 199–217). Cincinnati, OH: Anderson.

Lane, J., Mikelson, K.S., Sharkey, P.T., & Wissoker, D. (2001). Low-income and low-skilledworkers’ involvement in nonstandard employment. The Urban Institute. Retrieved fromhttp://www.urban.org/url.cfm?ID=410374

Latimer, J., Dowden, C., & Muise, D. (2005). The effectiveness of restorative justice prac-tices: A meta-analysis. The Prison Journal, 85, 127–144.

London, R.D. (2006). Paradigms lost: Repairing the harm of paradigm discourse in restor-ative justice. Criminal Justice Studies, 19(4), 397–422.

MacKenzie, D. (2000). Evidence-based corrections: Identifying what works. Crime andDelinquency, 46(4), 457–471.

MacKenzie, D., Browning, K., Skroban, S., & Smith, D. (1999). The impact of probation on thecriminal activities of offenders. Journal of Research in, Crime and Delinquency, 36, 423–454.

Maxwell, G., Kingi, V., Robertson, J., & Morris, G. (2004). Achieving effective outcomes inyouth justice: Final report to the Ministry of Social Development. Wellington: Ministryof Social Development.

Maxwell, G., & Morris, A. (2006). Youth justice in New Zealand: Restorative justice inpractice? Journal of Social Issues, 62(2), 239–258.

128 J.A. Humphrey et al.

Dow

nloa

ded

by [

] a

t 08:

40 1

8 Se

ptem

ber

2012

Maxwell, G., Morris, A., & Anderson, T. (1999). Community panel adult pre-trial diversion:Supplementary evaluation. Wellington: Institute of Criminology, Victoria University ofWellington.

McCold, P. (2003). An experiment in police-based restorative justice: The Bethlehem (PA)project. Policy, Practice and Research, 4(4), 379–390.

Minor, K.I., Wells, J.B., & Sims, C. (2003). Recidivism among federal probationers – pre-dicting sentence violations. Federal Probation, 67, 31–36.

Mitchell, C. (2001). Outcome-based planning: State partner and local communities workingtogether to improve the well-being of all Vermonters. State team for children, familiesand individuals. Waterbury, VT: State of Vermont, Agency of Human Services.

Morgan, K.D. (1994). Factors associated with probation outcome. Journal of Criminal Jus-tice, 22, 341–353.

Morgan, K.D. (1995). Variables associated with successful probation completion. Journal ofOffender Rehabilitation, 22(3/4), 141–153.

Morgan, K.D. (1996). Factors influencing probation outcome: A review of the literature. InT. Ellsworth (Ed.), Contemporary community corrections (pp. 327–340). ProspectHeights, IL: Waveland Press.

Perry, J. (1997). Reciprocity, and criminal justice using epigenetics rules to design a pro-gram. Vermont Law Review, 22, 407–423.

Perry, J. (1998). Community-tailored sentences: Creative penalties for non-violent offenders.Governing: Innovations in American Government. Author.

Perry, J. (1999). Corrections. In M. Sherman (Ed.), Vermont State Government since 1965.Burlington, VT: Center for Research on Vermont, University of Vermont and The Snel-ling Center.

Perry, J., & Gorczyk, J.F. (1997). Restructuring corrections: Using market research in Ver-mont. Corrections Management Quarterly, 1, 26–35.

Petersilia, J., Turner, S., Kahan, J., & Petersen, J. (1985). Granting felons probation: Publicrisks and alternatives. R-3186-NIJ The Rand Corporation.

Pew Center on the States. (2009). One in 31: The long reach of American corrections.Washington, DC: The Pew Charitable Trusts.

Reno, J. (1997). Regaining public confidence. State Government News, Publication of theCouncil of State Governments, 16–19.

Ridgeway, G. (2006). Assessing the effect of race bias in post-traffic stop outcomes usingpropensity scores. Journal of Quantitative Criminology, 22(1), 1–29.

Robinson, G. (2001). Power, knowledge, and ‘what works’ in probation. The Howard Jour-nal, 40(3), 235–254.

Rosenbaum, P.R., & Rubin, D.B. (1984). Reducing bias in observational studies using sub-classification on the propensity score. Journal of the American Statistical Association,79, 516–524.

Rosenbaum, P.R., & Rubin, D.B. (1985). Constructing a control group using multivariatematched sampling methods that incorporate the propensity score. American Statistician,39(1), 33–38.

Sampson, R., Laub, J., & Wimer, C. (2006). Does marriage reduce crime? A counterfactualapproach to within-individual causal effects. Criminology, 44(3), 465–510.

Schiff, M., Bazemore, G., & Erbe, C. (2001). Tracking restorative justice decisionmaking inthe response to youth crime: The prevalence of youth conferencing in the United States.Ft. Lauderdale, FL: The Community Justice Institute, Florida Atlantic University.

Scott, M.J. (1996). Righting wrongs. State Government News, Publication of the Council ofState Governments. October, 17–21.

Sims, B., & Jones, M. (1997). Predicting success or failure on probation: Factors associatedwith felony probation outcome. Crime and Delinquency, 43(3), 314–327.

Umbreit, M., Coates, R., & Vos, B. (2004). Restorative justice versus community justice:Clarifying a muddle or generating confusion? Contemporary Justice Review, 7, 91–100.

Umbreit, M., Coates, R., & Vos, B. (2007). Restorative justice dialogue: A multi-dimen-sional evidence-based practice theory. Contemporary Justice, Review, 10, 23–41.

Vito, G.F. (1986). Felony probation and recidivism: Replication and response. Federal Pro-bation, 50, 17–25.

Criminal Justice Studies 129

Dow

nloa

ded

by [

] a

t 08:

40 1

8 Se

ptem

ber

2012

Walther, L., & Perry, J. (1997). The Vermont reparative probation program. ICCA Journalon Community Corrections, 8, 26–34.

Whitehead, J.T. (1991). The effectiveness of felony probation: Results from an eastern state.Justice Quarterly, 8, 523–543.

Appendix 1. Recidivism patterns over five years for all, standard, and repara-tive probationers

A. All probationers YearNumber

recidivatedNumberat risk

Estimatedhazard rate

1 991 4792 0.2072 481 3801 0.1273 327 3320 0.0984 207 2993 0.0695 140 2786 0.050

Total w/New Offenses 2146 (44.8% of group total)>5 (censored cases) 2646

Total 4792

B. Standardprobationers

Year Numberrecidivated

Numberat risk

Estimatedhazard rate

1 530 2396 0.2212 227 1866 0.1223 172 1639 0.1054 104 1467 0.0715 72 1363 0.053

Total w/New Offenses 1105 (46.1% of group total)>5 (censored cases) 1291

Total 2396

C. Reparativeprobationers

Year Numberrecidivated

Numberat risk

Estimatedhazard rate

1 461 2396 0.1922 254 1935 0.1313 155 1681 0.0924 103 1526 0.0675 68 1423 0.048

Total w/New Offenses 1041 (43.4% of group total)>5 (censored cases) 1355

Total 2396

130 J.A. Humphrey et al.

Dow

nloa

ded

by [

] a

t 08:

40 1

8 Se

ptem

ber

2012