Determinants of Recidivism in Rhode Island’s 2009 Prison Population Vlad Konopelko, Lucian Drobot, Alex Gemma, David Rodin, Bill Garneau
Mar 21, 2016
Determinants of Recidivism in Rhode Island’s 2009 Prison Population
Vlad Konopelko, Lucian Drobot, Alex Gemma, David Rodin, Bill Garneau
Topic
• RI Recidivism study
• Recidivist = Repeat offender– 28% returned with new sentence – 34% were awaiting trial– 47% are for new crime rest for probation and parole violation
• Important to everyone
• Data availability
Objective
• Determine which factors impacts repeat offenders
• Identify factors that can be influenced through policies
Research History• “The Best Ones Come Out First! Early Release from Prison and Recidivism
A Regression Discontinuity Approach” Olivier Marie 2009• Building Criminal Capital vs Specific Deterrence: The Effect of Incarceration
Length on Recidivism. David S. Abrams 2010
Data Set
• Starting Data Set– 450,000 data points– 150 variables– 3700 Variables
• Ending Data Set– 47,000 data points– 28 Variables– 1670 Subjects
Removed Variables
• Redundant Variables– Length of stay, Total stay, % Time served
• Variables Insignificant to Our Study– Addresses, birthdays, admittance dates, etc…
• Incomplete records– 2000 Inmates did not have all the data points
Condensing the Data• Age Bracket
– 32 and Below– 33 and Above
• Employment – Under/Unemployed– Employed / Outside of workforce
• Housing Status– Homeless/ Living in a shelter– Program Transitional/ Temporary/Permanently residents
• Education– High school/GED +– Below high school and no GED
Logistic Regression Model Depending variable 0 – 1
The dependent variable is categorical with two possible values
It is based on the odds ratio:
odds ratio =
Example: odds ratio (for a 0.75 probability of interest)=0.75/(1-0.75)=3 (or 3 to 1)
Logistic Regression Model
Logistic Regression Model:
ln (odds ratio)= …
Logistic Regression Equation:
ln(estimated odds ratio)= …+
Logistic Regression Model
Determine
Determine estimated odds ratio
Determine estimated probability of an event of interest
Model Results Variable B S.E. Wald df Sig. Exp(B)
Step 1Total Sentence -.001 .000 23.950 1 .000 .999
Constant -.622 .072 75.434 1 .000 .537
Step 2
Age -.022 .006 16.050 1 .000 .978
Total Sentence -.001 .000 22.206 1 .000 .999
Constant .092 .190 .233 1 .629 1.096
Step 3
Age -.025 .006 19.942 1 .000 .975
Total Sentence -.001 .000 22.181 1 .000 .999
Citizenship .806 .205 15.458 1 .000 2.240
Constant -.534 .253 4.449 1 .035 .586
Step 4
Age -.028 .006 22.949 1 .000 .973
Total Sentence -.001 .000 22.901 1 .000 .999
Housing Status .336 .159 4.470 1 .035 1.399
Citizenship .811 .206 15.566 1 .000 2.250
Constant -.500 .254 3.875 1 .049 .607
32 and Under
Variable B S.E. Wald df Sig. Exp(B)
Age -.046 .020 5.351 1 .021 .955
Felony or Misdemeanor -.427 .156 7.498 1 .006 .652
Education .394 .161 5.939 1 .015 1.482
HousingStatus .505 .254 3.941 1 .047 1.657
Citizenship .804 .244 10.907 1 .001 2.235
Constant -.162 .513 .100 1 .752 .851
Less than GEDGED or Higher
Education
HomelessNot Homeless
Housing
UnemployedEmployed
Employment
Key Indicators
Policies 1
• 5 out of 28 variables– Single vs married
• For all: – Age: The higher the age the less likelihood.– Citizenship: US citizen are more likely to return