THE EFFECTS OF PRIVATE PRISON CONFINEMENT IN
MINNESOTA ON OFFENDER RECIDIVISM

Authors
Grant Duwe, Ph.D.
Research Director
Email: grant.duwe@state.mn.us
Valerie Clark, Ph.D.
Research Analyst Specialist
Email: valerie.clark@state.mn.us

1450 Energy Park Drive, Suite 200
St. Paul, Minnesota 55108-5219
651/361-7200
TTY 800/627-3529
www.doc.state.mn.us
March 2013
This information will be made available in alternative format upon request.
Printed on recycled paper with at least 10 percent post-consumer waste

Research Summary
Evidence has been mixed as to whether private prisons are more effective than stateoperated facilities in reducing recidivism. This study analyzes whether private prison
confinement in Minnesota has had an impact on recidivism by examining 3,532 offenders
released from prison between 2007 and 2009. Propensity score matching was used to
individually match a comparison group of 1,766 inmates who had only been confined in
state-run facilities with 1,766 offenders who had served time in a private prison facility.
Using multiple measures of recidivism and private prison confinement, 20 Cox regression
models were estimated. The results showed that offenders who had been incarcerated in a
private prison had a greater hazard of recidivism in all 20 models, and the recidivism risk
was significantly greater in eight of the models. The evidence presented in this study
suggests that private prisons are not more effective in reducing recidivism, which may be
attributable to fewer visitation and rehabilitative programming opportunities for offenders
incarcerated at private facilities.

Introduction
The number of private prisons has risen dramatically over the past three decades.
Private corporations had only a limited role in corrections at the beginning of the 1980s,
providing food service and healthcare, among other services (Selman and Leighton,
2010). By 2005, private corporations operated 415 prison facilities in addition to
providing other prison services to public facilities (Stephan, 2008). Determinate
sentencing practices and the War on Drugs led to unprecedented prison population
growth, which outpaced the capacity of U.S. prisons (Selman and Leighton, 2010; Austin
and Coventry, 2001). Thus, state and federal correctional agencies looked to private
facilities for bed space. Additionally, the political climate during the 1980s fostered a
movement towards privatizing many government functions, including corrections (Austin
and Coventry, 2001).
Between 2000 and 2005, the number of adult correctional facilities rose by nine
percent, and private facilities accounted for nearly all of this increase. Of the 153 new
prison facilities built during this time period, 151 were private (Stephan, 2008). State
prisoners have accounted for most of the growth in the number of offenders housed in
private prisons over the past decade (Sabol, Minton, and Harrison, 2007). Despite the
rapid increase in the number of private facilities and privately housed offenders over the
past three decades, inmates held in private facilities still make up only a small percent of
all prisoners. Offenders confined in private facilities account for 8 percent of all
prisoners; more than 16 percent of all federal prisoners are held in private facilities, while
nearly 7 percent of all state prisoners are in private facilities (West, Sabol, and
Greenman, 2010).

1

The debate over whether to privatize more prisons has become a fixture in many
state budget negotiations. The main argument used in favor of prison privatization is that
private institutions are more efficient. More than 30 cash-strapped states have turned to
private prisons under the assumption that they provide the same services and produce
better recidivism rates for less money than public facilities (Oppel, 2011). Proponents of
private prisons have claimed as much as 20 percent in savings over state-run prisons. The
difference in price, they argue, comes from the use of nonunion workers, which gives
them more control over wages and benefits. Further, the desire to maintain high profit
margins gives private prison corporations an incentive to act more efficiently (Cheung,
2004).
Opponents of prison privatization argue that claims made in favor of this
movement lack evidence (Government Accounting Office, 1996; Selman and Leighton,
2010; Cheung, 2004). Due to data limitations, as well as misleading use of data, few
studies have provided convincing evidence to support the claim that private prisons
provide the same or better services for less money (Government Accounting Office,
1996). All the while, private prisons generally accept only minimum or medium-security
offenders with few or no health and behavioral problems (Oppel, 2011; Arizona
Department of Corrections, 2011).
In addition to the lack of savings, opponents of privatization cite evidence
showing that private prisons are poorly run compared to public prisons (Cheung, 2004).
Austin and Coventry (2001) found a higher rate of inmate-on-inmate and inmate-on-staff
assaults at privately-run facilities compared to public facilities. However, other measures
of inmate misconduct, including riots and other inmate-led disturbances, are comparable.

2

More recently, many states and the federal government have scaled back their
reliance on private facilities (Cook, 2010; Shelden, 2010). Declining incarceration rates
across several states has created less demand for bed space. Further, the Great Recession
and budget crises have forced several states to crowd more prisoners into less space.
Private facilities in a handful of states, including Minnesota, now sit vacant (Cook, 2010;
Shelden, 2010).
Present Study
Prior to 2010, when prison population growth created shortages in prison beds at
state facilities, the Minnesota Department of Corrections (MnDOC) frequently housed
some of its inmates at the Prairie Correctional Facility (PCF) in Appleton, Minnesota.
The facility, which opened in 1996, once held as many as 1,200 Minnesota state prisoners
(Havens and Giles, 2009). Operated by Corrections Corporation of America (CCA), a
private prison company, the PCF closed in February 2010 due, in part, to slowed growth
in Minnesota’s prison population, which minimized the need to transfer offenders to nonMnDOC facilities.
This study evaluated whether private prison confinement had an impact on
recidivism among inmates released from Minnesota prisons between 2007 and 2009. Of
the 9,535 offenders incarcerated in MnDOC facilities and released to the community
during the 2007-2009 period, 1,766 (19 percent) spent at least a portion of their
confinement at the PCF. Due to eligibility criteria offenders had to meet in order to be
transferred to the PCF, propensity score matching was used to individually match the
1,766 offenders who spent time in the PCF with 1,766 offenders from the comparison
group pool (N = 7,769) who spent their entire confinement period in state correctional

3

facilities. In analyzing the data, Cox regression was used to determine whether
confinement at the PCF had an impact on recidivism.
In the next section, this report reviews prior research that has compared
recidivism outcomes and costs among private and public prisons. Next, it describes the
private prison experience in Minnesota in greater detail. Following a description of the
data and methods used in this study, the results from the statistical analyses are presented.
This report concludes by discussing the implications of the findings for correctional
policy and practice.
Prior Research Comparing Public and Private Prisons
Two major arguments made in favor of prison privatization—lower recidivism
rates and lower costs—have not received thorough scholarly attention. There have been
five major studies that compared the recidivism rates of public and private prisons
(Lanza-Kaduce, Parker, and Thomas, 1999; Lanza-Kaduce and Maggard, 2001; Farabee
and Knight, 2002; Bales et al., 2005; Spivak and Sharp, 2008) and one recent study that
compared the costs (Arizona Department of Corrections, 2011). In two separate studies
using the same data from Florida prisons, Lanza-Kaduce et al. (1999) and Lanza-Kaduce
and Maggard (2001) found that offenders released from private prisons had a lower risk
of recidivism compared to offenders released from public prisons. In the first study, the
authors matched approximately 200 male prisoners from medium security public and
private facilities. Individuals were matched based on race, type of offense, prior record,
and age. Within 12 months of release, the private prisoners were less likely to be arrested,
convicted of a new offense, and sent back to prison for a new offense compared to public
prisoners.

4

In the second study, Lanza-Kaduce and Maggard (2001) used a longer follow-up
period (four years). They replicated the results from the previous study using a relaxed
standard of significance (p < .10) and only for one measure of recidivism (Bales et al.,
2005). The authors found that private prisoners were less likely to be reincarcerated for
either a new offense or a technical violation compared to public prisoners.
Farabee and Knight (2002) found that the effects of public versus private prisons
varied based on gender but not age. They followed nearly 9,000 inmates released from
Florida prisons between 1997 and 2000 for up to three years. Recidivism was defined as a
new conviction and reincarceration. They found that male prisoners released from public
prisons did not have significantly different rates of recidivism for either measure of
recidivism. Among female prisoners, on the other hand, releasees from private prisons
were significantly less likely to be convicted of a new offense and less likely to be
reincarcerated compared to releasees from public prisons. Male offenders between the
ages of 18 and 24 released from public prisons did not have significantly different rates
of recidivism compared to their counterparts in private prisons.
Although the findings from the initial studies on Florida prisoners suggested that
private prison confinement reduced recidivism for some offenders (Farabee and Knight,
2002; Lanza-Kaduce and Maggard, 2001; Lanza-Kaduce et al., 1999), Bales and
colleagues (2005) reported that this effect washed out after making several
methodological improvements, including a more accurate measurement of private prison
exposure. Because many prisoners in states that have public and private facilities end up
spending some time in both, Bales et al. (2005) measured the extent to which offenders
were confined in public and private prisons. That is, they conducted multiple analyses

5

comparing groups based on where they were released and how much time they had spent
in public and private facilities. They followed a base sample of approximately 80,000
Florida inmates for up to 60 months after release. Overall, Bales and colleagues (2005)
found that private prison exposure had no effect on whether adult males, adult females, or
youthful offenders were reconvicted of a felony or reincarcerated for a new offense.
In the most recent study that compared the recidivism rates between public and
private prisons, Spivak and Sharp (2008) found that inmates released from private prisons
had higher rates of recidivism compared to releasees from public prisons. While the
authors replicated the Bales et al. (2005) study by using data from more than 23,000
inmates released from prisons in Oklahoma, they also extended the literature by
examining the proportion of time served in private and public facilities. Using a followup period that ranged between 36 and 84 months, Spivak and Sharp (2008) found that in
most of the models they tested, inmates released from private prisons were more likely to
return to prison compared to inmates released from public prisons.
None of the above studies compared the costs of private prisons to that of public
prisons. During the 1990s, the Government Accounting Office (GAO) (1991; 1996)
released two reports that examined whether private prisons save states and the federal
government money. After reviewing several unpublished studies that compared public
and private prison costs across several states, the GAO concluded that there is no
evidence that private prisons save money. Many of the studies they reviewed did not have
reliable data, and some offered misleading information. Of the few studies that have
provided conclusive evidence that private facilities save money over public facilities, the
savings were negligible. A study by the Bureau of Justice Assistance that compared the

6

costs of private and public prisons found that, at best, private prisons provide a savings of
only one percent compared to public facilities (Austin and Coventry, 2001). Moreover,
findings from several meta-analyses have shown that private prisons are no more costeffective than public prisons (Pratt and Maahs, 1999; Lundahl, Kunz, Brownell, Harris,
and Van Vleet, 2007).
A recent report from the Arizona Department of Corrections revealed that the
costs of private prisons are sometimes the same as, or greater than, the costs of running
public facilities (Arizona Department of Corrections, 2011). Private prisons have either
similar or higher costs in spite of the fact that in Arizona (and many other states) they
take only the healthiest and most well-behaved inmates (Oppel, 2011). Moreover, private
prisons usually offer less institutional programming and reentry services.
The general public’s admiration of private enterprise and distrust of government
contributes to a common assumption that private prisons are run more efficiently
compared to public prisons (Spivak and Sharp, 2008). In other words, private prisons are
thought to get the best results (lower recidivism) for the least amount of money. Although
many observers believe this proposition, few studies have actually compared the costs
and recidivism rates of public and private prisons, and the results have been mixed.
The present study contributes to the small body of literature on the relationship
between private prisons and recidivism in several ways. First, existing research has been
conducted in states that rely heavily on private prisons. Here, however, we examine the
performance of a private prison in a state—Minnesota—that has not historically relied on
private prisons. Second, this study extends prior research by using propensity score
matching to create a comparison group of offenders incarcerated only at state correctional

7

facilities. In doing so, we provide a counterfactual estimate of what would have likely
happened to the offenders incarcerated at the PCF had they only been incarcerated at state
correctional facilities. Third, existing research has measured recidivism as a reconviction,
reincarceration for a new offense, or any return to prison. We build upon prior research
by using four separate measures of recidivism (rearrest, reconviction, new offense
reincarceration, and technical violation revocation) to assess private prison performance.
Private Prison Confinement in Minnesota
Beginning in the late 1990s, the MnDOC began transferring male prisoners to the
PCF as a result of prison bed shortages created by prison population growth. The number
of prisoners transferred to the PCF, however, began to increase dramatically in 2004.
Indeed, the number of Minnesota offenders incarcerated at the PCF grew from less than
100 offenders on January 1, 2004, to nearly 1,200 (nearly 13 percent of the total prison
population) by January 1, 2008 (Minnesota Department of Corrections, 2004; 2008). A
little more than two years later, however, the MnDOC no longer housed any of its
prisoners at the PCF due to its closure.
The rise and fall in the number of Minnesota prisoners housed at the PCF was
largely due to historic shifts in Minnesota’s prison population. On January 1, 2000,
Minnesota’s prison population was a little more than 5,900 offenders (Minnesota
Department of Corrections, 2000). Six years later, the prison population had grown by
almost 50 percent, reaching nearly 8,900 offenders on January 1, 2006 (Minnesota
Department of Corrections, 2006). Much of this growth was due to the creation of a
felony driving while intoxicated (DWI) law in 2002 and a sharp increase in the number of
inmates incarcerated for methamphetamine offenses. After the “meth boom” subsided

8

and the DWI offender population plateaued, the total prison population began growing at
a much slower pace after 2006. Moreover, to accommodate the growth in the prison
population, the MnDOC began expanding bed space capacity at the Minnesota
Correctional Facility (MCF)-Faribault in 2008. As bed space shortages decreased in
response to the lack of growth and the MCF-Faribault expansion, so, too, did the
MnDOC’s need for PCF prison beds.
Although a relatively large number of Minnesota prisoners were confined at the
PCF, especially during the 2004-2009 period, offenders were required to meet a number
of criteria in order to be transferred to the PCF. For example, the PCF did not accept any
offenders over the age of 60, nor did it take any prisoners with serious medical conditions
or mental health disorders. Further, the PCF excluded offenders whose custody-level
classification was either secure or maximum, accepting only offenders who had a
medium or minimum custody level classification. Given that a prisoner’s classification is
determined largely by prior criminal history and behavior within the institution,
minimum- and medium-level inmates tend to have less serious criminal histories and/or
better behavior in prison than secure- and maximum-level offenders.
Overall, Minnesota prisoners confined at the PCF were relatively healthy
(mentally and physically), well-behaved inmates with limited criminal histories.
Nevertheless, compared to confining an inmate at a MnDOC facility, it cost about the
same to house an inmate at the PCF. For example, the MnDOC’s average marginal per
diem during the 2004-2009 period was nearly $65, which is the same as the daily rate the
MnDOC paid the PCF ($65) for housing an inmate (Minnesota Department of
Corrections, 2012).

9

Although confinement costs for the PCF and MnDOC facilities were roughly the
same, there were differences in the availability of programming. The PCF offered
inmates chemical dependency treatment, adult basic education, and some types of
vocational programming. While the MnDOC provided these types of programming, it
also offered other programming opportunities such as EMPLOY (an employment reentry
program), sex offender treatment, the Affordable Homes Program (a vocational
program), and the Minnesota Comprehensive Offender Reentry Plan (MCORP). These
programs have been shown to be effective in helping offenders find post-release
employment (Duwe, 2012a; Duwe, 2012b; Northcutt Bohmert and Duwe, 2012) or
reducing the risk of recidivism (Duwe, 2012a; Duwe, 2012b; Duwe and Goldman, 2009).
Like MnDOC facilities, the PCF offered visitation privileges to inmates. It is
worth noting, however, that Appleton (where the PCF is located) is about a three-hour
drive from the Twin Cities (Minneapolis and St. Paul), a seven-county metropolitan area
from which a majority of the prisoner population comes. Appleton is therefore about one
hour farther from the Twin Cities than the most distant MnDOC facility at Moose Lake.
Although it is unclear whether inmates were visited less frequently at the PCF due to the
lack of PCF visitation data, distance is presumed to reduce the likelihood of visitation
(Austin and Hardyman, 2004). Moreover, research on Minnesota prisoners has found that
prison visitation has a significant, albeit modest, association with reduced recidivism
(Duwe and Clark, 2011).
Data and Methodology
A retrospective quasi-experimental design was used to determine whether private
prison confinement had an impact on recidivism. The population for this study consisted

10

of 9,535 Minnesota prisoners released between 2007 and 2009. This time period was
selected because it captures most offenders who were confined at the PCF during its
period of optimum use (2004-2009). Because the PCF housed only male offenders,
female prisoners were excluded from this study. Similarly, this study excluded “shortterm offenders” who had a remaining term of imprisonment of 180 days or less at the
time they were sentenced to prison. Between July 2003 and June 2009, short-term
offenders were required to serve their prison time at a local jail. As a result, these
offenders were not confined at a MnDOC facility, nor were they eligible for transfer to
the PCF.
Dependent Variable
In this study, recidivism was defined as a 1) rearrest, 2) reconviction, 3)
reincarceration for a new sentence, or 4) revocation for a technical violation. It is
important to emphasize that the first three recidivism variables strictly measure new
criminal offenses. In contrast, technical violation revocations (the fourth measure)
represent a broader measure of rule-breaking behavior. Offenders can have their
supervision revoked for violating the conditions of their supervised release. Because
these violations can include activity that may not be criminal in nature (e.g., use of
alcohol, failing a community-based treatment program, failure to maintain agent contact,
failure to follow curfew, etc.), technical violation revocations do not necessarily measure
reoffending.
Recidivism data were collected on offenders through December 31, 2010. Given
that the offenders in this study were released between January 2007 and December 2009,
the follow-up time ranged from one year to four years with an average of 30 months.

11

Data on arrests and convictions were obtained electronically from the Minnesota Bureau
of Criminal Apprehension. Reincarceration and revocation data were derived from the
Correctional Operations Management System (COMS) database maintained by the
MnDOC. The main limitation with using these data is that they measure only arrests,
convictions or incarcerations that took place in Minnesota, although it is worth noting
that interstate recidivism is relatively low (Langan and Levin, 2002). As a result, the
findings presented later likely underestimate the true recidivism rates for the offenders
examined here.
To accurately measure the total amount of time offenders were actually at risk to
reoffend (i.e., “street time”), it was necessary to account for supervised release
revocations in the recidivism analyses. More specifically, for the three recidivism
variables that strictly measure new criminal offenses (rearrest, reconviction, and new
offense reincarceration), it was necessary to deduct the amount of time they spent in
prison for technical violation revocations from their total at-risk period. Failure to deduct
time spent in prison as a supervised release violator would artificially increase the length
of the at-risk periods for these offenders. Therefore, to achieve a more accurate measure
of “street time,” the time that an offender spent in prison as a supervised release violator
was subtracted from his at-risk period, but only if it preceded a rearrest, a reconviction, a
reincarceration for a new offense, or if the offender did not recidivate prior to January 1,
2011.
To illustrate, if an offender was released on January 1, 2007, he would have had
an at-risk period of 48 months given that December 31, 2010, was the end of the followup period. If it is assumed that this offender returned to prison for a technical violation

12

revocation on January 1, 2008, then his time to failure for a technical violation revocation
would be 12 months. If it is further assumed, however, that this offender was imprisoned
until June 30, 2008, for this technical violation revocation, our at-risk period calculations
in the rearrest, reconviction, and reincarceration analyses took into account the six
months this offender spent in prison from January 1-June 30, 2008. If this offender did
not recidivate with a new criminal offense, then his total at-risk period would have been
42 months instead of 48. If, however, this offender was rearrested for a new offense on
January 1, 2009, the time to rearrest would have been 18 months instead of 24.
Independent Variables
The main objective of this study is to evaluate the relationship between private
prison confinement and recidivism. As a result, the 1,766 offenders who served time in
the PCF were assigned a value of “1”, whereas the offenders in the comparison group
were given a value of “0”. Similar to the approach taken by Bales et al. (2005), two
additional measures of private prison exposure were created. These measures reflect the
fact that many offenders are imprisoned at multiple facilities during their sentences and
may not spend their entire terms of confinement in private facilities. For the first
measure, Private Prison Time, the total number of days offenders were incarcerated at the
PCF were counted. For the second measure, Private Prison Time Proportion, a variable
was created that calculated the proportion of time served at the PCF. For example, if an
offender’s term of imprisonment was 500 days and 200 of these days were served at the
PCF, then the Private Prison Time Proportion for this offender would be 0.40.

13

The independent, or control, variables included in the statistical models were
those that were not only available in the COMS database but also might have an impact
on recidivism and PCF selection. Table 1 describes the covariates used in the statistical
Table 1. Logistic Regression Model for Private Prison Selection
Predictors
Predictor Description
Coefficient
Minority
Age at Release (years)
Prior Supervision
Failures
Prior Convictions
LSI-R Score
Mental Health

Medical Health
Offense Type
Property
Drugs
Other
Criminal Sexual
Conduct
Felony DWI
New Court
Commitment
Metro Commit
Length of Stay
(months)
Supervision Type
ISR
CIP
Work Release
Discharge
Release Year
Constant
N
Log-likelihood
Nagelkerke R2

Minority = 1; White = 0
Offender age in years at time of release from prison
Number of prior revocations while under correctional
supervision
Number of prior felony convictions, excluding index
conviction(s)
Most recent Level of Service Inventory-Revised (LSIR) score prior to release
Mental health problems (major mental illness,
significant mood disorder, brain injury or is
intellectually impaired) = 1; no critical mental
problems = 0
Medical health problems (visual impairment, speech
impairment, mobility impairment, hearing loss,
paraplegic, etc.) = 1; no medical health problems = 0
Person offense serves as the reference
Property offense = 1; non-property offense = 0
Drug offense = 1; non-drug offense = 0
Other offense = 1; non-other offense = 0
Sex offense = 1; non-sex offense = 0
Felony DWI offense = 1; non-Felony DWI offense = 0
New commitment = 1; probation or release violator = 0
Twin Cities metropolitan area = 1; Greater Minnesota
=0
Number of months between prison admission and
release dates
Supervised release serves as the reference
Intensive supervised release (ISR) = 1; non-ISR = 0
Challenge Incarceration Program (CIP) = 1; non-CIP =
0
Work Release = 1; non-Work Release = 0
Discharge = 1; released to correctional supervision = 0
Year in which first released from prison for instant
offense

** p < .01
* p < .05

0.032
-0.010**

Standard
Error

-0.084**

0.026

0.010

0.008

-0.033**

0.005

-0.019

0.074

-0.279**

0.061

0.434**
1.131**
0.676**

0.111
0.099
0.108

-1.057**
1.398**

0.189
0.122

0.827**

0.074

-0.038

0.060

0.012**

0.001

-0.275**

0.092

0.160
0.218**
0.398

0.105
0.079
0.268

0.162**
-327.425
9,535
7968.569
0.187

0.035
69.581

models. Several measures commonly associated with recidivism risk were used,
including the offender’s race, age, number of prior supervision failures and prior

14

0.062
0.003

convictions, and Level of Service Inventory- Revised (LSI-R) score (Gendreau, Little, &
Goggin, 2006). Because private prisons, including the PCF, often do not accept inmates
with certain mental and medical conditions (Arizona Department of Corrections, 2011;
Oppel, 2011), several measures pertaining to medical and mental health were included.
Similar studies on private prison confinement and recidivism have found that offense
type (e.g., person offense, drug offense), length of sentence, and post-release supervision
significantly affect the outcome measure (Bales et al., 2005; Spivak & Sharp, 2008).
Thus, these measures were included in the analyses. Prior research has shown that
admission type (New Commit), county of commitment (Metro), and length of stay are
significant predictors of recidivism for Minnesota prisoners (Duwe, 2010; Duwe and
Clark, 2011), which is why they were included in this study. Measures relating to
participation in institutional programming (e.g., chemical dependency treatment),
visitation, or institutional discipline were excluded because these data were not available
for offenders while they were housed at the PCF.
Propensity Score Matching
PSM is a method that estimates the conditional probability of selection to a
particular treatment or group given a vector of observed covariates (Rosenbaum & Rubin,
1985). The predicted probability of selection, or propensity score, is typically generated
by estimating a logistic regression model in which selection (0 = no selection; 1 =
selection) is the dependent variable while the predictor variables consist of those that
theoretically have an impact on the selection process. Once estimated, the propensity
scores are then used to match individuals who entered treatment with those who did not.

15

Thus, an advantage with using PSM is that it can simultaneously “balance” multiple
covariates on the basis of a single composite score.
In matching private prison offenders with public prison offenders on the
conditional probability of entering the PCF, PSM reduces selection bias by creating a
counterfactual estimate of what would have happened to the private prison offenders had
they not been housed at the PCF. PSM has several limitations, however, that are worth
noting. First, and foremost, because propensity scores are based on observed covariates,
PSM is not robust against “hidden bias” from unmeasured variables that are associated
with both the assignment to treatment and the outcome variable. Second, there must be
substantial overlap among propensity scores between the two groups in order for PSM to
be effective (Shadish, Cook & Campbell, 2002); otherwise, the matching process will
yield incomplete or inexact matches. Finally, as Rubin (1997) points out, PSM tends to
work best with large samples.
Although somewhat limited by the data available, an attempt was made to address
potential concerns over unobserved bias by including as many theoretically-relevant
covariates (20) as possible in the propensity score model. In addition, this study later
demonstrates there was substantial overlap in propensity scores between the treated and
untreated offenders. Further, the sample size limitation was addressed by assembling a
large number of cases (N = 9,535) on which to conduct the propensity score analyses.
Matching Private and Public Prison Inmates
Propensity scores were calculated for the 1,766 private prison participants and the
7,769 non-participants in the comparison group pool by estimating a logistic regression
model in which the dependent variable was private prison incarceration. The predictors

16

were the 20 control variables used in the statistical analyses (see Table 1). The results
show a number of factors that predicted whether offenders were incarcerated at the PCF.
The Nagelkerke R2 value of 0.187 suggests the model explained nearly 19 percent of the
variability in entering the PCF.
In Table 1, the results show that the odds of serving time at the PCF were
significantly greater for younger offenders, offenders with fewer prior supervision
failures, offenders with lower LSI-R scores, offenders with fewer medical health
problems, and offenders admitted to prison as a new court commitment. Regarding
offense type, offenders incarcerated for property, drug, DWI, and other offenses were
significantly more likely to be housed at the PCF than offenders imprisoned for a person
offense. Sex offenders, however, were significantly less likely to be housed at the PCF.
Regarding post-release supervision, we also see that offenders confined at the PCF were
significantly less likely to be intensively supervised, although they were more likely to be
placed on work release. Lastly, release year was positively associated with entering the
PCF, which reflects the fact that the number of offenders entering the facility increased
over the period examined in this study.
As shown in Table 2, the difference in mean propensity score between private and
public prison offenders was statistically significant at the .01 level. Still, there was
substantial overlap in propensity scores. Indeed, the vast majority of offenders in both
groups (93 percent for private prison and 98 percent for public prison) had propensity
scores less than 0.50.
After obtaining propensity scores for the 9,535 offenders, a “greedy” matching
procedure that utilized a without replacement method was used to match the private

17

Table 2. Propensity Score Matching and Covariate Balance for Private Participation
Variable

Sample

Propensity Score

Total
Matched
Total
Matched
Total
Matched
Total
Matched
Total
Matched
Total
Matched
Total
Matched
Total
Matched
Total
Matched
Total
Matched
Total
Matched
Total
Matched
Total
Matched
Total
Matched
Total
Matched
Total
Matched
Total
Matched
Total
Matched
Total
Matched
Total
Matched
Total
Matched

Minority
Age at Release (Years)
Prior Supervision Failures
Prior Convictions
LSI-R Score
Mental Health
Medical Health
Property
Drugs
Other
Criminal Sexual Conduct
Felony DWI
New Court Commit
Metro
Length of Stay (Months)
ISR
CIP
Work Release
Discharge
Release Year

Total Private = 1,766
Total Public = 7,769
Matched Private = 1,766
Matched Public = 1,766

Private
Mean
28.56%
28.56%
47.06%
47.06%
35.03
35.03
0.74
0.74
4.89
4.89
23.82
23.82
18.40%
18.40%
35.84%
35.84%
13.93%
13.93%
40.49%
40.49%
14.04%
14.04%
2.15%
2.15%
15.80%
15.80%
81.20%
81.20%
50.06%
50.06%
29.46
29.46
13.02%
13.02%
13.36%
13.36%
19.65%
19.65%
1.13%
1.13%
2008.09
2008.09

18

Public Mean
16.23%
28.54%
49.30%
46.15%
34.89
35.23
1.21
0.71
5.05
4.79
27.22
23.43
26.08%
19.00%
46.09%
36.75%
21.11%
12.85%
22.14%
39.64%
15.25%
15.06%
10.22%
2.55%
7.50%
16.25%
61.17%
80.18%
54.00%
50.74%
21.79
30.03
26.73%
14.78%
4.85%
14.27%
11.38%
18.23%
1.24%
1.02%
2007.96
2008.08

Bias
(%)
75.04
0.12
3.66
1.49
1.19
1.74
28.98
2.05
3.19
2.06
39.31
4.64
15.47
1.26
17.22
1.55
15.92
2.57
32.09
1.42
2.81
2.37
31.18
2.19
20.28
1.00
38.41
2.12
6.44
1.11
30.45
1.95
29.85
4.20
22.83
2.16
18.11
2.94
0.84
0.87
13.21
1.01

Bias
Reduction
-99.84%
-59.35%
45.60%
-92.93%
-35.43%
-88.19%
-91.86%
-91.02%
-83.83%
-95.59%
-15.59%
-92.99%
-95.05%
-94.49%
-82.76%
-93.60%
-85.94%
-90.52%
-83.74%
3.08%
-92.32%

t test p
Value
0.00
0.96
0.09
0.59
0.61
0.52
0.00
0.40
0.15
0.47
0.00
0.09
0.00
0.67
0.00
0.58
0.00
0.35
0.00
0.61
0.20
0.39
0.00
0.44
0.00
0.71
0.00
0.44
0.00
0.69
0.00
0.53
0.00
0.13
0.00
0.44
0.00
0.28
0.72
0.74
0.00
0.82

prison offenders with the public prison inmates. Private prison offenders were matched to
public prison offenders who had the closest propensity score (i.e., “nearest neighbor”)
within a caliper (i.e., range of propensity scores) of 0.10. Matches were found for all
1,766 private prison offenders. Table 2 presents the covariate and propensity score means
for both groups prior to matching (“total”) and after matching (“matched”). In addition
to tests of statistical significance (“t test p value”), Table 2 provides a measure (“Bias”)
developed by Rosenbaum and Rubin (1985) that quantifies the amount of bias between
Bias =

100(X t - X c )

( S t2 + S c2 )
2
the treatment and comparison samples (i.e., standardized mean difference between
samples), where X t and St2 represent the sample mean and variance for the treated
offenders and X c and S c2 represent the sample mean and variance for the untreated
offenders. If the value of this statistic exceeds 20, the covariate is considered to be
unbalanced (Rosenbaum & Rubin, 1985).
As shown in Table 2, the matching procedure reduced the bias in propensity
scores between the private and public prison offenders by 99 percent. Whereas the p
value was 0.00 in the unmatched sample, it was 0.96 in the matched sample. In the
unmatched sample, there were nine covariates that were significantly imbalanced (i.e.,
the bias values exceeded 20). But in the matched sample, covariate balance was achieved
given that no covariates had bias values greater than 20. For example, in the unmatched
sample, the average LSI-R score for the 1,766 offenders in the private prison group was
23.82 versus an average score of 27.22 for the 7,769 offenders in the comparison group
pool. The difference in LSI-R scores between the two groups was statistically significant,

19

and the bias value was 39.31. After the matching procedure was used, however, the
average LSI-R score for the 1,766 offenders in the comparison group was 23.43, which
yielded a bias value of 4.64 and a t test value no longer significant at the .05 level.
Analysis
In analyzing recidivism, survival analysis models are preferable in that they
utilize time-dependent data, which are important in determining not only whether
offenders recidivate but also when they recidivate. As a result, this study uses a Cox
regression model, which uses both “time” and “status” variables in estimating the impact
of the independent variables on recidivism. For the analyses presented here, the “time”
variable measures the amount of time from the date of release until the date of first
rearrest, reconviction, reincarceration, technical violation revocation, or December 31,
2010, for those who did not recidivate. The “status” variable, meanwhile, measures
whether an offender recidivated (rearrest, reconviction, reincarceration for a new crime,
and technical violation revocation) during the period in which he was at risk to recidivate.
In the analyses presented below, Cox regression models were estimated for each of the
four recidivism measures.
Results
Recidivism rates are presented in Table 3 for offenders housed in the PCF
(private), offenders in the matched comparison group housed only in MnDOC facilities
(public), and all offenders housed only in MnDOC facilities (public comparison group
pool). The results show that the 7,769 offenders in the public comparison group pool had
higher rates for all four recidivism measures than the 1,766 offenders housed at the PCF.
Yet, after balancing the two groups on factors related to PCF selection and/or recidivism

20

risk, the results indicate that the private prison offenders had slightly higher recidivism
rates than those in the public prison comparison group.

Table 3. Recidivism Rates by Private and Public Prison
Recidivism
Private
Public
Rearrest
Reconviction
New Offense Reincarceration
Technical Violation Revocation
N

46.5%
31.0%
14.0%
33.2%
1,766

41.4%
25.8%
13.0%
31.0%
1,766

Public
Comparison Pool
54.5%
37.0%
18.7%
39.9%
7,769

Although these findings provide little indication as to whether private prison
confinement had a significant impact on recidivism, the observed recidivism differences
between the private and public prison offenders may be due to other factors such as time
at risk, amount of time incarcerated at a private facility, and proportion of prison term
served at a private facility. To statistically control for the impact of these other factors on
reoffending, Cox regression models were estimated for each private prison measure
across the four measures of recidivism. To determine model fit, the assumptions that the
hazards are proportional and that the relationships between the log hazard and covariates
are nonlinear were tested. Inspection of the residuals revealed that each of the Cox
regression models estimated in this study adequately fit the data.
The Impact of Private Prison on Recidivism
The results in Table 4 indicate that, controlling for the effects of the other
independent variables in the statistical model, offenders housed at the PCF for any period
of time had a significantly greater hazard ratio for two of the four recidivism measures
examined. In the rearrest model, the hazard ratio for the private prison variable was

21

1.133, which indicates that private prison confinement significantly increased the risk of
rearrest by 13 percent. In the reconviction model, the hazard ratio for the private prison
variable was 1.221, which suggests that private prison confinement increased the risk of
reconviction by 22 percent. Private prison confinement did not have a significant impact
on either new offense reincarceration or reincarceration for a technical violation
revocation.
Table 4. Cox Regression: Impact of Any Private Prison on the Hazard of Recidivism
Rearrest
Reconviction
Reincarceration
Private Prison
Minority
Age at Release (years)
Prior Supervision Failures
Prior Convictions
LSI-R Score
Mental Health
Medical Health
Offense Type
Property
Drugs
Other
Criminal Sexual Conduct
Felony DWI
New Court Commitment
Metro
Length of Stay (months)
Supervision Type
ISR
CIP
Work Release
Discharge
Release Year
Supervised Release Revocations
N
** p < .01
* p < .05

Hazard
Ratio
1.133*
1.278**
0.961**
1.169**
1.067**
1.028**
1.064
1.147*

SE

0.051
0.056
0.004
0.022
0.007
0.004
0.066
0.056

Hazard
Ratio
1.221**
1.108
0.960**
1.183**
1.055**
1.032**
1.101
1.207**

1.202
0.977
1.005
0.510*
0.987
1.060
1.352**
0.994**

0.100
0.091
0.096
0.299
0.110
0.068
0.055
0.001

0.922
0.538**
0.883
1.415
0.943
0.829**
3,532

0.083
0.106
0.070
0.209
0.034
0.051

SE

0.065
0.069
0.004
0.026
0.009
0.005
0.080
0.069

Hazard
Ratio
1.078
1.242*
0.963**
1.281**
1.033**
1.037**
1.558**
1.604**

1.338*
1.024
1.092
0.565
1.166
1.175
1.262**
0.992**

0.124
0.116
0.121
0.392
0.139
0.084
0.068
0.002

0.820
0.457**
0.854
2.116**
0.960
0.978
3,532

0.106
0.143
0.087
0.226
0.045
0.051

SE

Revocation

0.096
0.102
0.006
0.033
0.012
0.008
0.106
0.099

Hazard
Ratio
1.050
1.608**
0.973**
1.162**
1.018*
1.034**
1.337**
1.258**

0.060
0.065
0.004
0.024
0.009
0.005
0.074
0.064

1.593**
0.941
0.852
0.830
1.442
1.794**
1.277*
0.986**

0.178
0.174
0.184
0.525
0.200
0.131
0.100
0.003

1.017
0.997
1.141
1.788**
1.362*
0.969
1.093
0.999

0.122
0.109
0.112
0.210
0.128
0.077
0.064
0.001

0.654**
0.342**
0.973
3.670**
1.029
0.867*
3,532

0.163
0.253
0.125
0.264
0.068
0.067

1.718**
1.185
1.673**

0.091
0.116
0.079

0.990

0.039

3,494

The results also showed the hazard ratio was significantly greater for younger
offenders (all four measures), prior supervision failures (all four measures), prior
convictions (all four measures), offenders with higher LSI-R scores (all four measures),

22

SE

offenders with medical concerns (all four measures), minority inmates (three measures),
offenders committed from the Metro area (three measures), shorter lengths of stay in
prison (three measures), offenders with mental health concerns (two measures), inmates
who were released to no supervision (two measures), property offenders (two measures),
new court commitments (new offense reincarceration), sex offenders (technical violation
revocations), DWI offenders (technical violations revocations), offenders placed on work
release (technical violation revocations), and those released to ISR (technical violation
revocations). The risk (hazard) of revocation for a technical violation was significantly
lower, however, for CIP offenders (three measures), supervised release revocations (two
measures), sex offenders (rearrest), and offenders placed on ISR (new offense
reincarceration).
Of the 1,766 offenders who were confined at the PCF, their lengths of stay at that
facility ranged from one day to 2,912 days (about eight years) (SD = 8.61). The average
length of stay at the PCF for these offenders was 349 days (almost a year), whereas the
median length of stay was 295 days. In Table 5, the results that examined the impact of
total time spent at the PCF (in days) on the four recidivism measures are presented. The
findings indicate that private prison had a significant effect on only one of the four
recidivism measures, increasing the hazard for reconviction. More specifically, one
additional day in confinement at the PCF increased the risk of reconviction by 1.3
percent. Because the findings for the other covariates are similar to those presented in
Table 4, they will not be repeated here.

23

Table 5. Cox Regression: Impact of Time in Private Prison on the Hazard of Recidivism
Rearrest
Reconviction
Reincarceration
Revocation
Private Prison Time (Days)
Minority
Age at Release (years)
Prior Supervision Failures
Prior Convictions
LSI-R Score
Mental Health
Medical Health
Offense Type
Property
Drugs
Other
Criminal Sexual Conduct
Felony DWI
New Court Commitment
Metro
Length of Stay (months)
Supervision Type
ISR
CIP
Work Release
Discharge
Release Year
Supervised Release Revocations
N
** p < .01
* p < .05

Hazard
Ratio
1.003
1.276**
0.961**
1.169**
1.067**
1.028**
1.064
1.149*

SE

0.003
0.056
0.004
0.022
0.007
0.004
0.066
0.056

Hazard
Ratio
1.013**
1.104
0.959**
1.185**
1.056**
1.032**
1.105
1.217**

1.196
0.965
0.992
0.507*
0.983
1.062
1.355**
0.994**

0.100
0.091
0.096
0.299
0.110
0.068
0.055
0.001

0.916
0.547**
0.885
1.430
0.941
0.830**
3,532

0.083
0.107
0.070
0.208
0.034
0.051

SE

0.004
0.069
0.004
0.026
0.009
0.005
0.080
0.069

Hazard
Ratio
1.001
1.240*
0.963**
1.282**
1.033**
1.038**
1.557**
1.603**

1.322*
1.008
1.067
0.566
1.152
1.175
1.268**
0.991**

0.124
0.116
0.121
0.392
0.139
0.085
0.068
0.002

0.818
0.479**
0.859
2.131**
0.954
0.982
3,532

0.106
0.143
0.087
0.226
0.045
0.051

SE

0.007
0.102
0.006
0.033
0.012
0.008
0.106
0.099

Hazard
Ratio
1.001
1.607**
0.973**
1.162**
1.018*
1.034**
1.336**
1.259**

0.004
0.065
0.004
0.024
0.009
0.005
0.074
0.064

1.593**
0.933
0.848
0.829
1.437
1.796**
1.280*
0.986**

0.178
0.174
0.184
0.525
0.200
0.131
0.100
0.003

1.015
0.993
1.137
1.782**
1.361*
0.969
1.093
0.999

0.122
0.109
0.112
0.210
0.128
0.077
0.064
0.001

0.649*
0.348**
0.974
3.702**
1.028
0.867*
3,532

0.163
0.252
0.125
0.263
0.068
0.067

1.714**
1.189
1.675**

0.091
0.116
0.079

0.990

0.039

3,494

The 1,766 offenders housed at the PCF spent, on average, 40 percent of their term
of imprisonment at that facility. The median proportion of time served at the PCF was 38
percent. The minimum percentage of time served at PCF was less than one percent,
whereas the maximum was 92 percent (SD = 0.23). In Table 6, the results from the Cox
regression models that estimated the impact of proportion of prison time served at the
PCF on the four recidivism measures are presented. The proportion of prison time served
at the PCF significantly increased the hazard by 21 percent for rearrest and 52 percent for
reconviction. Put another way, the greater the proportion of private prison

24

SE

Table 6. Cox Regression: Impact of Private Prison Proportion on the Hazard of Recidivism
Rearrest
Reconviction
Reincarceration
Revocation
Private Prison Time Proportion
Minority
Age at Release (years)
Prior Supervision Failures
Prior Convictions
LSI-R Score
Mental Health
Medical Health
Offense Type
Property
Drugs
Other
Criminal Sexual Conduct
Felony DWI
New Court Commitment
Metro
Length of Stay (months)
Supervision Type
ISR
CIP
Work Release
Discharge
Release Year
Supervised Release Revocations
N
** p < .01
* p < .05

Hazard
Ratio
1.210*
1.276**
0.961**
1.169**
1.067**
1.028**
1.068
1.152*

SE

0.095
0.056
0.004
0.022
0.007
0.004
0.066
0.056

Hazard
Ratio
1.520**
1.108
0.959**
1.184**
1.056**
1.032**
1.109
1.219**

1.198
0.974
0.999
0.511*
0.992
1.061
1.355**
0.995**

0.100
0.091
0.096
0.300
0.110
0.068
0.055
0.001

0.923
0.547**
0.886
1.418
0.943
0.829**
3,532

0.083
0.106
0.070
0.208
0.034
0.051

SE

0.116
0.069
0.004
0.026
0.009
0.005
0.080
0.069

Hazard
Ratio
1.107
1.241*
0.963**
1.281**
1.034**
1.038**
1.559**
1.606**

1.332*
1.029
1.090
0.571
1.184
1.180
1.267**
0.992**

0.124
0.116
0.121
0.392
0.139
0.084
0.068
0.002

0.823
0.475**
0.858
2.054**
0.958
0.980
3,532

0.106
0.143
0.087
0.226
0.045
0.051

SE

0.175
0.102
0.006
0.033
0.012
0.008
0.106
0.099

Hazard
Ratio
1.056
1.607**
0.973**
1.162**
1.019*
1.034**
1.337**
1.259**

0.116
0.065
0.004
0.024
0.009
0.005
0.074
0.064

1.593**
0.938
0.851
0.831
1.445
1.798**
1.279*
0.986**

0.178
0.174
0.184
0.525
0.201
0.131
0.100
0.003

1.015
0.995
1.139
1.787**
1.363*
0.969
1.093
0.999

0.122
0.109
0.112
0.210
0.128
0.077
0.064
0.001

0.651**
0.348**
0.974
3.658**
1.028
0.867*
3,532

0.163
0.252
0.125
0.264
0.068
0.067

1.715**
1.190
1.675**

0.091
0.116
0.079

0.990

0.039

3,494

confinement time, the greater the risk of rearrest and reconviction. Proportion of
confinement time at the PCF did not have a significant effect, however, on the other two
recidivism measures (new offense reincarceration and technical violation revocations).
In an effort to address concerns that the three private prison measures used above
were insufficient to examine the effects of private prison confinement, two additional sets
of analyses were conducted. Private prison confinement was limited to a) prisoners who
served at least one year at the PCF and b) offenders whose percentage of time served at
the PCF was at least 50 percent of their total confinement time. Next, propensity score

25

SE

Table 7. Cox Regression: Impact of Private Prison Time in Excess of One Year on the Hazard of
Recidivism
Rearrest
Reconviction
Reincarceration
Revocation
Private Prison Time (> 1 year)
Minority
Age at Release (years)
Prior Supervision Failures
Prior Convictions
LSI-R Score
Mental Health
Medical Health
Offense Type
Property
Drugs
Other
Criminal Sexual Conduct
Felony DWI
New Court Commitment
Metro
Length of Stay (months)
Supervision Type
ISR
CIP
Work Release
Discharge
Release Year
Supervised Release Revocations
N
** p < .01
* p < .05

Hazard
Ratio
1.000
1.172
0.954**
1.230**
1.066**
1.024**
1.007
1.331**

SE

0.004
0.089
0.006
0.036
0.013
0.007
0.117
0.089

0.941
0.959
1.045
0.403
0.903
1.275*
1.298**
0.996

0.165
0.141
0.149
0.595
0.182
0.121
0.087
0.002

0.965
0.587*
0.908
1.464
0.967
0.814*
1,406

0.137
0.258
0.110
0.388
0.055
0.088

Hazard
Ratio
1.012*
1.071
0.958**
1.238**
1.072**
1.029**
0.950
1.345**

SE

0.005
0.112
0.007
0.045
0.016
0.008
0.144
0.110

Hazard
Ratio
1.007
1.360
0.954**
1.431**
1.066**
1.045**
1.482
1.696**

1.016
0.782
0.922
0.228
0.687
1.393*
1.164

0.197
0.176
0.188
1.018
0.228
0.153
0.108

0.994*
0.746
0.535
0.980
2.306*
0.981
1.017
1,406

0.003
0.178
0.371
0.135
0.395
0.074
0.090

SE

0.008
0.183
0.012
0.059
0.024
0.014
0.204
0.176

Hazard
Ratio
1.000
1.395**
0.971**
1.188**
1.019
1.035**
1.341*
1.119

SE

0.005
0.106
0.007
0.042
0.017
0.008
0.129
0.107

1.881
1.284
1.379
1.435
1.427
3.098**
1.270

0.352
0.333
0.348
1.058
0.390
0.291
0.177

1.128
0.958
1.229
1.493
1.462
0.999
1.026

0.194
0.171
0.175
0.379
0.209
0.137
0.102

0.996
0.814
0.937
1.283
1.658
0.881
0.807
1,406

0.005
0.290
0.610
0.215
0.726
0.123
0.124

1.001
1.850**
1.115
1.552**

0.002
0.147
0.289
0.125

1.049

0.065

1,392

analyses were conducted to create a comparison group for the 703 offenders who served
at least one year at the PCF and another comparison group for the 625 offenders whose
percentage of time served at the PCF was 50 percent or higher. After creating comparison
groups that were not significantly different from these two private prison groups, Cox
regression models were estimated for both private prison measures among the four
recidivism outcomes, resulting in eight additional models.
Consistent with the results shown above, the results show that private prison
confinement increased the hazard of recidivism for all eight models. The effects were

26

Table 8. Cox Regression: Impact of Private Prison Time (50% or more) on the Hazard of
Recidivism
Rearrest
Reconviction
Reincarceration
Revocation
Private Prison Time > 50%
Minority
Age at Release (years)
Prior Supervision Failures
Prior Convictions
LSI-R Score
Mental Health
Medical Health
Offense Type
Property
Drugs
Other
Criminal Sexual Conduct
Felony DWI
New Court Commitment
Metro
Length of Stay (months)
Supervision Type
ISR
CIP
Work Release
Discharge
Release Year
Supervised Release Revocations
N
** p < .01
* p < .05

Hazard
Ratio
1.281*
1.078
0.963**
1.216**
1.038**
1.026**
1.090
1.383**

SE

0.124
0.092
0.005
0.034
0.013
0.007
0.113
0.090

Hazard
Ratio
1.614**
0.947
0.964**
1.283**
1.028
1.029**
0.914
1.475**

1.147
1.004
1.157
0.249
1.130
1.335*
1.328**
0.991**

0.147
0.142
0.138
0.726
0.220
0.119
0.092
0.003

1.056
0.572*
0.917
1.909**
1.043
0.877
1,250

0.139
0.250
0.115
0.240
0.058
0.087

SE

0.151
0.112
0.007
0.038
0.015
0.008
0.137
0.109

Hazard
Ratio
1.448
1.076
0.969**
1.299**
1.020
1.050**
1.196
2.146**

1.055
0.960
1.034
0.356
0.990
1.565**
1.352**
0.988**

0.177
0.172
0.170
0.736
0.276
0.143
0.112
0.004

0.841
0.407*
0.965
2.811**
1.152
1.061
1,250

0.178
0.374
0.139
0.255
0.073
0.084

SE

0.235
0.174
0.010
0.048
0.021
0.013
0.187
0.167

Hazard
Ratio
1.073
1.696**
0.969**
1.232**
1.006
1.023**
1.432**
1.272*

0.156
0.116
0.007
0.039
0.017
0.009
0.136
0.115

2.163*
1.358
1.409
1.097
2.304
2.436**
1.421*
0.986*

0.308
0.309
0.309
0.793
0.434
0.240
0.176
0.007

1.343
1.043
1.233
2.080
2.060**
1.099
0.951
1.000

0.191
0.192
0.187
0.444
0.254
0.144
0.114
0.003

0.640
0.570
1.010
7.434**
1.265*
0.948
1,250

0.312
0.613
0.226
0.325
0.118
0.110

1.969**
1.124
1.860**

0.159
0.290
0.134

1.011

0.069

1,218

statistically significant, however in only three of the eight models. Total private prison
time served (i.e., 12 months or more) significantly increased the hazard of reconviction
(see Table 7), whereas the percentage of private prison time served measure (i.e., 50
percent or more) significantly increased the risk of rearrest and reconviction (see Table
8).
Conclusion
The findings showed that private prison incarceration was associated with a
greater risk of recidivism in all 20 Cox regression models that were estimated. This

27

SE

association was statistically significant, however, in only 8 of the 20 models. In
particular, all five private prison measures significantly increased the risk of reconviction,
whereas three measures (any private prison, private prison time proportion, and offenders
who served at least half of their time in the PCF) had a significant effect on rearrest.
None of the private prison variables, however, had a significant impact on either
reincarceration measure (new offense reincarceration and technical violation revocation).
While the findings suggest that time spent at the PCF did not have a beneficial impact on
recidivism outcomes, it should be emphasized that PCF confinement did not significantly
increase risk consistently across all recidivism measures examined. Moreover, the
magnitude of increased recidivism risk was relatively modest (13 percent for rearrest and
22 percent for reconviction) in the models that analyzed any exposure to private prison.
Findings from the three most recent evaluations (including this one) on prisoners
in three different states suggest that confinement in private prisons does not lead to
improved recidivism outcomes. For example, in the most rigorous study on Florida
prisoners to date, Bales and colleagues (2005) found no difference in recidivism between
public and private facilities. Spivak and Sharp (2008), meanwhile, reported that private
prison confinement among Oklahoma prisoners was significantly associated with
increased recidivism in most of the models they estimated. Like Spivak and Sharp
(2008), we found that private prison confinement significantly increased offender
recidivism risk in some, but not all, of the models that were tested.
The recidivism results observed for private prisons may be attributable to a lack
of visitation and rehabilitative programming in comparison to state-operated facilities.
Although PCF programming and visitation data were not available, which is perhaps the

28

most significant limitation with this study, the PCF offered fewer programming
opportunities for offenders compared to MnDOC facilities. To be sure, the PCF provided
offenders with programming to address chemical dependency, educational, and
vocational needs. The MnDOC facilities, however, offered a greater variety of
programming, some of which has been demonstrated to increase employment (Duwe,
2012b; Northcutt Bohmert and Duwe, 2012) and lower recidivism (Duwe, 2012a; Duwe,
2012b; Duwe and Goldman, 2009). Moreover, even though the PCF provided offenders
with visitation opportunities, the facility’s distance from Minnesota’s major population
center (Twin Cities) may have resulted in fewer visits, at least in comparison to MnDOC
facilities. Given that visitation has a significant, although modest, impact on recidivism
(Duwe and Clark, 2011), fewer visits may have contributed to slightly higher recidivism
rates for those who were confined at the PCF.
A cost-benefit analysis was outside the scope of this evaluation, although it is
worth reiterating that the daily rate the MnDOC paid the PCF was roughly the same as
the MnDOC’s marginal per diem. As such, the PCF did not incarcerate these offenders
less expensively than the MnDOC. Yet, due to PCF eligibility criteria, the MnDOC
confined offenders who are, compared to those placed at the PCF, generally more
expensive to incarcerate (i.e., older, less healthy, and more likely to have behavioral
problems). Moreover, the higher recidivism rates observed among the PCF offenders
relative to the comparison group may have resulted in more crimes committed postrelease, producing increased reincarceration and victimization costs to the state.
Confining offenders at the PCF, then, may have ultimately cost the state more money
than if they had been able to remain at a MnDOC facility.

29

As noted earlier, however, there was no room for these offenders at MnDOC
facilities. When prison bed shortages arise, corrections agencies generally have three
options: 1) construct a new facility, 2) add bed space capacity to an existing facility
and/or 3) transfer inmates to private or public (local, state, or federal) correctional
facilities where space is available. Whereas the first two options are geared towards
addressing anticipated long-term population growth, transferring offenders to other
correctional facilities provides a viable alternative to address rapid, short-term growth. In
2008, the MnDOC began to gradually expand the bed space capacity at the MCFFaribault to accommodate increases in the prison population over the long term. Prior to
achieving a more long-lasting solution, however, the PCF provided the MnDOC with a
release valve to help manage prison bed space shortages in the interim.
As this evaluation suggests, private prisons can offer correctional agencies a
valuable resource—prison bed space—during periods of sharp population growth. The
value of this resource declines, however, if bed space is available in state-operated
facilities. More specifically, the evidence from this evaluation and prior studies indicates
that private prisons are not a superior alternative to state-run prisons. The findings from
this study suggest, if anything, that private prisons produce slightly worse recidivism
outcomes among the healthiest and best-behaved inmates for the same amount of money.
Given the small number of existing studies, however, more research is needed
from other states that utilize private prisons to draw a more definitive conclusion
regarding the relative effectiveness of private and public prisons. To that end, it is
recommended that future research should not only compare recidivism outcomes between
offenders confined at private and state-operated prisons, but that it should also link these

30

outcomes to the provision of programming at both types of facilities. Future studies
should also take the costs of recidivism into account when examining the cost
effectiveness of private and public prisons.

31

References
Arizona Department of Corrections (2011). FY 2010 Operating Per Capita Cost Report:
Cost Identification and Comparison of State and Private Contract Beds. Phoenix,
AZ: Arizona Department of Corrections, Bureau of Planning, Budget and
Research.
Austin, J., & Coventry, G. (2001). Emerging Issues on Privatized Prisons. Monograph.
Washington, DC: U.S. Department of Justice, Office of Justice Programs, Bureau
of Justice Assistance.
Austin, J., & Hardyman, P.L. (2004). The Risks and Needs of the Returning Prisoner
Population. Review of Policy Research, 21, 13-29.
Bales, W.D., Bedard, L.E., Quinn, S.T., Ensley, D.T., & Holley, G.P. (2005). Recidivism
of Public and Private State Prison Inmates in Florida. Criminology and Public
Policy, 4, 57-82.
Cheung, A. (2004). Prison Privatization and the Use of Incarceration. Washington, DC:
The Sentencing Project.
Cook, N. (2010). “How the Recession Hurts Private Prisons.” Newsweek, Accessed
August 11th, 2011. (http://www.thedailybeast.com/newsweek/2010/06/30/howthe-recession-hurts-private-prisons.html).
Duwe, G. (2010). Prison-based chemical dependency treatment in Minnesota: An
outcome evaluation. The Journal of Experimental Criminology 6: 57-81.
Duwe, G. (2012a). Evaluating the Minnesota Comprehensive Offender Reentry Plan
(MCORP): Results from a randomized experiment. Justice Quarterly, 29: 347383.
Duwe, G. (2012b). The benefits of keeping idle hands busy: The impact of a prisoner
reentry employment program on post-release employment and offender
recidivism. Crime & Delinquency. DOI: 10.1177/0011128711421653.
Duwe, G., & Clark, V. (2011). Blessed be the social tie that binds: The effects of prison
visitation on offender recidivism. Criminal Justice Policy Review doi:
10.1177/0887403411429724

32

Duwe, G., & Goldman, R. (2009). The impact of prison-based treatment on sex offender
recidivism: Evidence from Minnesota. Sexual Abuse: A Journal of Research and
Treatment, 21, 279-307.
Farabee, D., & Knight, K. (2002). A comparison of public and private prisons in Florida:
During- and post-prison performance indicators. Los Angeles, CA: Query
Research.
Gendreau, P., Little, T. & Goggin, C. (2006). A Meta-Analysis of the Predictors of Adult
Offender Recidivism: What Works! Criminology, 34(4), 575-608.
Government Accounting Office. (1991). Private Prisons: Cost Savings and BOP’s
Statutory Authority Need to be Resolved. Report to the Chairman, Subcommittee
on Regulation, Business Opportunities and Energy, Committee on Small
Business, House of Representatives. Washington, DC: General Government
Division, General Accounting Office.
Government Accounting Office. (1996). Private and Public Prisons: Studies Comparing
Operational Costs and/or Quality of Service. Report to the Subcommittee on
Crime, Committee on the Judiciary, House of Representatives. Washington, DC:
General Government Division, General Accounting Office.
Havens, C., & Giles, K. (2009). “Minnesota may use private prison in Appleton.” Star
Tribune, Accessed August 11th, 2011.
(http://www.startribune.com/local/east/72081417.html).
Langan, P.A., & Levin, D.J. (2002). Recidivism of Prisoners Released in 1994.
Washington, DC: U.S. Department of Justice, Bureau of Justice Statistics.
Lanza-Kaduce, L., & Maggard, S. (2001). The long-term recidivism of public and private
prisoners. Unpublished paper presented at the National Conference of the Bureau
of Justice Statistics and Justice Research and Statistics.
Lanza-Kaduce, L., Parker, K., & Thomas, C.W. (1999). “A Comparative Recidivism
Analysis of Releasees from Private and Public Prisons.” Crime & Delinquency,
45, 28-47.
Lundahl, B., Kunz, C., Brownell, C., Harris, N., & Van Vleet, R. (2007). Privation
privatization: A meta-analysis of cost effectiveness and quality of confinement
indicators. Salt Lake City, UT: Utah Criminal Justice Center.

33

Minnesota Department of Corrections (2000). Adult Inmate Profile (January 1, 2000).
http://www.doc.state.mn.us/aboutdoc/stats/adultprofile1-1-00.htm.
Minnesota Department of Corrections (2004). Adult Inmate Profile (January 1, 2004).
http://www.doc.state.mn.us/aboutdoc/stats/pdf/AdultProfile01-01-2004.pdf.
Minnesota Department of Corrections (2006). Adult Inmate Profile (January 1, 2006).
http://www.doc.state.mn.us/aboutdoc/stats/documents/January2006ProfileCard_fi
nal.pdf.
Minnesota Department of Corrections (2008). Adult Inmate Profile (January 1, 2008).
http://www.doc.state.mn.us/aboutdoc/stats/documents/AdultInmateProfile01-012008.pdf.
Minnesota Department of Corrections (2012). Financial Services Unit.
Northcutt Bohmert, M., & Duwe, G. (2012). Minnesota’s Affordable Homes Program:
Evaluating the effects of a prison work program on recidivism, employment and
cost avoidance. Criminal Justice Policy Review, 23, 327-351.
Oppel, Richard A. 2011. “Private Prisons Found to Offer Little in Savings.” New York
Times, Accessed May 18th, 2011.
(http://www.nytimes.com/2011/05/19/us/19prisons.html).
Pratt, T.C. & Maahs, J. (1999). Are private prisons more cost-effective than public
prisons? A meta-analysis of evaluation research studies. Crime & Delinquency,
45, 358-371.
Rosenbaum, P.R., & Rubin, D.B. (1985). Constructing a control group using multivariate
matched sampling methods that incorporate the propensity score. The American
Statistician, 39, 33-38.
Rubin, D.B. (1997). Estimating causal effects from large data sets using propensity
scores. Annals of Internal Medicine, 127, 757-763.
Sabol, W.J., Minton, T.D., & Harrison, P.M. (2007). Prison and Jail Inmates at Midyear
2006. Bulletin. Washington, DC: U.S. Department of Justice, Office of Justice
Programs, Bureau of Justice Statistics.
Selman, D., & Leighton, P. (2010). Punishment for Sale: Private Prisons, Big Business,
and the Incarceration Binge. Lanham, MD: Rowman & Littlefield Publishers.

34

Shadish, W.R., Cook, T.D., & Campbell, D.T. (2002). Experimental and quasiexperimental designs for generalized causal inference. Boston: Houghton
Mifflin.
Shelden, R. (2010). The Prison Industry. Research Brief. San Francisco, CA: Center on
Juvenile and Criminal Justice.
Spivak, A.L., & Sharp, S.F. (2008). Inmate Recidivism as a Measure of Private
Prison Performance. Crime & Delinquency, 54, 482-508.
Stephan, J.J. (2008). Census of State and Federal Correctional Facilities, 2005. Report.
Washington, DC: U.S. Department of Justice, Office of Justice Programs, Bureau
of Justice Statistics.
West, H.C., Sabol, W.J., & Greenman, S.J. (2010). Prisoners in 2009. Bulletin.
Washington, DC: U.S. Department of Justice, Office of Justice Programs, Bureau
of Justice Statistics.

35