The criminogenic effects of imprisonment: Evidence from state panel data, 1974–2002

Abstract

The heavy reliance on the use of incarceration in an attempt to address the crime problem has resulted in a dramatic growth in the number of state prisoners over the past 30 years. In recent years, however, a growing concern has developed about the impact that large numbers of offenders released from prison will have on crime rates. Using a state panel data set for 46 states from 1974 to 2002, this study demonstrates that although prison population growth seems to be associated with statistically significant decreases in crime rates, increases in the number of prisoners released from prison seem to be significantly associated with increases in crime. Because we control for changes in prison population levels, we attribute the apparent positive influences on crime that seem to follow prison releases to the criminogenic effects of prison.
Policy makers should continue to serve the public interest by carefully considering policies that are designed to reduce incarceration rates and thus assuage the criminogenic effects of prison. These policies may include changes in sentencing, changes in probation and/or parole practices, or better funding of reentry services prerelease and postrelease.

Full text

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 1 21-AUG-07 9:49
THE CRIMINOGENIC EFFECTS OF
IMPRISONMENT: EVIDENCE FROM STATE
PANEL DATA, 1974–2002
LYNNE M. VIERAITIS
TOMISLAV V. KOVANDZIC
University of Texas at Dallas
THOMAS B. MARVELL
Justec Research
Research Summary:
The heavy reliance on the use of incarceration in an attempt to address
the crime problem has resulted in a dramatic growth in the number of
state prisoners over the past 30 years. In recent years, however, a grow-
ing concern has developed about the impact that large numbers of
offenders released from prison will have on crime rates. Using a state
panel data set for 46 states from 1974 to 2002, this study demonstrates
that although prison population growth seems to be associated with sta-
tistically significant decreases in crime rates, increases in the number of
prisoners released from prison seem to be significantly associated with
increases in crime. Because we control for changes in prison popula-
tion levels, we attribute the apparent positive influences on crime that
seem to follow prison releases to the criminogenic effects of prison.
Policy Implications:
Policy makers should continue to serve the public interest by carefully
considering policies that are designed to reduce incarceration rates and
thus assuage the criminogenic effects of prison. These policies may
include changes in sentencing, changes in probation and/or parole
practices, or better funding of reentry services prerelease and
postrelease.
KEYWORDS: Incarceration, Prison Release, Prison Population
Over the past three decades, U.S. crime policy has mirrored the philo-
sophical shift from rehabilitation to punishment. This shift is reflected in
policies that are designed to send more offenders to prison for longer peri-
ods of time (e.g., truth-in-sentencing laws that mandate that offenders
serve at least 85% of their time, three strikes laws, and mandatory mini-
mums). The heavy reliance on the use of incarceration in an attempt to
address the crime problem has resulted in a dramatic growth in the num-
ber of state prisoners over the past 30 years. In 1980, the number of state
VOLUME 6 NUMBER 3 2007 PP 589–622
R

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 2 21-AUG-07 9:49
590 VIERAITIS, KOVANDZIC, & MARVELL
prisoners equaled 305,458. By the end of 2004, this number had grown to
1.24 million, which is an increase of more than 400% (U.S. Bureau of Jus-
tice Statistics, 2006).
Whether the “imprisonment binge” of the past 25 years has worked at
lowering crime rates remains a hotly debated topic, although most
research on prison population and crime that uses regression analysis sug-
gests a certain degree of success (Levitt, 1996; Marvell and Moody, 1994,
1997, 1998; but see DeFina and Arvanites, 2002 and Kovandzic and Vier-
aitis, 2006). Although it is difficult to empirically disentangle whether the
impact of incarceration on crime rates is attributable to incapacitation or
deterrence, most analysts emphasize the incapacitation effects of incarcer-
ation (e.g., Marvell and Moody, 1994).
Other research, however, has identified many of the unintended or col-
lateral consequences of relying on incarceration, which thus suggests that
success is relative (Austin and Irwin, 2001; Clear, 1996, 1997; Clear et al.,
2001; Mauer, 1999; Mauer and Chesney-Lind, 2002; Petersilia, 2000, 2001,
2003; Rose and Clear, 1998). Some consequences are primarily monetary:
for example, the financial costs associated with building and operating
prisons and the health-care expenditures associated with an aging prison
population and with inmates suffering from infectious diseases. More pris-
ons were built in the 20 years from 1980 to 2000 than in the entire history
of prisons in the United States (Irwin, 2005), and state prison expenditures
increased 150% from FY1986 to FY2001 (Stephan, 2004). Other conse-
quences are not necessarily monetary and occur postrelease. For example,
some ex-prisoners are legally barred from voting, from receiving public
assistance including welfare benefits and public housing, from obtaining a
driver’s license and thus the ability to access jobs that require driving, or
from retaining custody of their children (Petersilia, 2003; Travis, 2002).
In addition, criminologists have begun to identify the impact on commu-
nities when large numbers of residents, disproportionately racial minori-
ties residing in disadvantaged neighborhoods, are removed and then
returned (Rose and Clear, 1998; Clear et al. 2001). This research suggests
that high incarceration and return rates may disrupt the social networks of
a community by affecting family formation, reducing informal control of
children and income to families, and lessening ties among residents. These
factors, which lead to neighborhood instability and low informal social
control, have been linked to higher crime rates in previous research on
social disorganization and crime (Bellair, 1997; Markowitz et al., 2001;
Sampson et al., 1997).
Another prospective consequence of mass incarceration, and the focus
of this study, is the potential effect on crime when large numbers of
offenders are released from prison. Some scholars have suggested, for
example, that the routinization and restriction of prison life may increase

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 3 21-AUG-07 9:49
CRIMINOGENIC EFFECTS OF IMPRISONMENT 591
the likelihood of reoffending postrelease when offenders find it difficult to
transition to life on the outside, especially as the amount of time served
increases (Carceral, 2004; Irwin, 2005; Sykes, 1958). Other scholars have
suggested that the experience of prison can increase crime postrelease by
making it more difficult for offenders to reenter society because of the
lack of needed services (Austin, 2001; Lynch and Sabol, 2001; Petersilia,
2003). It is also plausible that the prison experience can increase crime by
providing inmates with an education in crime as criminals learn from each
other how to commit new crimes or how to improve their techniques.
Whether the experience of prison leads to increases in crime is a crucial
issue today because the number of prisoners being released is unprece-
dented in the history of the United States. Between 1980 and 2003, the
number of state prison inmates released from prison grew from 154,107 to
612,185—a rise of 297%. Although the gap between prison admissions and
releases has remained relatively stable over this time period (an average
ratio of 1.13), the gap has closed significantly in recent years. By 2003, the
ratio of admissions to releases had declined to almost 1 (1.03).
1
This study examines the potential criminogenic effects of prison release
on crime by using annual state panel data for the period 1974 to 2002 for
46 states. The next section summarizes the relevant criminological theo-
ries, discusses how prisoner releases could impact crime, and addresses the
methodological issues that researchers confront in estimating such effects.
The next section describes at length the data and panel regression proce-
dures used. The last two sections present the results and discuss the policy
implications that emerge from our empirical work.
THEORY
PRISONS AS CRIMINOGENIC
The experience of prison itself may increase the likelihood of reoffend-
ing postrelease. According to Gendreau et al. (1999:1), the “barren, inhu-
mane, and psychologically destructive” nature of imprisonment may
increase the likelihood of crime after release. Although one may debate
the degree to which the pains associated with imprisonment are intended
by those who design penal practices, the fact remains that imprisonment
causes harm to prisoners (Clear, 1994; Irwin, 2005; Sykes, 1958). This is
harm that Irwin (2005) argues may make it more difficult for prisoners to
“achieve viability, satisfaction, and respect” when they are released. For
example, the routinization and restriction of prison life interferes with the
capacity of an inmate to exert power and control over his or her life. With
1. Interestingly, law enforcement officials and others have begun to attribute
recent rises in violent crime rates to the return of so many unrehabilitated prisoners to
the streets (Petersilia, 2003; Rashbaum, 2002; Winship, 2002).

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 4 21-AUG-07 9:49
592 VIERAITIS, KOVANDZIC, & MARVELL
few exceptions, inmates follow a rigid routine every day for as long as they
are incarcerated and must comply with a set of rules governing all aspects
of their lives and activities (Carceral, 2004; Irwin, 2005; Sykes, 1958). The
degree to which inmates are affected by these facets of prison life will
depend, in part, on the length of their incarceration and the type of institu-
tion to which they are assigned. As the amount of time served increases, so
might the deleterious effects of prison life on inmates increase. In addi-
tion, the routinization and control of prison life is unlike life on the
outside, which makes adjustment difficult for some prisoners once they are
released (Irwin, 2005). The pains of imprisonment also include the isola-
tion from family, relatives, and friends. Although inmates have visiting
and mailing privileges to varying degrees, links to persons outside the
institution are usually weak, and the weakening of these links may be
exacerbated as the length of incarceration increases (Sykes, 1958).
The experience of prison also complicates the ability of an offender to
reenter society successfully because prison may not provide needed ser-
vices. Although many inmates participate in prison programs to increase
their educational levels and vocational skills or to address substance abuse
and/or mental health problems, participation levels are relatively low
(Austin, 2001; Lynch and Sabol, 2001; Petersilia, 2003). Moreover, the
recent reductions in prison programming and in participation by inmates
have increased the number of prisoners who are ill prepared for life
outside prison. According to Lynch and Sabol (2001), the number of pris-
oners returning to society not having participated in educational, voca-
tional, drug/alcohol treatment, and prerelease programs has increased in
recent decades. From 1991 to 1997, participation in educational programs
declined from 43% to 35%; the number of prisioners participating in voca-
tional programs declined from 31% to 27% over the same time period. In
addition, despite the high rate of substance use by state prisoners, rates of
participation in drug/alcohol treatment programs have also declined.
Levels of drug treatment postadmission were lower for state prisoners in
1997 (10%) than they were in 1991 (25%) (Mumola, 1999). Moreover, in
1997, only 12% of prisoners who were about to be released participated in
prerelease planning (Lynch and Sabol, 2001). As a result, released prison-
ers already disadvantaged by their labels of ex-prisoners are ill prepared
and unlikely to find legitimate employment (Uggen, 1999).
The most significant barrier to successful reintegration is the inability of
an inmate to secure a place to live and a job. According to Petersilia
(2003:120), “parole officials say that finding housing for parolees is by far
their biggest challenge, even more difficult and important than finding a
job.” The cost of housing and the legal restrictions against convicted
offenders residing in public housing are significant barriers to securing
housing for persons released from prison (Irwin, 2005; Petersilia, 2003;

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 5 21-AUG-07 9:49
CRIMINOGENIC EFFECTS OF IMPRISONMENT 593
Travis, 2005). With respect to finding a job, released prisoners face tre-
mendous obstacles as well. Research indicates that many employers would
not hire an ex-prisoner (Holzer et al., 2002). Thus, in addition to the fact
that offenders are barred from certain occupations, those occupations for
which they are eligible are difficult to secure. Moreover, employers who
are willing to hire ex-inmates typically offer lower wages and fewer bene-
fits (Petersilia, 2003). In addition to the barriers to employment that are
associated with the stigma of ex-inmate, released prisoners typically have
low levels of education, little job experience, problems with substance
abuse, and mental health issues; plus, most return to disadvantaged com-
munities that offer little access to legitimate work (Holzer et al., 2002;
Petersilia, 2003).
The prison experience also may provide inmates with an education in
crime and may reinforce criminal identities. The “prison as school” argu-
ment suggests that housing criminals together in close proximity for long
periods of time results in numerous educational opportunities. Criminals
make “contacts” or associations that may facilitate crime upon release
(i.e., they may learn from each other how to commit new crimes or how to
improve their techniques) (Letkemann, 1973). Young, first-time offenders
may be at particular risk as they are exposed to more experienced inmates
who can influence their lifestyle and help solidify their criminal identities.
Thus, upon release offenders are more educated in terms of crime com-
mission and may have new contacts as a result of networking in prison that
can facilitate their return to crime.
Another factor that may contribute to a positive association between
prison release and crime rates is whether prison officials have control over
who and when a prisoner is released. By 2000, 16 states had abolished
discretionary parole for all offenders and 4 additional states had abolished
parole for certain violent offenders (U. S. Bureau of Justice Statistics,
2001). In 1977, 72% of those released from state prisons had served an
indeterminate sentence and had a parole board decide their release
(Greenfeld, 1995). These figures declined over the next two decades. In
1980, nearly 55% of state inmates were released via discretionary parole,
19% were released via mandatory parole, and just over 14% were released
unconditionally through the expiration of their sentences. By 2003, state
inmates released via discretionary parole had declined to 22%, mandatory
parole releases had increased to 36%, and unconditional releases had
increased to almost 20% (Glaze and Palla, 2005). As Petersilia (2003)
notes, this trend can be harmful because offenders are released with no
supervision and there is no control over who gets released. Under determi-
nate sentencing, once offenders have served their time, they are released.
In states with discretionary parole, prison officials can continue to incar-
cerate an inmate who is deemed to be at risk for offending upon release.

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 6 21-AUG-07 9:49
594 VIERAITIS, KOVANDZIC, & MARVELL
Moreover, because most state prisoners today serve presumptive
sentences, 90% of state inmates can estimate their probable release date
(Greenfeld, 1995) and prison inmates have less incentive to participate in
programs (Petersilia, 2003).
In addition, postrelease supervision has become increasingly difficult as
the numbers of persons released from prison increases. In 2004, 85% of
parolees were under active supervision that required them to regularly
report to a parole authority in person, by mail, or by telephone (Glaze and
Palla, 2005). This type of supervision has increased since 1995 (78%),
whereas parolees on inactive status, excluded from regular reporting but
still on parole, declined (Glaze and Palla, 2005). According to Petersilia
(2003), U.S. parolees are supervised on caseloads averaging 66 cases to
one parole office, although the ideal caseload is 35–50 cases. In part
because of the increasing numbers of parolees on active supervision,
parole in most states is focused mainly on control-oriented activities
(Petersilia, 2003).
PRISONER RELEASE AND REVERSE INCAPACITATION
To the extent that incarceration reduces crime through incapacitation,
increases in the number of offenders released from prison may lead to
increases in crime by simply reducing the number of offenders behind
bars, i.e., by lowering prison levels. Although such effects are entirely
plausible, they are unlikely to be operating here because during the time
period included in our analyses, prison admissions have generally out-
paced prison releases. For example, from 1977 to 1998, the average ratio of
annual state prison admissions to state prison releases was 1.13 or about 11
new admissions for every 10 releases (authors’ analysis of U.S. Bureau of
Justice Statistics data). Thus, prison releases are unlikely to have led to
higher crime rates through reverse incapacitation because those released
from prison have generally been replaced by new prison admits. However,
if, as Petersilia (2003) predicts, the rate of releases begins to outpace the
rate of admissions, the reverse incapacitation argument becomes more
plausible.
It is also important to note that prison release may increase crime if
those released from prison are more serious offenders than those being
admitted, independent of any criminogenic effects of prison itself. That is,
it may not be the deleterious effects of the prison environment or reverse
incapacitation (in terms of fewer admits) as discussed above but the mere
fact that those being released are different (i.e., more criminal) than those
being admitted. If true, an increase in prison release would increase crime
rates regardless of changes in overall prison levels. In other words, an
increase in crime would occur because high-rate offenders (releasees) are

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 7 21-AUG-07 9:49
CRIMINOGENIC EFFECTS OF IMPRISONMENT 595
replaced with low-rate offenders (new prison admits). Although theoreti-
cally plausible, recent research by Raphael and Stoll (2005) suggests that
this result is unlikely because admittees are largely similar to releasees
along key demographic and prior criminal history measures.
PRISONS AS GENERATING DETERRENCE
Prison releases may have no impact on crime if the experience of having
gone to prison generates specific deterrence. Several factors exist that may
reduce the likelihood of criminality when offenders are released. First, the
experience of prison may serve to deter the offender from committing
crimes once released for fear of being reincarcerated (i.e., specific deter-
rence). The “pains of imprisonment,” including the loss of freedom, rigid-
ity of prison life, loss of contacts with friends and family, and
stigmatization, may convince offenders that prison is a place to which they
wish never to return.
Having been deterred, released offenders do not commit crime, and in
turn the crime rate does not increase or decrease but remains stable.
Another possible reason that prison releases may not impact the crime
rate is that most offenders “age-out” of crime, and thus offenders who are
older at release may be more likely to desist from crime. According to The
Sentencing Project (2005), the average time served in prison rose by 30%
from 1995 to 2001 and 1 of every 23 inmates in prison is age 55 or older, an
85% increase since 1995. As offenders serve longer sentences, the average
age of releasees will increase.
Although prison release will neither increase nor decrease crime
through specific deterrence, it could impact crime through general deter-
rence. Prison release may decrease crime if released offenders make
potential offenders aware of the experience of prison and this awareness
convinces them not to engage in crime. Although theoretically plausible,
some criminologists have argued that the increased use of incarceration
has lessened the general deterrent potential of prison because it has
become a normal experience for so many citizens (Clear, 1996; Cook,
1998; Mauer, 1999). If true, at best prison release would have no impact on
crime through general deterrence, and at worst prison release would
increase crime if potential offenders were convinced, by an increasing
number of prison releasees, that prison is not costly.
PRIOR RESEARCH
Although a considerable body of individual-level research examines the
link between prison releases and crime (summarized in Sherman et al.,
2002), macrolevel research on the possible links between prison release
and crime must be performed to test these assumptions. This is partly true

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 8 21-AUG-07 9:49
596 VIERAITIS, KOVANDZIC, & MARVELL
because it is useful to have multiple approaches to testing a given hypothe-
sis. Perhaps more importantly, macrolevel analysis enables estimation of
the net effects of prison release on crime. Individual-level research can
identify determinants of postrelease failure, but these measures of failure
are not very useful for assessing the total amount of crime committed by
releasees. Also, if recent releasees communicate the pains of imprison-
ment to prospective criminals, then the deterrent effects of prisons would
not be limited to releasees and might not differ between the two. For these
reasons the aggregate net impact of crime-increasing and crime-decreasing
effects of prison releases can be quantified only through macrolevel
research.
A review of the literature revealed only two published studies to date
that have examined the macrolevel effects of prison releases on crime.
Kovandzic et al. (2004) regressed homicide rates on prison release rates,
prison population levels, and numerous control variables using state panel
data for the period 1975 to 1999. The results of their study provided no
evidence of a significant positive relationship between prison releases and
homicide. Because the authors controlled for the potential reverse inca-
pacitation effects of prison release by including prison population as a
regressor, their results provided little evidence for the “prisons as crimi-
nogenic” thesis.
Raphael and Stoll (2004) used state panel data from 1977 to 1999 to
examine the relationship between year-to-year changes in the number of
inmates released from prison and subsequent changes in violent and prop-
erty crime rates.
2
The authors also controlled for changes in the number of
new court commitments to prison in order to assess the relative propensity
of recently released inmates to commit crime upon release. According to
their findings, a net increase in the number of inmates released from
prison from the previous year was significantly and positively related to
changes in most index crimes, although the magnitudes seemed to be
slightly less than the crime-reduction impacts resulting from new court
commitments. The authors also found that the net impact of prison
releases and new court commitments on crime varied considerably over
2. We should note that Raphael and Stoll (2004) were actually interested in esti-
mating the impact of ex-offenders and prospective offenders (i.e., those never incarcer-
ated) on crime rates and not so much the impact of prison releases on crime. Because
data for these two populations were not observable, especially in levels, they approxi-
mated changes to these populations by examining year-to-year changes in the number
of offenders flowing into and out of prison. Specifically, the authors measured changes
in the ex-offender population in a given calendar year by subtracting the number of
conditionally released inmates returned to prison (total prison admissions minus new
court commitments) from the total number of inmates released from prison. Changes to
the population of never incarcerated offenders were measured as the total number of
persons admitted to prison for the first time.

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 9 21-AUG-07 9:49
CRIMINOGENIC EFFECTS OF IMPRISONMENT 597
time, with larger impacts observed for the late 1970s/early 1980s and
smaller impacts observed for the mid-1980s/early 1990s and mid-to-late
1990s. These results are consistent with their thesis that the marginal effect
on crime of more recent releases and court commitments would be less
than the comparable effect during the 1970s when incarceration rates were
much lower (and thus the most dangerous and criminally active offenders
were probably incarcerated first). Lastly, the Raphael and Stoll findings
indicated that the net impact of prison releases on crime was at least par-
tially influenced by the sentencing schemes used by different states. Spe-
cifically, they found that net releases have moderate-to-large statistically
significant impacts on crime in states with weakened or no parole board
but have little or no impact on crime in states with strong parole boards.
DATA AND METHODS
The current study is based on analyses of annual panel data for the
period 1974 to 2002 for 46 states. Panel data offer distinct advantages over
conventional time-series or cross-sectional studies. The most important
advantage is the ability to include state and time fixed effects in the crime
equations to mitigate omitted variable bias. The state and time fixed
effects are discussed further below. Second, the high number of degrees of
freedom allows us to enter numerous variables to control for confounding
factors, that is, variables that might be correlated with both crime and
prison release rates. Third, the high number of degrees of freedom pro-
vides for greater statistical power (Wooldridge, 2000:409) and thus makes
it possible to detect more modest effects of recent prison releases on crime
rates.
CRIME RATES
The dependent variables are the rates per 100,000 state populations of
murder and non-negligent manslaughter, robbery, aggravated assault, for-
cible rape, burglary, larceny, and motor vehicle theft. Data for each crime
category were taken from the Federal Bureau of Investigation (FBI) Uni-
form Crime Reports (1976–2004), as prepared and distributed by the
National Archive of Criminal Justice Data. State crime counts are based
on all reporting agencies and estimates for unreported areas (Federal
Bureau of Investigation, 2005:490–497).
PRISON RELEASES
For each state, the prison release variable is measured as the total num-
ber of inmates released back into society during the calendar year who had

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 10 21-AUG-07 9:49
598 VIERAITIS, KOVANDZIC, & MARVELL
been sentenced to terms of more than a year.
3
Similar to the crime rate
variables, the prison release variable is based on the rate per 100,000 state
population. Calendar year data are available for 1974 through 2002. Prison
release data for 1974 to 1998 were taken from the U.S. Bureau of Justice
Statistic’s (BJS) National Prisoner Statistics data series (NPS-1), which is
available online at http://www.ojp.usdoj.gov/bjs/data/corpop01.csv.
Unpublished prison release data for 1999 to 2002 were made available to
the authors by BJS staff.
The prison release data, according to the definition set by BJS, are the
releases of prisoners who were under prison jurisdiction and who were
originally sentenced to more than a year. A problem is that several states
changed their definitions of releases to comply with the BJS standard.
These changes can affect prison release data. For example, a state adopt-
ing a jurisdiction count of prisoners would start including prisoners held in
local jails due to prison overcrowding, which would artificially increase the
number released. When definitional changes were major, we deleted data
by subtracting the difference between prison admissions and releases from
the difference between prison populations in the current and prior years.
If reporting is consistent, the result is zero, but that was not always the
case. Most discrepancies were small, probably from recounting. But when
the difference reached 10% of the prison population, we considered it a
problem that needed to be addressed. Usually these discrepancies
occurred during the early years, and we included the state but used data
only after the problems occurred.
4
Prison releases include inmates released unconditionally or condition-
ally, inmates out on appeal or bond, escapees and AWOL prisoners, and
those released back into society for unspecified reasons. Unconditional
releases are mainly from expiration of sentence but also include inmates
who were released from prison because of court order, good time, commu-
tation, or pardon. Conditional releases are primarily inmates released to
parole, intensive supervision, work releases, or some form of community
corrections program. Because our concern is with the impact of prisoners
3. Coefficient estimates for the prison release variable are very close to those
reported in Table 2 when lagging the prison release variable by 1 year (to account for
any lagged effects) or taking the average value of the current and prior year.
4. Data used for Alabama starts with 1981, Alaska 1978, Delaware 1978, Ken-
tucky 1988, Louisiana 1977, Mississippi 1981, Missouri 1975, Montana 1980, Nebraska
1982, New Jersey 1988, New Mexico 1981, Oklahoma 1978, Pennsylvania 1978, Tennes-
see 1989, Vermont 1978, and Virginia 1980. Four states (Connecticut, Oregon, Texas,
and West Virginia) were deleted from the analysis because problems continued until
recent years. In addition, the Alaska 1994 data are missing, and we estimated it to be
the average of 1993 and 1995 numbers.

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 11 21-AUG-07 9:49
CRIMINOGENIC EFFECTS OF IMPRISONMENT 599
released back into society, we do not include prisoners released because of
death or inmates transferred to other institutions.
CONTROL VARIABLES
In addition to the state and time fixed effects, we include numerous con-
trol variables that theory and prior research suggest are causally antece-
dent to both crime rates and punishment levels. Failing to control for
factors that have opposite or same sign effects on both prison release and
crime rates could suppress (i.e., mask any positive impact of prison
releases on crime) or lead to spurious or partially spurious results for the
prison release variable, respectively. The specific control variables
included in each crime regression were the percent of the civilian labor
force unemployed; total employment rate; military employment rate; con-
struction employment rate; real per-capita income (divided by the con-
sumer price index); percent of the population living below the poverty
line; percent of the population comprising African-American men; percent
of the population residing in metropolitan areas; percent of the population
with a bachelor’s degree or higher; percent of the population aged 15 to
19, 20 to 24, 25 to 29, 30 to 34, and 35 to 44 years; and current-year prison
population.
5
The age structure variables are important because they probably affect
both prison releases and crime in a positive direction; thus, failing to con-
trol for changes in age structure could lead to spurious results for the
prison release variable. The age groups used here are consistently those
with the highest arrest rates for crime (Federal Bureau of Investigation,
2005:290–291). This suggests that crime rates should increase as these age
cohorts grow, although many studies do not support that hypothesis (Land
et al., 1990; Marvell and Moody, 1991). Age structure is also an important
determinant of prison population. Marvell and Moody (1997) found that
age structure, especially the 25–34 age group, was positively related to
prison population growth. Such findings are in accord with data on the age
of inmates in prison (U.S. Bureau of Justice Statistics, 2001:10–11).
Because the age structure of a state is likely to impact the size of its prison
population, it is also likely to impact its prison release rates; states with
greater numbers of younger persons have a larger pool of prisoners eligi-
ble to be released from prison. Consequently, one might expect prison
release rates to increase as the number of individuals in these age groups
increases.
The second set of variables controls for changes in economic trends,
which numerous macrostructural theories and prior research suggest are
5. The data are for the end of the year, and we estimate the prison population
over the year by averaging the current and prior year numbers.

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 12 21-AUG-07 9:49
600 VIERAITIS, KOVANDZIC, & MARVELL
related to both punishment levels and crime (Chiricos and Delone, 1992;
Greenberg and West, 2001; Land et al., 1990). Several criminological theo-
ries, including strain/deprivation, social disorganization, Marxist theory,
and macrostructural theory, contend that economic distress has a positive
impact on crime, and research provides support for the effects of economic
deprivation on crime, especially homicide (see reviews in Chiricos, 1987;
Land et al., 1990; Vieraitis, 2000). With respect to the connection between
prisoners and economic conditions, the basic argument is that prisons are
an effective way to manage populations (e.g., unemployed and marginal
workers) perceived as threatening during economic downturns and that
the business community will devalue potential laborers when unemploy-
ment rates are high (Cappell and Sykes, 1991; Hale, 1989; Parker and Hor-
witz, 1986; Rusche and Kirchheimer, 1939; Sabol, 1989; Speiglman, 1977).
Other criminologists have suggested that judges may fear the threat posed
by unemployed workers who may be considered at greater risk for
returning to crime (Box, 1987; Greenberg, 1977) and, therefore, sentence
them more harshly. Thus, theory suggests that prison administrators and
parole boards release fewer prison inmates during economic downturns.
6
In sum, these considerations lead to the prediction that worsening eco-
nomic conditions are positively related to crime rates and negatively
related to prison releases. Thus, failing to control for changes in economic
trends might suppress any positive impact of prison releases on crime.
7
Racial heterogeneity is also likely to affect crime and incarceration
rates. African Americans, for example, were nearly six times more likely
than whites to be murdered in 2000 and seven times more likely than
whites to commit a homicide (U.S. Bureau of Justice Statistics, 2004). With
respect to prison population, by year-end 2000, African Americans made
up nearly two thirds of all inmates, with incarceration rates for African
Americans roughly eight times that of whites (U.S. Bureau of Justice Sta-
tistics, 2001:11).
The final control variable added to the crime regressions is prison popu-
lation (as of December 31 of each year), which is the average of the figures
from the current and prior years. As discussed, although the prison popu-
lation is unlikely to be an intervening variable in the prison release–crime
relationship because prison levels have seldom declined during the time
period studied, prison population may play an important role as a poten-
tial suppressor variable because of its positive relationship with prison
releases and negative relationship with crime rates. With respect to prison
6. It is also possible that budget restraints lead to more releases during economic
downturns.
7. The military employment rate and construction employment rate are included
because these jobs are held primarily by young men, i.e., those in the crime-prone age
group.

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 13 21-AUG-07 9:49
CRIMINOGENIC EFFECTS OF IMPRISONMENT 601
releases, prison population growth over the past decade has led to unprec-
edented growth in the number of subsequent offenders released from
prison (Travis and Lawrence, 2002). At the same time, prison population
growth is considered by many to have played a critical role in the crime
reduction of the 1990s (e.g., Levitt, 2004), and research generally supports
the more prisoners–less crime thesis (Devine et al., 1988; Levitt, 1996;
Marvell and Moody, 1994, 1997; but see also DeFina and Arvanites, 2002;
Kovandzic and Vieraitis, 2006). Consequently, failing to control for
changes in prison population levels could suppress, at least partially, any
positive impact of prison releases on crime rates. As we will show, the
coefficients on the prison release variable differ significantly when drop-
ping prison population from the crime regressions.
Poverty data were obtained from the Bureau of the Census website:
http:www.census.gov/hhes/www/poverty.html. Data on state-level unem-
ployment were taken from the Bureau of Labor Statistics website: http://
www.bls.gov/sae/home.htm. Data on personal income, real welfare pay-
ments, military employment, and construction employment were obtained
from the Bureau of Economic Analysis website: http://www.bea.doc.gov/
bea/regional/reis/. Percent of the population with college degrees or higher
and residing in metropolitan areas is linear interpolation of decennial cen-
sus data, as reported in various editions of the Statistical Abstracts of the
United States.
8
Age group and racial heterogeneity data were obtained
directly from the U.S. Bureau of the Census on computer disk. For the
source and notes on corrections made to the prison population data, see
Marvell and Moody (1998).
ANALYTIC METHODS
To assess the influence of recent prison releases on crime rates, we fol-
low conventional strategies for panel data and estimate a fixed-effects
model. The fixed-effects model requires adding a binary dummy variable
for each state and year (except the first to avoid collinearity problems)
(Wooldridge, 2000:420). The state fixed effects are an integral part of the
fixed-effects approach because they control for persistent differences
across states; that is, they control for the collective effect of stable state-
specific factors that cause crime to vary from state to state. The time fixed
effects (i.e., year dummies) control for national time trends that are com-
mon to all states in a year. Because the analysis includes both state and
time fixed effects, the parameter estimates for the prison release variable
are based solely on within-state changes over time.
8. Dropping the linearly interpolated variables from the statistical models pro-
duces results virtually identical to those reported in Table 2.

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 14 21-AUG-07 9:49
602 VIERAITIS, KOVANDZIC, & MARVELL
Finally, we control for state-specific time trends by including a separate
linear trend variable for each state. Controlling for state-specific time
trends has become standard practice in panel data analysis (Ayres and
Donohue, 2003; Marvell and Moody, 1996, 2001). The state-specific time
trends control for unobserved factors that affect the time-series behavior
of crime that can differ from state to state and depart from the nationwide
trends captured by the time fixed effects.
9
It should be noted, however,
that the state trend variables do not control for trends in a state that occur
in a nonlinear fashion and that depart from nationwide trends.
The choice of functional form is always a difficult one. Similar to most
previous prison crime research, we decided to use a log–log model, also
referred to as a constant elasticity model, in which crime, prison releases,
and the control variables are expressed in their natural logs (Wooldridge,
2000). The log–log model provides us with the benefit of allowing the coef-
ficient on the prison release variable to be interpreted as an elasticity: the
percent change in the crime rate expected from a 1% change in the prison
release rate, holding all other factors fixed (Wooldridge, 2000). It is also
useful to note that because we estimate a log–log model the crime elastic-
ity for prison releases will remain constant throughout, hence the alterna-
tive name, the constant elasticity model (Wooldridge, 2000:44–45). In
other words, the constant elasticity model implies diminishing marginal
returns; that is, the proportional change in crime from proportional
changes in prison release rates is constant regardless of actual prison
release levels.
The Im, Pesaran, and Shin (2003) heterogeneous panel unit root test
indicates no evidence of unit roots, except for auto–theft, which seems to
be a nonstationary random walk.
10
Because the results for auto theft were
largely similar when we reestimated the auto theft regression in first-dif-
ferences (which generates a stationary series), we report the results in
levels to be consistent with the other crime specifications. It is natural to
expect the presence of serial correlation (errors within states are tempo-
rally correlated) and panel-level heteroskedasticity (each state has its own
variance) in a panel data set; indeed, application of the Wooldridge (2000)
test for serial correlation and the modified Wald statistic for group-wise
9. An anonymous reviewer suggested that we also estimate a quadratic time-
trend model by entering both a linear trend and a quadratic trend (linear trend
squared). The quadratic term is a proxy for omitted state factors that are changing at an
ever faster rate. Adding the quadratic trend has little impact on the findings, except that
the coefficient for prison releases increases by nearly a factor of two in the homicide
model and is roughly halved and no longer statistically significant in the burglary
model.
10. The panel unit root tests were estimated with fixed effects, with and without
trends, and a 1-year lag.

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 15 21-AUG-07 9:49
CRIMINOGENIC EFFECTS OF IMPRISONMENT 603
heteroskedasticity by Greene (2003) suggests that they are both present
here. Another likely violation of the independence of error terms assump-
tion in the context of state panel data is contemporaneous correlation of
error terms across states caused by common shocks in a given year. To
examine this possibility we calculated the Breusch–Pagan (BP) test statis-
tic (1980) for detecting contemporaneous correlation in the residuals. The
null hypothesis of cross-sectional independence was rejected in each crime
specification. To ensure valid statistical inference, we follow one standard
method for removing serial correlation (and arguably the preferred
method) by including a 1-year lag of crime in each crime specification to
mitigate serial correlation and use panel-corrected standard errors
(PCSEs), which are robust in the presence of panel heteroskedastic errors
and contemporaneously correlated error terms (Beck and Katz, 1995,
1996).
11
We also follow the standard procedure in panel data analysis of
weighting the regressions by state population to account for the greater
year-to-year variability in crime rates in smaller states.
12
Failing to weight
the crime specifications by state population leads to smaller states domi-
nating the results. Examination of collinearity diagnostics developed by
Belsley et al. (1980) revealed no serious collinearity problems for the
prison release variable, although problems exist for some control vari-
ables, mainly for the age structure and employment variables that change
slowly over time and are highly correlated with the state dummy variables.
Thus, it is necessary to use caution in the interpretation of results for these
variables.
Table 1 lists the variables included in each crime model. In addition to
the variable name and a brief description, the mean, overall, and within-
state standard deviations are also shown. Estimation was carried out in
Stata, version 9.0.
RESULTS
Table 2 presents estimates of the average impact of prison release rates
on each of the FBI’s seven index crimes, using regression procedures for
panel data discussed above. Again, it is worth noting that because we
include prison population as a control variable, the coefficients for the
prison release rate variable reflect the effect of incarceration on crime
11. In addition to correcting for serial correlation, the lagged dependent variables
also serve to mitigate omitted variable bias by capturing any temporal dynamic effects
that may exist (Moody, 2001).
12. The appropriate weights were calculated using the BP test. The weights are as
follows: homicide (pop.*1), rape (pop.*.7), robbery (pop.*.5), assault (pop.*.6), bur-
glary, larceny, and auto-theft (pop*.4). Coefficient estimates for the prison release vari-
able were not qualitatively affected when reestimating the crime equations without
weighting or weighting by total state population.

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 16 21-AUG-07 9:49
604 VIERAITIS, KOVANDZIC, & MARVELL
TABLE 1. SUMMARY STATISTICS
Standard Standard
Deviation Deviation
Variable Mean (Overall) (Within State)
Dependent Variables
Homicide rate 6.75 3.79 1.67
Rape rate 31.46 14.33 9.31
Robbery rate 134.55 101.58 38.69
Assault rate 259.41 146.44 79.91
Burglary rate 1,081.80 415.51 278.49
Larceny rate 2,734.42 764.23 457.00
Motor vehicle theft rate 394.95 207.74 106.69
Primary Independent Variable
Prison release rate 113.88 71.60 56.44
Control Variables
Incarceration rate 207.83 140.64 115.23
% aged 15 to 19 8.31 1.29 1.21
% aged 20 to 24 8.17 1.23 1.14
% aged 25 to 29 7.85 1.11 0.99
% aged 30 to 34 7.58 1.15 0.99
% aged 35 to 44 13.47 2.43 2.31
% metropolitan 65.79 21.76 2.75
% black males 4.55 4.31 0.41
% college graduate or higher 18.34 5.54 4.52
% living below poverty line 12.91 4.21 1.94
Per-capita income 4,425.40 907.65 660.52
% unemployed 5.99 2.06 1.72
Total employment rate 53,010.49 6,470.41 5,195.35
Military employment rate 1,380.02 1,227.76 460.58
Construction employment rate 2,895.49 733.38 535.86
All values are expressed in their original units (i.e., before logging). All rate variables are per
100,000 state population.
after partialling out any crime-enhancing effects caused by reverse inca-
pacitation effects. We should also note that the coefficient on prison
releases (and prison population) pertains only to criminals who stay in
state. To the extent that offenders migrate across state lines upon release
from prison, the total effect of prison releases on crime is likely to be
larger (Marvell and Moody, 1998). To conserve space, the coefficients for
the state, year, and individual state-trend variables are not shown. Instead,
we report the results of three separate F tests that test the joint signifi-
cance of each group of dummy variables. As observed in Table 2, all three
sets of dummy variables are highly significant in each crime specification.
Additional analyses explore the potential simultaneous relationship
between prison release rates and crime rates and potential differential
effects of prison release rates across states by creating separate prison
release variables for each state. We also examine the robustness of our
results by varying the model specifications in Table 2.

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 17 21-AUG-07 9:49
CRIMINOGENIC EFFECTS OF IMPRISONMENT 605
TABLE 2. THE ESTIMATED IMPACT OF PRISON RELEASES ON CRIME RATES:
EVIDENCE FROM STATE PANEL DATA, 1974 TO 2002
Dependent Variables: Natural Logs of the Crime Rate per 100,000 state population
Motor Vehicle
Homicide Rape Robbery Assault Burglary Larceny Theft
Coefficient tCoefficient tCoefficient tCoefficient tCoefficient tCoefficient tCoefficient t
Prison release rate 0.045 2.11 –0.031 2.53 0.064 4.37 0.005 0.38 0.032 3.54 0.013 1.63 0.023 1.36
Incarceration rate –0.206 4.93 –0.077 3.47 –0.224 8.80 –0.031 1.15 –0.122 6.45 –0.064 4.66 –0.183 6.01
% aged 15 to 19 –0.28 0.67 0.778 3.52 0.671 2.88 0.797 4.21 0.445 2.58 0.151 1.17 0.833 2.96
% aged 20 to 24 0.085 0.29 –0.385 2.46 –0.335 1.88 –0.537 3.89 –0.220 2.07 –0.211 2.28 –0.359 2.23
% aged 25 to 29 –0.101 0.49 0.042 0.35 –0.159 1.26 –0.09 0.92 0.049 0.69 –0.036 0.55 –0.041 0.36
% aged 30 to 34 –0.265 1.02 0.134 0.92 0.037 0.22 0.102 0.81 0.155 1.62 0.095 1.20 0.093 0.60
% aged 35 to 44 –0.829 1.91 –0.440 2.13 –1.08 4.10 –0.625 4.04 –0.117 0.81 –0.133 1.31 –0.785 3.63
% metropolitan –0.164 0.30 0.092 0.42 –0.56 2.04 –0.523 2.41 0.072 0.58 0.069 0.76 –0.256 1.56
% black males 0.688 3.03 –0.109 1.15 0.389 3.21 0.263 3.04 0.146 2.30 0.074 1.48 0.238 2.38
% college graduate 0.104 0.48 –0.037 0.19 0.054 0.31 0.012 0.07 0.273 3.09 0.213 3.53 0.118 1.01
Poverty rate –0.021 0.56 0.022 0.93 0.034 1.34 0.005 0.23 0.014 0.86 0.010 0.81 0.009 0.36
Per-capita income –0.155 0.50 –0.182 1.27 0.07 0.42 0.207 1.44 0.144 1.24 0.053 0.79 0.209 1.50
% unemployed –0.016 0.45 0 0.01 0.017 0.68 –0.012 0.63 0.050 4.02 0.035 3.80 –0.012 0.55
Total employment rate 0.236 0.53 0.759 3.52 0.593 2.39 –0.050 0.23 –0.176 1.04 0.022 0.18 –0.438 1.55
Military employment rate 0.329 4.89 0.104 2.40 0.136 2.91 0.018 0.40 –0.008 0.31 0.036 1.54 0.203 4.61
Construction employment rate 0.280 3.12 –0.002 0.04 0.036 0.74 0.019 0.44 0.042 1.28 0.073 3.02 0.167 2.92
Lagged dependent variable .353 7.04 .540 13.41 .636 19.75 .697 18.16 .665 19.57 .667 20.86 0.722 20.55
F statistic for variable groups
State dummies 7.09 18.04 71.15 22.05 13.40 14.73 24.59
Year dummies 230.84 2,131.47 984.10 1,342.89 2,600.21 3,102.85 1,928.76
State trends 8.57 9.25 29.13 20.71 16.04 19.61 21.40
N1,225 1,225 1,225 1,225 1,225 1,225 1,225
NOTES: The dependent variables are the natural log of each index crime rate per 100,000 state residents. The data set comprises annual state-level data (not including
Connecticut, Oregon, Texas, and West Virginia) from 1974 to 2002. All regressions use weighted least squares, where the weighting is a function of state population. The
number of observations in each crime equation is 1,225. In all cases, estimation allows for panel-level heteroskedasticity and contemporaneous correlation of error terms.
Although not shown, state, year, and state-trend dummies were included in all crime specifications. Coefficients that are significant at the 0.10 level are underlined.
Coefficients that are significant at the 0.05 level are displayed in bold. Coefficients that are significant at the 0.01 level are both underlined and displayed in bold.

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 18 21-AUG-07 9:49
606 VIERAITIS, KOVANDZIC, & MARVELL
The results in Table 2 provide strong support for the “prisons are crimi-
nogenic” thesis. With the exception of rape, the coefficients for the prison
release rate variable are in the expected positive direction and are statisti-
cally significant at conventional significance levels, except for assault, lar-
ceny, and auto theft.
13
The null findings for assault are not all that
surprising given changes in reporting and recording practices for these
crimes during the study period examined here (Reiss and Roth, 1993). As
noted, the use of a double-logged model allows us to interpret the coeffi-
cients for the prison release variable as an elasticity. The estimated elastic-
ities for homicide, robbery, and burglary with respect to prison releases
are 0.045, 0.064, and 0.032, respectively. That is, for each 1% increase in
the prison release rate, homicides, robberies, and burglaries increase by
0.045%, 0.064%, and 0.032%, respectively. We should note, however, that
the elasticities reported above are understated because much of the impact
of prison releases eventually comes through the lagged dependent varia-
ble. To estimate the full, long-term elasticity, we followed the recommen-
dation of Hamilton (1994:19–20) and divided the short-run coefficient for
the prison release variable by one minus the sum of the coefficient on the
lagged dependent variable. The long-term elasticities for homicide, rob-
bery, and burglary after the adjustment are 0.070, 0.176, and 0.095,
respectively.
14
To estimate the crime-enhancing effects of releasing one prisoner at the
margin, we multiplied the long-term elasticity for each crime category by
the ratio of the average rate for that crime category to the average rate for
13. An anonymous reviewer suggested we reestimate the regressions without
lagged dependent variables as it is well known that the fixed effects estimator is biased
and inconsistent when a lagged dependent variable is included, although the bias
decreases as T gets large (Nickell, 1981). Our review of the econometric literature on
panel data suggests the bias is limited to panel data sets where T is small and, therefore,
should not be a serious issue here where T is larger than in typical panel data sets. In
any event, we reestimated the regressions in Table 2 using robust fixed-effects standard
errors to examine the robustness of the findings using PCSEs. The coefficients and stan-
dard errors for the prison release variable were somewhat larger than those reported
above, especially for homicide and robbery, but the overall findings generally remained
the same. The main exceptions occurred for rape, burglary, and larceny. The coefficient
for prison releases is now nearly significant at the 0.05 level for larceny, is only signifi-
cant at the 0.10 level for burglary, and is no longer statistically significant, even at the
more generous 0.10 level, for rape.
14. Although not the focus of this article, the elasticities for prison populations are
similar to those reported in previous studies using state panel data (Levitt, 1996; Mar-
vell and Moody, 1994). However, these estimates are likely to overstate the impact of
imprisonment if one accounts for the criminogenic effects of prisons reported above.
Because the elasticities for prison release with regard to crime are roughly one fourth
that on prison population, this suggests that estimates of incapacitation overstate the
impact of imprisonment by roughly 25%, or perhaps about 14% if you take into
account the relative numbers of prisoners and releases.

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 19 21-AUG-07 9:49
CRIMINOGENIC EFFECTS OF IMPRISONMENT 607
prison releases. The ratio of murder, rape, robbery, assault, burglary, lar-
ceny, and auto theft to prison releases are 0.06, 0.28, 1.18, 2.28, 9.50, 24,
and 3.47, respectively. The result is an increase of slightly over two index
crimes per year per additional release in each state. Most increases occur
in the property crime categories. Of course, this number would increase to
over four index crimes per prison release if we adjusted the FBI crime
figures for underreporting. On the other hand, this figure would decrease
if we did not include crime elasticities for crime types in which the prison
release variable was not statistically significant at conventional significance
levels.
TESTING FOR POTENTIAL REVERSE INCAPACITATION
EFFECTS
As noted, it is unlikely that prison releases lead to increases in crime
rates by lowering prison populations (i.e., via reverse incapacitation) as
prison admissions have generally outpaced prison releases, at least during
the study period examined here. In any event, we reestimated the regres-
sions in Table 2 while dropping prison population as a control variable. If,
in fact, prison releases increase crime by lowering prison populations (i.e.,
if prison release is an antecedent variable), then the coefficients for the
prison release variable should become larger in the positive direction and
achieve greater statistical significance as the prison release variable now
reflects both the effects of prisonization and reverse incapacitation.
Although not shown, the results of these estimations provide no evidence
that prison releases increase crime by lowering prison populations. In fact,
the coefficients for the prison release variable decreased by roughly half
for most crime types, and only two estimates (robbery and burglary)
remained statistically significant at the 0.05 level. Instead, these results
support our earlier suspicion that prison population probably serves as a
suppressor variable in the prison release–crime relationship (more prison-
ers leads to more releases), and thus, failing to control for changes in
prison population levels would have masked, at least partially, the crimi-
nogenic effects of prison on crime.
SIMULTANEITY BIAS
Although the results in Table 2 provide strong evidence for the crimi-
nogenic effects of incarceration on crime, they might understate the total
impact of incarceration on crime if state justice systems respond to
increases in crime by curtailing the number of inmates released from
prison. That is, our results do not account for the possibility of simultane-
ous effects, and thus, the coefficients for prison release presented in Table
2 may be lower bound estimates of the contribution of prison releases to

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 20 21-AUG-07 9:49
608 VIERAITIS, KOVANDZIC, & MARVELL
crime. We explore this possibility by using the Granger causality test. The
Granger test is an econometric procedure used to explore causal direction
and to determine the ability of one variable to predict another (Wool-
dridge, 2000). The logic behind the Granger causality test is nicely summa-
rized by Marvell and Moody (1994:122):
[T]he fundamental notion underlying the [Granger] test is that if X
causes Y, then lagged values of X will be significant in a regression of
Y on its own lagged values and the lagged values of X.
The Granger test has a drawback in that it cannot rule out contemporane-
ous causation. In this situation, however, this drawback is not likely to be a
serious problem. It can be assumed that if crime rates have a current-year
impact on the number of inmates released from prison, crime rates must
also have a 1-year lagged impact on prison releases. The reason is that it
takes time for policy makers and prison officials to learn of changes in
crime trends, provide funds to incarcerate offenders for longer periods of
time, and change standards inmates must meet to be eligible for release.
15
Also, because we conduct the Granger test in levels (as opposed to first-
differences), any contemporaneous causation would be reflected in the 1-
year lag caused by serial correlation (correlation between current and
prior year crime rates). For these reasons, therefore, the absence of a 1-
year lagged impact of crime on the prison release variable implies the
absence of a current-year impact. To examine whether crime rates
“Granger-cause” prison releases, we regressed prison release rates on 1-
year and 2-year lags of crime rates and its own lagged values.
16
If the
lagged crime rates were jointly significant (as determined by the F test),
then crime rates Granger-caused prison releases.
The results of the Granger test for each crime type are presented in
Table 3. The results of the Granger test show little evidence that crime
rates Granger-cause prison releases. Although the F statistic for the signif-
icance of the two lagged crime rate variables as a group is highly signifi-
cant for homicide, robbery, larceny, and auto theft, the coefficients for the
lagged crime rate variables are in the unexpected positive direction. This
finding suggests increases in these crime categories lead to increases in
prison releases. The impact is delayed until the second lag; the coefficient
on the 1-year lag is seldom significant. The most likely explanation for this
finding is that increases in crime, all else equal, eventually lead to
15. We should also note that increases in current-year crime would have no bear-
ing on the roughly one-quarter of prisoners who are released from prison each year by
virtue of having served their entire sentence.
16. The Granger test was initially conducted using three lags. The second and third
lags were then dropped if they were not significant and if the significance level of the F
test did not decline. This resulted in two lags of crime in the prison release regressions.

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 21 21-AUG-07 9:49
CRIMINOGENIC EFFECTS OF IMPRISONMENT 609
TABLE 3. GRANGER ANALYSIS OF THE IMPACT OF
CRIME RATES ON PRISON RELEASES
Coefficients on Crime Rates Lagged
1 and 2 Years (Independent Variables)
1-year lag 2-year lag F Value
Independent Variable Coefficient tCoefficient tValue Probability
Homicide 0.022 0.87 0.09 3.64 17.31 0.00
Rape –0.041 –0.99 0.045 1.07 1.48 0.48
Robbery 0.067 1.53 0.06 1.46 10.61 0.01
Assault –0.022 –0.50 0.045 1.00 1.11 0.57
Burglary 0.050 0.65 0.113 1.48 6.93 0.03
Larceny 0.174 1.72 0.215 2.12 23.29 0.00
Motor vehicle theft –0.067 –1.35 0.126 2.57 7.44 0.02
NOTES: Prison releases are regressed on prison releases lagged 1 and 2 years, the crime rate
lagged 1 and 2 years, and the control variables listed in Table 1. Only the results for lags of
crime are presented, and the F value is for the two lags. All regressions use weighted least
squares, where the weighting is the square root of state population. The number of
observations in each prison release equation is 1,179. In all cases, estimation allows for panel-
level heteroskedasticity and contemporaneous correlation of error terms. Coefficients that
are significant at the 0.10 level are underlined. Coefficients that are significant at the 0.05
level are displayed in bold. Coefficients that are significant at the 0.01 level are both
underlined and displayed in bold.
increases in prison admissions, which in turn lead to increases in the num-
ber of prisoners released from prisons as state governments attempt to
make room for newly admitted felons. It is unlikely that the significant
positive coefficients observed for the prison release variable in the homi-
cide, robbery, and burglary specifications in Table 2 are from the positive
reverse causation of crime on prison releases. As discussed, a priori, one
would not expect a current-year impact because the process of changing
prison release standards in response to rising crime takes considerable
time. More importantly, the coefficients on the 1-year lag for these crimes
are never statistically significant (even at the more generous 0.10 level). It
is reasonable to assume that if there is no effect at the 1-year lag, there is
also no current-year impact. If there was an impact, it would be evident in
the 1-year lag because of the serial correlation between current-year and
prior-year crime levels. In all, there is no evidence that state policy makers
and prison officials respond to increases in crime rates by reducing the
number of inmates released from prison. More importantly, there is little
reason to believe that the parameter estimates of the effects of prison
releases on crime rates reported in Table 2 were overestimated or underes-
timated because of the presence of simultaneity bias.

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 22 21-AUG-07 9:49
610 VIERAITIS, KOVANDZIC, & MARVELL
INDIVIDUAL STATE RESULTS
Based on the evidence presented in Table 2, there is strong evidence to
support the claim that incarceration increases crime rates through the pro-
cess of prisonization. But the panel regressions in Table 2 have assumed a
single, homogeneous effect of prison releases on crime rates in all states.
If, for example, the postrelease criminal behavior of former inmates on
crime rates varies significantly across states, then the pooled regressions
presented in Table 2 are misspecified. The dangers of estimating a single
aggregated effect are particularly acute in this case because of the vast
differences across states in the discretion granted to correctional authori-
ties in determining which offenders to release and when. As noted by
Raphael and Stoll (2004), states operating under an indeterminate sen-
tencing system generally provide parole boards with complete discretion
over release and parole decisions, whereas those operating under a deter-
minate sentencing system have either abolished or severely limited the dis-
cretionary power of their parole boards. Theoretically, such different
sentencing systems may lead to the differing effects that the experience of
incarceration can have on criminal activity. One might hypothesize, for
example, that the criminogenic effects of prisonization are less in states
with discretionary parole release practices, because it encourages offend-
ers to participate in prison reintegration programs in order to increase the
chance of early parole (Raphael and Stoll, 1994). To the extent that partic-
ipation in such programs provides offenders with needed services (e.g.,
education, job training, or life-skills classes), the effects of prisonization on
future criminal behavior may be lessened.
One way to remedy this problem is to change the specification to esti-
mate a state-specific effect for prison releases in each state. In other
words, we include in our panel regressions for each crime category a sepa-
rate prison release variable for each state. Table 4 presents all of these
estimates for all seven crime categories. Table 4 results reject the more
constrained specifications of the aggregate regressions, which implicitly
assumed that the impact of prison releases was constant across states.
Indeed, for each crime type, we were able to reject the hypothesis that the
46 state-specific prison release variables were jointly equal. This result sug-
gests that the panel regressions presented in Table 2, which assumed uni-
form impacts of prison releases on crime for all states, is too restrictive.
With the exception of rape, the coefficients on the state-specific prison
release variables suggest that the number of states that experience signifi-
cant increases in crime rates because of increases in prison releases is
greater than the number of states that experience significant decreases.
For example, although 16 states seem to experience statistically significant
increases in robbery because of increases in the number of released

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 23 21-AUG-07 9:49
CRIMINOGENIC EFFECTS OF IMPRISONMENT 611
TABLE 4. THE ESTIMATED IMPACT OF PRISON RELEASES ON CRIME RATES:
STATE-SPECIFIC ESTIMATES
Motor Vehicle
Homicide Rape Robbery Assault Burglary Larceny Theft
Coef. tCoef. tCoef. tCoef. tCoef. tCoef. tCoef. t
Alabama .106 1.25 –.156 –1.76 .001 0.01 .039 0.30 –.039 –0.72 –.040 –0.91 .106 1.25
Alaska –.075 –0.34 –.027 –0.21 .314 2.17 .023 0.26 .076 1.17 .062 1.42 –.075 –0.34
Arizona –.120 –1.17 –.113 –1.73 –.107 –1.58 –.033 –0.58 –.034 –0.58 –.022 –0.34 –.120 –1.17
Arkansas .010 0.05 –.050 –0.31 .133 0.97 –.004 –0.05 –.032 –0.41 –.014 –0.23 .010 0.05
California .068 1.12 –.125 –7.05 .116 2.87 .097 3.87 .028 0.85 .006 0.25 .068 1.12
Colorado .200 0.96 .147 1.23 .259 2.32 –.059 –0.55 .204 3.46 .046 1.06 .200 0.96
Delaware .298 1.68 –.113 –0.62 –.071 –0.46 –.199 –2.00 –.008 –0.09 –.063 –0.81 .298 1.68
Florida .076 1.88 –.100 –3.52 .138 4.08 –.001 –0.03 .069 3.04 .023 1.32 .076 1.88
Georgia –.164 –1.27 .272 2.61 –.013 –0.20 .071 1.74 .094 2.05 .051 0.87 –.164 –1.27
Hawaii –.060 –0.56 –.096 –1.89 –.075 –1.71 –.077 –2.48 –.048 –1.71 –.048 –2.48 –.060 –0.56
Idaho .325 0.69 –.003 –0.02 .354 1.50 –.154 –0.98 .026 0.26 .080 1.32 .325 0.69
Illinois .042 0.54 –.046 –0.40 .121 1.19 .128 1.48 .074 1.45 –.003 –0.09 .042 0.54
Indiana .076 0.48 –.069 –0.91 .105 0.89 –.127 –0.96 .076 1.76 .037 1.27 .076 0.48
Iowa .138 0.53 –.339 –1.79 –.349 –1.65 .097 0.73 –.008 –0.07 –.050 –0.68 .138 0.53
Kansas .293 0.99 –.003 –0.03 .259 3.00 –.006 –0.08 .126 2.02 –.022 –0.35 .293 0.99
Kentucky –.435 –1.81 .013 0.07 .199 1.29 –.020 –0.05 .030 0.30 –.073 –1.06 –.435 –1.81
Louisiana .224 1.38 –.057 –0.71 .269 2.61 .108 1.63 .076 1.48 .178 3.71 .224 1.38
Maine –.328 –1.32 .248 2.51 .078 0.75 –.104 –1.22 .069 1.32 –.010 –0.27 –.328 –1.32
Maryland .096 1.29 .136 1.97 .116 2.06 –.062 –1.16 .081 2.03 .121 3.73 .096 1.29
Massachusetts .039 0.37 –.002 –0.05 .006 0.10 .129 1.85 .045 1.23 .007 0.23 .039 0.37
Michigan –.333 –4.95 –.034 –0.54 –.035 –0.41 .042 0.71 .068 1.38 .037 1.17 –.333 –4.95
Minnesota .119 0.43 –.044 –0.14 –.073 –0.45 –.104 –1.08 –.003 –0.03 –.046 –0.95 .119 0.43

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 24 21-AUG-07 9:49
612 VIERAITIS, KOVANDZIC, & MARVELL
TABLE 4. CONTINUED
Motor Vehicle
Homicide Rape Robbery Assault Burglary Larceny Theft
Coef. tCoef. tCoef. tCoef. tCoef. tCoef. tCoef. t
Mississippi .328 1.50 –.128 –0.79 .174 0.94 –.173 –0.82 .061 0.59 .072 0.96 .328 1.50
Missouri .182 1.34 .107 1.21 .365 3.05 –.087 –0.68 .032 0.33 –.091 –1.25 .182 1.34
Montana –.881 –1.69 .127 0.43 .367 1.25 .902 2.81 –.002 –0.01 .000 0.00 –.881 –1.69
Nebraska –.353 –1.59 –.012 –0.07 –.222 –1.73 .130 0.87 –.035 –0.57 –.042 –0.93 –.353 –1.59
Nevada –.326 –2.10 .050 0.68 –.040 –0.49 –.341 –2.15 –.003 –0.06 .078 1.76 –.326 –2.10
New Hampshire .574 2.21 .097 0.77 .273 2.18 –.325 –1.61 –.075 –0.91 .070 1.92 .574 2.21
New Jersey –.061 –0.58 .086 0.86 –.013 –0.16 –.076 –1.59 .116 1.75 –.002 –0.04 –.061 –0.58
New Mexico .017 0.07 –.054 –0.52 –.099 –1.15 –.032 –0.38 .022 0.40 –.045 –1.09 .017 0.07
New York .427 4.81 –.091 –2.37 .160 3.06 –.017 –0.36 .155 4.54 .042 1.38 .427 4.81
North Carolina .038 0.97 .057 2.62 .056 1.84 .006 0.27 .049 2.21 .036 2.63 .038 0.97
North Dakota 1.29 1.39 .324 1.72 .771 2.79 –.276 –1.52 .138 0.99 –.088 –1.47 1.29 1.39
Ohio .064 0.48 .050 0.92 .161 2.13 .048 0.44 –.018 –0.30 –.059 –1.50 .064 0.48
Oklahoma .004 0.02 .034 0.57 .268 3.44 –.111 –2.06 .059 1.14 .018 0.36 .004 0.02
Pennsylvania –.213 –1.76 –.226 –2.17 –.090 –0.68 –.268 –1.57 –.164 –2.40 –.095 –1.49 –.213 –1.76
Rhode Island .228 2.15 .053 0.75 .116 2.02 –.040 –0.70 .058 1.38 .071 2.24 .228 2.15
South Carolina –.084 –0.39 –.092 –0.66 .081 0.96 –.020 –0.22 .062 1.24 –.016 –0.47 –.084 –0.39
South Dakota 1.57 2.02 .058 0.29 –.367 –1.26 .165 0.91 –.038 –0.40 –.025 –0.26 1.57 2.02
Tennessee .082 0.77 –.090 –0.66 .222 2.83 .179 4.16 .123 2.00 .108 2.11 .082 0.77
Utah .065 0.25 .080 0.64 –.033 –0.25 –.068 –0.58 –.111 –1.01 .052 0.56 .065 0.25
Vermont .365 1.39 .017 0.14 .130 0.70 .070 0.77 .031 0.52 –.039 –0.50 .365 1.39
Virginia .048 0.77 –.029 –0.62 .055 0.95 –.028 –0.68 .001 0.04 .043 1.46 .048 0.77
Washington .372 1.87 –.214 –1.60 –.051 –0.50 .085 1.32 .003 0.02 .079 1.67 .372 1.87
Wisconsin –.033 –0.19 .224 2.14 –.164 –1.01 .036 0.27 .062 0.88 .010 0.16 –.033 –0.19
Wyoming .272 0.57 –.125 –0.68 –.264 –2.05 .022 0.14 –.170 –1.41 –.008 –0.12 .272 0.57

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 25 21-AUG-07 9:49
CRIMINOGENIC EFFECTS OF IMPRISONMENT 613
TABLE 4. CONTINUED
Motor Vehicle
Homicide Rape Robbery Assault Burglary Larceny Theft
Coef. tCoef. tCoef. tCoef. tCoef. tCoef. tCoef. t
Summary for 46 Prison Release Variables
Neg. & Sig. 5 8 4 4 2 1 4
Neg. & Not Sig. 9 19 13 23 14 21 11
Pos. & Sig. 7 6 16 5 10 8 8
Pos. & Not Sig. 25 13 13 14 20 16 23
NOTES: These results are the coefficients on state-specific prison release variables, which are created by multiplying the aggregate prison release
variable by the state dummy variables. Otherwise, the model specification is the same as estimated in Table 2, and the results for other variables are
similar. In all cases, estimation allows for panel-level heteroskedasticity and contemporaneous correlation of error terms. Coefficients that are
significant at the 0.10 level are underlined. Coefficients that are significant at the 0.05 level are displayed in bold. Coefficients that are significant at the
0.01 level are both underlined and displayed in bold.

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 26 21-AUG-07 9:49
614 VIERAITIS, KOVANDZIC, & MARVELL
inmates, only 4 states experience a statistically significant decrease. The
second column of Table 4 reveals a disparity for rape. Eight states experi-
ence a statistically significant decrease in rape, whereas only six states
exhibit a statistically significant increase. Out of the 322 estimated impacts
in Table 4 (46 states by seven crime categories), 60 exhibited statistically
significant increases in crime because of increases in the number of
released inmates, whereas only 28 exhibited statistically significant
decreases. In all, the evidence for prisonization effects on crime is strong
and seems to overwhelm any negative impacts of incarceration on crime
through deterrence and/or rehabilitation.
DISCUSSION
The purpose of our study was to examine the impact of prison release
on crime rates. As explained, we hypothesized that prison release may
have a direct and positive impact on crime if prisoners commit more
crimes than they would have had they not gone to prison (i.e., prisons are
“criminogenic”), or prison release reverses the effects of incapacitation by
reducing the prison population or replacing releasees with less serious
offenders. Using a state panel data set for 46 states from 1974 to 2002, our
analyses indicate that increases in the number of prisoners released from
prison seem to be significantly associated with increases in crime. Because
we control for changes in prison population levels, we attribute the appar-
ent positive influences on crime that seem to follow prison releases to the
criminogenic effects of prison.
Whether the pains associated with imprisonment are deliberate (i.e.,
purposely designed by administrators) and/or warranted (i.e., the principle
of least eligibility), the result is that imprisonment causes harm to prison-
ers. Moreover, the evidence presented here suggests that the general pub-
lic suffers harm through increases in crime. Although this study does not
allow us to draw definitive conclusions about the pathways from prison
release to crime rates, research offers several explanations as to how incar-
ceration may increase the likelihood of crime after release.
The nature of prison life makes the transition from prison to society
very difficult. Unlike life on the outside, prison life is highly routinized,
restrictive, and isolating, and thus interferes with inmates’ capacity for
power and control over their lives (Carceral, 2004; Irwin, 2005; Travis,
2005). The daily routine of prison life and the adaptive behaviors one
develops are likely inconsistent with social, family, and work routines
(Irwin and Austin, 1997). The transition to life on the outside is also com-
plicated by inmates’ sometimes lengthy absence from family, relatives, and
friends. To the extent that the social support provided by significant others
increases the likelihood of successful reentry and reintegration, years of

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 27 21-AUG-07 9:49
CRIMINOGENIC EFFECTS OF IMPRISONMENT 615
limited contact and isolation can severely hamper an ex-inmate’s access to
support upon release.
The experience of prison also complicates an offender’s ability to suc-
cessfully reenter society because it creates obstacles to and leaves them ill-
prepared to securing housing and work (Petersilia, 2003). Prisons may
limit access or fail to provide programs designed to address deficiencies in
education and vocational skills or to address substance abuse and/or
mental health problems (Austin, 2001; Lynch and Sabol, 2001; Petersilia,
2003). Many ex-prisoners have low levels of educational attainment,
lengthy histories of substance abuse, and mental health problems. Accord-
ing to BJS data, more than two thirds of all state inmates did not receive a
high-school diploma (Harlow, 2003), more than half met the DSM-IV cri-
teria for drug dependence or abuse (Mumola and Karberg, 2006), and
more than half had a mental health problem (James and Glaze, 2006).
Moreover, legal restrictions that prohibit offenders from qualifying for
public housing or securing employment in certain occupations also limit
their access to housing and work. For example, a felony record can bar an
ex-prisoner from employment in licensed or professional occupations, spe-
cifically jobs in health care, skilled trades, and in several states, public sec-
tor employment (Western et al., 2001).
The most damaging factor to an offender’s ability to reenter society and
avoid crime may be the stigma of the label “ex-prisoner.” Although an
offender’s status as “ex-prisoner” interferes with his or her attempts to
exercise his or her legal rights, access services, and secure housing, perhaps
the most serious consequence is its impact on the probability of employ-
ment and income. Research demonstrates that a criminal record is a signif-
icant obstacle to finding work and greatly reduces one’s earning potential
(Pager, 2003; Western, 2002; Western and Beckett, 1999). Although obsta-
cles to securing employment and decent earnings may be caused, in part,
by the characteristics of offenders independent of the prison experience
and label of “ex-prisoner,” prior research suggests otherwise (see Western
et al., 2001 for a discussion).
Ethical considerations about the treatment of inmates aside, it has
become increasingly clear that policy makers must begin to address the
unintended consequences of mass incarceration as a crime-control strat-
egy. As the number of returning prisoners increases, policy makers will be
faced with a serious policy dilemma—increase funding for research, devel-
opment, and support of programs that increase the likelihood of successful
reentry and reintegration of ex-prisoners into society, or build more pris-
ons to house an increasing number of returning prisoners. The latter
approach may simply delay the current problem for future generations to
deal with, whereas the former depends heavily on identifying the pathways

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 28 21-AUG-07 9:49
616 VIERAITIS, KOVANDZIC, & MARVELL
to recidivism and on identifying existing programs or developing new pro-
grams that can ease or eliminate the “pains of imprisonment” that increase
the likelihood of recidivism and push crime rates upward. For example, a
long line of research suggests that ex-offenders who secure good jobs and
establish stable marriages pose less threat to public safety than those who
remain single and unemployed (see Uggen et al., 2005 for a discussion).
The public’s interest, in terms of safety, may be best served by policies
designed to increase the rate of successful prisoner reentry. Our study sug-
gests that, in addition, to the harm caused to prisoners and communities as
identified in previous research (Austin and Irwin, 2001; Clear, 1994; 1997;
Clear et al., 2001; Irwin, 2005; Mauer, 1999; Mauer and Chesney-Lind,
2002; Petersilia, 2003; Rose and Clear, 1998; Travis, 2005), our policy of
mass incarceration causes harm to the public through increases in crime as
the result of prison releases. Policy makers should continue to serve the
public interest by carefully considering policies designed to reduce incar-
ceration rates and thus assuage the criminogenic effects of prison. These
policies may include changes in sentencing, probation and/or parole prac-
tices, or better funding of reentry services prerelease and postrelease.
REFERENCES
Austin, James
2001 Prisoner reentry: Current trends, practices, and issues. Crime and
Delinquency 47:314–334.
Austin, James and John Irwin
2001 It’s About Time: America’s Imprisonment Binge. Stamford, Conn.:
Wadsworth.
Ayres, Ian and John J. Donohue III
2003 Shooting down the more guns, less crime hypothesis. Stanford Law
Review 55:1193–1314.
Beck, Nathaniel and Jonathan N. Katz
1995 What to do (and not to do) with time-series cross-section data in
comparative politics. American Political Science Review 89:634–647.
1996 Nuisance vs. substance: Specifying and estimating time-series-cross-section
models. Political Analysis 6:1–36.
Bellair, Paul E.
1997 Social interaction and community crime: Examining the importance of
neighbor networks. Criminology 35:677–704.
Belsley, David A., Edward Kuh, and Roy E. Welsh
1980 Regression Diagnostics. New York: Wiley.
Box, Stephen
1987 Recession, Crime and Punishment. Totawa, N.J.: Barnes and Noble.

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 29 21-AUG-07 9:49
CRIMINOGENIC EFFECTS OF IMPRISONMENT 617
Cappell, Charles L. and Gresham Sykes
1991 Prison commitments, crime, and unemployment: A theoretical and
empirical specification for the United States, 1933-1985. Journal of
Quantitative Criminology 7:155–199.
Carceral, K.C.
2004 Behind a Convict’s Eyes: Doing Time in Modern Prison. Belmont, Calif.:
Wadsworth.
Chiricos, Theodore G.
1987 Rates of crime and unemployment: An analysis of aggregate research
evidence. Social Problems 34:187–212.
Chiricos, Theodore G. and Miriam Delone
1992 Labor surplus and punishment: A review and assessment of theory and
evidence. Social Problems 39:421–446.
Clear, Todd R.
1994 Harm in American Penology: Offenders, Victims, and Their Communities.
Albany, N.Y.: University of New York Press.
1996 Backfire: When incarceration increases crime. Journal of the Oklahoma
Criminal Justice Research Consortium 3. Available online: http://www.doc.
state.ok.us/DOCS/OCJRC/OCJRC96/Ocjrc96.htm.
1997 Ten unintended consequences of the growth in imprisonment. Correc-
tional Management Quarterly 1:25–31.
Clear, Todd, Dina R. Rose, and Judith A. Ryder
2001 Incarceration and the community: The problem of removing and
returning offenders. Crime and Delinquency 47:335–351.
Cook, Philip J.
1998 The epidemic of youth gun violence. In U. S. Department of Justice,
Perspectives on Crime and Justice: 1997-1998 Lecture Series, vol. 2.
Washington, D.C.: National Institute of Justice. Available online: http://
www.ncjrs.org.
DeFina, Robert H. and Thomas M. Arvanites
2002 The weak effect of imprisonment on crime: 1971-1998. Social Science
Quarterly 83:635–653.
Devine, Joel A., Joseph F. Sheley, and M. Dwayne Smith
1988 Macroeconomic and social-control policy influences on crime rate
changes, 1948-1985. American Sociological Review 53:407–420.
Federal Bureau of Investigation
2005 Crime in the United States, 2004. Uniform Crime Reports. Washington,
D.C.: U.S. Government Printing Office.
Gendreau, Paul, Claire Goggin, and Francis T. Cullen
1999 The effects of prison sentences on recidivism. Report to the Department
of the Solicitor General Canada. Available online: http://www.sgc.gc.ca.
Glaze, Lauren E. and Seri Palla
2005 Probation and Parole in the United States, 2004. Washington, D.C.:
Bureau of Justice Statistics.
Greenberg, David F.
1977 The dynamics of oscillatory punishment processes. Journal of Criminal
Law and Criminology 68:643–655.

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 30 21-AUG-07 9:49
618 VIERAITIS, KOVANDZIC, & MARVELL
Greenberg, David F. and Valerie West
2001 State prison populations and their growth, 1971-1991. Criminology
39:615–654.
Greene, William H.
2003 Econometric Analysis. Upper Saddle River, N.J.: Prentice-Hall.
Greenfeld, Lawrence A.
1995 Prison Sentences and Time Served for Violence. Washington, D.C.:
Bureau of Justice Statistics.
Hale, Chris
1989 Unemployment, imprisonment, and the stability of punishment hypothe-
sis: Some results using cointegration and error correction models. Journal
of Quantitative Criminology 5:169–186.
Hamilton, James D.
1994 Time Series Analysis. Princeton, N.J.:Princeton University Press.
Harlow, Caroline Wolf
2003 Education and Correctional Populations. Washington, D.C.: Bureau of
Justice Statistics.
Holzer, Harry J., Steven Raphael, and Michael A. Stoll
2002 Can employers play a more positive role in prisoner reentry? Working
Discussion Paper for the Urban Institute Reentry Roundtable, Washing-
ton, D.C., March 20–21.
Im, Kyong So, Hashem Pesaran, and Yongcheol Shin
2003 Testing for unit roots in heterogeneous panels. Journal of Econometrics
115:53-74.
Irwin, John
2005 The Warehouse Prison: Disposal of the New Dangerous Class. Los
Angeles, Calif.: Roxbury.
Irwin, John and James Austin
1997 It’s About Time: America’s Imprisonment Binge. Belmont, Calif.: Wad-
sworth.
James, Doris J. and Lauren E. Glaze
2006 Mental Health Problems of Prison and Jail Inmates. Washington, D.C.:
Bureau of Justice Statistics.
Kovandzic, Tomislav V. and Lynne M. Vieraitis
2006 The effect of county-level prison population growth on crime rates.
Criminology & Public Policy 5:213–244.
Kovandzic, Tomislav V., Thomas B. Marvell, Lynne M. Vieraitis, and Carlisle E.
Moody
2004 When prisoners get out: The impact of prison releases on homicide rates,
1975-1999. Criminal Justice Policy Review 15:212–228.
Land, Kenneth C., Patricia L. McCall, and Lawrence E. Cohen
1990 Structural covariates of homicide rates: Are there any invariances across
time and social space? American Journal of Sociology 95:922–963.
Letkemann, Peter
1973 Crime as Work. Englewood Cliffs, N.J.: Prentice-Hall.

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 31 21-AUG-07 9:49
CRIMINOGENIC EFFECTS OF IMPRISONMENT 619
Levitt, Steven D.
1996 The effect of prison population size on crime rates: Evidence from prison
overcrowding litigation. Quarterly Journal of Economics 111:319–351.
2004 Understanding why crime fell in the 1990s: Four factors that explain the
decline and six that do not. Journal of Economic Perspectives 18:163–190.
Lynch, James P. and William J. Sabol
2001 Prisoner reentry in perspective. Crime Policy Report, vol. 3. Washington,
D.C.: The Urban Institute.
Markowitz, Fred E., Paul E. Bellair, Allen E. Liska, and Jianhong Liu
2001 Extending social disorganization theory: Modeling the relationship
between cohesion, disorder, and fear. Criminology 39:293–320.
Marvell, Thomas B. and Carlisle E. Moody
1991 Age structure and crime rates: The conflicting evidence. Journal of
Quantitative Criminology 7:237–273.
1994 Prison population growth and crime reduction. Journal of Quantitative
Criminology 10:109–140.
1996 Specification problems, police levels, and crime rates. Criminology
34:609–646.
1997 The impact of prison growth on homicide. Homicide Studies 1:205–233.
1998 The impact of out-of-state prison population on state homicide rates:
Displacement and free rider effects. Criminology 36:513–535.
2001 The lethal effects of three-strikes laws. Journal of Legal Studies
30:89–106.
Mauer, Marc
1999 Race to Incarcerate. New York: The New Press.
Mauer, Marc and Meda Chesney-Lind
2002 Invisible Punishment: The Collateral Consequences of Mass Imprison-
ment. New York: The New Press.
Moody, Carlisle E.
2001 Testing for the effects of concealed weapons laws: Specification errors and
robustness. Journal of Law and Economics 44:799-813.
Mumola, Christopher J.
1999 Substance Abuse and Treatment, State and Federal Prisoners, 1997.
Washington, D.C.: Bureau of Justice Statistics.
Mumola, Christopher J. and Jennifer C. Karberg
2006 Drug Use and Dependence, State and Federal Prisoners, 2004. Washing-
ton, D.C.: Bureau of Justice Statistics.
Nickell, Stephen
1981 Biases in dynamic models with fixed effects. Econometrica 49:1417-1426.
Pager, Devah
2003 The mark of a criminal record. American Journal of Sociology
108:937–975.
Parker, Robert N. and Allan Horowitz
1986 Unemployment, crime, and imprisonment: A panel approach. Criminol-
ogy 24:751–753.

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 32 21-AUG-07 9:49
620 VIERAITIS, KOVANDZIC, & MARVELL
Petersilia, Joan
2000 When prisoners return to the community: Political, economic, and social
consequences. Sentencing and Corrections 9:1–8. Washington, D.C.: US
Department of Justice.
2001 Prisoner reentry: Public safety and reintegration challenges. The Prison
Journal 81:360–375.
2003 When prisoners come home: Parole and prisoner reentry. Oxford, U.K.:
Oxford University Press.
Raphael, Steven and Michael A. Stoll
2004 The effect of prison releases on regional crime rates. Brookings-Wharton
Papers on Urban Affairs 2004:207–256.
Rashbaum, William K.
2002 Falling crime in New York defies trend. The New York Times (November
29):B1.
Reiss, Albert J. and Jeffrey A. Roth
1993 Understanding and Preventing Violence. Washington, D.C.: National
Academy Press.
Rose, Dina R. and Todd R. Clear
1998 Incarceration, social capital, and crime: Implications for social disorgani-
zation theory. Criminology 36:441–479.
Rusche, George and Otto Kirchheimer
1939 Punishment and Social Structure. New York: Columbia University Press.
Sabol, William J.
1989 The dynamics of unemployment and imprisonment in England and Wales,
1946-1985. Journal of Quantitative Criminology 5:147–168.
Sampson, Robert J., Stephen W. Raudenbush, and Felton Earls
1997 Neighborhoods and violent crime: A multilevel study of collective
efficacy. Science 277:918–924.
The Sentencing Project
2005 New incarceration figures: growth in population continues. Available
online: http://www.sentencingproject.org.
Sherman, Lawrence W., David P. Farrington, Brandon C. Welsh, and Doris Layton
MacKenzie
2002 Evidence Based Crime Prevention. New York: Routledge.
Speiglman, Richard
1977 Prison drugs, psychiatry, and the state. In David F. Greenberg (ed.),
Corrections and Punishment. Beverly Hills, Calif.: Sage.
Stephan, James J.
2004 State prison expenditures, 2001. Washington, D.C.: Bureau of Justice
Statistics. Available online: http://www.ojp.usdoj.gov/bjs/pub/pdf/spe01.pdf.
Sykes, Gresham
1958 The Society of Captives: A Study of a Maximum Security Prison.
Princeton, N.J.: Princeton University Press.

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 33 21-AUG-07 9:49
CRIMINOGENIC EFFECTS OF IMPRISONMENT 621
Travis, Jeremy
2002 Invisible punishment: An instrument of social exclusion. In Marc Mauer
and Meda Chesney-Lind (eds.), Invisible Punishment. New York: The
New Press.
2005 But They All Come Back: Facing the Challenges of Prisoner Reentry.
Washington, D.C.: Urban Institute Press.
Travis, Jeremy and Sarah Lawrence
2002 Beyond the Prison Gates: The State of Parole in America. Washington,
D.C.: The Urban Institute.
Uggen, Christopher
1999 Ex-offenders and the conformist alternative: A job quality model of work
and crime. Social Problems 46:127–151.
Uggen, Christopher, Sara Wakefield, and Bruce Western
2005 Work and family perspectives on reentry. In Jeremy Travis and Christy
Visher (eds.), Prisoner Reentry and Crime in America. New York:
Cambridge University Press.
U.S. Bureau of Justice Statistics
2001 Prisoners in 2000. Bureau of Justice Statistics Special Report, August
2001. Washington, D.C.: U.S. Government Printing Office.
2004. Homicide Trends in the United States. Bureau of Justice Statistics,
September 2004. Available online: http://www.ojp.usdoj.gov/bjs/homicide/
homtrnd.htm.
2006 Reentry trends in the U.S.: Releases from state prison. Available online:
http://www.ojp.usdoj.govbjs/reentry/releases.htm.
Vieraitis, Lynne M.
2000 Income inequality, poverty, and violent crime: A review of the empirical
evidence. Social Pathology 6:24–45.
Western, Bruce
2002 The impact of incarceration on wage mobility and inequality. American
Sociological Review 67:526-546.
Western, Bruce and Katherine Beckett
1999 How unregulated is the U.S. labor market: The penal system as a labor
market institution. American Journal of Sociology 104:1030–1060.
Western, Bruce, Jeffrey R. Kling, and David F. Weiman
2001 The labor market consequences of incarceration. Crime and Delinquency
47:410–427.
Winship, Christopher
2002 End of a miracle? Crime, faith, and partnership in Boston in the 1990s.
Unpublished manuscript.
Wooldridge, Jeffrey M.
2000 Introductory Econometrics. Belmont, Calif.: South-Western Publishing.
Lynne M. Vieraitis is an associate professor in the Criminology Program at the Uni-
versity of Texas at Dallas. She conducts research in the areas of inequality and violent
crime and criminal justice policy. She received her Ph.D. degree in criminology from
Florida State University in 1999.

\\server05\productn\C\CPP\6-3\CPP312.txt unknown Seq: 34 21-AUG-07 9:49
622 VIERAITIS, KOVANDZIC, & MARVELL
Tomislav V. Kovandzic is an associate professor in the Criminology Program at the
University of Texas at Dallas. His research interests include gun-related violence, crimi-
nal justice policy, and inequality and violent crime. He received his Ph.D. degree in
criminology from Florida State University in 1999.
Thomas B. Marvell is a lawyer-sociologist and director of Justec Research.