Estimating a Dose–Response Relationship Between Time Served in Prison and Recidivism

Abstract

Objectives:

Estimate the dose–response relationship between time served in prison and offenders’ odds of recidivism.

Methods:

Using a large, representative sample of adult offenders released from prison under postrelease supervision in the state of Ohio, we examine the relationship between the length of time these offenders served in prison and their odds of recidivism during the year following their release. Multivariate logistic regression and analyses involving propensity score matching for ordered doses are both used to estimate the time served–recidivism relationship.

Results:

Analyses of these data revealed that offenders confined for longer periods of time had lower odds of recidivism, but these odds were only substantively lower for those offenders who served the longest periods of time in prison. Findings suggest the inverse effect of time served was not realized until after offenders have been confined for at least five years.

Conclusion:

Study findings indicate that the specific deterrent effect of prison sentences may be limited, and sentences less than five years may be reduced in order to save costs without a substantial threat to public safety.

Keywords

prison incarceration time served offender recidivism

Over the past several decades, incarceration rates in the United States have increased by over 200 percent (Blumstein and Beck 2005; Clear 1994; Glaze 2010; Irwin and Austin 2001), and the merits and various effects of this increased use of incarceration have been the subject of considerable academic and practical debate (see, e.g. summer (2010) issue of Daedalus, the February (2011) issue of Criminology & Public Policy, and the December (2007) issue of the Journal of Quantitative Criminology). Incarceration has several purposes (e.g., retribution), but researchers have typically focused on estimating the incapacitative or deterrent effects of imprisonment on crime (e.g., Clear 2008; Donohue 2007; Nagin 1998; Piquero and Blumstein 2007). Regarding specific deterrent effects, studies have found that over 60 percent of individuals released from prison are rearrested and approximately 45 to 50 percent are returned to prison within three years (e.g., Langan and Levin 2002; Pew Center for the States 2011). Findings such as these have prompted a number of scholars to argue that imprisonment does not reduce offenders’ odds of recidivism and, in fact, may even increase them (e.g., Gendreau, Little, and Goggin 1996). Yet empirical evidence regarding the effects of imprisonment or length of imprisonment on recidivism is fairly sparse, and the evidence that does exist is insufficient to draw firm conclusions regarding causal effects (Nagin, Cullen, and Jonson 2009). Given the frequency with which incarceration is used in the United States, an understanding of the effects of incarceration on offenders’ odds of recidivism is critical for informing theory and public policy.

In this study, we focus on the relationship between time served in prison and recidivism, what medical researchers term a dose–response relationship (i.e., different lengths of imprisonment function as different dosage levels of a treatment; Loughran et al. 2009; Nagin et al. 2009). In one of the better studies to date, Loughran et al. (2009) recently offered evidence of a null relationship between time served and recidivism among a sample of juvenile delinquents. We build upon their findings by investigating the time served–recidivism relationship with a representative sample of adult offenders released from prison under postrelease supervision in Ohio. Juvenile delinquents are developmentally different than adult offenders, and incarcerated delinquents often serve less time than imprisoned adults. Thus, the effect of time served on the recidivism rates of adult offenders may differ from the effect observed for juvenile delinquents. Further, there are over 1.6 million adult offenders incarcerated in state and federal correctional facilities (Guerino, Harrison, and Sabol 2011) compared to the approximately 81,000 juvenile delinquents confined in state or privately operated residential facilities (Hockenberry, Sickmund, and Sladky 2011), and so a determination of the effect of time served on recidivism among adults is of paramount importance to correctional policy and practice.

Perspectives on the Impact of Time Served on Recidivism

Scholars have outlined three perspectives on the impact of imprisonment on recidivism: (1) prison is punitive and deters future offending; (2) prison is criminogenic and increases offending; and (3) prison has no impact on offending behavior. We describe each of these perspectives briefly here (for more detailed reviews, see Gendreau, Goggin, and Cullen 1999; Nagin et al. 2009).

The prisons as punishment perspective emphasizes that prison is an unpleasant experience, and therefore the costs associated with imprisonment reduce the expected utility of committing crime (Nagin et al. 2009). Individuals’ estimates of the certainty that sanctions (imprisonment) will be imposed and the severity of those sanctions should they be imposed form the basis for their calculation of the costs of committing crimes that are weighed alongside the benefits associated with these acts (Becker 1968; Grasmick and Bursick 1990). Since the 1970s, the use of incarceration as a response to crime has increased, simultaneously increasing the certainty that individuals who commit a crime will receive a prison sentence (Blumstein and Beck 2005; Langan 1991; Nagin 1998). Imprisonment is perceived to be more severe than noncustodial sanctions, and longer prison sentences are perceived to be more severe than shorter prison sentences (Nagin et al. 2009). The criminal law also prescribes harsher penalties for individuals with a criminal history. Thus, individuals who have been released from prison may be deterred because they might expect the certainty of receiving a prison sentence and the severity of that prison sentence to be greater should they be arrested for a new offense (Nagin 1998; Nagin et al. 2009).¹

Scholars who argue that prison is criminogenic point to the dehumanizing experience of incarceration and the adaptation process to the prison subculture (Gendreau et al. 1999; Nagin et al. 2009). Ethnographic studies of prison environments have described an inmate culture emphasizing (or tolerating) violence and opposition toward legal authority that develops in response to the pains of prison life (Sykes 1958). In contrast, some scholars have argued that these subcultural values are “imported” into prisons and simply reflect subcultural values found in lower class neighborhoods (Irwin and Cressey 1962). Regardless of the origins of the value system, inmates who assimilate into the inmate subculture may have a more difficult time adjusting to life outside of prison (Irwin and Cressey 1962; Toch, Adams, and Grant 1989). Some scholars have referred to prisons as “schools of crime” because the prison culture may teach and reinforce antisocial values and behaviors, increasing the odds of recidivism for individuals who are imprisoned or individuals who serve more time (Gendreau et al. 1999; Nagin et al. 2009).

The minimalist perspective emphasizes that imprisonment has no impact on offending. Inmates have deficits in their abilities to cope with problems prior to coming to prison, which often contribute to criminal solutions to those problems (Johnson 2002; Zamble and Porporino 1988). Imprisonment alone does not correct inmates’ coping deficits (Gendreau et al. 1999; Zamble and Porporino 1988). Therefore, inmates go into a “psychological deep freeze” where their outside behavioral repertoire is stored away, only to be reactivated upon release (Zamble and Porporino 1988). In other words, individuals exhibit continuity in criminal behavior, regardless of whether they are incarcerated or how long they are incarcerated.

The extant literature on imprisonment and recidivism does not offer strong support for any of the perspectives described above. After reviewing the evidence regarding the time served–recidivism relationship, Nagin et al. (2009) found that several studies have revealed a deterrent effect associated with longer prison terms, while a few studies have uncovered a criminogenic effect. The majority of studies, however, have not found a relationship between time served and recidivism. Nagin et al. (2009) also identified several problems with the existing studies. First, most studies did not isolate the causal effect of time served on recidivism; time served was only included as a control variable. Second, most studies have failed to adequately address the potential selection bias inherent in estimating the causal effect of time served on recidivism (i.e., individuals confined for different periods of time may differ in some systematic way that may also be related to their odds of recidivism). Third, a number of the studies have focused on juvenile delinquents. Since juveniles are developmentally different from adults, and delinquents are often confined for shorter periods of time than adult offenders, the findings from those studies may not be generalizeable to adult offenders.

In one of the more rigorous studies to date, Loughran et al. (2009) examined the time served–recidivism relationship among a sample of juvenile delinquents. They addressed the possibility of selection bias by using propensity score matching (PSM) to “balance” groups of delinquents who were confined for different amounts of time on a number of theoretically relevant covariates. Loughran et al. (2009) found no meaningful differences in the recidivism rates of the groups of delinquents who had served different amounts of time.

We build upon and extend Loughran et al.’s (2009) important findings by examining the relationship between time served and recidivism among a representative sample of adult offenders released under postrelease supervision in the state of Ohio. Examination of the time served–recidivism relationship among adult offenders is important for several reasons. First, juveniles only represent a small fraction of the individuals incarcerated in state or federal facilities in the United States (compare Hockenberry et al. 2011 with Guerino et al. 2011). Second, the juvenile justice system is built upon the idea that delinquents should be treated differently and separately from adult offenders, and the fact that juveniles are still developing raises the possibility that they may respond to similar stimuli differently than adults (Loughran et al. 2009). Finally, incarcerated juvenile delinquents are also typically confined for shorter periods of time than imprisoned adult offenders, and so it seems plausible that deterrent effects may be more likely to be realized among adults where the severity of the punishment is often greater.

Correlates of Time Served and Postrelease Recidivism

Estimation of a dose–response relationship between time served in prison and recidivism with observational data requires identification, measurement, and “control” of the covariates that could confound the relationship (Berk 2004; Nagin et al. 2009). Nagin et al. (2009) recommended a minimum set of covariates that should be included in any study of the time served–recidivism relationship: age, sex, race, committing offense type, and prior record. Younger offenders may be more likely to recidivate because they often have fewer prosocial relationships and are less likely to be involved in conventional activities (MacKenzie and De Li 2002). African American offenders (particularly males) may be more likely to recidivate due to the overrepresentation of minority offenders from economically and socially disadvantaged neighborhoods (Rose and Clear 1998), where feelings of resentment and hostility toward legal authority are pervasive among residents (Anderson 2001; Sampson and Bartusch 1998). If African American offenders do not hold much respect for the law or legal authority, deviance may be more likely. The inverse relationship between age and recidivism is well documented (Gendreau et al. 1996; Langan and Levin 2002; Mackenzie et al. 1999; Nagin et al. 2009; Rosenfeld, Wallman, and Fornango 2005); however, evidence regarding the effect of race on recidivism is mixed (DeJong 1997; Gendreau et al. 1996; Griffin and Armstrong 2003; Huebner, Varano, and Bynum 2007; Langan and Levin 2002; MacKenzie et al. 1999; MacKenzie and De Li 2002; Rosenfeld et al. 2005). Age, sex, and race have all been linked to longer prison sentences (e.g., Steffensmeier, Ulmer, and Kramer 1998).

Offenders convicted of more serious crimes and offenders who have longer prior records are more likely to be sentenced to longer periods of confinement (Nagin et al. 2009). Researchers have also uncovered that offenders’ committing offenses and prior record are important correlates of recidivism (Dejong 1997; Gendreau et al. 1996; Langan and Levin 2002; MacKenzie et al. 1999; MacKenzie and De Li 2002; Nagin et al. 2009; Visher and Courtney 2007). The salience of continuity in offending behavior is well documented (e.g., Gendreau et al. 1996; Laub and Sampson 2003; Sampson and Laub 1993). Regarding offense type, researchers have found that offenders convicted of drug or property offenses (as opposed to other offenses) are more likely to recidivate (Langan and Levin 2002; Rosenfeld et al. 2005; Solomon, Kachnowski, and Bhati 2005).

Other potentially important covariates may be offenders’ level of education, as well as whether they have been designated a sexual offender or gang member. Researchers have uncovered a link between defendants’ level of education and sentence severity (e.g., Wooldredge, Griffin, and Rauschenberg 2005). Offenders with higher levels of education may also be less likely to recidivate because they have more social capital and may have a greater commitment to conventional behavior (Clear 2008; Hirschi 1969; Lochner and Moretti 2004). Offenders designated as sexual offenders are generally perceived as more dangerous and therefore may be more likely to be sentenced to longer terms of imprisonment; these offenders may also have higher odds of recidivism upon release (Huebner and Bynum 2006; Levenson et al. 2007; Lin, Grattet, and Petersilia 2010). Many states have added sentencing enhancements for gang-related crimes (Esbensen et al. 2001). Gang membership may also increase the risk of recidivism among offenders because gangs can provide a reference group of antisocial peers (Huebner et al. 2007).

The type of institution in which offenders were confined along with the area in which they were released may also influence offenders’ odds of recidivism. Different types of facilities often hold inmates who have been sentenced to different periods of confinement (Irwin 2005). Confinement in a higher security prison may restrict opportunities for program participation as well as restrict social contact with the outside (Bales and Mears 2008; Mears and Bales 2009). For these reasons, along with the exposure to the harsher more sterile environment found in higher security prisons, offenders who are released from higher security institutions may also have higher odds of recidivism (Chen and Shapiro 2007; Mears and Bales 2009). Area of release is technically a covariate that is measured after offenders have served their time, however, offenders are typically released back to the community from which they were removed (Clear 2008). Thus, area of release may also proxy differences in the areas in which offenders lived prior to their imprisonment. Offenders’ area of residence has been linked to harsher sentences and higher odds of recidivism (Kubrin and Stewart 2006; Wooldredge 2007).

Method

The data used for this study were collected as a part of a larger project designed to evaluate the effects of a change to the Ohio Department of Rehabilitation and Correction (ODRC) parole violation sanction policy. The target population for the study included all the offenders released under postrelease supervision in Ohio during a three-month period before (October–December, 2003) and after (August–October, 2005) the policy was implemented statewide. The larger study revealed that the change to the policy had no effect on various measures of offender recidivism (see Martin and Van Dine 2008), and so the two samples were combined for the purposes of the current study.

Sample

The samples used for the larger study were each selected from a list of all the offenders released under postrelease supervision in Ohio during the periods mentioned above. Female offenders were selected with certainty, while male offenders were selected randomly with the goal of 95 percent confidence intervals for parameter estimates. Male offenders were also oversampled by 20 percent to account for unusable cases (e.g., interstate compacts), cases with missing data, and so forth. These procedures resulted in 1,040 and 1,012 offenders for the two samples, respectively, and a combined sample of 2,052 offenders. For analytical purposes, normalized weights were created to adjust for the oversampling of female offenders.

Information regarding each offender was collected from a number of official sources (e.g., case files), which were cross referenced against each other in order to increase the reliability of the data. Offenders were followed for a full year after their release or, if applicable, until the date they recidivated (as defined below). The decision to follow offenders for only one-year after their release was based on practical constraints imposed by the time required to collect the information necessary for the larger project, and because in Ohio the majority of offenders typically only serve one to one and a half years under postrelease supervision.²

From the sample of 2,052 offenders, cases were removed if they were released to detainers, interstate compact cases, or had missing data on the measures of recidivism or risk described below. These procedures reduced the sample used here to 1,989 offenders released under supervision in the state of Ohio. Comparisons between population parameters provided by the ODRC (age, race, gender, and committing offense type) and the corresponding sample statistics suggested that the sample was not significantly different from the target population (p ≤ .05).

Measures

All of the measures included in the final models are described in Table 1. The measures are described for the full sample and the samples of recidivists and nonrecidivists. Recidivism was measured with a dichotomous indicator of whether an offender was rearrested for a new felony offense during the study period. Rearrest was chosen over other measures (e.g., reconviction) in order to avoid problems associated with measures that require further procession into the criminal justice system (Maltz 1984). The measure was restricted to only arrests for new felonies in order to isolate the effect of time served on reoffending (i.e., some “arrests” occur for technical violations).³ Official measures of recidivism may underestimate offenders’ actual offending behavior (MacKenzie et al. 1999); however, official measures of recidivism have also been considered valid indicators of offender behavior (Farrall 2005). Nonetheless, the limitations of the measure used here should be kept in mind when interpreting the findings.

Table 1.

Description of Weighted Sample and Comparisons Between Recidivists and Non recidivists.

			Recidivists		Nonrecidivists
	Mean	SD	Mean	SD	Mean	SD	SBS^a
Outcome variable
Felony rearrest	.25	(.43)
Independent variables
Natural log of time served (in months)	3.21	(1.20)	2.89	(1.08)	3.32	(1.22)	−37.32
Age (in years)	30.25	(9.87)	28.33	(8.49)	30.89	(10.21)	−27.26
Female	.09	(.29)	.04	(.21)	.11	(.31)	−26.44
African American	.52	(.50)	.58	(.50)	.50	(.50)	16.00
High risk	.17	(.37)	.25	(.43)	.14	(.35)	28.06
Medium risk	.41	(.49)	.50	(.50)	.38	(.49)	24.24
Felony level (1–5)	3.25	(1.31)	2.98	(1.34)	3.34	(1.28)	−27.43
Sentenced for property offense	.22	(.42)	.28	(.45)	.20	(.40)	18.79
Sentenced for drug offense	.15	(.35)	.17	(.38)	.14	(.35)	8.21
Gang member	.16	(.36)	.22	(.41)	.14	(.34)	21.24
Sex offender	.19	(.39)	.14	(.35)	.21	(.41)	−18.36
GED	.24	(.43)	.22	(.42)	.25	(.43)	−7.08
≥high school diploma	.28	(.45)	.20	(.40)	.30	(.46)	−23.20
Institution of release > close security	.19	(.39)	.22	(.42)	.17	(.38)	12.55
Institution of release minimum security	.13	(.34)	.10	(.30)	.14	(.35)	−12.34
Akron region	.15	(.36)	.15	(.35)	.15	(.36)	−12.34
Cincinnati region	.22	(.41)	.23	(.42)	.21	(.41)	4.81
Columbus region	.14	(.35)	.13	(.34)	.14	(.35)	−2.92
Lima region	.10	(.30)	.07	(.26)	.11	(.31)	−13.88
Mansfield region	.08	(.28)	.08	(.27)	.08	(.27)	.00
Chillicothe region	.08	(.28)	.08	(.26)	.09	(.28)	−3.65
N =	1,989		472		1,517

Note. ^a Based on comparisons between the descriptive statistics for offenders rearrested for a felony (recidivists) and offenders not rearrested for a felony (nonrecidivists); standardized bias statistic (SBS) >20 indicates significant difference.

The primary independent variable of interest, time served, was measured with the number of months each offender served in prison. For analyses that involved the metric scale of this measure, the natural log of time served was used because the distribution was skewed right. The other covariates included in the analyses were age at commitment, sex (female), race (African American), risk level (high risk, medium risk, low risk), felony level of committing offense (range 1–5), committing offense type (sentenced for property offense, drug offense), official designation as a gang member, or sex offender, education (< than a GED/high school diploma, GED, > high school diploma), type of institution of release (institution of release minimum security, medium security, ≥ close security), and ecological region of release (Akron region, Cincinnati region, Columbus region, Lima region, Mansfield region, Chillicothe region).^4,5,6

Risk level was measured with the ODRC’s additive static risk assessment, which is primarily comprised of indicators of offenders’ criminal history (e.g., prior convictions). Gang membership was retrieved from ODRC prison records and indicates participation in a security threat group. Designation as a sex offender reflects whether an offender had ever been convicted of a sex offense. For the measures of risk level, education, institution of release, and region of release, low risk, < than a GED/high school diploma, institution of release medium security, and released to Cleveland region were treated as the reference categories in the analyses.

Statistical Analyses

Prior to the analyses, recidivists were compared to nonrecidivists using the standardized bias statistic (Rosenbaum and Rubin 1985). The results of these tests are contained in Table 1. Recidivists differed from nonrecidivists on the measures time served, age, female, high risk, medium risk, felony level, gang member, and ≥ high school diploma.

The majority of the extant studies of the time served–recidivism relationship have used multivariate regression to control for potentially confounding covariates (Nagin et al. 2009), and so we used multivariate logistic regression here.⁷ However, researchers have identified a number of problems associated with using multivariate regression techniques to estimate causal effects (see, e.g. Berk 2004; Guo and Fraser 2010; Nagin et al. 2009; Oakes and Johnson 2006), and so we also followed methods used by Loughran et al. (2009) in their study of the time served–recidivism relationship among juvenile delinquents (see also Joffe and Rosenbaum 1999; Zanutto, Lu, and Hornik 2005).⁸ First, we created discrete ordered categories of time served reflecting different dosages. These categories included the quintiles of the distribution of time served (≤6 months, 7–16 months, 17–32 months, 33–77 months, and ≥78 months) and an alternative categorization (≤12 months, 13–24 months, 25–36 months, 37–60 months, and ≥61 months). Bivariate analyses of the relationships between the covariates described in Table 1, and each of these time served categories revealed that 12 of the 20 covariates described in Table 1 were not balanced across the groups. Next, an ordinal logit model that included all of the covariates described in Table 1 was used to generate a balancing score (hereafter a propensity score) for each case in the sample. Cases were then subclassified into five equally sized strata based on the estimated propensity score (Rosenbaum and Rubin 1984). After stratifying the cases by propensity scores, the covariates were assessed for balance within the subclasses. In the event any of the covariates were still imbalanced, the propensity score was reestimated by adding interaction terms to the ordinal logit model until an acceptable level of balance was achieved. The final step in this portion of the analyses involved estimating treatment effects, along with stratum weighted odds of recidivism for each group defined by their dosage of time served (Yanovitzky, Zanuto, and Hornik 2005; Zanutto et al. 2005). The estimated odds of recidivism for each group were then used to graph dose–response curves (Loughran et al. 2009).⁹

Results

Based on the data used for this study, the median length of stay for offenders released from prisons under postrelease supervision in Ohio was approximately 24 months. In the year following their release, 25 percent of these offenders were rearrested for a new felony offense. Table 2 presents the results from the multivariate regression analysis of time served on felony rearrest. The coefficient for time served generated from the model is negative and statistically significant, indicating that individuals who served longer sentences in prison have lower odds of recidivism. The odds of recidivism among offenders who served the typical amount of time in prison (roughly 2 years) were .23, compared to .35 for offenders who served six months and .29 for offenders who served one year.¹⁰ Offenders who served 5 years in prison had a 17 percent chance of recidivating compared to 12 percent for offenders who served 10 years.

Table 2.

Logistic Regression of Felony Rearrest.

	Coef.	(SE)
Constant	.94	(.35)
Natural log of time served	−.44^**	(.07)
Age	−.04^**	(.01)
Female	−.89^**	(.25)
African American	.20	(.23)
High risk	1.29^**	(.17)
Medium risk	.73^**	(.13)
Felony level	.06	(.06)
Sentenced for property offense	.29^*	(.14)
Sentenced for drug offense	−.03	(.17)
Gang member	.20	(.15)
Sex offender	−.06	(.16)
GED	−.25	(.14)
≥High school diploma	−.37^**	(.14)
Institution of release ≥ close security	.11	(.14)
Institution of release minimum security	−.13	(.19)
Akron region	−.12	(.19)
Cincinnati region	−.10	(.16)
Columbus region	−.27	(.19)
Lima region	−.57^*	(.23)
Mansfield region	−.16	(.48)
Chillicothe region	−.20	(.24)
N =	1,989

Note. ^** p ≤ .01. ^* p ≤ .05

The results of the PSM analyses where cases were subclassified and matched across dosage categories of time served are contained in Table 3 and Figure 1. For the analysis involving dosage categories defined by the quintiles of the distribution of time served, we achieved balance across categories within each of the five strata for all but one of the covariates.¹¹ We were able to balance all of the covariates in the analysis involving the alternative dosage categories. Achieving this level of balance required the addition of three interaction terms to each of the ordinal logit models used to create the propensity scores. It is also worth noting that this level of balance is better than what would be expected had randomization been used (p ≤ .05).

Figure 1.

Estimated odds of felony rearrest by dosage level of time served. Estimated odds of felony rearrest (standard error in parentheses) at each dosage level after subclassifying on the estimated balancing score (± 1 standard error).

Table 3.

Estimated Dosage Effects of Time Served on Odds of Felony Rearrest.

Dosage (in months)	Group size	Group size (less time)	ATT (vs. less time)	(SE)	Group size (more time)	ATT (vs. more time)	(SE)
<6	395	—	—	—	1,594	−.042	(.06)
7–16	401	395	.003	(.14)	1,193	.007	(.08)
17–32	396	796	−.065	(.12)	797	.106	(.11)
33–77	400	1,192	−.082	(.07)	397	.107	(.12)
≥78	397	1,592	−.142	(.02)	—	—	—
≤12	723	—	—	—	1,266	−.014	(.06)
13–24	343	723	.005	(.08)	923	.079	(.06)
25–36	225	1,066	−.037	(.07)	698	.099	(.07)
37–60	232	1,291	−.073	(.05)	466	.054	(.05)
≥61	466	1,523	−.054	(.02)	—	—	—

Note. Estimated odds of felony rearrest (standard error in parentheses) at each dosage level after subclassifying on the estimated balancing score.

Table 3 contains the estimated average treatment effect for the treated (ATT) for each dosage category relative to the effects of serving less time and more time. The estimates of the ATT and corresponding standard errors were computed using the procedures outlined by Yanovitzky et al. (2005).¹² For the initial categorization of dosages (quintiles), serving between 7 and 16 months was the dosage category associated with the highest odds of recidivism. Offenders who served less than 7 months and offenders who served more than 16 months each had lower odds of recidivism than offenders who served between 7 and 16 months, although the treatment effects were relatively small (vs. less time = .003; vs. more time = .007). Offenders who served less than 7 months had lower odds of recidivism than offenders who served more time. Offenders within each of the dosage categories that involved serving more time than 16 months (17–32, 33–77, and ≥78) had lower odds of recidivism compared to offenders who served less time, but higher odds of recidivism compared to offenders who served more time (with the obvious exception of ≥78). However, if we consider that an ATT needs to be approximately 2 standard errors (SEs) above or below zero to constitute a significant difference between the dosage group and the comparison group, then only the ATT associated with serving at least 78 months (6.5 years) was a meaningful effect. Offenders who served at least six and a half years had lower odds of recidivism than offenders who served less time.

The estimates of the treatment effects for the alternative dosage categories revealed a similar pattern. The category of offenders who served between 13 and 24 months in prison had the highest odds of recidivism; offenders who served less time and offenders who served more time had lower odds of recidivism, although the treatment effects associated with these comparisons were relatively modest (vs. less time = .005; vs. more time = .079). Offenders who served less than 13 months had lower odds of recidivism than offenders who served more time. Offenders who received dosages of 25–36 and 37–60 months had higher odds of recidivism than offenders who served more time, but these offenders and those who served ≥61 months had lower odds of recidivism compared to offenders who served less time. Only the treatment effect associated with serving at least 61 months was a meaningful effect, however. Offenders who served over five years in prison had lower odds of recidivism than the offenders who served less time.

Figure 1 contains a dose–response curve that might illustrate these findings more clearly. The estimates reported in Figure 1 are the odds of recidivism for offenders who received each dosage of prison time based on the alternative categorization.¹³ The estimates were derived by calculating a weighted average of the odds of recidivism within each stratum for each dosage category. Standard errors were derived using the formula offered by Zanutto et al. (2005). Although this estimate of the SE is not unbiased, researchers have found it to be a reasonable approximation (e.g., Zanutto et al. 2005).

Figure 1 shows that the odds of recidivism increased as length of confinement increased to roughly two years. After two years, however, there was a reduction in the odds of recidivism as the amount of time served in prison increased. Examination of the SEs shows that most of these fluctuations were not significant. Only the odds of recidivism associated with serving at least five years in prison was significantly different from the odds associated with serving different lengths of time.

Discussion and Conclusions

Incarceration has several purposes (e.g., retribution, deterrence), and the findings from this study of the relationship between time served in prison and recidivism among a large sample of offenders released from prison under supervision in Ohio have theoretical and practical implications regarding the specific deterrent effect of length of imprisonment. The findings derived from the two analyses were consistent; there was an inverse relationship between time served and offenders’ odds of recidivism. On their face, these findings refute the prisons as criminogenic perspective. A closer examination of the findings from the analyses involving PSM for ordered doses, however, revealed that the magnitude and substantive interpretation of this finding may not unilaterally support the specific deterrent hypothesis either.

Based on the results of the regression analysis, the distance between the odds of recidivism for offenders who served six months and offenders who served the typical amount of time in prison (approximately two years) was about .12. The distance between the odds of recidivism for offenders who served the typical amount of time in prison and offenders who served five years was about .08. Thus, we can conclude that the change in the odds of recidivism associated with serving different amounts of time, while statistically significant, was not very large in magnitude.

The analyses involving PSM for ordered doses revealed that the time served–recidivism relationship did not conform to a perfectly linear dose–response curve. Offenders’ odds of recidivism decreased incrementally as the amount of time served increased beyond two years. However, only the dosage of serving at least five years was associated with a significant difference in offenders’ odds of recidivism. Thus, the analysis involving the PSM for ordered doses was more insightful because we were able to unpack the significant inverse relationship of time served on recidivism and determine more specifically what amount of time served was associated with a significant reduction in offenders’ odds of recidivism (e.g., > five years). Accordingly, we recommend using PSM for ordered doses when analyzing the dose–response relationship between time served in prison and recidivism with observational data (see also Loughran et al. 2009; Nagin et al. 2009).

The majority of extant studies of the time served–recidivism relationship have not found a relationship between length of confinement and offenders’ odds of recidivism (Nagin et al. 2009). In our study, we did observe an inverse relationship between time served and recidivism. It is possible that our findings are unique to the data examined in this study. It is also possible that our findings were due to the methods used in this study; most of the existing studies have relied on multivariate regression analysis. Our study also focused on adults as opposed to juveniles; it may be that longer stays in confinement deter adults, while the effect of different dosages of confinement is null for juveniles, who are still developing. An additional explanation, however, could be our choice of outcome. In this study, recidivism was defined as any arrest for a new felony, which excludes arrests for minor crimes and technical violations of parole. Other studies that have used new felonious behavior as the measure of recidivism have also revealed an inverse relationship between time served and recidivism (e.g., Bales and Meares 2008; Maguire, Flanagan, and Thornberry 1988). It could be that longer prison stays deter offenders from committing more serious crimes that might be more likely to receive a lengthy prison term. In other words, length of confinement may have little or no deterrent effects on offenders’ odds of committing minor crimes or technical violations because the odds of offenders receiving a lengthy prison sentence for these behaviors are not very great. In Ohio, for example, offenders cannot be sentenced to prison for misdemeanor offenses and offenders who commit a technical violation of the conditions of their release are only eligible to serve the remainder of their period of postrelease control (mandatory parole) in prison. An important direction for future research could be how time served affects different measures of recidivism among adult offenders released from prison.

The findings from the analysis involving PSM for ordered doses also raise an alternative explanation for the inverse relationship between time served and recidivism—an incapacitation effect. Recall that we did not observe a meaningful effect of time served on recidivism until offenders had served at least five years in prison. It could be that the offenders who received this dosage of imprisonment were incapacitated until they aged out of the peak years of offending. An incapacitation effect would be harder to detect in studies of juvenile delinquents (e.g., Loughran et al. 2009) because incarcerated juveniles typically serve less time than adults and juveniles released from confinement facilities are still entrenched in their crime-prone years. The use of PSM for ordered doses ensured that groups of offenders were equivalent (at least statistically) prior to imprisonment. However, the obvious link between time served and age ensures that adult offenders who receive different dosages of time served are not equivalent after they are released. This fact does not refute the inverse effects of time served observed here but simply raises the question whether the mechanism underlying the observed effect was deterrence or incapacitation and maturation. Thus, an equally important line of research could be an examination of the mechanisms by which the prison experience shapes offenders’ perceptions and impacts their future offending patterns (Nagin 1998; Paternoster 1987).

Finally, it is important to note that the measures used in this study were created with information from official sources. Even though attempts were made to increase the reliability of the measures by cross referencing the information across multiple sources, the information was potentially subject to some discretionary recording by parole officials. All inferences drawn from this study are, in part, dependent on the accuracy of the information contained within the official records. It is also worth reiterating that the data used for this study were collected for a larger project designed to examine the effects of a parole violation sanction policy. As such, the sample only included offenders released under supervision, which might restrict the generalizability of the findings to some extent. The decision to only follow offenders for one year after their release may also limit the generalizability of the findings. Although the majority of offenders released under supervision in Ohio are only monitored one to one and a half years, this may not be the case in other jurisdictions (e.g., indeterminate sentencing states). On the other hand, Langan and Levin (2002) observed that over 65 percent of all offenders who were rearrested within a three-year period were arrested in the first year of their release, and so the problem may be less of a concern.

Although the finding of a deterrent effect for longer stays in prison may be of theoretical import, the policy implications of this finding may be less significant. Our analyses revealed that the effects of lengthier terms are not realized until after an offender has served at least five years. It is also worth noting that the offenders within this group typically served between 10 and 11 years, and nearly half of the offenders in the group served more than 10 years. Thus, many of the offenders in the group who served at least five years would have aged out of the peak offending years by the time they were released, which suggests that our finding could, in large part, be attributed to the processes of incapacitation and maturation. Recall that although the groups of offenders we compared were equivalent when they were incarcerated, the group of offenders who served at least five years was, on average, much older than the other groups of offenders when they were released. In light of these findings, it may be worthwhile to examine the odds of recidivism among offenders who serve different amounts of time beyond five years. Such a study might shed light on whether related findings can be attributed to deterrence or incapacitation and maturation. Regardless of whether our findings were attributable to deterrence or incapacitation, however, the question that policymakers must ask is, “is it worth it?” The financial, individual, and social costs of incarcerating an individual for five years, 10 years, or longer are beyond the scope of this article (but see, e.g., Clear 2007). However, we can infer that individuals’ odds of recidivism decline by .10 to .15 when an offender serves five years or more compared to one to two years. These reductions must be considered alongside the costs of incarcerating an individual for the additional amount of time. On the other hand, it may very well be that some of the prison sentences between the typical sentence (∼ 2 years) and five years could be reduced without any meaningful compromise to public safety. Thus, a more cost-effective policy question may be, “how can we reduce the length of time most offenders serve without significantly compromising public safety?”

Footnotes

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Notes

References

Anderson

Elijah

. 2001. “Going Straight: The Story of a Young Inner-city Ex-convict.” Punishment and Society 3:135–52.

Bales

William D.

Mears

Daniel P.

. 2008. “Inmate Social Ties and the Transition to Society: Does Visitation Reduce Recidivism?” Journal of Research in Crime and Delinquency 45:287–321.

Becker

Gary S.

1968. “Crime and Punishment: An Economic Approach.” The Journal of Political Economy 76:169–217.

Berk

Richard A.

2004. Regression Analysis: A Constructive Critique. Thousand Oaks, CA:Sage.

Blumstein

Alfred

Beck

Allen J.

. 2005. “Reentry as a transient state between liberty and recommitment.” Pp. 50–79 in Prisoner Reentry and Crime in America, edited by Travis

Visher

. New York: Cambridge University Press.

Chen

M. Keith

Shapiro

Jesse M.

. 2007. “Do Harsher Prison Conditions Reduce Recidivism? A Discontinuity-based Approach.” American Law and Economics Review 9:1–30.

Clear

Todd R.

1994. Harm in American Penology: Offenders, Victims, and Their Communities. Albany: SUNY Press.

Clear

Todd R.

2007. Imprisoning Communities: How Mass Incarceration Makes Disadvantaged Neighborhoods Worse. New York, NY: Oxford University Press.

Clear

Todd R.

2008. “The Effect of High Imprisonment Rates on Communities.” Pp. 97-132 in Crime and Justice: A Review of Research, Vol 37, edited by Tonry

. Chicago, IL: University of Chicago Press.

10.

Dejong

Christina

1997. “Survival Analysis and Specific Deterrence: Integrating Theoretical and Empirical Models of Recidivism.” Criminology 35:561–75.

11.

Donohue, John J. 2007. Assessing the Relative Benefits of Incarceration: The Overall Change Over the Previous Decades and the Benefits on the Margin. New Haven, CT: Yale University.

12.

Esbensen

Finn-Aage

Thomas Winfree

Taylor

Terrance J.

. 2001. “Youth Gangs and Definitional Issues: When is a Gang a Gang, and Why does it Matter?” Crime and Delinquency 47:105–30.

13.

Farrall

Stephen

2005. “Officially Recorded Convictions for Probationers: The Relationship with Self-report and Supervisory Observations.” Legal and Criminological Psychology 10:121–31.

14.

Gendreau

Paul

Goggin

Claire

Cullen

Francis T.

. 1999. The Effects of Prison Sentences on Recidivism. Ottawa, ON: Solicitor General Canada.

15.

Gendreau

Paul

Little

Tracy

Goggin

Claire

. 1996. “A Meta-analysis of the Predictors of Adult Offender Recidivism: What Works?” Criminology 34:575–607.

16.

Glaze

Lauren E.

2010. Correctional Populations in the United States, 2009 (Bulletin) . Washington, DC: Bureau of Justice Statistics.

17.

Goodman

Richard A.

Mercy

James A.

Layde

Peter M.

Thacker

Stephen B.

. 1988. “Case-control Studies: Design Issues for Criminological Applications.” Journal of Quantitative Criminology 4:71–84.

18.

Gottfredson

Michael

Hirschi

Travis

. 1990. A General Theory of Crime. Stanford, CA: Stanford University Press.

19.

Grasmick

Harold G.

Bursik

Robert J.

. 1990. “Conscience, Significant Others, and Rational Choice: Extending the Deterrence Model.” Law and Society Review 24:837–61.

20.

Griffin

Marie L.

Armstrong

Gaylene S.

. 2003. “The Effect of Local Life Circumstances on Female Probationers' Offending.” Justice Quarterly 20:213–39.

21.

Guerino

Paul

Harrison

Paige M.

Sabol

William J.

. 2011. Prisoners in 2010. U.S. Dept. of Justice, Office of Justice Programs, Bureau of Justice Statistics, Washington, DC.

22.

Guo

Shenyang

Fraser

Mark W.

. 2010. Propensity Score Analysis: Statistical Methods and Applications. Thousand Oaks, CA: Sage.

23.

Hanushek

Eric A.

Jackson

John E.

. 1977. Statistical Methods for Social Scientists. New York: Academic Press.

24.

Heckman

James J.

2005. “The Scientific Model of Causality.” Sociological Methodology 35:1–97.

25.

Hirschi

Travis

. 1969. Causes of Delinquency. Berkeley: University of California Press.

26.

Hockenberry

Sarah

Sickmund

Melissa

Sladky

Anthony

. 2011. Juvenile Residential Facility Census, 2008: Selected Findings. Office of Juvenile Justice and Delinquency Prevention, Office of Justice Programs, U.S. Department of Justice, Washington, DC.

27.

Huebner

Beth M.

Bynum

Timothy S.

. 2006. “An Analysis of Parole Decision Making using a Sample of Sex Offenders: A Focal Concerns Perspective.” Criminology 44:961–91.

28.

Huebner

Beth M.

Varano

Sean P.

S.Bynum

Timothy

. 2007. “Gangs, Guns, and Drugs: Recidivism Among Serious, Young Offenders.” Criminology and Public Policy 6:187–222.

29.

Irwin

John

. 2005. The Warehouse Prison: Disposal of the New Dangerous Class. Los Angeles, CA: Roxbury Press.

30.

Irwin

John

Austin

James

. 2001. It's About Time: America's Imprisonment Binge. 3rd ed. Belmont, CA: Wadsworth.

31.

Irwin

John

Cressey

Donald

. 1962. “Thieves, Convicts, and the Inmate Culture.” Social Problems, 10:142–55.

32.

Joffe

Marshall M.

Rosenbaum

Paul R.

. 1999

“Invited Commentary: Propensity Scores.”

American Journal of Epidemiology 150:327–33.

33.

Johnson

Robert

2002. Hard Time: Understanding and Reforming the Prison. Belmont, CA: Wadsworth.

34.

Kubrin

Charis E.

Stewart

Eric A.

. 2006. “Predicting Who Reoffends: The Neglected Role of Neighborhood Context in Recidivism Studies.” Criminology 44:165–97.

35.

Langan

Patrick A.

1991. “America's Soaring Prison Population.” Science 251:1568–73.

36.

Langan

Patrick A.

Levin

David J.

. 2002. “Recidivism of Prisoners Released in 1994.” Washington, DC: Bureau of Justice Statistics.

37.

Laub

John

Sampson

Robert

. 2003. Shared Beginnings, Divergent Lives: Delinquent Boys to Age 70. Cambridge MA: Harvard University Press.

38.

Levenson

Jill S.

Brannon

Yolanda N.

Fortney

Timothy

Baker

Juanita

. 2007. “Public Perceptions about Sex Offenders and Community Protection Policies.” Analyses of Social Issues and Public Policy 7:1–25.

39.

Lin

Jeffrey

Grattet

Ryken

Petersilia

Joan

. 2010. “‘Back-end Sentencing’ and Reimprisonment: Individual, Organizational, and Community Predictors of Parole Sanctioning Decisions.” Criminology 48:759–95.

40.

Lochner

Lance

Moretti

Enrico

. 2004. “The Effect of Education on Crime: Evidence from Prison Inmates, Arrests, and Self-reports.” The American Economic Review 94:155–89.

41.

Loftin

Colin

McDowall

David

. 1988. “The Analysis of Case-control Studies in Criminology.” Journal of Quantitative Criminology 4:85–98.

42.

Loughran

Thomas.A.

Mulvey

Edward P.

Schubert

Carol A.

Fagan

Jeffrey

Piquero

Alex R.

Losoya

Sandra H.

. 2009. “Estimating a Dose-response Relationship between Length of Stay and Future Recidivism in Serious Juvenile Offenders.” Criminology 47:699–739.

43.

MacKenzie

Doris Layton

Browning

Katherine

Skroban

Stacy B.

Smith

Douglas A.

. 1999. “The Impact of Probation on the Criminal Activities of Offenders.” Journal of Research in Crime and Delinquency 36:423–53.

44.

MacKenzie

Doris Layton

Spencer De

. 2002. “The Impact of Formal and Informal Social Controls on the Criminal Activities of Probationers.” Journal of Research in Crime and Delinquency 36:423–53.

45.

Maguire

Kathleen E.

Flanagan

Timothy J.

Thornberry

Terence P.

. 1988. “Prison Labor and Recidivism.” Journal of Quantitative Criminology 4:3–18.

46.

Maltz

Michael D.

1984. Recidivism. Orlando, FL: Academic Press.

47.

Martin

Brian

Dine

Steve Van

. 2008. Examining the Impact of Ohio’s Progressive Sanction Grid. Washington, DC: National Institute of Justice Award, Office of Justice Programs.

48.

Mears

Daniel P.

Bales

William D.

. 2009. “Supermax Incarceration and Recidivism”. Criminology, 47:1131–66.

49.

Nagin

Daniel

. 1998. “Criminal Deterrence Research at the Outset of the Twenty-first Century.” Pp. 1–42 in Crime and Justice: A Review of Research, Vol. 23, edited by Tonry

. Chicago, IL: University of Chicago Press.

50.

Nagin

Daniel

Cullen

Francis T.

Jonson

Cheryl Lero

. 2009. “Imprisonment and Reoffending.” Pp. 115–200 in Crime and Justice: A Review of Research, Vol. 38, edited by Tonry

. Chicago, IL: University of Chicago Press.

51.

Oakes

J. Michael

Johnson

Pamela Jo

. 2006. “Propensity Score Matching for Social Epidemiology”. Pp. 364–86 in Methods in Social Epidemiology, edited by Oakes

J.M.

Kaufman

Jay S.

. San Francisco, CA: Jossey-Bass/Wiley.

52.

Paternoster

Raymond

. 1987. “The Deterrent Effect of the Perceived Certainty and Severity of Punishment: A Review of the Evidence and Issues.” Justice Quarterly 4:173–217.

53.

Piquero

Alex R.

Alfred

Blumstein,

. (2007). “Does Incapacitation Reduce Crime?” Journal of Quantitative Criminology 23:267–86.

54.

Pew Center for the States. 2011. State of Recidivism: The Revolving Door of America’s Prisons. Washington, DC: The Pew Charitable Trusts.

55.

Rose, Dina R. and Todd R. Clear. 1998. “Incarceration, Social Capital and Crime: Implications for Social Disorganization Theory.” Criminology 36:441-79.

56.

Rosenbaum

Paul R.

Rubin

Donald B.

. 1984. “Reducing Bias in Observational Studies using Subclassification on the Propensity Score.” Journal of the American Statistical Association 78:516–24.

57.

Rosenbaum

Paul R.

Rubin

Donald B.

. 1985. “Constructing a Control Group using Multivariate Matched Sampling Methods that Incorporate the Propensity Score.” American Statistician 39:33–38.

58.

Rosenfeld

Richard

Wallman

Joel

Fornango

Robert

. 2005. “The Contribution of Ex-prisoners to Crime Rates.” Pp. 80–104 in Prisoner Reentry and Crime in America edited by Travis

Visher

. New York: Cambridge University Press.

59.

Sampson

Robert J.

Laub

John H.

. 1993. Crime in the Making: Pathways and Turning Points through life. Cambridge: Harvard University Press.

60.

Sampson

Robert J.

Bartusch

Dawn Jeglum

. 1998. “Legal Cynicism and (subcultural?) Toleration of Deviance: The Neighborhood Context of Racial Differences.” Law and Society Review 32:777–804.

61.

Solomon

Amy L.

Kachnowski

Vera

Bhati

Avinash

. 2005. Does Parole Work? Analyzing the Impact of Postprison Supervision on Rearrest Outcomes. Washington, DC: The Urban Institute.

62.

Steffensmeier

Darrell

Ulmer

Jeffrey

Kramer

John

. 1998. “The Interaction of Race, Gender, and Age in Criminal Sentencing: The Punishment Cost of Being Young, Black, and Male.” Criminology 36:763–97.

63.

Sykes

Gresham M.

1958. The Society of Captives. Princeton, NJ: Princeton University Press.

64.

Toch

Hans

Adams

Kenneth

Douglas Grant

. 1989. Coping: Maladaptation in Prisons. New Brunswick, NJ: Transaction.

65.

Visher

Christy A.

Courtney

Shannon M. E.

. 2007. One Year Out: Experiences of Prisoners Returning to Cleveland (Returning Home Policy Brief). Washington, DC: The Urban Institute.

66.

Wooldredge

John

. 2007. “Neighborhood Effects on Felony Sentencing.” Journal of Research in Crime and Delinquency 44:238–63.

67.

Wooldredge

John

. 2009. “Short- versus Long-term Effects of Ohio’s Switch to More Structured Sentencing on Extra-legal Disparities in Prison Sentences in an Urban Court.” Criminology and Public Policy 8:285–312.

68.

Wooldredge

Griffin

Timothy

Rauschenberg

Fritz

. 2005. “Sentencing Reform and Reductions in the Disparate Treatment of Felony Defendants.” Law and Society Review 39:835–74.

69.

Yanovitzky

Itzhak

Zanuto

Elaine

Hornik

Robert

. 2005. “Estimating Causal Effects of Public Health Education Campaigns using Propensity Score Methodology.” Evaluation and Program Planning 28:209–20.

70.

Zamble

Edward

Porporino

Frank J.

. 1988. Coping, Behavior, and Adaptation in Prison Inmates. New York: Springer-Verlag.

71.

Zanutto

Elaine

Hornik

Robert

. 2005. “Using Propensity Score Subclassification for Multiple Treatment Doses to Evaluate a National Antidrug Media Campaign.” Journal of Educational and Behavioral Statistics 30:59–73.

72.

Zimring

Franklin E.

Hawkins

Gordon

. 1995. Incapacitation: Penal Confinement and the Restraint of Crime. New York: Oxford University Press.