Short- and long-term effects of imprisonment on future felony convictions and prison admissions

Fig. 1 shows the probability of ever having been admitted to prison over time since release by race for three related measures of future imprisonment: admission for any reason, admission for a technical violation, and admission for a new sentence (arising from conviction for a new felony). It compares outcomes among those sentenced to prison with our comparison group, those sentenced to probation, showing much higher rates of all three imprisonment outcomes among prisoners compared with probationers for both racial groups at all time periods. Imprisonment for technical violations among prisoners is concentrated in the first 2 y after release, whereas imprisonment for new sentences increases gradually over the 5-y period for both prisoners and probationers. In addition, comparing the rates of prison admissions for technical violations to admissions for new sentences shows that technical violations are a larger component of prison admissions for former prisoners than for probationers at almost all time points. Among probationers, prison admissions for new sentences are more common than prison admissions for technical violations.

We leverage the natural experiment based on the random assignment of judges to examine the degree to which such prisoner–probationer differences in recidivism are causal effects of a prison sentence rather than preexisting differences between those sentenced to prison vs. probation. Table 1 shows instrumental-variables (IV) estimates of the effects of prison compared with probation on future admissions to prison and new felony convictions, measured at 1, 3, and 5 y after release (comparable ordinary least-squares estimates are provided in SI Appendix, Table S5; full model results are provided in SI Appendix, Table S6 A–D). Being sentenced to prison rather than probation increases the probability of future imprisonment quite dramatically for both racial groups at all follow-up times. In the first year since release, the effect is 8 percentage points among nonwhites and 10 percentage points among whites. These effects grow considerably at 3 y since release, to 18 percentage points among nonwhites and 19 percentage points among whites, and eventually reach 20 percentage points for both racial groups at 5 y since release.

Table 1 also shows that these effects are driven primarily by imprisonment for technical violations of community supervision rather than sentences for new felony crimes. (At longer follow-up times, it is possible for an individual to experience both types of prison admissions, which explains why the any admission effect is not simply the sum of the effects for the two admission types.) Whereas the effect of a prior prison sentence on being sentenced to prison in the future for a technical violation is significant at all time periods for both racial groups, reaching 18 percentage points for nonwhites and 14 percentage points for whites after 5 y, the effect on future imprisonment for a new felony is smaller (and not always significant), 4 percentage points for nonwhites and 9 percentage points for whites after 5 y. None of the differences by race in the estimates in Table 1 is statistically significant. Collectively, these results present strong evidence that prison’s revolving door is a causal effect of imprisonment rather than an artifact of preexisting differences between prisoners and probationers and that imprisonment for technical violations is a key mechanism driving this effect.

One explanation for why technical parole violations are the main pathway through which people sentenced to prison are readmitted after their release is that parole supervision is more intensive than probation supervision and it increases the detection and punishment of low-level offending or violation behavior. An alternative explanation is that some prosecutors may decide not to charge individuals on parole for low-level felony crimes because such individuals can be reimprisoned more easily for a technical violation. If we can assume that prosecutors would almost always prosecute medium- or high-severity felonies but that they might not prosecute low-level felonies when the suspect is already being sent back to prison for a technical violation, we can test this hypothesis by analyzing the effects of imprisonment on new felony convictions by the severity of the new crime. Under this hypothesis, we would expect prison sentences to have negative effects on less severe felonies (if prosecutors refrain from charging low-level felonies when the person could be returned to prison for a technical violation) and null or positive prison effects on more severe felonies.

We created categories of felony severity based on the maximum prison sentence for the specific crime committed: low-severity felonies are those with maximum sentences of 0–48 mo, medium are 49–72 mo, and high severity are 73 or more months. (Because the maximum sentence is set by law by the Michigan state legislature, this measure of felony severity reflects the assessment of the state legislature. We have explored whether results vary with the definitions of the severity categories. Results with categories defined at 0–36, 37–60, and 61+ are very similar.) We examine convictions for any felony, medium- or high-severity felonies, or high-severity felonies only. (The medium–high-severity felony variable is coded 1 for medium- or high-severity felonies and zero for low-severity felonies or those that were not convicted of any felony. The high-severity felony variable is coded 1 for high-severity felonies and coded zero for medium- and low-severity felonies and those with no felony convictions.) Fig. 2 shows descriptive statistics on new felony outcomes by race and time since release. The differences between prisoners and probationers on felony convictions are smaller than the differences for imprisonment but still sizeable for both racial groups.

Results of our IV analysis of prison sentences’ effects on future felony convictions appear in the bottom panel of Table 1. Being sentenced to prison rather than probation had no effect on conviction for any new felony (the general measure that does not specify severity) for either racial group at any of the time periods after release. Our severity-specific felony conviction analysis shows that prison reduces the probability of having a medium–high- or high-severity felony, but only for nonwhites and at 3 or 5 y after release from prison. Because there is a slight negative effect on medium–high-severity felonies and a smaller negative effect on felonies of any severity (which is not statistically different from zero) at both 3 and 5 y, we can also infer by differencing these two estimates that there is a slight positive effect of prison on low-severity felonies after release. In sum, the results of our postrelease felony conviction analysis lead us to conclude that prison has no effect on serious criminal offending among whites and small effects among nonwhites, decreasing slightly both medium- and high-severity felonies (while increasing slightly low-severity felonies).

Considering together the results of our analysis of both postrelease prison admissions and felony convictions, we found that the strongest relationship between prison sentences and recidivism is their positive effects on future prison admissions due to technical violations. Moreover, since we see null and negative effects of prison on medium- and high-severity felonies and a slight positive effect of prison on low-severity felonies, our findings do not support the hypothesis that the positive effects on technical violations are a by-product of decisions not to prosecute low-level felonies among former prisoners on parole. Finally, it is also striking in Table 1 that being sentenced to prison increases one’s probability of a prison admission for a new felony while simultaneously decreasing or having no effect on the probability of a conviction for a new felony. We suspect this reflects the role of previous imprisonment in sentencing decisions, an additional aspect of prison’s revolving door.

Thus far, we have presented results for outcomes measured postrelease. Such outcomes remove any effect of a prison sentence on future crime and imprisonment due to incapacitation during the initial prison sentence, one of the primary ways prison is theorized to affect recidivism. To examine incapacitation effects, we consider outcomes measured since the sentence date, thus building in any incapacitation effect of imprisonment. To gauge the degree to which postsentence outcome measures are the product of incapacitation, we must consider the time spent in prison among those sentenced to prison. The median time in prison among prisoners in this sample is 28 mo for both nonwhites and whites (the mean is 38 mo for both groups). At 1 y after sentence, 86% of nonwhite with a prison sentence and 87% of whites are still incarcerated on the initial prison sentence. At 3 y after sentence, these figures are 48.9% (nonwhites) and 45.6% (whites). At 5 y, these figures are 32.5% (nonwhites) and 27.5% (whites).

Table 2 shows IV effect estimates by race and time since the original sentence (for full model results, see SI Appendix, Tables S7 A–D; descriptive statistics on outcomes are in SI Appendix, Table S1 and Figs. S2 and S3). This analysis counts time that people sentenced to prison spent incarcerated, so it includes incapacitation effects, especially in the earlier time periods, before most individuals were released from prison. The estimated effects of imprisonment reported in Table 2 are more negative compared with the results in Table 1. In the first year after sentence, there are no effects among whites and only small effects among nonwhites, for whom a prison sentence reduces the probability of a prison admission for a new sentence by about 2 percentage points. These estimates reflect the incapacitation of prisoners and the low rate of prison admission among probationers (3%). Three years after sentence, the effect of a prison sentence on future admission to prison is not statistically different from zero among nonwhites, but this reflects counterbalancing positive effects on new admissions for technical violations (an approximately 6-percentage point increase) and negative effects on admissions for new sentences (4-percentage point decrease). Among whites, there is an approximately 7-point effect on prison admissions, which is entirely due to an effect on admissions for technical violations. At 5 y after the sentence, any incapacitation effects of the initial prison sentence have been swamped by greater prison admissions among those initially sentenced to prison. Both nonwhites and whites show positive effects of prison vs. probation on future prison admissions (10 percentage points among nonwhites and 14 percentage points among whites). These effects appear to be entirely due to admissions for a technical violation among nonwhites, and primarily due to technical violations among whites, with a 12-percentage point effect on admissions for technical violations and 4-percentage point effect on admissions for new sentences. Only the 3- and 5-y estimates for prison admissions for new sentences are statistically different by race. These results provide further evidence of the power of prison’s revolving door: sentencing an individual to prison rather than probation increases his or her probability of entering prison again at some point within 5 y of the sentence by 10–14 percentage points, swamping the short-term negative effects of incapacitation.

The bottom panel of Table 2 shows after-sentence results for new felony convictions. Effect estimates for 1 y after sentence reflect entirely the incapacitation effect of imprisonment because the probability of a prisoner from either racial group being convicted of a new felony in the first year after sentence is effectively zero (SI Appendix, Table S1). In other words, these effects are driven by the new felony behavior of probationers. These estimates show an 8-percentage point reduction in the probability of being convicted of any new felony among nonwhites and a 5-percentage point reduction among whites. This includes a 5-percentage point reduction in medium- and high-severity felonies and 2-percentage point reduction in high-severity felonies among nonwhites, with 4- and 2-percentage point reductions, respectively, among whites. The prevention of new felonies from a prison sentence is largest at 3 y at 19 percentage points among nonwhites (12 for medium–high and 6 for high severity) and 10 percentage points among whites (6 for medium–high and 2 for high severity). By 5 y after sentence, the preventative effect of prison falls again, to 15 percentage points for nonwhites (11 for medium–high and 5 for high severity) and 9 percentage points for whites (5 for medium–high and zero for high severity). In sum, our comparison of convictions for new felonies between prisoners and probationers in the years after their sentences leads us to conclude that there is an important incapacitation effect of prison on felony offending, but such effects are modest in size. It is certainly not the case that each individual imprisoned would have committed a felony had he or she remained in the community.

The increase in incarceration in the late 20th and early 21st centuries in the United States raises important questions about both the causes of the rise and its consequences for the prevention or generation of serious criminal behavior. This study has leveraged a natural experiment using the random assignment of judges to criminal cases in Michigan from 2003 to 2006, with individuals followed through September 2013. Because judges vary in their penchant for sentencing offenders to prison rather than probation, our research design allows us to isolate variation in prison vs. probation sentences that is independent of offender characteristics, both observed and unobserved, and thereby generate causal estimates of the effect of being sentenced to prison on future prison admissions and felony convictions.

With regard to the rise in incarceration, our results demonstrate that imprisonment leads to future imprisonment. In other words, prison’s figurative revolving door has real causal force, rather than being the simple consequence of imprisonment of individuals at higher risk for future offending. For example, being sentenced to prison rather than probation increases the probability of a future prison admission within 3 y after release by 18–19 percentage points. These results imply that the rise in incarceration was to some degree self-generating, as imprisonment creates more imprisonment. Although the estimated effect sizes are not different by race, substantially higher rates of imprisonment among nonwhites mean that the consequences of these effects in the aggregate are much more severe for racial minorities, particularly African-Americans, who are imprisoned at over five times the rate as whites nationwide (20).

Our results also demonstrate that the majority of this reimprisonment effect is generated by parole violations rather than prison admissions for new felony convictions. Postprison parole supervision surveilles and punishes, and in so doing increases incarceration. In other words, the rise in incarceration was in part a self-perpetuating process resulting from the workings of the criminal justice system itself. The most common reasons for parole violations in Michigan include failure to report (included in 73% of violation reports), failure to complete required programming (61%), substance abuse (50%), moving residence without permission including absconding (50%), curfew violations or entering restricted areas (49%), associating with felons (29%), and possession of weapons other than firearms (29%). However, those most associated with revocation to prison are moving residences/absconding, nonfirearms weapons possession, failure to maintain registration as a sex offender, substance abuse, and driving without permission (21).

With regard to the crime preventive or criminogenic effects of imprisonment, our results demonstrate that there is a moderate incapacitation effect of imprisonment while an offender is in prison. For instance, a prison sentence reduces the probability of being convicted of a new felony by 5–8 percentage points in the first year after sentence. The reduction for medium–high-severity felonies is 4–5 percentage points and for high-severity felonies, the reduction is about 2 percentage points. These effects rise at 3 y after release and then fall slightly at 5 y after release. However, our results find only small effects of imprisonment on new felony convictions after the prisoner has returned to the community (at most a reduction of 3–5 percentage points, although such effects are concentrated among high-severity felonies). We find no evidence that imprisonment increases overall criminal behavior after release, at least when criminal behavior is measured as convictions for any new felony. Finally, it is possible that some of the crime-preventative effects of the initial prison sentence on more severe felonies could be due to incapacitation effects of the type documented in Table 1 occurring during subsequent imprisonment. However, this cannot be the entire explanation. Note that if future prison admissions were creating a large incapacitation effect on new felonies, we would expect the new felony effect sizes to increase rather than decrease between years 3 and 5 as more and more prisoners were returned to prison.

Our results have important policy implications. First, probation sentences might be employed more frequently as an alternative to incarceration. The cost savings associated with probation is large relative to the incapacitation effect of imprisonment. In addition, a prison sentence does little to reduce criminal offending after release relative to offending by probationers. Second, because parole violations play a substantial role in the growth of the prison system through “back-end” sentencing, diverting technical parole violators from reimprisonment to alternative sanctions may be an important mechanism for shrinking the size of the prison system. However, more research is needed on which alternative sanctions are likely to be most effective at reducing offending and assisting ex-offenders in escaping involvement in the criminal justice system. Such a policy recommendation also depends critically on whether individuals returned to prison on parole violation would have engaged in serious offending had they remained in the community, a hypothesis that warrants further scrutiny.

We alert the reader to the limitations of this study. First, our results come from a single state and may not be generalizable to other states with different criminal justice policies and practices. The importance of returns to prison for parole violations is likely a direct consequence of the relative rates of parole and probation revocations in Michigan. Compared with state corrections practices across the nation in the same time period, Michigan’s rates of probation revocations to incarceration were slightly lower than national averages while its rates of parole revocations to prison were much lower than national averages. In 2007, Michigan’s probation revocation rate was 0.5% while the national average was 2%, and Michigan’s parole revocation rate was 6.6% while the national average was 11.4%. [These values were calculated from Bureau of Justice Statistics counts for Michigan and “All States” (23). Denominators are total number of individuals on parole or probation on January 1, 2007, plus all individuals entering that year (tables 2 and 3 in ref. 23). Parole numerator is revocations to prison without new sentence (table 7 in ref. 23). Probation numerator is incarcerations under current sentence (table S3 in ref. 23).] Thus, our findings regarding the sizeable impact of technical revocations on reimprisonment were found in a state with parole revocations rates that were lower than the national average. The estimated revolving-door effect estimated here might well be small relative to what would be found nationwide. Furthermore, our results come from a particular historical period when rates of reimprisonment for technical violations of parole were high, especially among African-Americans (22).

Second, we are unable to examine racial groups other than whites and blacks due to low numbers of Latinos and Asians in Michigan. Third, we can only examine criminal offending that is both known to law enforcement and results in prosecution and conviction. In addition, we focus on criminal offending that occurs in the community, the explicit focus of most studies of incapacitation. Our data do not allow for reliable measurement of crimes committed in prison, which typically only result in arrest or prosecution for the most serious crimes. Most criminal activity in prison is handled through internal disciplinary procedures. Fourth, our data do not currently allow for long-term follow-up.

Fifth, causal effect estimates from an IV analysis are “local average treatment effects.” This means we are estimating the effect of incarceration in prison compared with probation among individuals for whom the randomly assigned judge made the difference between prison and probation (SI Appendix, Methods). Those are individuals who are on the margin between prison and probation. Our estimates do not provide average treatment effects for all individuals sentenced to prison in Michigan. As a result, they should not be interpreted as informative regarding radical policy changes such as decarceration on a massive scale, which would surely involve individuals who are far from the margin at which the effects in this paper are estimated.

We thank Charley Chilcote and Paulette Hatchett at the Michigan Department of Corrections for facilitating access to the data and for advice on their use. This study was funded by a grant from the National Science Foundation (SES1061018), with additional support from center grants from the Eunice Kennedy Shriver National Institute of Child Health and Human Development to the Population Studies Centers at the University of Michigan (R24 HD041028) and at University of California, Berkeley (R24 HD073964).