Classifying Prisoner Returns: A Research Note

Transcription

1 Research Note Classifying Prisoner Returns: A Research Note Justice Research and Policy 2016, Vol. 17(1) ª The Author(s) 2016 Reprints and permission: sagepub.com/journalspermissions.nav DOI: / journals.sagepub.com/home/jrx Gerald G. Gaes 1, Jeremy Luallen 2, William Rhodes 2, and Jared Edgerton 2 Abstract Scholars often use administrative corrections data to identify the reasons that offenders return to prison, though such data usually obscure more complex processes underlying the cause of a return. This article describes the procedural nuances that make it difficult to record prison return paths and discusses these limitations. We focus on data elements and recording practices commonly found in administrative databases and discuss whether and how researchers may use these data to reliably identify/classify returns. We provide empirical demonstrations of these arguments using publicly available data and conclude that more extensive data are often needed to accomplish this objective. Keywords prisons, recidivism, return, classification, revocation A common practice in criminal justice evaluation is to measure recidivism as a return to prison. Scholars also expand upon this definition by further decomposing the type of return. This expansion most often employs a typology that distinguishes new court commitments from revocation events, where revocation events themselves are further broken down into finer groupings: (a) revocations that occur as the result of a technical violation of the conditions of supervision or (b) revocations that result from an arrest for a new crime. Although this taxonomy is conceptually straightforward, many administrative databases do not contain a sufficient amount of information to classify these return paths. This limits the policy insights that can be derived from studies of prison programs, reentry services, community supervision effectiveness, or other criminal justice interventions that use prisoner returns as the key outcome measure. 1 1 Florida State University, Tallahassee, FL, USA 2 ABT Associates, Cambridge, MA, USA Corresponding Author: Jeremy Luallen, ABT Associates, 55 Wheeler St., Cambridge, MA 02138, USA. jeremy_luallen@abtassoc.com

2 Gaes et al. 49 In this article, we lay out a framework for thinking through these issues in an effort to bring clarity to the subject. Our motivation comes from our work with the National Corrections Reporting Program (NCRP), a Bureau of Justice Statistics (BJS) data collection of individual records of prison and community supervision admissions and releases. These records and other sources of BJS data have been prominently featured in the literature on reentry and community reintegration (e.g., Mears & Cochran, 2015; Petersilia, 2003; Petersilia et al., 2008; Travis, 2005; Travis & Visher, 2005). Our work with these data led us into an investigation of the criminal justice processes that affect the prison return paths and we thought it was important to document these processes for analysts who may use NCRP or similar sources of data in the future. We are also struck that the conventional wisdom on returns a very high proportion are violations also motivates the importance of the discussion. Much of the conventional wisdom derives from scholarly exegesis of prison returns using available national estimates of return types; however, proper interpretation of this assertion requires that one understand that a large proportion of violations stem from an arrest. Scholars such as Burke and Tonry (2006), Mears and Cochran (2015), Petersilia (2003), Travis (2005), and Travis and Visher (2005) have relied on BJS sources that for the most part do not distinguish revocation returns based on an arrest as opposed to a purely technical violation. As we show later in this article, revocations often result from a combination of purely technical violations and arrests. We return to this problem later in the article. Most important to this discussion is that accurate and reliable return classifications have important implications for policy analysis. Knowing the type of return to prison is an important dimension when conducting analysis of interventions intended to reduce recidivism, whether those interventions occur prior to release or while the person is under postrelease supervision. Any modifications to postrelease supervision policy would be informed by accurate return classification. These types of research questions motivated the two examples we provide later in this article in which researchers had to gather supplemental administrative data to make sense of the classification of prison returns (Grattet, Petersilia, & Lin, 2008; Jones, 2016). Even the seemingly mundane task of recording time served will depend upon an analyst s ability to separate prison terms initiated for violations of supervision (typically much shorter) than those resulting from new criminal actions. Though administrative databases often contain variables meant to distinguish the reasons for returning to prison, such measures often discount or obscure the complex processes underlying the cause(s) of a return. 2 Simply put, databases are not designed with the intricacies of identification in mind. Nevertheless, researchers must use these data to inform their investigations, facing few (readily available) alternatives. In this article, we provide a detailed discussion of the procedural nuances that make it difficult to reliably record the prison return paths and by implication make it more difficult to inform postprison supervision policy. Because this article is mainly intended for researchers using administrative data, we focus on data elements and recording practices commonly found in administrative databases. We discuss whether and how researchers may use these data to reliably classify the type of return to prison. We also provide empirical demonstrations of these ideas using publicly available data from the NCRP. 3

3 50 Justice Research and Policy 17(1) Overall, the purpose of this article is to provoke thought on these issues and to challenge analysts to be circumspect in how they approach these data. We hope to encourage researchers to recognize limitations in administrative databases that can lead to future improvements. The remainder of this article is organized as follows. First, we define terms. Then we describe the revocation process and specify its complexity. Next, using NCRP, we demonstrate some of the problems these complexities create. We follow this demonstration with a discussion of research that has successfully disentangled the reason for returns and conclude with a summary of the issues. Prison Return Definitions Part of the confusion in the scholarly literature on prison returns are the nuances in definitions and the changing structure of community supervision. Parole supervision has been replaced by many different forms of community supervision for an offender serving a community supervision term after his or her term of prison (Petersilia, 2003; Piehl & LoBuglio, 2005; Stemen & Rengifo, 2012). We have adopted the more generic terminology postconfinement community supervision (PCCS) to include parole, mandatory release supervision, and even a split sentence that includes both a term of prison and a subsequent term of probation. Throughout this article, we will use many of these terms because that is the way they are referenced in community supervision scholarship. National estimates of postprison community supervision only make sense if all forms of postrelease supervision are included. A second point of confusion is terminology associated with the revocation returns. We often see in the literature a distinction between new court commitments and violations. Even if someone is under community supervision, if he or she commits a crime and is sentenced to prison for that crime, scholars make the distinction between this new conviction return and community supervision revocations. However, the latter can include both violations of community supervision from either a new unprosecuted arrest, or what is often referred to a purely technical violation such as a failed drug test, missed appointment, or failure to attend rehabilitation services that were a condition of supervision. The confusion promulgated in the literature is that the technical violation can be based on a new arrest, a pure technical violation, or both. We suspect this confusion misleads some naive readers to think that technical violators have failed a drug test, missed an appointment, or did not show up for their rehabilitation assignment, when in fact they were arrested for a new crime. To reduce this confusion, we adopt the following terminology: Revocation violation or violation the more general term for a return to prison based on a purely technical violation, an arrest, or both. Purely technical violation a community supervision breach based on a violation of a condition of supervision that does not include a new arrest, prosecuted, or unprosecuted. New court commitment a return on a new sentence that can occur whether or not a person is under postrelease community supervision.

4 Gaes et al. 51 Figure 1. Flow of release and revocation decisions for offenders under postconfinement community supervision. We next turn to the complications arising from the revocation process itself. Classification Challenges for Prison Returns Complexities in the Revocation Process Ambiguity in interpreting prisoner returns is born out of complexities inherent to the revocation process. To understand how, first consider, as a starting point, an idealized version of criminal justice processing around revocation proceedings shown in Figure 1. Though it is a reduced-form figure, its elements require explanation. Figure 1 begins in block (1) with an offender that has been sentenced for a crime, completed at least part of their prison term and is about to be released. Movement to (2) indicates that a community corrections officer has detected misconduct, either a new crime or a purely technical violation of supervision conditions. If a new crime has occurred, the offender is detained pending a revocation hearing as shown in (3). After the revocation hearing in (4), an offender might be released in (5) or may be returned to prison to serve additional time in (6). If a new crime led to a conviction, then the

5 52 Justice Research and Policy 17(1) return would be based on a new conviction; however, a conviction may also trigger a violation of the prior crime so that the offender is serving both a new commitment and a violation term concurrently or consecutively. 4 Importantly, the heavily shaded blocks (1), (3), and (6) identify times when an offender might be observed in prison administrative records. For offenders who are not released to some form of PCCS, returns are convictions for new crimes avoiding the revocation paths. Figure 1 represents how researchers generally wish to describe the elements of the revocation process when classifying returns, separating revocations from new convictions; however, it is overly simplistic and diverts attention from important details. By contrast, consider the more detailed depictions of revocation processes in North Carolina as shown in Figures 2 and 3. Figure 2 shows the flow of events for revocations based on new crimes and Figure 3 shows the flow of revocations based on purely technical violations. Each figure shows the event sequence, hearings that occur during the process and the likely place of confinement during each event. While the specifics of process timing, criminal justice agents, place of confinement, and rules for process and sanctions vary from state to state, the essential components remain the same across states. Both figures also show that due process requires preliminary and final revocation hearings. Most offenders waive their rights to preliminary hearings when a pending revocation is based on allegations of a new crime because this will shorten their prison stay. One of the many important takeaways from these diagrams is that when there is an allegation of a new crime, depending on the offense, and depending on the offender s willingness to waive preliminary hearings, either a concurrent or consecutive sentence revocation term will be added to any new term. We return to the importance of this distinction later on in the article Recording Admission Reasons The sequence of events in the revocation processes outlined above makes it difficult to classify prison admissions. Consider an example case where an offender completes his or her sentence following conviction for a robbery, and while on postconfinement community supervision (PCCS), the offender commits a burglary. Process dictates that once this offender is arrested, their community supervision is revoked and they are returned to prison or jail while the revocation process is completed. Meanwhile, he or she is subsequently convicted for the burglary, sentenced for that new crime, and begins a new term while in custody under the revocation. Now the question, why has this person returned to prison? To an analyst, the new crime and resulting arrest is likely the important event; the revocation is simply incidental to that event. However, to an administrator recording the admission of an inmate into custody, the reason is based on facts known at the time of commitment. The inmate has not yet been convicted at the time they are admitted to prison so the reason is vague. How is this to be recorded in an administrative system?

6 Gaes et al. 53 Figure 2. Revocation process for crimes committed during postconfinement community supervision, North Carolina. When a new crime is committed while under supervision, authorities will often delay revocation hearings until the disposition of the new crime is resolved (Binnall, 2012). In such cases, offenders convicted of new crimes will receive a new sentence while under PCCS and will also receive either a concurrent or consecutive violation term for the supervision violation (Binnall, 2012). Since both of these events occur

7 54 Justice Research and Policy 17(1) Figure 3. Revocation process for a technical violation during postconfinement community supervision, North Carolina.

8 Gaes et al. 55 after a person has been detained, the only legitimate status at the time of admission is admission pending the results of a revocation hearing. Some administrative systems allow for this possibility, but in most jurisdictions such codes are not often used. Instead, codes usually indicate revocations but provide little or no additional information. In many cases, an offender is returned to prison during the revocation process, so his or her status is both pending a revocation and pending a conviction. Subsequent to initial admission, status changes over time. Ultimate disposition will depend on the court ruling and the disposition of the revocation hearings, all of which can occur during a continuous term of imprisonment. To make this more complicated, any stage of this process (prehearing, preconviction, conviction, sentencing, etc.) may be served in a prison or a jail, depending on the state and the circumstances. 5 For example, in California, prior to the Realignment Act, a person could be held in either a jail or a state prison (reception center) during the revocation process. After the Realignment Act, a revocation could only be served in a jail (Grattet, February 1, 2015, personal communication). In Pennsylvania, a person convicted of a new crime or purely technical violation while under parole supervision can serve time in a prison, jail, or special violation center similar to a halfway house. 6 This variation in physical placement represents a complication for measurement since many state administrative prison databases are not well equipped to record the status of offenders spending time in jail. In extreme cases, the revocation itself may be lost entirely. Consider one last convolution. In our scenario, the person was originally committed for a robbery, was released, committed a burglary, was revoked, and was also sentenced to a term of prison for the burglary. In this hypothetical, the revocation resulted in a consecutive revocation term but was served concurrently with the new burglary sentence. How do we classify this person s current term? Is the offender a robber/burglar? A revoked robber/burglar? A robber, then a burglar? Is his or her time served for one or more of these classes? The best we might do with most administrative databases is to classify the offense as a composite of instant and return crimes, possibly gauging the seriousness levels of these combinations. Revocations in Lieu of Prosecutions Prosecutorial discretion is another complicating factor in classifying returns to prison. In cases where incarceration may be achieved through revocation, prosecutors may decline to prosecute offenders for new crimes because the process is costly, time consuming, and uncertain. However, that is not to say that the new crime goes unaccounted for. Additional time served may be included as part of the revocation itself, so that even in cases where admissions are technically limited to revocation as the cause, the properties of the prison term reflect additional unspecified sanctions. Disentangling these nuances requires a system of records with all of these discrete events: the type of revocation misbehavior, including whether there was an arrest, the disposition of the arrest, and the disposition of the revocation. Few administrative databases are up to this task.

9 56 Justice Research and Policy 17(1) Determinate and Indeterminate Sentencing States Another complication arises from the distinction between states having either determinate or indeterminate sentencing structures. These different structures confound admission classification in two ways. 7 The first relates to sentencing itself. In some states with indeterminate sentencing, offenders returned to prison for violations will simply complete all or part of their extant sentence there is no new court commitment. In those states, the offender is returned to prison to serve a new term and is thus resentenced. This resentencing may appear as a new commitment though it truly represents a return for a violation. Second, in jurisdictions where a determinate sentence leads to separate prison and community supervision terms, it seems likely that reliance on revocations in lieu of prosecution would be more common in states with indeterminate sentencing relative to those with determinate sentencing. In indeterminate sentencing states, the time remaining on the original sentence is often lengthy because time served is much shorter than the maximum sentence imposed. The prosecutor has more to work with when fashioning what he or she sees as an appropriate prosecutorial response. In determinate sentencing states, where returning to prison for a revocation of the technical conditions governing community supervision is restricted to short periods by law, prosecutors may be less inclined to rely on revocations in lieu of prosecution. 8 State Agencies Responsible for PCCS There is also state variation in agencies responsible for supervising people under PCCS and different agencies and procedures involved in the revocation determination (Piehl & LoBuglio, 2005). 9 Piehl and LoBuglio found that in some states a person on PCCS can be supervised by more than one agency. This complex structure of PCCS results in equally complex administrative databases each under the control of different agencies. Evidence From Administrative Data In the preceding discussion, we argue that the revocation process can create ambiguity in the classification of the type of admission. This ambiguity is fostered by the sequence and timing of revocation events, the separation of jail and prison systems, the place of confinement during the sequence of revocation events, sentencing structures, prosecutorial discretion, and complexities in agencies responsible for supervision. In spite of these limitations, however, data sets exist which attempt to standardize classifications of these events for accounting and analytic purposes. One such data set is the NCRP, a data collection program operated by the BJS. NCRP is a compilation of administrative data from corrections agencies across the country. In this section, we describe the NCRP data and provide empirical demonstrations that suggest difficulties exist in the application of admission classifications. 10 The NCRP is a data collection program that compiles admission and release records for offenders entering and exiting state prison in states throughout the

10 Gaes et al. 57 United States. It has been operating since 1983; however, more recently the data have been reconstituted starting in 2000 to create a series of reliable, longitudinal state data files that track individual offenders into and out of prison over time. Moreover, the NCRP reports information on offender movements onto and off of community supervision, though these data are less extensive and currently cover fewer states. Overall, the NCRP includes data from all 50 states, though only a subset of these data are used to construct longitudinal prison records (called term files ) and longitudinal PCCS records (or PCCS term files ). Importantly, data used to populate the NCRP come directly from the same administrative sources used to record information on prisoner status and prisoner movements within states. This quality makes them useful for gauging how states characterize admission status and how these characterizations align with patterns of reporting. 11 We are able to use these data to demonstrate evidence of the signs and symptoms of the classification problems we ve described up to now. The remainder of this section provides such illustrations. Distribution of Prison Admission Codes Consider first the distribution of NCRP prison admissions codes across all the states contributing to NCRP over a 12-year period, We found in the NCRP, five states report that more than 96% of admissions are new court commitments: Florida, Idaho, Maryland, North Carolina, and Washington. 13 At face value, this finding is hard to believe, especially if states provide even a small degree of monitoring in the postconfinement period. One would expect that at least some modest number of revocations are occurring. Other states report between 18% and 89% new court commitments. This variation in reporting across other states implies (but does not prove) diverse reporting practices or limited postrelease supervision periods. Time Served as a Diagnostic Measurements of time served provide a second, useful diagnostic. Within an offense category, time served for revocations should be, on average, much less than time served for new court commitments. There should be less variability in time served among revocations as well. We test this proposition in four states where data quality is known to be exceptionally high: South Carolina, Georgia, New York, and California. Results are reported in Figure 4. Each graph in Figure 4 shows a series of box plots. Each set of box plots illustrates time served calculations across five general categories of offenses (burglary [B], aggravated assault [AA], automobile theft [AT], armed robbery [R], and grand larceny [L]), where each category is further divided by commitment type (1 ¼ revocation/ other commitment and 2 ¼ new court commitment). The graphs show these offense types in pairs with revocations to the left of the court commitments for a particular offense. The horizontal white lines indicate the median value of the time served. The

11 58 Justice Research and Policy 17(1) Figure 4. The distribution of time served by admission type across four states. B ¼ burglary (B2 court commitment and B1 is other commitment); AA ¼ aggravated assault; AT ¼ auto theft; R ¼ armed robbery; L ¼ grand larceny. dark vertical bars show the 25 75% range. The whiskers show the range exclusive of some outside values. Data are truncated at 5 years to improve resolution of the graphs. Results from New York and California show what we expect from states that impose relatively short prison terms for violations of the conditions of supervision. The distribution of time served following revocation is fairly uniform regardless of the original offense. The same uniformity does not appear for new commitments, as expected, because some crimes are more serious than others and criminal history heavily influences sentencing practices. The impression is that admission codes for New York and California are relatively accurate although (1) time served following revocation is sometimes lengthy, suggesting that these are mixtures of terms that may include violations but other admission types as well that would have longer periods of time served and (2) time served following a new commitment is sometimes very short, even for serious crimes, suggesting that these are not all terms for new commitments. A different pattern appears in South Carolina and Georgia. In South Carolina, there is little distinction between the distribution of time served following a revocation and time served following a court commitment, with the exception of armed robbery. Even for armed robbery, the overlap is considerable. The distinction between the

12 Gaes et al. 59 distribution of time served for court commitments and other admissions is clearer in Georgia, but again there is considerable overlap. Additionally, the length of time served following a revocation is typically longer in South Carolina and Georgia than in New York and California. Perhaps these differences reflect real differences across the states in the severity of sanctions triggered by a revocation; we cannot be sure. Still, it appears that a revocation (as recorded in the NCRP) means something different in South Carolina and Georgia than it means in New York and California. As we logically showed in the previous section, revocation return terms are often aggregations of new sentences and revocation sanctions. Therefore, differences in the time served distributions we observed may reflect state-by-state variation in both how states aggregate these terms and how states code these terms. Time Elapsed Since Prison Release as a Proxy for Admission Type A third test of the data is to examine how the distribution of NCRP prison admission and supervision release codes compare to measures of how much time has elapsed since an offender s last release from prison. We compare the release date from the PCCS (community supervision) to the admission date for the prison term. If the PCCS release date means that the revocation process has ended and a revocation occurred, then that date should be very close to the subsequent prison admission date if someone is returned to prison rather than jail or a community alternative. We test this idea using NCRP data from Pennsylvania, a state for which we have a sufficiently long time series of community supervision records. Results are reported in Table 1. Table 1 shows the crosstabulation of return types using the PCCS datedefined definition (column 1 of Table 1) and state-reported prison admission codes (columns 2 4). For simplicity in reporting, comparisons are limited to three classes of offenses: violent crimes, property crimes, and drug crimes. 14 The percentages are column percentages. Table 1 shows three important things. The first is that there is a lot of agreement between PCCS release/prison admission, date-defined admission definition, and prison admission codes. Among cases with known prison admission codes (the first two columns), overall agreement rises to 98% for new admissions and 81% for revocation admissions. The second is that prison records with admission codes that are other and unknown tend to be more often revocation admissions according to the admission type based on the PCCS date-defined information. PCCS records give us another window into the supervision release, prison admission sequence. This is not definitive because of the possibility that people released from PCCS on a revocation do not return immediately to prison, and perhaps for long periods of time. People who are undergoing a prolonged revocation process may be serving that period in a jail or community alternative place of confinement. This may be why there is more agreement with new commitments than revocations. The third point is that misclassification, when it occurs, tends to run in one direction overclassifying new admissions as revocations. Assuming the PCCS date-based definition is correct, Table 2 uses another approach to compare PCCS releases and

13 60 Justice Research and Policy 17(1) Table 1. Comparison of PCCS Release Date-Defined Admissions to Prison Admission Codes. Admission Type Using Prison Admission Codes Admission Type Using PCCS Records (Date Defined) New Admissions Revocation Admissions Other/ Unknown a Violent crimes New admissions 10,380 (98.2%) 719 (20.9%) 172 (13.9%) Revocation admissions 185 (1.8%) 2,726 (79.1%) 1,065 (86.1%) Property crimes New admissions 5,908 (97.5%) 417 (19.4%) 492 (34.4%) Revocation admissions 152 (2.5%) 1,735 (80.6%) 937 (65.6%) Drug crimes New admissions 8,784 (95.2%) 636 (19.4%) 759 (44.7%) Revocation admissions 131 (4.8%) 2,642 (80.6%) 979 (56.3%) Note. PCCS ¼ postconfinement community supervision. a The other/unknown category includes admissions that could not be unambiguously classified as new admissions versus revocation admissions and admissions that had missing codes. Table 2. Comparison of PCCS Release Date-Defined Admissions to Admissions Based on the 1-Year Rule. Admission Type Using Time Elapsed Since Release (Return Within 1 Year) Admission Type Using PCCS Records (Date Defined) New Admissions Revocation Admissions Violent crimes New admissions 10,987 (81.4%) 284 (16.2%) Revocation admissions 2,512 (18.6%) 1,464 (83.8%) Property crimes New admissions 6,563 (79.7%) 254 (18.0%) Revocation admissions 1,669 (20.3%) 1,155 (82.0%) Drug crimes New admissions 9,916 (80.2%) 263 (16.7%) Revocation admissions 2,444 (19.8%) 1,308 (83.3%) Note. PCCS ¼ postconfinement community supervision. prison admissions. It shows the crosstabulation of return types using the PCCS datedefined definition to the definition based on a 1-year rule of elapsed time since release. Under this scenario, if someone returns within 1 year of release, we assume they are a revocation return, otherwise a new court commitment. The logic of this test is based on the idea that revocations resulting from violations tend to occur within 1 year of release and that most admissions after 1 year are likely to be new court commitments, though the 1-year mark is an arbitrary one and could easily be replaced with some other length of elapsed time.

14 Gaes et al. 61 Table 2 shows 80 84% of cases agree. Other cases that do not agree overwhelmingly fall into the category of cases where supervision was revoked sometime after a year of supervision had passed. There are also cases where offenders return within 1 year but did not appear to be on supervision. These cases could also have occurred if, for example, an offender moved out of state after their release. Overall, this table suggests that identification of admission type using the 1-year rule underestimates revocation returns especially since we have confidence in the Pennsylvania prison admission codes. But we must also assume that the PCCS release date is accurate and means the revocation has been resolved. We also ran another test of the 1-year rule based on the prison admission codes in two states that are known to have reliable data, California and New York. 15 We examined the crosstabulation of the 1-year results from all returns from 2000 to 2011 with the prison admission codes that distinguish new court commitments and revocations ignoring any other type of admission such as transfers from other jurisdictions. Of the 992,479 returns to California prison over this timespan, 31% were new court commitments and 69% were revocations according to the 1-year rule. According to the prison admission codes, 10% of returns were new court commitments and 90% were revocations. Assuming the prison admission codes are correct, the 1-year rule overcounts new court commitments and undercounts revocations by a wide margin. We find similar results in New York. Of the 166,683 returns over the period , 63% of the returns were new court commitments and 37% were revocations according to the 1-year rule. Prison admission codes tell a different story 26% were new court commitments and 74% were revocations. Choosing alternative thresholds for elapsed time would change the nature of the classification errors; however, any optimal threshold would still have large classification errors. States with less reliable admissions codes require some alternatives such as the PCCS date-defined definition where it is available. In lieu of that approach where there is no alternative, analysts might adopt a 1-year rule acknowledging this classification will introduce error. Supplementing Prison Admissions Data: Evidence From a Study of Revocations in Wisconsin and California Data from the NCRP illustrate how administrative data derived solely from state prison systems are often insufficient for accurately classifying admissions in many cases. At the same time, it is possible for researchers to overcome these challenges by pooling data across multiple sources. Recently, Jones (2016) was able to supplement Wisconsin prison admissions data with new court commitment data. Offenders with prison revocation admissions may eventually receive a new sentence, but that disposition is often unknown at the time of admission. Jones was able to identify new sentences that occurred after revocation admissions, by identifying offenses committed within the 6 months prior to the revocation admission with associated convictions that occurred after the revocation admission. Prior to this linking with new data, during the period of , the ratio of revocation-only to revocations with new sentences was about After the linking

15 62 Justice Research and Policy 17(1) with new data, the ratio over the same period was about For example, prior to the new information in 2015, there were about 4,000 revocation-only admissions and about 1,000 revocations with new sentences. After the linking, there were about 3,000 revocation-only admissions and about 2,000 revocations with new sentences portraying a very different picture of admissions types. Grattet, Petersilia, and Lin (2008) provide an illuminating illustration of the effort and data quality that is required to successfully decompose prison returns into revocation types and new court commitments. Grattet et al. (2008) found that At the end of three years, 66 percent of all California parolees had been returned to a California prison, 27 percent for a new criminal conviction and 39 percent for a technical or administrative violation, which can result from new crimes or violations of the conditions of parole. On any given day, six out of ten admissions to California prisons are returning parolees. (p. 5) To decompose admissions into this distinct set of types, they compiled a unique database that merged information from 12 different national and local databases. These data gave them sufficient information to evaluate the return pathways. To bolster data on violations they had to merge three distinct databases from the California Department of Corrections and Rehabilitation. As the authors indicate, A comprehensive picture of violations was only possible by merging information from all three systems (Grattet et al., 2008, p. 68). However, they also needed data from California s criminal history depository to confirm whether a return was based on a crime resulting in a new court commitment without referral to the Parole Board. During the scope of their evaluation, there were 151,750 violation reports containing information on...whether criminal violation cases were successfully prosecuted in court as opposed to being referred to the parole board, and whether cases heard by the Parole Board were returned to prison or continued on parole (Grattet et al., 2008, p. 11). With these data, Grattet et al. could also decompose returns into purely technical violations which were mostly for absconding missing an appointment qualifies as absconding or for new crimes. Of the total 151,750 violations, 80.5% were returned to prison and 19.5% remained under supervision. Of the 122,141 violations resulting in a return, 25.7% of the offenders had a new sentence, 57.6% had been detained due to an arrest for a crime but received a revocation without a new sentence. Many of these arrests were for drug possession and drug use offenses. There were also unprosecuted serious crimes as well that led to a revocation rather than a new court commitment. The remaining 16.7% were revoked on strictly a purely technical violation. With the ability to decompose prison returns into new court commitments, purely technical violations without a new arrest, and violations with a new arrest, along with a rich set of covariates involving the offenders, the agents, and the community, Grattet, Petersilia, and Lin were able to model many aspects of the back end process of prisoner returns (Grattet & Lin, 2014; Grattet, Lin, & Petersilia, 2011; Grattet, Petersilia, Lin, & Beckman, 2009; Lin, Grattet, & Petersilia, 2010, 2012).

16 Gaes et al. 63 Their work highlights the effort required to develop a prerequisite database to address policy analyses that require important, foundational distinctions among return events. National Estimates of Violations of Supervision The California study is emblematic of a large body of research showing that a high proportion of prison returns are violations of PCCS (Blumstein & Beck, 2005; Burke & Tonry, 2006; Hughes, Wilson, & Beck, 2001; Petersilia, 2003; National Academy of Sciences [NAS], 2014). As the Grattet et al. (2008) research demonstrates, a person whose supervision is revoked is often arrested for a new crime; however, in lieu of prosecution, authorities use the supervision revocations process, saving criminal justice resources that would be involved in prosecution, pretrial detention/supervision, sentencing, and potentially longer prison stays. While an unprosecuted arrest may qualify as grounds for a technical violation, we think it is misleading to represent revocation returns as purely technical violations, lumping together unprosecuted crimes with failing a drug test, absconding, and missing a parole appointment. Pfaff (2015) used the Bureau of Justice 2004 Survey of Inmates in State Correctional Facilities (SISCF) to evaluate revocations. The SISCF surveys a nationally representative sample of inmates and reflects cross-sectional or prevalence measures of self-reported data. 16 Pfaff found that for inmates whose parole was revoked, 68.3% claimed it was for a new arrest. Since inmates were allowed to self-report more than one reason for the revocation, those who reported other technical violations, such as a failed drug test, could also report a new arrest or offense. Pfaff (2015) notes...revocation for a new offense arguably the least technical of all violations is reported by 26% of those who also failed a drug test, 29% of those who also failed to report to a drug test, 31% of those who also failed to report to treatment, 36% of those who also failed to report to other treatment, and so on. Moreover, for over half of those who reported both a new offense violation and a drug test violation, the new offense was something more serious than a drug crime either a violent or property offense. (p. 190) Thus, the premise of many scholars that violations are a large proportion of prisoner returns may be technically true; however, this hides the fact that many of the revocation returns are more than benign, purely technical violations. Conclusion We have described how the revocation process and state-by-state structural differences confound our understanding of the decomposition of prison returns into revocations and new court commitments. The ambiguity is fostered by the sequence and timing of revocation events, the separate jail and prison systems in most states, and placement of offenders during the sequencing of revocation events. We have also demonstrated how these difficulties translate into measurement problems. Administrative data that record different admission types must be treated with circumspection.

17 64 Justice Research and Policy 17(1) This is a caution to analysts and policy makers who blithely read revocation statistics without digging into either the meaning or the veracity of the estimates. A better coding of prisoner returns would lead to a better understanding of patterns of recidivism and supervision noncompliance. 17 There are also differences among states in many other dimensions including the sentence length threshold for admission to a prison or jail, sentencing structures, length of supervision, capability of information technology systems, and state coding conventions. Further complications arise from the aggregate terms that result from some combination of the revocation term and a new conviction, and the ambiguity in the meaning of a supervision release date. Additional complexity comes from determinate sentencing where revocations and new commitments both result in new sentences, so the admission code is not informative. Prosecutors have discretion to charge a new crime if someone is arrested while under supervision or allowing the revocation term to serve as a punishment expedient. This means that variability in state revocation rates is dependent on this level of discretion as much as any state variation in supervision capabilities or heterogeneity in the prisoners released to community supervision. Efforts by Grattet et al. highlight the importance of decomposing returns into their most elemental parts. Only a precise typology can inform policy applications and drive meaningful insights. However, it is also true that sufficiently discriminating data sets may not always be available. They can only be derived from sophisticated state recording systems or researchers with high degrees of access to multiple data streams. Thus, we are forced to conclude that information on prison admission status from standalone, administrative data sources is not likely to be enough to properly classify the reason for prison entry and that any sensible set of classification rules must incorporate at least some additional information, for example, information on supervision status. Analysts who have no other recourse than to analyze NCRP or similar administrative databases can use either the prison admission codes or the 1-year rule, but must acknowledge the problems in any estimates that differentiate revocation from returns with reconvictions. This is not a general indictment of the quality of NCRP data. It is the only data set that can be used to analyze individual prison and PCCS terms over a large number of states over a long period of time. NCRP has been crucial to address issues involving mass incarceration (Pfaff, 2008, 2014a, 2014b, 2015) different operationalization of recidivism (Rhodes et al., 2014), and the primary drivers of prison aging (Luallen & Kling, 2014) among other important topics. Toward a National Estimate of Return Types Are Information Technology Systems Up to the Task? The NCRP is one of the primary instruments for making national estimates of prison returns. All of the primary summaries of research on returns use some combination of NCRP, the SISPC, the Annual Parole Survey, and the National Prisoner Statistics tools funded and administered by the BJS. 18 The research we cited focusing on returns (Blumstein & Beck, 2005; Burke & Tonry, 2006; Hughes et al., 2001; Petersilia, 2003;

18 Gaes et al. 65 NAS, 2014), either indirectly or directly, use these estimates in an explication of mass incarceration or desistance from crime. The NAS (2014) report The Growth of Incarceration in the United States: Exploring Causes and Consequences discusses the role of the expansion of parole populations and parolees exposure to reconfinement as a result of a violation. 19 The NAS report reflects the other research in this genre showing the significant contribution of violations to the rise in mass incarceration. The NAS report cites BJS research by Carson and Sabol (2012) and Glaze and Bonczar (2011) indicating that nationally about 20% of all prison admissions in 2010 were violations of PCCS and that violations had doubled during the period The Carson and Sabol report shows that, of all sentenced state prison admissions in 2011, 34% were parole and mandatory release supervision violators. In all three of these sources, there is recognition that the violations are a composite of both purely technical breaches and unprosecuted arrests. Despite that recognition, there is no appraisal of the implications of this decomposition to address supervision policy and practices. Knowing whether revocations are for unprosecuted arrests or technical violations would most likely suggest alternative policies for community control and reintegration services. Yet, at least at the national level, this is a data vacuum. Can this void be addressed? While the NCRP instructions are clear on the requirements for assigning codes associated with a new commitment or revocation, information technology (IT) systems used to record these events do not necessarily record information in a way that allows an analyst or programmer to unambiguously meet the NCRP requirements. Furthermore, the NCRP instructs the submitting state to represent only physical admissions. How states actually assign these codes will depend on whether they actually record all of these changes of status in their databases or whether the IT systems simply update the data to reflect the most recent status. The programming effort required to apply the NCRP rules to cull the data for an NCRP submission is not a trivial exercise, even if the databases contain all of the events. Looking toward the future, it seems reasonable to think that more comprehensive and useful accountings of admission codes, release codes, and admission and release dates in administrative data will evolve over time; however, improvements may be slow or difficult to achieve. Electronic prison recordkeeping systems focus more on admission dates than on admission reasons, and record keepers may not know the reason for admission at the time of admission. In addition, many IT systems record physical admissions and releases separately from information on sentencing and revocation information and aligning the two is no simple task. We can always uncover a return and with the addition of supervision records, we can always ascertain if the return occurred while someone was under supervision, assuming accuracy of these latter records. Admission types for most new court commitments are likely to be accurate. Other return types are less so. As we have shown, the timing of events naturally implies that we may never escape some return classification ambiguity. Returns are difficult to characterize because of the timing of the revocation process and the loss of information, when some or all of the revocation process occurs while the person is located in a jail. Lost events in sequences also

19 66 Justice Research and Policy 17(1) confound our ability to classify offense types. In some cases, the best we might do is to indicate a person is serving an aggregate term based on the offense associated with the original sentence and the offense associated with the revocation, if indeed, there was an arrest involved in the revocation. Part of the ambiguity stems from the black hole created in states which have separate prison and jail jurisdictions, and the jail records are not available. This reality calls for a national system of jail records, but that is the topic of another paper. Authors Note Points of view in this document are those of the authors and do not represent the official position of the U.S. Department of Justice. The authors are responsible for any errors in the article. Acknowledgments We would like to thank all the individuals who have supported the development of this article and contributed to its improvement. Specifically, we would like to thank William Sabol and E. Ann Carson of the Bureau of Justice Statistics and members of the Abt Associates Journal Authors Support Group. Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. Funding The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by Grant Nos BJ-CX-K067 and 2015-R2-CX-K135 awarded by the Bureau of Justice Statistics, Office of Justice Programs, U.S. Department of Justice. For this work, Thomas Rich served as a project director along with principal investigators William Rhodes and Gerald G. Gaes. Notes 1. A reviewer asked us to embed our discussion in a broader context of the quality of other criminal justice data collections and the social construction of key variables such as race. We acknowledge that there is a wider viewpoint to be taken, but limiting the scope of our discussion and analysis allows us to focus on the issues and potential remedies for this very important dimension of prison recordkeeping. 2. Data compiled for the National Corrections Reporting Program (NCRP) represents one such data source. It is partly the subject of this article. 3. NCRP is the largest and most comprehensive compilation of these types of records in the United States. 4. There are variations of the sequences shown in this figure. In some jurisdictions, an offender may be allowed to remain in the community during the supervision revocation process. In other jurisdictions, offenders are sent halfway back and spend a short term in a halfway house or serve revocation terms in a halfway house in lieu of returning to prison or jail to await the revocation hearing. The purpose of Figure 1 is to represent the most prevalent scenario across the states.