Supervisory Pretrial Services Administrator
Christopher T. Lowenkamp
Office of Probation and Pretrial Services
Administrative Office of the U.S. Courts
Pretrial Services Risk Assessment Literature and PTRA
Federal Risk Assessment
IN 2009 THE ADMINISTRATIVE Office of the U.S. Courts and the Office of Federal Detention Trustee (a Justice Department agency charged with administering and controlling the costs of pretrial detention in the federal system) published Pretrial Risk Assessment in the Federal Court, which recommended that the federal pretrial services system develop and implement an actuarial risk assessment tool. Ever since then, the system has been moving towards that goal. Lowenkamp and Whetzel (2009) have already detailed the process followed and described the tool that was ultimately developed, the Pretrial Risk Assessment (PTRA). As of August 2011 the tool had been fully implemented in almost all districts; therefore, it seems an appropriate time to assess the tool in light of the available pretrial risk assessment literature, determine how implementation has proceeded in the federal system to this point, and assess the ultimate impact, if any, of the tool on the federal pretrial services system. Perhaps the most important question is whether the tool has begun to affect officer recommendations for release/detention and/or release rates in districts where it has been operational for a year or more.
One of the few areas in which pretrial services research initially led the way before our counterparts in post-conviction was risk assessment, with devices utilized in several of the larger cities, including Washington, D.C. (Toborg, Yezer, Tseng, & Carpenter, June 1984) and New York (Ares, Rankin, & Sturz, 1963), long before post-conviction risk assessment devices had been introduced there. Unfortunately, while such risk assessment tools remained in use in those cities, they did not spread to other pretrial services agencies as rapidly as risk assessment tools did in post-conviction organizations. In addition, assessing risk for offenders differs significantly from assessing risk for defendants, making it impossible for post-conviction and pretrial services to share risk assessment devices. For example, pretrial services risk assessment devices focus on failure-to-appear, which is not a focus of post-conviction tools; and post-conviction risk assessment devices focus on long-term recidivism, something that has not historically been a primary concern of pretrial services. Therefore, at least theoretically, there is little crossover between the two disciplines in the area of risk assessment.
While the literature does not show a lot of work on risk assessment in pretrial services, a comparison with the post-conviction risk assessment literature finds that this is the evidence-based practice area in pretrial services that has received the greatest research attention and there are some studies of excellent quality that cover a wide range of issues (e.g., Toborg, Yezer, Tseng & Carpenter, 1984; Goldkamp & Gottfredson, 1988; Goldkamp & Vilcica, 2009; Levin, 2006). For example, Toborg, Yezer, Tseng, and Carpenter provide an excellent discussion of selectivity bias. First, arrested defendants are detained; because of this detention, their propensity for pretrial arrest and failure-to-appear cannot be observed. Research on the first form of bias is fairly common in the literature and is discussed in most research on pretrial services risk assessment initiatives. However, rarely seen (but discussed in Toborg et al.) is the second form of selectivity bias, which affects defendants released under different scenarios: some without any restriction and others released on various bond types or with various conditions that are based on individual characteristics (Toborg, Yezer, Tseng & Carpenter, 1984:102). Toborg, Yezer, Tseng and Carpenter have done an excellent job of addressing this very important issue, which needs further discussion in the literature.
Typically, defendants who are released on supervision in the federal system are given a "laundry list" of conditions. In fact, beginning in 2009, fully 99 percent of federal defendants who were released had one or more conditions in place. There has been little to no research establishing that any one condition or any combination of conditions helps achieve the desired goals of appearing in court and not committing an offense while on pretrial release. In addition, the conditions frequently result in technical violations, which can cause the defendant to be returned to the court for additional hearings, which can result in the defendant's detention or in modification/addition to the conditions. If these conditions cannot be demonstrated to increase the likelihood that the defendant will appear in court as required and/or reduce new offenses committed by the defendant while on pretrial release, then the significant investment of pretrial release agencies and courts in these conditions and their enforcement is ineffective and unwise. In fact, preliminary research in the federal system seems to indicate that these conditions do not increase positive outcomes; instead, such conditions may increase negative outcomes (VanNostrand & Keebler, 2009:31–33). In addition, a recently completed literature review that focused on release conditions and supervision in all pretrial services entities (federal, state, county, and local) concluded that the evidence for the utility of such conditions is at best weak and in many cases nonexistent. "It must be acknowledged that research in this area is very limited and that more is needed" (VanNostrand, Rose, & Weibrecht, 2010:34). Therefore, more refined testing of the use of pretrial release conditions seems warranted.
Toborg et al.'s research, which used a team-developed risk assessment tool, produced the following very interesting findings: use of the tool led to more defendants being released on less restrictive conditions and with no increase in failure-to-appear or re-arrest rates (Toborg et al., 1984:105); the risk prediction tool that Toborg, et al developed increased release rates by 12 percent with no appreciable increase in failure-to-appear or re-arrest rates (Toborg et al., 1984:58); and, finally, the tool is more accurate for appearance in court than for safety (Toborg et al., 1984:73). Risk tools, while tremendously useful in improving agency decision making and ultimately release recommendations, have limitations. For instance, they are good at identifying groups of defendants who present various risks, but they cannot be totally accurate at the individual level (Toborg et al., 1984:111). Therefore, agencies need to convey to line staff, as the federal system has done, that the tool should not be followed blindly; in addition, they should permit officers leeway to override the tool after staffing with a supervisor or some similar methodology.
Finally, Toborg et al reported one finding that has yet to be replicated: pretrial services supervision had no effect on controlling the risk of nonappearance (Toborg et al., 1984:73). Further research is needed either to disprove or strengthen this finding due to the potential significance it carries if confirmed in other jurisdictions with other risk assessment tools. Once the new tool is fully implemented in the federal system, this finding will be tested.
Goldkamp and Gottfredson studied three urban jurisdictions and presented a seemingly simple conclusion: that judicial involvement is essential to the successful implementation of a risk assessment device (Goldkamp & Gottfredson, 1988:129). That simple conclusion also emerges in Goldkamp's most recent research on pretrial risk assessment (Goldkamp & Vilcica, 2009:129–30). A finding that spans 22 years, appears in multiple jurisdictions, and derives from different research partners would seem to hold potential as a strong replicable result. Goldkamp and Gottfredson identified some ways to contribute to successful implementation through ongoing training, assessment of the officer's use of the tool, and annual or bi-annual certification of the officer's skills in using the tool. In addition, the Goldkamp and Gottfredson study confirmed the major findings of Toborg et al's earlier research. Unfortunately, the federal pretrial services risk assessment was not developed with judicial input. The early data emerging from its implementation in the federal system suggests that the lack of judicial input has made the tool less effective in increasing officer recommendations for release and achieving increased rates of defendants released in the federal system. However, as implementation has progressed, opportunities for judicial training and input have been identified and implemented. The input from judicial officers is likely to continue and even grow, providing opportunity for further refinement, based on judicial input, in the future.
One of the strengths of the Goldkamp and Vilcica research is that it squarely takes on one of the most enduring "urban legends" of pretrial services risk assessment research. Most pretrial services agencies, including the federal system, continue to capture data on and analyze the variable "community ties." While some of the fascination with this variable stems from its identification as an important variable in the "granddaddy" of all pretrial services research, the original Vera project, the "community ties" variable likely endures because of its tremendous "face validity." Its inclusion in the small number of long-standing important pretrial services variables is certainly not warranted by the research results of the last 20 years. However, most researchers merely ignore the variable of community ties, since the analysis does not bear out its value (see, for example, Administrative Office of the United States Courts, 1979; VanNostrand, 2003; VanNostrand & Keebler, 2009; Winterfield, Coggeshall, & Harrell, 2003). Goldkamp and Vilcica take on the lack of value of community ties for pretrial risk assessment and ultimately remove it from its lofty perch.
Goldkamp's analysis of factors influencing judicial decisions at the pretrial release decision, however, found that contrary to the intended effect of Vera-type information-based reform procedures community ties items did not play a significant role in shaping judges' actual pretrial custody decisions–and were not helpful predictors of defendant risk. (Goldkamp & Vilcica, 2009: p. 124).
In the research used to develop the PTRA, a "community ties" variable known as "Time in Area" was tested and found to have no predictive value in the PTRA model.
A seemingly "obvious" issue not found in virtually any other research on the topic of risk assessment is the importance of including judicial officers in the development, implementation, and ongoing use of a risk assessment device. Only Goldkamp and Vilcica's findings discuss this issue, not to mention endorsing the strong role it played in the Philadelphia research: "As a judicially developed and adopted policy, it stands alone in the nation in the first years of the 21st century–one might argue, in isolation–as an empirically informed approach to the problem of judicial discretion at the bail stage" (Goldkamp & Vilcica, 2009:129–30). As mentioned earlier, the federal system is currently implementing a risk assessment tool without judicial involvement, which could be impacting the acceptance and use of the tool in the federal system, although opportunities for judicial training and input have progressively increased in the course of implementation.
Given Goldkamp and Vilcica's vision of pretrial justice and their desire to improve the pretrial release process and reduce judicial discretion, it is almost shocking that they missed the importance of pretrial detention and made their tool detention neutral (Goldkamp & Vilcica, 2009:134). This is especially true since Philadelphia has operated pretrial services under federal court supervision due to jail overcrowding at various times during the 20-plus years encompassed by the guideline project in Philadelphia. Reducing unnecessary pretrial detention needs to be a core principle for pretrial services and judicial officers, given the negative consequences of pretrial detention at subsequent phases of the criminal justice system. The negative impact of detention on defendants has previously been documented.
Defendants incur significant costs when they are detained while their case is pending (Goldkamp & Gottfredson,1985:17; Foote, 1954; Klein, 1997:292; Williams, 2003; Rankin, 1964; Tarturo & Sedelmaier, 2009:218). Research has consistently shown that a defendant's pretrial release or detention status impacts case disposition and ultimately sentence. Proponents of pretrial release argue that several factors contribute to that outcome: 1) detainees have reduced access to their attorneys, which limits the defendant's ability to fully participate in the preparation of the defense case; 2) detention exerts great pressure on the defendant to plea bargain the disposition of the case; and 3) detention creates negative perceptions of the detainee in the minds of the court/jury who convict and/or sentence the defendant. Goldkamp concluded that the effect of detention was more limited, in that its only demonstrated impact was an increased risk of receiving a sentence of incarceration when compared to released defendants.
Levin's research revealed that a defendant's odds of failing to appear in a county that uses a quantitative risk assessment are .40 times lower than the odds faced by a defendant appearing in a county that uses qualitative risk assessment (Levin, 2006:10). In addition, if the county uses some mix of quantitative and qualitative measures, defendants are still less likely to fail to appear than if they used qualitative alone (Levin, 2006:10). Finally, if the county uses some mix of quantitative and qualitative measures, defendants are also less likely to be rearrested (Levin, 2006:11).
The literature on pretrial services risk assessment clearly establishes several important premises: "objective risk assessment produces more non-cash release recommendations" (Cooprider, 2009:15); "Notwithstanding a broader definition of 'pretrial failure' and cutting field contacts in half, violation rates declined or remained stable since the implementation of objective risk assessment" (Cooprider, 2009:15); and predictive items identified in pretrial services risk assessment research change over time and therefore must be re-validated on an ongoing basis to ensure their integrity and effectiveness (e.g., VanNostrand, 2003; VanNostrand & Keebler, 2009; Siddiqi, 2002).
The need for validation of pretrial risk assessment tools is illustrated by the following example of an established risk assessment finding that is likely to change. Risk prediction research in the city of New York for the past 20 years has established with relative consistency the predictive value of having a telephone in the residence of the defendant. Given the move in the past decade from the dominance of landline technology to increasing reliance on cell phone technology, it seems unlikely that future research will continue to find great predictive value for a landline phone in the defendant's residence (Siddiqi, 2002:2). Fortunately for citizens in New York City, the agency providing pretrial services has an excellent research operation that re-validates its risk prediction tool every three to five years as warranted. Ongoing re-validation is an essential step for all pretrial risk assessments and it will be an ongoing component of the federal risk assessment program.
One of the major recommendations of the Office of Federal Detention Trustee research is that the pretrial services system should develop and implement an actuarial risk assessment tool VanNostrand & Keebler, 2009). The Pretrial Services Risk Assessment tool was constructed using the same data employed in the Office of Federal Detention Trustee research. The Office of Probation and Pretrial Services hired a staff person proficient in the development of actuarial devices and ultimately developed the tool internally. The result is an objective, actuarial instrument that provides a consistent and valid method of predicting risk of failure-to-appear, new criminal arrest, and technical violations that lead to revocation while on pretrial release. The instrument contains 11 scored and 9 unscored items. The unscored items are rated as either A or B and do not contribute to the current overall risk score. However, they will be analyzed for future revisions aimed at improving the predictive value of the tool. The scored items are given a number of points (0, 1, or 2), which are then added up to produce an overall score. When administered correctly, the Pretrial Services Risk Assessment provides a score that allows for classification into a risk category. Those risk categories are then associated with rates of failure-to-appear, new criminal arrest, and technical violations leading to revocation (See Table A).
When a defendant or material witness is arrested or summoned to appear before the court for an initial appearance, the magistrate judge typically requires a pretrial services report based on the investigation conducted by the pretrial services officer. The officer interviews the defendant to gather information for the report, which contains: defendant case information, including residence; family ties; employment history; financial resources; health (including mental health and substance abuse histories); and criminal history. Based on this information, the officer will provide the court with an assessment of whether or not the defendant is likely to appear for court proceedings in the future and whether he or she presents a danger to the community. Finally, the last section of the report provides the officer's recommendation to the court for the release or detention of the defendant. Once an officer has been trained on the new risk prediction tool, that recommendation should routinely be based on the Pretrial Services Risk Assessment, although the officer can depart from the tool's recommendation after staffing the results with his or her supervisor.
Once developed, the Pretrial Services Risk Assessment tool was piloted in several districts and formal implementation of the tool began in January 2010. As of June 2011, there were 72 districts "live," using the tool on "all" cases; 87 districts trained and certified in using the Pretrial Services Risk Assessment tool; and 6 districts yet to be trained.1 In terms of numbers of officers, as of June 1,742 pretrial services officers had been trained and 1,338 certified to use the tool effectively. National implementation was completed in almost all 93 districts by August 2011.
The certification process should also provide a measure of the effectiveness of the tool and the quality of the training; however, all of the necessary data elements are not available in a format that allows them to be extracted and analyzed. The certification process requires the user to successfully complete two of three risk assessment tool scenarios. At this writing no officer has failed to be certified through that process; however, data is not available on whether all officers passed after completing just the two scenarios. Were that the case, it would indicate successful training and the tool's effectiveness. While trainers can be encouraged that no trained officers have failed to be certified, conclusions beyond that are not warranted without the actual data.
The pretrial risk assessment was first implemented in two pilot districts (Nebraska and North Carolina Western), where forms, policies, and procedures were also implemented and tested. The pilot focused primarily on issues such as ease of use to insure that the national implementation went forward with as few complications as possible. That pilot led to such changes to the tool as clarification of definitions, modification of the tool form, etc. However, those changes were for the most part routine and implementation progressed without major problems.
The only remaining concern in the implementation is the pace of districts using the tool on all cases. The federal system averages about 26,000 pretrial investigations and reports per quarter. Unfortunately, the PTRA is averaging about 4,000 per quarter, leaving 22,000 reports without PTRA scores. That pace must improve significantly and quickly. OPPS has begun tracking investigations/reports and PTRA scores by district to enable us to quickly identify and address districts that are not producing a PTRA for every report submitted.
The implementation of the tool has generated great debate over the finding (represented in the scores of "1" for defendants charged with violent offenses) that violent defendants in fact perform better than most other defendants in terms of re-arrest, failure-to-appear, and technical violations leading to revocation of pretrial release. However, the results found in the federal study are consistent with other similar findings: "defendants charged with more serious offenses and violent have not posed a high risk of failure pending trial in past research" (Austin, Krisberg & Litsky, 1984:30; VanNostrand & Keebler, 2009:21; Toborg et al., 1984:56). Pretrial services officers can be quick to convert long rap sheets into detention recommendations, citing public safety; however, in many cases the evidence does not support that conclusion.
Early results from the first three months of implementation in the two pilot districts (Nebraska and North Carolina Western) showed that the tool increased officer recommendations in favor of release, which was a desired goal of the risk assessment tool. The early pilot results showed no impact from the tool on actual release rates. In the first year of operation, after excluding immigration cases (since the vast majority of those defendants are detained), for the January 1, 2010 through December 31, 2010 timeframe the Western District of North Carolina increased recommendations for release 13.5 percent and increased defendants actually released 6.1 percent. In the first year of operation, again excluding immigration cases and focusing on the one-year period from January 1, 2010, through December 31, 2010, the District of Nebraska increased recommendations for release 2.0 percent and increased defendants actually released 1.4 percent. These are significant trends for the pilot districts and we can hope that those trends will continue nationwide as use of the tool becomes more widespread and ultimately universal.
The largest issue facing the federal pretrial services system is unnecessary pretrial detention. As Table B shows, significant percentages of low-risk defendants (PTRA 1 & PTRA 2) have been detained over the past five years and in all likelihood will continue to be detained unless two things occur: pretrial services officers prepare strong, factual, and accurate pretrial services reports that contain strong release recommendations and (where appropriate) release packages designed to protect the community and ensure pretrial justice for the defendant, and 2) United States magistrate judges assume a higher level of risk in selecting defendants for release than they been willing to assume in the past five years. For magistrate judges, such a change in practice would itself be risky, since they must apply the Bail Reform of 1986 to the more than 100,000 defendants who appear before them each year. Such openness to pretrial release is not for the faint of heart. Luckily the position is staffed by some of the most talented and dedicated public servants to ever serve in the federal justice system.