Predicting Success in a Large Sample of Federal Pretrial Offenders

Abstract

The present study analyzed a host of risk variables related to successful completion of a federal pretrial release program within a large and demographically diverse sample of pretrial defendants. Significant differences were found between variables predicting overall success of pretrial release, including a history of failures to appear and escape behavior. Logistic regressions revealed differences among certain ethnic groups for predicting successful outcome, with Whites significantly more likely than Blacks to succeed. Differences in contributing variables were also found between ethnic groups. Implications for future studies of general risk assessment across ethnic groups are discussed.

Keywords

ethnicity pretrial release risk assessment

The courts and criminal justice system are increasingly engaging in risk assessment and management strategies throughout the adjudication process. This includes the time prior to sentencing a defendant, when determining the level of supervision, and upon sentencing or supervised release. As a result, the literature regarding risk assessment and management has expanded immensely over the last 30 years. Risk assessment strategies continue to offer a cost-effective, objective, and resourceful way to identify the individuals that are at risk to engage in harmful behavior. A survey by the National Institute of Corrections (2003) found that of the 73 community corrections agencies surveyed, all but two included some type of offender risk assessment in their programs. Additionally, a recent meta-analysis has identified over 120 different risk assessment tools available for courts, prisons, clinics, rehabilitation centers, and other settings in which individuals have the potential to cause harm to themselves or others (Singh, Grann, & Fazel, 2011). Among these, there are dozens of instruments for assessing violence risk and general risk in cases such as juvenile delinquency, sexual violence, supervision status, and other areas related to the criminal justice system. However, seldom do these approaches consider the racial and ethnical factors in making these determinations, and rarely do they do so in a sample of pretrial defendants.

Risk Assessment and Ethnicity

Despite the growing demand for risk assessment and management, concerns have been raised regarding their utility and proper use. There are hundreds of relevant factors from which empirical questions have been addressed. Among these questions is whether risk assessments are valid for evaluating offending across groups of different ethnic or racial backgrounds (for a systematic review, see Singh et al., 2011). Despite the growing body of research on risk assessment, there continues to be conflicting findings on the predictive validity of risk assessment tools across different ethnic backgrounds. At one end of the risk assessment literature, studies have shown differences in predicted outcomes between different racial/ethnic groups (Schwalbe, Fraser, Day, & Cooley, 2006; Gavazzi, Yarcheck, & Lim, 2005; Långström, 2004; Fass, Heilbrun, Dematteo, & Fretz, 2008; Singh et al., 2011). For example, Whiteacre (2001) examined racial/ethnic differences in classification errors with the Level of Service Inventory–Revised (LSI-R), a widely used classification instrument designed to predict correctional program performance and recidivism. Results of the study showed a tendency toward more classification errors for Blacks than Whites or Latinos. Gavazzi et al. (2005) found that White status offenders in juvenile court displayed significantly greater risk levels across a variety of domains than other ethnic groups. Additionally, Fass, Heilbrun, Dematteo, and Fretz (2008) examined the validity of the Correctional Offender Management Profiling for Alternative Sanctions (COMPAS) and LSI-R in predicting rearrests of urban ethnic minority offenders within 1 year of community release. They found differences in predictive ability between the instruments for a sample of White, Latino, and Black offenders and found that Blacks were more likely to be overclassified (erroneously predicted to be rearrested when they were not), while Whites and Latinos were more likely to be underclassified (erroneously predicted to be successful when they were rearrested).

At the other end of the risk assessment literature, researchers have provided evidence suggesting little to no differences in predictive validity rates between ethnic backgrounds (Brennan, Dieterich, & Ehret, 2009; Edens & Cahill, 2007; Holsinger, Lowenkamp, & Latessa, 2003; Fujii, Tokioka, Lichton, & Hishinuma, 2005). Holsinger et al. (2006) examined the LSI-R for differences in risk between a sample of Native American and White offenders and found modest predictive validity for the entire sample of offenders, with nonsignificant results in predictive validity for the sample of Native American offenders. Brennan et al. (2009) evaluated the predictive validity of the COMPAS, a fourth-generation risk and needs assessment instrument, and found that it was equally reliable in predicting outcome for both Whites and Blacks in presentence and probation cases.

Risk Assessment and Corrections

Corrections agencies previously implemented risk assessment instruments that were empirically derived from population samples different from their own target populations. This is often because developing instruments specifically for target populations can be too costly for agencies (Latessa, Lemke, Makarios, Smith, & Lowenkamp, 2010). With this reality, and because no current risk assessment instrument is universally applicable across offender populations, researchers are encouraged to test risk assessment instruments on specific target populations so as to strengthen the predictive validity of such instruments (Latessa et al., 2010).

Probation, parole, and pretrial release are divisions of the criminal justice system where the courts have actively evaluated the risk of an offender and piloted new instruments on various populations. Lowenkamp, Lemke, and Latessa (2008) constructed and validated a pretrial risk instrument by testing it on data from a sample of adult pretrial offenders. Eight theoretical constructs were chosen to comprise the risk and need domains: criminal history, pretrial supervision, drug/alcohol use, employment, residence/transportation, medical and mental health, criminal thoughts/attitude, and criminal association. Data were then collected via structured interviews and questionnaires on a sample of pretrial defendants. From the data collected, six items were identified as being statistically significant predictors of rearrest or failure to appear (FTA): age of defendant at first arrest, FTA in last 24 months, prior jail incarcerations, reported illegal drug use, drug-related problems, and employment status. Residential stability and FTAs in the past 2 years were later added for face validity despite their nonsignificance (Lowenkamp et al., 2008). Together, these eight items were incorporated into a risk assessment instrument and were used to classify risk-level cutoffs. The instrument was then tested and validated. Although the selected items were found to have good predictive validity for FTA or new arrest, a major limitation noted in the study was the small sample size of only 342 adult offenders. Ethnic differences were also left unexplored (Lowenkamp et al., 2008).

Pretrial Services

Although much of the research on risk assessment comes from incarcerated populations, relatively little research has been conducted on the population of pretrial offenders. Pretrial Services is a U.S. federal agency with the appointed task of gathering information about a defendant and making an informed recommendation for or against pretrial release based on the defendant’s risk of harming himself or the community and the defendant’s risk of failing to appear for court. Once the decision is made that a defendant be released, the defendant is supervised by an appointed officer and certain conditions of release are imposed to ensure the safety of the community and the defendant’s appearance on court date (Lowenkamp et al., 2008). A defendant’s failure to comply with officers’ guidelines may result in punitive measures or termination from the program.

Federal Pretrial Services generally follows the risk principle in its philosophy of correctional supervision, which states that a defendant’s intensity of treatment and supervision must be proportional to the defendant’s level of risk (Lowenkamp et al., 2008). To determine a defendant’s level of risk, Pretrial Services considers the criminal history, current offense, and other factors related to a defendant. These factors include number of felony convictions, foreign ties, residence status, citizenship, pending felonies or misdemeanors, current offense type, and current drug problems. Furthermore, research shows that matching a defendant’s level of risk with intensity of supervision effectively reduces recidivism (Lowenkamp et al., 2008). Inappropriate levels of supervision can negatively impact the supervision process and place burden on the federal justice system. Although much of the corrections population is low risk, a large portion of offenders are being given intensive treatment, which can be ineffective and costly when applied incorrectly (Austin, 2006). Hanley (2006) examined offenders in intensive supervision programs and found that intensive supervision reduced recidivism for high-risk offenders but not for low-risk offenders. Bonta, Wallace-Capretta, and Rooney (2000) evaluated treatment effectiveness for offenders ranging from low to high risk in intensive community supervision. They found that high-risk offenders showed lower recidivism when matched with intensive levels of treatment, while low-risk offenders showed higher recidivism when matched with intensive treatment. As a result, it is clear that matching a defendant with the appropriate level of supervision can have a significant effect on the arrestees and the success of the supervisory system. In an era of increasing fiscal constraints, working effectively and efficiently with limited public resources is imperative. It is not simply a matter of matching high-risk offenders with the appropriate level of supervision, but it is important to match every offender with the appropriate level of intensity.

Criteria for Predicting Failure or Success

There are several ways pretrial defendants can fail the supervision process. One is by failing to appear for a court-ordered appointment. An empirical analysis by Maxwell (1999) examined the relationship between a judge’s decision to release defendants and risk of FTA among those pretrial defendants. Results showed mixed relations between FTA behavior and defendants’ gender, offense type, prior convictions, and ethnicity. In addition, there was no congruence found between release decisions and FTA risk. For example, defendants charged with property offenses showed a higher risk of FTA but had higher rates of release on recognizance (Maxwell, 1999).

Absconding is also something corrections officers must be cautious about. Absconders are those who fail to comply with officers during the supervisory period by failing to appear for appointments with their appointed officers and by cutting off communication with their officers. Mayzer, Gray, and Maxwell (2004) found that absconders from probation were less educated and had more felonies in their criminal record than successful probationers. They found that race, sex, employment stability, residential instability, supervision level, and prior misdemeanors were the best predictors of absconding (Mayzer et al., 2004). However, Mayzer et al.’s findings are on probation rather than pretrial release.

Because of the scarcity of risk assessment and management research on pretrial populations, judges and pretrial officers often employ discretion in the decision-making process. Use of discretion in the federal system without objective measures has had alarming consequences. Studies have suggested that consideration of extralegal factors by judges can lead to regional variation and can interfere with accurate decision-making in an offender’s disposition (Wu, 2010). Past research has indicated that race and ethnicity have an effect on defendants’ dispositions pretrial (Hartford, Carey, & Mendonca, 2007; Petee, 1994; Pinals, Packer, Fisher, & Roy-Bujnowski, 2004). For instance, Freiburger, Marcum, and Pierce (2010) examined the relationship between ethnicity and sentencing outcome in a sample of pretrial drug offenders and found that Black defendants were less likely to receive pretrial release than defendants of other ethnic backgrounds. In addition, Pinals et al. (2004) found that Blacks were significantly more likely than Whites and Latinos to be referred for an inpatient evaluation in a strict-security facility regardless of level of risk.

Some jurisdictions have even included mental health history and prior treatment in their release decisions. A study by Feder (1994) showed that psychiatric history of a defendant had a significant impact on the pretrial release decision. Disconcertingly, researchers are unsure as to the nature of the relationship between mental health and risk of future offending (see Feder, 1991a, 1991b; Grann, Danesh, & Fazel, 2008; Lovell, Gagliardi, & Peterson, 2002; Olver, Stockdale & Wormith, 2011). It has even been suggested that there may be no unique differences in terms of risk of reoffending between those offenders who have received psychiatric treatment and those who have not. Feder (1991a) examined the community adjustment of offenders who were given psychiatric treatment during incarceration prior to their release. Results showed high rates of rearrest during release. However, it was suggested that recidivism was more closely tied to variables associated with criminality (e.g., education, age, occupation, criminal history) than to mental illness (Feder, 1991b). Lovell et al. (2002) reviewed archival data on mentally ill offenders released from prison in 1996 and 1997. While no causal inferences between mental illness and recidivism were made, the authors identified nonclinical risk factors that predicted future offense, and these risk factors were found to be similar to those identified in widely used risk assessment instruments (Lovell et al., 2002).

In summary, more research is needed to clarify the relationship between ethnicity and supervision outcome as well as psychiatric treatment history and risk of reoffending. Despite the uncertainty of whether these factors are good predictors of a defendant failing or succeeding pretrial supervision, judges and other legal decision makers have included such factors in their determinations.

To maximize the accuracy in decisions of risk and match offenders with appropriate levels of supervision, corrections agencies must take the important safeguard of objectively and empirically identifying variables that predict program attrition (Olver et al., 2011). Olver et al. (2011) identified a number of common predictors of attrition in their meta-analysis. These predictors were employment, age, ethnicity, education, criminal history (prior offenses, prior violent offenses, prior nonviolent offenses, and institutional offenses), substance abuse (SA), and mental health concerns. A majority of these selected variables have been identified in previous studies on offender supervision outcome (Demuth & Steffensmeier, 2004; Gavazzi et al., 2005; Mayzer et al., 2004; Schwalbe, Fraser, Day, & Cooley, 2006).

Objectives

Using the risk factors of recidivism and attrition identified in Olver et al.’s (2011) meta-analysis and those variables included in the risk assessment instrument piloted by Pretrial Services (Lowenkamp et al., 2008) as a foundation, we have selected groups of variables taken from a sample of pretrial defendants who have undergone supervised release: current SA problem, prior felony arrest, prior drug conviction, prior violent felony, pending felony, prior FTA, prior absconding, prior escapes, psychiatric treatment history, age, gender, current employment, length of current residency, education level, and ethnicity. Four of our chosen variables overlap with the risk and need variables identified by Lowenkamp et al. (2008) for developing their pretrial risk assessment instrument: prior FTAs, employment status, length of residency, and current drug problems. Seven of our chosen variables also overlap with the risk factors for attrition identified by Olver et al. (2011): ethnicity, employment status, education level, age, prior offenses, prior violent offenses, and SA problems. This study seeks to determine the variables that have the greatest relationship with a defendant failing or succeeding supervision, and the moderating effects these variables have on Black, Latino, Asian, and White ethnic groups. We hypothesize that defendants with a current SA problem, one or more prior felony arrests, prior drug convictions, prior violent felonies, and pending felonies are at a higher risk of failing supervision. Regarding past behavior within the corrections community, we hypothesize that defendants with histories of escapes, prior absconding, and prior FTAs are also at a higher risk of failing supervision (Lowenkamp et al., 2008; see Mayzer et al., 2004). As it has been evidenced that psychiatric treatment history is not a unique predictor for future criminal behavior among offenders (Feder, 1991a, 1991b; Grann et al., 2008; Lovell et al., 2002; Olver et al., 2011), we hypothesize that a history of psychiatric treatment will not be an important factor for predicting supervision outcome. Finally, we hypothesize that gender, age, education, employment, and residency are significant predictors of supervision outcome. Specifically, we hypothesize that being a male defendant with no more than a high school education, who is unemployed, has had a residency of less than 100 months, and is a minority poses a significantly higher risk of failing supervision (Fujii et al., 2005; Lowenkamp et al., 2008; see Olver et al., 2011; Singh et al., 2011).

Method

Participants

Background information from a total of 4,449 (4,260 male, 724 female, M age = 32.18) pretrial defendants was gathered from a single Midwestern district. Participants in the sample were White non-Latino (38.5%), White Latino (36.1%), Black non-Latino (20.6%), Native American (3.8%), Asian (0.8%), and Black Latino (0.2%). In addition, defendants currently were charged with a variety of crimes (61.1% drug, 9.2% immigration law, 10.1% theft/fraud, 11.1% firearms, 5.5% violent, 2.9% other). Similarly diverse criminal histories were also evident (81.2% one or more prior felony convictions, 72.6% one or more prior violent felony convictions, 59% at least one FTA, 96.2% no prior escapes). Additional variables analyzed included demographic variables, residency and employment, and criminal history. Univariate statistics used for these demographic variables are shown in Table 1.

Table 1:

Summary of Demographic Variables of Participants

	Total	White	Black	Latino
Outcome type
Success	87.5%	93%	82.9%	70.5%
Failure	12.5%	7%	17.1%	29.5%
Prior felony arrests
None	32.5%	31%	15.5%	41%
One or more	67.5%	69%	84.5%	59%
Prior drug convictions
No drug convictions	55.1%	53.5%	40.9%	62%
One or more convictions	44.9%	46.5%	59.1%	38%
Prior violent felonies
No violent felony arrests	72.4%	75.7%	53.1%	80.2%
One or more arrests	27.6%	24.3%	46.9%	19.8%
Pending felonies
One or more	81.2%	76.5%	76.9%	87.4%
None	18.8%	23.5%	23.1%	12.6%
Age (M)	32.19	35.56	30.02	30.15
Gender
Male	85.5%	78.3%	90%	91%
Female	14.5%	21.7%	10%	9%
Employed at initial appearance
Not employed	57%	52.7%	62.4%	56.9%
Employed	43%	47.3%	37.6%	43.1%
Time of residence in area
100 months or less	52.6%	40%	45.6%	29%
More than 100 months	46.3%	60%	54.4%	71%
Prior psychiatric treatment
Yes	7.5%	10.7%	5.2%	3.1%
No	92.5%	89.3%	94.8%	96.9%
Substance abuse problem
Yes	74.9%	22%	20.2%	35.4%
No	24.4%	78%	79.8%	64.6%
Education
High school diploma	63.3%	78.6%	60.5%	36.5%
No high school diploma	36.7%	21.4%	39.5%	63.5%
Prior FTA
No prior FTAs	58.8%	59.3%	32.8%	72.8%
One or more prior FTAs	41.2%	40.7%	67.2%	27.2%
Prior absconding
No absconding	95.6%	95.6%	93.5%	96.6%
One or more	4.4%	4.4%	6.5%	3.4%
Prior escapes
No escapes	96.2%	96.3%	94.3%	97.8%
One or more escapes	3.8%	3.7%	5.7%	2.2%

Note: FTA = failure to appear.

Procedures

Access to information was granted by U.S. Pretrial Services via their Probation and Pretrial Services Automation and Case Tracking System (PACTS). Federal probation and pretrial services continue to develop a comprehensive system for predicting defendant outcomes and for allocating among jurisdictions more effective strategies for reducing offender recidivism. As a result, officers regularly assemble all offender information in a comprehensive database after collecting the information from defendants through pretrial and presentence interviews and criminal background investigations. PACTS is a federal database developed in 2001 for research and record keeping purposes that contains criminal, behavioral, and demographic records of all persons charged with offenses in the federal courts. Information is entered into the database annually by pretrial caseworkers, who gather the information of each defendant in their caseload via background investigations, which include pretrial interviews, bail reports, contacting treatment providers and collaterals, as well as collaboration from other institutions familiar with the defendant, such as penitentiaries, treatment providers, and federal agencies (FBI, DEA, ATF, etc.). During pretrial interviews, pretrial officers meet with arrestees and gather information such as age, gender, immigration history, family ties, family history, education, current employment, date of last employment, current income, current residence status, previous residency, prior drug use, current drug use, prior hospitalization, prior psychiatric hospitalization, current mental and physical health, criminal history, history of incarceration, probation history, and type of charged offense. The information gathered from investigations is used to assist the federal courts in determining conditions of release.

The data set used for this analysis was extracted from PACTS and consists of all persons charged with criminal offenses in federal court between FY 2001 and FY 2008 who were processed by the U.S. Pretrial Services Office of a Midwestern district. After reviewing the ethical guidelines with pretrial officers and receiving institutional review board (IRB) approval, a formal written proposal was sent to Pretrial Services in a Midwestern district that stated the intentions of the study and requested access to defendant background information. After reviewing the request, Pretrial Services granted access to a data set extracted from PACTS, and variable information from the data set was recoded and analyzed. Certain variables were withheld from the data set to keep anonymity of defendants, such as names, addresses, and social security numbers.

Analysis

Data were analyzed using binary logistic regression as our independent variable (success or failure in the program) was a binary variable for which we were trying to determine the relevant predictors. In addition, the following variables were used and coded as the following: supervision outcome (0 = failure, 1 = success), prior felony arrests (0 = no prior arrests, 1 = one or more prior arrests), prior drug convictions (0 = no prior convictions, 1 = one or more prior convictions), prior violent felonies (0 = none, 1 = one or more), pending felonies (0 = none, 1 = one or more), age (numerical), gender (0 = male, 1 = female), employed at initial appearance (0 = employed, 1 = not employed), time of residence in area (0 = 100 months or less, 1 = more than 100 months), prior psychiatric treatment (0 = no, 1 = yes), SA problem (0 = no, 1 = yes), education (0 = high school diploma, 1 = no high school diploma), ethnicity (0 = White, 1 = non-White), prior FTA (0 = no prior FTAs, 1 = one or more prior FTAs), prior absconding (0 = no prior absconding, 1 = prior absconding), and prior escapes (0 = no prior escapes, 1 = one or more prior escapes).

Results

Binary logistic regressions were used to examine four groups in our sample: all defendants, White defendants only, Black defendants only, and Latino defendants only. In our entire sample, 12.5% of the defendants failed supervision and 87.5% were successful.

For all defendants, a logistic regression examined success type against each of the 15 independent variables (see Table 1 for a complete list of the variables) and was significant, X²(15) = 118.868, p < .001. Specifically, age (X² = 72.453), gender (X² = 18.194), SA problem (X² = 13.476), ethnicity (X² = 62.933), felony arrest (X² = 9.170), drug conviction (X² = 5.773), violent felony history (X² = 9.016), prior FTA (X² = 28.405), prior escapes (X² = 6.155), and education level (X² = 12.408) were significant contributors to the model (see Table 2 for beta and Odds Ratios). Specifically, being of a younger age, being male, having an SA problem, being of ethnic minority, having at least one prior FTA, having one or more prior escapes, and failing to graduate from high school all increased the chances of a supervision failure. Follow-up chi-square statistics indicated that there were significant differences between Whites and other ethnic groups. Success rates for White defendants were significantly higher, with a 92.7% chance of success (X² = 53.994) compared to 71.2% for Blacks (X² = 15.748) and 83.2% for Latinos (X² = 37.851).

Table 2:

Summary of Significance Values of Contributing Variables (Success or Failure)

	B	Wald	Sig.	Exp(B)
Felony arrest	−.009	.001	.972	.991
Drug conviction	−.222	1.143	.285	.801
Violent felony	−.092	.200	.655	.912
Pending felonies	.075	.095	.758	1.078
Age	.018	4.287	.038	1.018
Gender	.708	9.196	.002	2.030
Employed	−.263	2.668	.102	.769
Residence in area	.001	1.247	.264	1.001
Prior psychiatric treatment	−.045	.022	.883	.956
SA problem	−.422	4.337	.037	.656
Education level	−.509	9.688	.002	.601
Ethnicity	−.844	23.586	.000	.430
FTA	−.411	5.558	.018	.663
Prior absconding	−.138	.065	.798	.871
Prior escapes	−.911	4.046	.044	.402

Note: FTA = failure to appear; SA = substance abuse.

For Whites, 7% failed supervision and 93% were successful. A logistic regression examined success type against each variable and was significant (X² = 53.994, p < .001). It was revealed that the significant contributing factors for success type for Whites were gender (X² = 8.410), SA problem (X² = 16.380), education level (X² = 5.470), prior felony arrests (X² = 12.410), prior drug convictions (X² = 5.133), prior violent felony convictions (X² = 6.041), prior FTAs (X² = 21.010), and prior escapes (X² = 7.277). Specifically, being male, being unemployed, having an SA problem, and having at least one prior FTA increased the likelihood of failure. For Blacks, 17.1% failed supervision and 82.9% were successful. Logistic regression results were also significant (X² = 15.748, p < .002). Significant contributing factors for Blacks were age (X² = 37.569), SA problem (X² = 4.032), prior felony arrest (X² = 5.448), prior absconding (X² = 4.885), and prior FTAs (X² = 14.448). Specifically, being of a younger age, having at least one prior FTA, and having a history of absconding increased the likelihood of failure. For Latinos, 29.5% failed supervision and 70.5% were successful. No significant contributing factors were found for Latinos (X² = .329, p > .300).

Discussion

The present study sought to identify the most predictive factors, within a sample of pretrial defendants, for a defendant failing or succeeding supervision. Logistic analyses revealed an overall model and found that age, gender, SA problem, ethnicity, a history of FTAs, prior escapes, and education level predicted supervision failure. Overall, results from the logistic regression examining success type for all defendants indicated that younger age, being male, having an SA problem, being of ethnic minority, having at least one prior FTA, having one or more prior escapes, and failing to graduate from high school all increased the chances of a supervision failure.

Results were mixed given our original hypotheses. Our hypothesis that criminal history variables would have an impact on supervision completion was not supported at all. None of the four criminal history variables were predictive of supervision failure in the overall model. Having at least one prior felony arrest, a drug conviction, violent felony arrest, and/or a pending felony did not significantly affect the likelihood of a defendant failing supervision.

There was only partial support for past corrections behavior predicting failure. Contrary to our hypothesis that defendants with a history of absconding would be more likely to fail supervision, prior absconding did not significantly contribute to the logistic regression model examining success type for all defendants in our sample. Still, as expected, having a history of prior escapes and FTAs did significantly contribute to the model. Having at least one prior escape and/or prior FTA significantly decreased the likelihood of success.

Regarding SA and psychiatric treatment history, existence of an SA problem was predictive of supervision failure but not previous psychiatric treatment. Our hypothesis that demographic variables (gender, age, employment, education, and residency) outside of ethnicity would have an impact on likelihood of failing was partially supported. Results clearly indicated that age, ethnicity, gender, and education all had a significant impact on supervision success or failure. However, contrary to our hypothesis, residency and employment were not significant predictors.

It is routinely mentioned that the best predictor of future behavior is past behavior, so it comes as no surprise that a history of FTAs or escaping suggested supervision failure. However, the fact that criminal history outside of these variables had nothing to do with success or failure on supervision was not expected. Overall, the sample was split relatively evenly on prior drug convictions. However, most of the current sample had a previous felony arrest (67.5%). Few had prior violent felonies (27.6%), but over 80% of the current sample had a pending felony arrest at this time. Although none of the single variables predicted supervision completion, it could be that an accumulation of factors related to a more severe criminal history could be predictive of supervision failure. Furthermore, a linear relationship may exist between some of these variables and supervision completion so that, while the presence of a felony arrest is not indicative of supervision failure, the more felony arrests a person experiences, the more likely he is to fail supervision. It also could be that because many pretrial defendants enter the federal system because of felony charges, certain pending felonies are given more weight than others in the pretrial decision. The high base rate of pending felonies may only be indicative of crimes that are less significant in predicting risk (e.g., nonviolent felony charges).

A current SA problem was related to supervision failure but not psychiatric history. These findings could be suggestive of differences in the base rates of either variable, the variables themselves, or the fact that current versus historical differences were inherent in the measurement of these two variables. It is not surprising that the more proximal variable, current SA, was significant in the model, and past research has shown that SA is a significant risk predictor (Lowenkamp et al., 2008; Olver et al., 2011). It was also expected that psychiatric treatment history was not predictive; however, it should be cautioned that the measurement of this variable does not specify the length of treatment received by the offender or the severity of the offender’s psychiatric issues.

The results of the demographic variables were mixed in terms of our original hypothesis. The relative strength of these factors to predict program success was surprising, whereas their individual contributions were not. Age and gender produced two of the three highest Odd Ratios, with men twice as likely to fail supervision as women. Women typically receive more lenient release decisions, as women are viewed as less dangerous and less of a threat to the community (Demuth & Steffensmeier, 2004). In addition, younger defendants were more likely to fail supervision. This finding suggests a linear relationship between age and pretrial supervision outcome. Previous research on pretrial defendants has shown that younger defendants (aged 32 or younger) are more likely to fail supervision, and these results have been incorporated in the development of pretrial screening instruments (Lowenkamp et al., 2008). In keeping with our results, it appears that it would be appropriate to include age to aid in decisions of release.

However, our results show no significant effect between age and White defendants (see Table 3). This could be due to differences in average age between the groups, with the mean age of White defendants (M = 35.56) being higher than that of Black defendants (M = 30.02). On the contrary, our results could suggest that some screening instruments that include a specific age as a cutoff but do not incorporate ethnicity (which consists of a large number of screening instruments put in practice today by officers and legal decision makers) are not entirely sufficient in determining the level of danger some groups of defendants pose as compared to others. Furthermore, previous research on age related to release decisions has shown that there is an inverted-U relationship with the likelihood of detention. Among the samples of defendants in studies carried out in the past, younger and older defendants are less likely to be detained (see Demuth & Steffensmeier, 2004). This could be because younger defendants are perceived as less dangerous by decision makers despite a higher likelihood of supervision failure, which would appear inappropriate according to our results. Taken together with those less formally educated and with more fluid residence status, a picture of a less educated, younger male offender with fewer community ties (as suggestive of the residence status) emerges as suggestive of a greater risk for failing to complete pretrial release successfully.

Table 3:

Summary of Significance Values of Contributing Variables (White Defendants)

	B	Wald	Sig.	Exp(B)
Felony arrest	−.363	3.219	.073	.483
Drug conviction	−.011	.001	.974	.989
Violent felony	−.151	.211	.646	.860
Pending felonies	.224	.412	.521	1.251
Age	−.006	.169	.681	.994
Gender	.876	4.525	.033	2.400
Employed	−.774	6.895	.009	.461
Residence in area	.236	.662	.416	1.266
Prior psychiatric treatment	−.728	3.219	.073	.483
SA problem	−1.161	6.492	.011	.313
Education level	.346	.973	.324	1.414
Ethnicity	—	—	—	—
FTA	−.716	6.111	.013	.489
Prior absconding	19.034	.000	.998	1.847
Prior escapes	−1.153	3.376	.066	.316

Note: FTA = failure to appear.

Our hypothesis that there would be differences in factors predicting outcomes among ethnic groups was supported. Both White defendants and Black defendants produced a significant pattern of factors related to supervision success, whereas Latino defendants did not (see Tables 3, 4 and 5). White and Black defendants shared SA and one or more FTAs as significant predictors. The fact that SA problem and prior FTAs were significant predictors in our results for White and Black defendants is in congruence with the pretrial screening instrument piloted in Lowenkamp et al.’s (2008) article, which includes both variables in its risk calculation. Despite this similarity, there were clear differences between White and Black defendants. Gender and employment predicted success for White defendants but not for Black defendants. A prior absconding predicted failure for Black defendants but not White defendants. In addition, it is surprising that Latino defendants did not display any significant pattern, even when considering they were the second largest group in our sample following White defendants. A plausible reason for this result might concern the unique social and economic differences that exist between these three ethnic groups. A number of differences can be considered, such as immigration status, with many Latino defendants in this sample being immigrants from a foreign country. Specifically, over 97% of the Black and White participants in this sample were U.S. citizens, compared to only 23.1% of the Latino participants. Although immigration status by itself does not indicate a higher likelihood of criminal involvement, immigrants are more prone to arrests and incarceration than are citizens (Hagan, 1999). Jackson (1997) examined the profiles of White, Latino, and Black inmates housed in a federal correctional system and found considerable differences in criminal and socioeconomic background between the groups, especially between Blacks and Latinos. Specifically, Latino defendants had significantly lower education levels than the other two groups. The Latino group in our study was not spread across certain variables in a similar manner to the other ethnic groups. Similar to Jackson’s (1997) finding, the education levels for Latino defendants in our sample were fairly low, at 63.5% for no high school diploma. This is much larger a percentage than what was shown for Black defendants (39.5%) and White defendants (21.4%).

Table 4:

Summary of Significance Values of Contributing Variables (Black Defendants)

	B	Wald	Sig.	Exp(B)
Felony arrests	−.850	2.610	.106	.427
Drug conviction	−.093	.068	.794	.911
Violent felony	.037	.012	.914	1.038
Pending felonies	.220	.212	.645	1.246
Age	.065	8.819	.003	1.067
Gender	.263	.308	.579	1.300
Employed	.226	.536	.464	1.254
Residence in area	−.031	.010	.920	.969
Prior psychiatric treatment	.419	.281	.596	1.520
SA problem	−.280	.490	.484	.756
Education level	−.629	2.404	.121	.533
Ethnicity	—	—	—	—
FTA	−.913	6.365	.012	.401
Prior absconding	−1.612	3.916	.048	.200
Prior escapes	.243	.06	.813	1.275

Note: FTA = failure to appear; SA = substance abuse.

Table 5:

Summary of Significance Values of Contributing Variables (Latino Defendants)

	B	Wald	Sig.	Exp(B)
Felony arrests	.556	1.318	.251	1.743
Drug convictions	−.972	3.726	.054	.378
Violent felonies	−.220	.174	.677	.802
Pending felonies	.819	1.292	.256	2.268
Age	.015	.715	.398	1.016
Gender	1.060	5.338	.021	2.887
Employed	.017	.003	.959	1.017
Residence in area	.197	.271	.602	1.218
Prior psychiatric treatment	.724	.667	.414	2.063
SA problem	−.092	.072	.789	.912
Education level	.526	.945	.331	1.691
Ethnicity	—	—	—	—
Prior FTAs	.315	.572	.449	1.370
Prior absconding	−.328	.060	.806	.721
Prior escapes	−1.885	1.630	.202	.152

Note: FTA = failure to appear; SA = substance abuse.

This discrepancy still does not explain for the lack of predictability among Latino defendants, as education level was not a significant contributor for White or Black defendants independently, unless education is a unique contributor to the Latino population. Also worth taking note of is the high percentage of Latinos who had resided at their current location for more than 100 months (71%). This is a considerably high base rate for Residency in Area compared to the other defendants in our sample. Still, this variable was not found to be significant in our entire sample. In addition, the Latino defendants in our study had a higher failure rate in supervision outcome as compared to Whites and Blacks (see Table 1). This calls into question the predictive accuracy of a “one size fits all” model of risk assessment for ethnic groups, reflected in the higher failure rates and lack of significant results in predicting outcomes for Latino defendants.

Demuth (2003) examined data on the processing of felony defendants at the pretrial release stage. After examining racial/ethnic and criminal history variables, the author found that Hispanics were more likely than Whites to be detained. This was due to their inability to pay higher bail amounts compared to Whites, rather than racially differential decision-making (Demuth, 2003). Specifically, the author noted a “triple disadvantage” at the pretrial release stage for Hispanic defendants: that they are more likely to have to pay bail, have the highest bail amounts, and are the group least able to pay bail (Demuth, 2003). What Demuth’s (2003) study brings to mind is that there are other extralegal factors that can influence judicial decision-making and outcomes at multiple stages of the criminal justice system. It could be the case that more extralegal characteristics influence outcomes of Latino defendants than other ethnic groups at different stages of the pretrial process, thus leading to variability. For example, they could be given harsher restrictions for release than Whites or Blacks, resulting in higher failure rates. Of particular note is the near significance of drug convictions for Latino defendants (p = .054). This should be noted, as research has shown that Latino groups are more often charged and incarcerated for drug crimes than are other ethnic groups and are particularly more likely to have been imprisoned for drug trafficking (McGovern, Demuth, & Jacoby, 2010). Nevertheless, we can only speculate on these possibilities given the limited scope of our available data. More research needs to be done on the types of variables that can differentially impact release outcomes for Latino defendants.

There were also clear differences between White and Black defendants in our sample. Although employment is factored into some screening instruments (see Lowenkamp et al., 2008), employment was found to be an insignificant contributor for Black defendants (see Table 4). It is arguable that indiscriminate screening of all ethnic groups as opposed to each ethnic group as unique from one another can lead to misrepresentation and inaccurate decision-making. The fact that certain variables were pertinent for Black defendants but not Whites and vice versa in our results only bolsters this reasoning.

Implications for Future Research on Risk Assessment

Recently, questions have been raised in the corrections community as to whether decisions regarding a defendant’s release or detainment in the pretrial process are based on accurate assessments of a defendant’s risk of failing to appear for court or reoffending. Demuth and Steffensmeier (2004) analyzed pretrial release data on a sample of White, Black, and Latino defendants and found that White females were most likely to be granted pretrial release, whereas Black and Latino defendants were more likely to be detained than White defendants. In addition, they found that male Latinos were most likely to be detained of all the groups. In another study, Freiburger et al. (2010) found that being Black increased the chances of being detained. The results of both of these studies are similar to our findings that gender and ethnicity have a moderating effect on supervision outcome, particularly that being a non-White male defendant significantly increases the likelihood of failing supervision. As a result, one could argue that appropriate decisions are being made in regard to White and Black defendants but not Latino defendants according to our results. Perhaps there are other factors uniquely related to the Latino population that are more predictive of their supervision outcome that were not included in our model. Or it could be that certain predictor variables included in our model that are also considered by officers and judges in the decision-making process are not valid predictors for Latino defendants but only reliable for the White and Black populations. More research is needed to answer the question of how the Latino population is unique from other populations in the corrections community.

The current study goes further and suggests that specific factors perform differently in these decisions across ethnicities and suggest that a one-size-fits-all model to risk assessment may not be appropriate in making pretrial release decisions. In order to improve decision making in these contexts, pretrial officers may be wise to look at specific factors that are predictive of supervision failure within a given ethnic group. One instrument that has been utilized in certain districts is the Pretrial Assessment Tool (PAT). The PAT places defendants under a certain risk category according to their scores under itemized risk domains. For each domain, points are given to defendants depending on where they fall under scored and unscored items. A total score is calculated and it is then determined which risk category they fall under (Latessa et al., 2010). The PAT does not include race or ethnicity as a function of risk. In some districts, these scores have been used in pretrial recommendations. Another limitation is that studies examining race and ethnicity have clearly indicated that group differences are frequently a result of socioeconomic differences and not ethnicity. However, the results held when measures associated with socioeconomic status, such as education, residence stability, and employment history, were included. Future studies could include income as a specific variable, but socioeconomic status is not simply a matter of household income. Nonetheless, results should be considered within the context of the wider literature on criminal justice outcomes and ethnicity.

As the results of our study and studies such as Heilbrun, Douglas, and Kento (2009) demonstrate, there may be different pathways to violations for some groups relative to others. This holds implications for actuarial judgments. The stability of these judgments across samples can vary as a function of the numerous features of a study (Heilbrun et al., 2009). The results of this study suggest ethnic differences in outcome of pretrial release and suggest that the factors considered when assessing release decisions differ between ethnicities. Results are not definitive about whether ethnic disparities are due to individual differences among ethnic groups or systemic factors that impact these groups. Perhaps our models for assessing risk across a variety of situations are not as applicable on an individual level in situations other than ethnicity.

Limitations

Still, our study has its limitations. The lack of size and unequal distribution in our samples may have accounted for the lack of significance values for some variables relative to others. Absconding was found to be insignificant in our model examining all defendants. However, it should be pointed out that only 4.4% of our sample had absconded in the past, while 95% had not. Another limitation is that because our variables of interest were dichotomous and already subject to interpretation (as they were entered into the PACTS database by officers before we had access), we were limited in our ability to determine the meaning behind our results. For example, we were unable to interpret the severity or type of psychiatric concern or length of psychiatric treatment for our psychiatric treatment history variable. Additionally, we were unable to interpret the types of felony arrests or pending felonies (violent or nonviolent) defendants had at the time of the pretrial interview.

Conclusion

There is very little research in the risk assessment literature on ethnicity and how it plays a role in the corrections community. Our results show that there may be different pathways to outcome for some ethnic groups relative to others. These results hold implications for actuarial judgments of risk, as certain risk factors may be less applicable for some groups of offenders relative to others, leading to questions about the generalizability and predictive validity of risk assessment instruments in the criminal justice system. More research is needed to understand the possible moderating role ethnicity plays in risk assessment and how this factor determines the predictive accuracy of some risk assessment instruments that are being used today in the corrections community.

Footnotes

Authors’ Note:

We would like to thank the staff at Pretrial Services for generously allowing us access to their defendant data and for giving us the opportunity to use their resources to conduct our study. Please note that the U.S. Pretrial Services Office for the District of Nebraska or the U.S. District Court do not endorse any statements in this article or the specific findings.

Matthew Fennessy graduated from Creighton University in Omaha, Nebraska with a Bachelor of Arts degree in psychology. He is currently working toward a Master of Arts in the Industrial/Organizational Psychology Program at Hofstra University in Hempstead, New York.

Matthew T. Huss, PhD, MLS, is a professor at Creighton University in Omaha, Nebraska. He also is a graduate of the University of Nebraska Law and Psychology and Clinical Psychology training programs. He is the author of over 60 different scholarly publications and a textbook on forensic clinical psychology, Forensic Psychology: Research, Practice, and Applications. His primary research interests focus on the assessment of violence risk, domestic violence, psychopathy, and sexual violence. In addition, he has significant interests in training and education in law and psychology.

References

Austin

(2006). How much risk can we take? The misuse of risk assessment in corrections. Federal Probation, 70, 58-63.

Bonta

Wallace-Capretta

Rooney

(2000). A quasi-experimental evaluation of an intensive rehabilitation supervision program. Criminal Justice and Behavior, 27, 312-329.

Brennan

Dieterich

Ehret

(2009). Evaluating the predictive validity of the COMPAS risk and needs assessment system. Criminal Justice and Behavior, 36, 21-40.

Demuth

(2003). Racial and ethnic differences in pretrial release decisions and outcomes: A comparison of Hispanic, black, and white felony arrestees. Criminology, 41, 873-907.

Demuth

Steffensmeier

(2004). The impact of gender and race-ethnicity in the pretrial release process. Social Problems, 51, 222-242.

Edens

J. F.

Cahill

M. A.

(2007). Psychopathy in adolescence and criminal recidivism in young adulthood: Longitudinal results from a multiethnic sample of youthful offenders. Assessment, 14, 57-64.

Fass

T. L.

Heilbrun

Dematteo

Fretz

(2008). The LSI-R and the COMPAS: Validation data on two risk-needs tools. Criminal Justice and Behavior, 35, 1095-1108.

Feder

(1991a). A comparison of the community adjustment of mentally ill offenders with those from the general prison population: An 18-month follow-up. Law and Human Behavior, 15, 477-493.

Feder

(1991b). A profile of mentally ill offenders and their adjustment in the community. Journal of Psychiatry and the Law, 19, 79-98.

10.

Feder

(1994). Psychiatric hospitalization history and parole decisions. Law and Human Behavior, 18, 395-410.

11.

Freiburger

Marcum

Pierce

(2010). The impact of race on the pretrial decision. American Journal of Criminal Justice, 35, 76-86.

12.

Fujii

D. M.

Tokioka

A. B.

Lichton

A. I.

Hishinuma

(2005). Ethnic differences in prediction of violence risk with the HCR-20 among psychiatric inpatients. Psychiatric Services, 56, 711-716.

13.

Gavazzi

Yarcheck

Lim

(2005). Ethnicity, gender, and global risk indicators in the lives of status offenders coming to the attention of the juvenile court. International Journal of Offender Therapy and Comparative Criminology, 49, 696-710.

14.

Grann

Danesh

Fazel

(2008). The association between psychiatric diagnosis and violent re-offending in adult offenders in the community. BMC Psychiatry, 8, ArtID 92.

15.

Hagan

J. A.

(1999). Sociological criminology and the mythology of Hispanic immigration and crime. Social Problems, 46, 617-632.

16.

Hanley

(2006). Appropriate services: Examining the case classification principle. Journal of Offender Rehabilitation, 42, 1-22.

17.

Hartford

Carey

Mendonca

(2007). Pretrial court diversion of people with mental illness. The Journal of Behavioral Health Services & Research, 34, 198-205.

18.

Heilbrun

Douglas

Kento

(2009). Violence risk assessment: Core controversies. In Skeem

Douglas

Lilienfeld

(Eds.), Psychological science in the courtroom: Consensus and controversy (pp. 333-357). London: The Guilford Press.

19.

Holsinger

Lowenkamp

Latessa

(2006). Exploring the validity of the Level of Service Inventory–Revised with Native American offenders. Journal of Criminal Justice, 34, 331-337.

20.

Jackson

(1997). Differences in the background and criminal justice characteristics of young Black, White and Latino male federal prison inmates. Journal of Black Studies, 27, 494-509.

21.

Långström

(2004). Accuracy of actuarial procedures for assessment of sexual offender recidivism risk may vary across ethnicity. Sexual Abuse: Journal of Research and Treatment, 16, 107-120.

22.

Latessa

E. J.

Lemke

Makarios

Smith

Lowenkamp

C. T.

(2010). The creation and validation of the Ohio Risk Assessment System (ORAS). Federal Probation, 74, 16-22.

23.

Lovell

Gagliardi

Peterson

(2002). Recidivism and use of services among persons with mental illness after release from prison. Psychiatric Services, 53, 1290-1296.

24.

Lowenkamp

Lemke

Latessa

(2008). The development and validation of a pretrial screening tool. Federal Probation, 72, 2-9.

25.

Maxwell

(1999). Examining the congruence between predictors of ROR and failures to appear. Journal of Criminal Justice, 27, 127-141.

26.

Mayzer

Gray

Maxwell

(2004). Probation absconders: A unique risk group? Journal of Criminal Justice, 32, 137-150.

27.

McGovern

Demuth

Jacoby

(2010). Racial and ethnic recidivism risks: A comparison of postincarceration, and reincarceration among White, Black, and Hispanic releases. The Prison Journal, 89, 309-327.

28.

National Institute of Corrections. (2003). Offender assessment. Topics in community corrections annual issue. Washington, DC: U.S. Department of Justice.

29.

Olver

Stockdale

Wormith

(2011). A meta-analysis of predictors of offender treatment attrition and its relationship to recidivism. Journal of Consulting and Clinical Psychology, 79, 6-21.

30.

Petee

(1994). Recommended for release on recognizance: Factors affecting pretrial release recommendations. The Journal of Social Psychology, 134, 375-382.

31.

Pinals

Packer

Fisher

Roy-Bujnowski

(2004). Relationship between race and ethnicity and forensic clinical triage dispositions. Psychiatric Services, 55, 873-878.

32.

Schwalbe

Fraser

Day

Cooley

(2006). Classifying juvenile offenders according to risk of recidivism: Predictive validity, race/ethnicity, and gender. Criminal Justice and Behavior, 33, 305-324.

33.

Singh

Grann

Fazel

(2011). A comparative study of violence risk assessment tools: A systematic review and meta-regression analysis of 68 studies involving 25,980 participants. Clinical Psychology Review, 31, 499-513.

34.

Whiteacre

(2001). Testing the Level of Service Inventory–Revised (LSI-R) for racial/ethnic bias. Criminal Justice Policy Review, 17, 330-342.

35.

(2010). Immigration, race/ethnicity, and differential sentencing: An examination of the effects of citizenship status on sentencing outcomes in federal courts. Dissertation Abstracts International: Section A. Humanities and Social Sciences, 70(7-A), 2743.