Introduction

The pretrial justice system, diversionary programs excluded, in the United States exists for a variety of reasons. These include the need to: a) to physically detain (i.e., jail) defendants to ensure that they stand trial for accused crimes and/or to prevent them from causing further harm and b) to release lower-risk individuals back into society before trial to offset the public burdens associated with jail administration [1]. Approximately sixty-three percent of jail inmates in the US are pretrial detainees [2]. Of those, most will be released prior to their trial via some form of legal bond after a court sets bail (i.e., a dollar amount necessary to facilitate release prior to trial). The bond itself is a legal matter and is intended to ensure that the defendant shall appear in court to stand trial on a future date. Such bonds may be backed by a financial or non-financial guarantee to the court. Financially secured bonds may be backed by cash (or sometimes property) delivered by or on behalf of the defendant or via a privately obtained second-party surety (e.g., a commercial bail bondsman or an attorney). Non-financial bonds may include some form of government funded supervision mechanism (e.g., a pretrial services unit) or simply be the defendant’s promise to appear (i.e., personal recognizance release). While each of these methods of pretrial release is commonplace, little is known regarding their efficacy in terms of preventing failure to appear in court.

Commercial bail bonds, a primary mechanism of pretrial release in the United States, are currently mired in controversy, as they have been at times in the past. In the past few years, increased attention to the pretrial release process, the effects of financially backed release mechanisms on the truly indigent, and potential alternatives have started to call the efficacy and future of financially secured bonds into question. While the aforementioned issues should be and need to be discussed, statements concerning the efficacy of commercial bail bonds are premature as only a few studies have evaluated the ability of commercial bail bonds to achieve the goal of pretrial release: to ensure a defendant released pretrial returns to all subsequent court dates. Of these studies, none have directly compared failure to appear rates for defendants released on commercial bonds to the most commonly applied non-financial alternative, pretrial services release bonds.

The purpose of this study was to provide a balanced empirical evaluation of the performance of varying pretrial release mechanisms in terms of the likelihood of court appearance leading to bond forfeiture for one large urban county in Texas (USA). The availability of a particular type of bond can vary across jurisdictions and can depend on the severity of the accused offense(s), the defendant’s criminal history, and can also be driven by judicial discretion. This reality makes inter-jurisdictional assessments of release mechanisms problematic and caution must be taken in any attempt to make comparisons regarding the performance or utility of one mechanism of pretrial release over another.

Pretrial release itself constitutes the time at which a defendant is released from jail prior to a court hearing where the determination of guilt will be adjudicated. However, there is considerable variation in how a defendant may experience the pretrial justice component of the justice process. In some cases, a defendant undergoes some form of supervised release, or monitoring, while other times he or she may be released without any condition other than a promise to appear in court. Unfortunately, it is this period of time within the criminal justice process that little is known in terms of evidence-based practices and “what works” scholarship. Very few studies, for example, have focused on how different release mechanisms impact defendant outcomes, such as failure-to-appear (FTA) in court, pretrial misconduct, bond forfeiture, recidivism, and so on. It is argued here that understanding the pretrial justice period is critical to enhancing public safety, optimizing the criminal justice process, reducing the burdens of incarceration, and ultimately saving taxpayer dollars.

The practical importance of assessments, such as the present study, is simple: operating a jail is incredibly expensive and jurisdictions are continually challenged by the need to reduce the financial burden of operating jails (i.e., reducing the number jail beds in use), while upholding the constitutional rights of defendants. For example, in Dallas County, Texas, which was home to over 2.4 million inhabitants as of 2014, the cost of jailing an inmate for one night is in excess of fifty (US) dollars. In Dallas County, the average daily jail population generally varies between five and six thousand inmates. Conservatively, and after defendants are released via a pretrial release mechanism, the county spends over $250,000 per day operating the jail, or about $91.3 million annually. Hidden social costs aside, it is easy to see that pretrial release is necessary to operating a criminal justice system; without it, the justice system could not function due to limited jail capacity.

This reality is particularly problematic given that between twenty-five and thirty-three percent of defendants FTA in court post-release from jail to a point at which the bond is forfeited by the court [3] [4], which is costly and burdensome. It is estimated that the procedural cost of a single FTA is roughly $1,700, based on a 2013 dollar value [5]. Beyond the monetary costs associated with FTA, defendants who miss court are not available to receive justice should they be found guilty.

Thus, it makes sense to consider the possibility of modifying justice procedures to maximize the probability of court appearance. There is not, however, a solid understanding of the factors that influence FTA and bond forfeiture or whether the release mechanism, in and of itself, may play a role in the probability of such an occurrence. To complicate matters further, there is considerable inter-jurisdictional variation in how some pretrial release mechanisms are practiced, thereby calling into question studies comparing the utility between different bond types while relying on nationally aggregated data [4], for example. The present study contributes to the body of literature about pretrial release by targeting the first step of the pretrial release process via an assessment of whether distinct pretrial release practices influence FTA in and of themselves in one major metropolitan US jurisdiction and among comparable defendants.

Methods

Relying on matched samples to isolate pretrial release mechanism effects

The effect of one mechanism over another was assessed via a quasi-experimental design (counterfactual) known as propensity score matching (PSM). PSM provides a useful means for transparently establishing equivalence between a treatment and a matched control group, for the purpose of assessing mean differences on an outcome variable, here whether FTA occurred. In short, both treatment and control groups are matched on a vector of conditions or attributes, ideally balancing the probability of receiving the treatment (i.e., release mechanism) or not, analogous to an experimental research design, but not a replacement for an experiment. The PSM method in which matched controls who are statistically indistinguishable from those who received the treatment are identified is a random process that can deem contextual differences between treated versus control groups ignorable [20], [21], [22]. PSM is well positioned to study differences in effectiveness between pretrial release mechanisms in terms of FTA for a situation where randomized trials are not feasible, but where processes specific to assignment to one mechanism over another can be statistically modeled using observed archival data and variables that likely confound the raw probability of a defendant being released by one mechanism over another [23]. Indeed, such methods are common within the criminological literature [24], [25].

Through this approach, the assumption of “conditional independence” must be satisfied. This occurs only when (1) a reasonable degree of certainty can be empirically established and, (2) that any underlying processes driving the likelihood of assignment to one treatment over another is ignorable once the propensity for treatment is adequately controlled via temporally preceding features correlated to receiving the treatment itself. By way of this assumption, propensity scores can be used to match “treated” defendants to similarly situated defendants who were equally likely to have received that same “treatment” (i.e., release mechanism), but did not. Propensity scores are thus defined as “the conditional probability of assignment to a particular treatment given a vector of observed covariates” [20] (p.41).

An important aspect of the current application of PSM lies in the fact that it involves an assessment of multiple treatments simultaneously. Though relatively rare, such an approach has been used in similar studies (e.g., see [4]; [26], [27]. In this study, the likelihood of FTA was explored across multiple mechanisms of pretrial release for defendants at an equivalent stage in the criminal justice process. Since there were multiple treatments under study, comparisons were made from one release-type to another for every possible combination of treatments, respectively. The goal was to end up with a conditioned estimate of the “treatment effect” for one release type over all other types separately. This was the difference in the mean probability for defendants failing to appear between two specific release mechanisms. Again, these comparisons were based on statistically matched defendants who were equally likely to have received the treatment. In the end, comparisons were made not between all defendants released by way of a particular method, but only between statistically matched pairs (i.e., apples-to-apples).

The matching procedure itself is carried out by identifying one or more equally likely untreated counterparts to every treated case in terms of estimated propensity score. For the analysis at hand, the 1-to-1 nearest neighbor algorithm was employed, without replacement [28]. Further, a condition for matching one case to another lies in the distance between a treated case and an untreated case on the score itself, know as the “caliper.” For present purposes, the caliper was conservatively set at 0.001.

Within the context of exploring differences in the effect of various pretrial release mechanisms on the probability of court appearance, PSM offers a transparent method for which problems surrounding the counterfactual are addressed. This protocol accounts for selection bias that otherwise may influence whether a defendant is released via one mechanism over another. Through a counterfactual design such as this, the experimental design is approximated with regard to the fundamental of equal probability of selection to the treatment, or control group, otherwise known as “balance.”

The first step in the process of creating comparable groups of defendants for each release mechanism combination was to estimate the propensity scores on a vector of covariates, all of which were outlined in the preceding section, by way of probit regression. Plainly, these propensity scores are the estimated probabilities of being released via one mechanism over another for the offense of record. As noted, each alternative comparison was assessed resulting in a total of 12 probit regression models for six different pairings of release mechanisms (e.g., commercial vs. cash, cash vs. commercial, pretrial services vs. commercial, commercial vs. pretrial services, and so on) and for each stratification (i.e., all defendants, felonies, misdemeanors, drug offenses, DWI offenses, and larceny offenses). These stratifications were included to understand the presence and nature of any variations in efficacy stemming from offense type. Inclusion of drug offenses, DWI offenses, and larceny offenses was the result of the frequency of these charges which lent well to matching. Other specific offense types were available (i.e., homicide) but it occurred too infrequently to be match effectively.

Two models were assessed for each mechanism comparison within each stratification in order to ensure differences were not artifacts of the PSM algorithm. Helland and Tabarrok [4] summarize the efficacy of this process when matching across multiple treatments. Once these propensity scores were estimated, the overlap in propensity scores between treatment and matched control groups for each analysis were visually examined via box-whisker plots to assess the potential for finding matches for treated cases. Fig 1 is a visual depiction of the propensities to be released via the top mechanism for both mechanisms. For example, the upper left of the figure presents the distribution of propensities for defendants that were released via commercial bonds or pretrial services to be released via commercial bonds. The visual interpretation of these box-whisker plots is to ensure that there is overlap between these mechanisms in the propensity for release. In the absence of sufficient overlap, there would not be sufficient cases available to match. This would result in few if any cases being matched across mechanism, especially when using a caliper of 0.001. The selection of box-whisker plots is provided to demonstrate the presence of an overlap across mechanisms and offense types. In each case, there was adequate overlap for matched cases to be identified.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Propensity score overlap between groupings, by stratification.

Box-whisker plots illustrating overlap between groupings on the propensity for being released via a particular mechanism.

https://doi.org/10.1371/journal.pone.0182772.g001

The following step involved an assessment of covariate balance across each comparison group post-estimation of propensity scores between treated cases and their matched controls. Independent samples t-tests were carried out for all covariates between treated versus untreated and treated versus matched controls with the goal of demonstrating that the matching procedure effectively removed bias between those treated and the controls as a result of the matching procedure. Due to the volume of comparisons for the present analysis, the t-test results specific to balancing are omitted herein, however, the matching procedure effectively removed all covariate differences in each stratification and grouping post-matching (i.e., t was less than 1.96 in each instance).

Upon establishing balance, tests for differences between the final outcomes were calculated via t-tests for each matched group comparison. Sensitivity of the results was assessed via the Mantel and Haenszel [29] test statistic (MH; commonly referred to as mhbounds) as it applies to assessing the impact of unmeasured covariates in propensity score estimation for binary outcomes. The MH statistic represents the magnitude, in the form of an odds ratio, that an unmeasured covariate would need to be represented by in order to potentially confound a treatment effect–more detail on sensitivity in PSM can be found in DiPrete and Gangl [30]. Estimated treatment effects and sensitivity analyses for each comparison grouping are discussed in the following section.

Results

This study explored the efficacy of four pretrial release mechanisms on FTA by assessing differences in the probability of FTA between a series of pairings of Dallas County, Texas defendants released via either of one type of mechanism to another, until all possible comparisons were established. These four mechanisms included attorney bonds, cash bonds, commercial bonds, and pretrial services bonds. Rates of FTA were first compared between mechanism pairings for all defendants. Crime types were then broken down to parse out efficacy for felony and misdemeanor offenses, as well as offense-specific categories. Comparisons were created with each mechanism acting as a treatment and a control in order to address differences in FTA rates resulting from disparate ranges of effects. Table A in S1 Appendix presents the summary statistics for all defendants released via each of the four mechanisms and across each strata assessed.

Table 1 presents conditional differences between release types resulting from the PSM analysis. These PSM findings are presented to illustrate the conditional differences in FTA rates between those treated and their matched controls for all releases, felonies, and misdemeanors, respectively. On the diagonal of these tables are the unadjusted FTA rates for each release type. These raw statistics are presented for reference and are suggestive of commercial bonds being the more successful in terms of reducing the probability of FTA compared to any other release type studied. The off-diagonal statistics are the conditional mean difference in FTA rates (i.e., the treatment effects) between those released via a particular mechanism versus a single alternative. This is identified by the left-hand column, compared to a single alternative, identified by the top row of the table. Note that the number range discussed (if statistically significant) reflects the estimated difference for matching based on an inverted treatment outcomes (e.g., commercial vs. attorney and attorney vs. commercial), which is important to consider given that matching may not be equivalent for the inverse of a treatment condition [27].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Charge levels: Average treatment effects: Failure to appear (Unconditioned rates on the diagonal).

https://doi.org/10.1371/journal.pone.0182772.t001

When assessing differences using data from all defendants, only the comparisons involving commercial bonds were found to be statistically significant. In this study, commercial bond defendants tended to fail to appear at a significantly lower rate compared to each alternative. This difference was largest between commercial bonds and pretrial services bonds, which ranged between 0.12 and 0.15. The smallest reduction was between commercial and cash release mechanisms. Commercial bonds produced a FTA rate 0.10 lower than cash bonds and a rate of 0.12 lower than attorney bonds. Comparisons between other bond mechanisms did not produce statistically significant differences. These general findings suggest that commercial bonds may be more effective at compelling a defendant to return to court appearances than other mechanisms, when equivalent defendants are compared. Analyses then focused specifically on felonies and misdemeanors, respectively.

For felonies and misdemeanors, a similar pattern of findings emerged. Commercial bond defendants were less likely to fail to appear compared to alternative mechanisms, respectively. This trend again continued for comparisons using misdemeanor offenses with one exception. Results indicated a statistically significant difference between cash bonds and pretrial services bonds with cash bond defendants being more likely to appear in court. These analyses provide additional evidence that those released via commercial bonds tended to have lower FTA rates compared to other mechanisms across both felony and misdemeanor charges.

To supplement the above analyses, comparisons were made on FTA rates for specific charge/offense types where data were amply available. These results are presented in Table 2. Specifically, these involved defendants charged with drug offenses, driving while intoxicated, and larceny (theft), respectively. In general, defendants released on commercial bonds again tended to FTA at lower rates compared to other mechanisms for these crime specific comparisons. Regarding those defendants charged with a drug related crime, those released via commercial bonds were significantly more likely to appear in court for adjudication. No other between-mechanism comparisons were found to be significantly different.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Specific crime types: Average treatment effects: Failure to appear (Unconditioned rates on the diagonal).

https://doi.org/10.1371/journal.pone.0182772.t002

Comparisons analyzing defendants charged with driving while intoxicated produced evidence of similar reductions from commercial bonds when compared to cash bonds and pretrial services bonds. However, conditional differences between commercial bonds and attorney bonds for driving while intoxicated were not significant.

The differences in FTA rates between mechanisms for defendants charged with larceny were similar to driving while intoxicated defendants. There was a statistically significant difference between commercial bonds when compared to cash bonds and pretrial services bonds. In both cases, defendants released on a commercial bond had lower rates of FTA compared to the other mechanism. The comparison between attorney bonds and commercial bonds produced partial support for a difference. There was a significant difference when attorney bonds were matched to commercial bonds with attorney bonds acting as the treatment and commercial bonds acting as the control, though the reverse was not as such, lending only partial support to this comparison.

Overall, these findings tend to suggest that when comparing similar defendants in equivalent contexts, those released on a commercial bail bond for initial release from jail pending trial are consistently less likely to fail to appear resulting in bond forfeiture. This difference was found to be present across varying strata of offenses. However, it should be noted that these analyses may be sensitive to unmeasured effects, as is the case in most counterfactual studies. Sensitivity analysis indicated that unmeasured variables with modest odds ratios could have impacted matching, however, the bulk of the findings were robust to sensitivity assessments (i.e., bounds statistics). An overview of the sensitivity of the above results is discussed in the following section.

In all counterfactual research designs, it is important to take into consideration the potential for hidden bias (i.e., that which may stem from unmeasured variables). In the present case, data were limited to official records, thus leaving some potentially important features associated with FTA left unmeasured such as personal beliefs, situational information, and court room workgroup variables. One approach to assessing the potential for an impact from such unmeasured bias, if indeed it had been available, is to estimate bounds statistics (see [28], [30], [31], and [32]. Following the protocol outlined by Rosenbaum [32], gamma coefficients (MH) were estimated for each matching procedure, resulting in dozens of estimates in this case. These results are presented in Table 3. For each statistically significant finding, gamma ranged between 1.4 and 4.0 between groupings, suggesting that depending on the particular matching procedure, an omitted feature would need have a conditional effect, in terms of an odds ratio, in excess of 1.4 to offset the finding of any of the statistically significant difference tests presented in this study (see Table 3). On average, gamma was 2.4, suggesting less of a concern for the influence of omitted variables and that these results are relatively robust to hidden bias.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. The effects of pretrial release mechanism on failure to appear (FTA) for four pretrial release mechanisms in Dallas County, Texas.

https://doi.org/10.1371/journal.pone.0182772.t003

Discussion and conclusion

The present study contributes to the literature by offering a comparative assessment of the utility of varying pretrial defendant release mechanisms in terms of FTA in court, leading to bond forfeiture. What is unique about the present study is that it is the first to assess these effects between similarly situated defendants using comprehensive archival records from Dallas County, Texas. The focus on a single jurisdiction is important as to avoid cross-jurisdiction comparisons where justice systems vary considerably and impact the likelihood and availability of any given release mechanism. In general, the findings are robust in suggesting that comparable defendants released via a commercial bond tend to be more likely to appear in court as compared to attorney, cash, and pretrial services bonds. This finding held across many strata of offense categories, however, these effects should be taken in context.

All pretrial release mechanisms are not equal in the resources available to compel a defendant to return to court. Helland and Tabarrok [4] outline these differences, which cover many domains. First, the commercial bail industry is permitted to require co-signers and collateral. This practice, however, is not permitted in other pretrial release mechanisms. Second, commercial bond agencies regularly contact defendants to remind them of court appearances. These reminders can increase the likelihood of appearing for court [19]. Third, commercial bond agencies collect detailed information on the defendant that can be used to help locate him/her in the event the defendant absconds and can always refer to the co-signers for information. Bail bond agencies gather information on the defendant’s family, employment, and residence, for example [4]. Finally, in the event a defendant absconds, commercial bond agencies have extraordinary tools available to return the defendant to court including arrest, imprisonment, and transportation across state lines [33]. These differences (e.g., the use of bounty hunters) demonstrate the unequal field of promoting court appearance. None of the mechanisms evaluated here are provided all of these tools to encourage defendants appear for trial.

Commercial bond agencies can even petition the court to remove them as the bonding agent in cases where the defendants are not cooperating with the terms of their bail. This is done with an affidavit-to-go-off-bond (ATGOB). In the event that the court accepts the ATGOB, the liability of the bonding agency is removed and an arrest warrant is put out for the defendant [34].

Despite the lower FTA rate of commercial bail, each mechanism plays an important role pretrial release. This is particularly true for pretrial services bonds as this service allows indigent defendants to be released pretrial in the absence of financial collateral. Financial restrictions should not preclude an individual from pretrial release, especially when pretrial detention affects criminal justice system processing. Nevertheless, the criminal justice system needs to balance the competing aims of equitable pretrial release while ensuring a defendant’s appearance in court.

Considering that over sixty percent of jail inmates in the US are being held pretrial [2], it is paramount that the criminal justice system has mechanisms to relieve this burden without wasting tax dollars. It is also important, however, for defendants to appear in court. FTA in court not only permits the defendant to escape judgment but also has an inherent expense to the criminal justice system. Block and Twist estimate that each FTA cost the criminal justice system approximately $1,273 or closer to $1,700 in 2009 [5]. In Dallas, this equates to an approximated cost of $12,466,100 for the 7,333 FTAs for 2009 alone, or nearly $35,000 per day. Considering the substantial cost associated with FTA in a single county court, there is considerable fiscal incentive to reduce FTA in addition to the pursuit of justice.

While relatively robust, the current study is not without limitations. First, these analyses stem from a single jurisdiction in a single state. Second, while this study was able to determine the efficacy of bail mechanisms across various offenses, the low frequency of some violent crimes, such as sexual assault, prevented further granularity in the analysis. Third, release on own recognizance bonds were rarely applied in Dallas County during the data collection and thus were prevented from inclusion in the study. It should be noted that a pretrial services release in Dallas County, during the study period, was akin to release on recognizance bonds.

Finally, accurately measuring FTA and bond forfeiture is challenging. The complexities associated with capturing FTA are outlined in Dallas County Bail Bond Task Force’s report [34]. At the most basic level, a FTA would occur when a defendant does not show up to a scheduled court proceeding requiring the defendant’s presence. The bailiff would record a FTA if the defendant does not appear within a reasonable time after being called. In Dallas County, defendants and attorneys are allowed to appear at-will. This increases the difficulty of identifying FTA and challenges the prosecutor to be cognizant of appearances and the necessity to request bond forfeiture. Furthermore, the judges and court staff maintain information of FTA, instead of the bailiff, as part of court records, but there is no requirement for collecting or maintaining such data.

In Texas, the court decides when to forfeit a bond and there is no mandated guideline or criteria for forfeiture per se. For example, a defendant may fail to appear many times prior to the court actually forfeiting the bond. In another court in the same jurisdiction, a judge may have a strict forfeiture policy where a single FTA results in forfeiture. A possible solution for this discrepancy would be to require the court to record appearance information in addition to bond forfeiture.

Despite these limitations, this study increases the knowledge of the efficacy of different pretrial release mechanisms, especially when parsing out the effect by offense type. Here, commercial bonds consistently produce the lowest rate of FTA and forfeiture compared to other mechanisms. Effectiveness of commercial bonds aside, it is important to acknowledge each mechanism has its merits, especially with indigent defendants.

It should be noted that the authors are not advocating for or against any mechanisms in particular. The current study is an analysis of pretrial release options available in Dallas County, Texas in 2008. As such, the results may not be generalizable to other periods in time or other jurisdictions. Instead, it is important to acknowledge that all of these mechanisms coexist as independent components in a complex pretrial release process. It would be difficult to predict how the pretrial release system would function after any major change to these mechanisms. Simply ending the use of one or more mechanisms in favor of another could have unforeseen consequences resulting in drastically different efficacy. Until more is understood about pretrial release, failure to appear, each of these mechanisms individually, and how they operate in the system together, caution is warranted prior to any policy decisions. This is especially true considering the pretrial services release department operating in Dallas County, Texas during 2008 lacked many standard components of pretrial services agencies.

Future research should focus similar analyses on alternative jurisdictions. There is significant potential for between jurisdiction variation based on laws, court practices, individual discretion, and program quality. Future research should also look at the conditions imposed by pretrial release mechanisms to attempt to identify the most significant predictors of successful appearance. This knowledge could increase the use of evidence-based practices in all pretrial release mechanisms.

Supporting information

Acknowledgments

This manuscript resulted from a larger study funded by Dallas County, Texas (USA) and carried out by the Center for Crime and Justice Studies at the University of Texas at Dallas. Special thanks to the Dallas County Criminal Justice Advisory Board, the Dallas County Commissioners Court, Hon. Dr. Elba Garcia, Hon. Dr. Teresa Daniel, Mr. Ron Stretcher, Mr. Dwayne Steele, Mr. Mark Crooks, Dr. James W. Marquart, and The University of Texas at Dallas for their support.

Research reported in this manuscript was supported by Dallas County, Texas [Court Order 2014-0669]. The content is solely the responsibility of the authors and does not necessarily represent the official views of Dallas County, or the State of Texas. Special gratitude is extended to (Hon.) Dr. Elba Garcia, (Hon.) Dr. Teresa M. Daniel, Ron Stretcher, Duane Steele, Dr. Michael Noyes, Hon. Clay Jenkins, Dallas County Criminal Justice Advisory Board (CJAB) and the Texas Department of Public Safety.