Page 1

CJA

NEW YORK CITY CRIMINAL JUSTICE AGENCY, INC. NEW YORK CITY CRIMINAL USTICE AGENCY

Jerome E. McElroy Executive Director

PRETRIAL FAILURE TO APPEAR AND PRETRIAL RE-ARREST AMONG DOMESTIC VIOLENCE DEFENDANTS IN NEW YORK CITY

Richard R. Peterson, Ph.D. Project Director and Director, Research Department

FINAL REPORT

September 2006

52 Duane Street, New York, NY 10007

(646) 213-2500


PRETRIAL FAILURE TO APPEAR AND PRETRIAL RE-ARREST AMONG DOMESTIC VIOLENCE DEFENDANTS IN NEW YORK CITY

Richard R. Peterson, Ph.D. Project Director and Director, Research Department Research Assistance: Justin P. Bernstein Senior Research Assistant Raymond P. Caligiure Graphics and Production Specialist Steve Mardenfeld Research Assistant Information Systems Programming: Barbara Geller Diaz Associate Director, Information Systems Wayne Nehwadowich Senior Programmer/Analyst Administrative Support: Annie Su Administrative Associate

September 2006

This report can be downloaded from http://www.cjareports.org Š 2006 NYC Criminal Justice Agency, Inc.

When citing this report, please include the following elements, adapted to your citation style: Peterson, Richard R. 2006. Pretrial Failure to Appear and Pretrial Re-Arrest Among Domestic Violence Defendants in New York City. New York: New York City Criminal Justice Agency, Inc.


ACKNOWLEDGEMENTS

This report could not have been completed without the assistance of colleagues at CJA, Judges, members of the District Attorneys’ offices and others. The author extends special thanks to Dr. Qudsia Siddiqi, Senior Research Analyst at CJA and to Dr. Mary T. Phillips, Deputy Director of the Research Department at CJA for their advice and comments on the report. Dr. Siddiqi’s prior research on failure to appear and pretrial re-arrest and Dr. Phillips’ prior research on release and bail decisions were valuable resources for conducting the research described in this report. The author appreciates the research assistance of Justin P. Bernstein, who prepared the data for analysis and made editorial comments on the final draft. The author also thanks Annie Su who prepared numerous drafts of the figures and appendices, Raymond P. Caligiure who created tables, and Steve Mardenfeld who checked the final draft of the manuscript. The author extends special thanks to Barbara Geller Diaz who did the programming to extract re-arrest data from the CJA database and to Wayne Nehwadowich who programmed the case processing data. The author also acknowledges other colleagues at CJA who provided advice, information and editorial suggestions: Jerome E. McElroy, Executive Director of CJA, and Mari Curbelo, Esq., Geraldine Ferrara, Esq., Marian Gewirtz, Peter Kiers, Elyse J. Revere, Frank Sergi, and Dr. Freda F. Solomon. The author would also like to thank Amy Barasch, Esq., Deirdre Bialo-Padin, Esq., Abena Darkeh, Esq., Hon. Matthew J. D’Emic, Prof. Jo Dixon, Peter Glick, Esq., Scott Kessler, Esq., Karen Kleinberg, Esq., Elisa Koenderman, Esq., Ms. Sharon Lastique, Hon. John M. Leventhal, David J. Levin, Ph.D., Wanda Lucibello, Esq., Hon. Deborah Stevens Modica, Audrey Moore, Esq., Hon. Esther M. Morgenstern, Joseph Muroff, Esq., Ms. Amanda Voytek and Michael Yavinsky, Esq. for their assistance with the research. Finally, thanks to the New York State Division of Criminal Justice Services (DCJS) for providing supplemental data. The methodology, findings and conclusions of the study, as well as any errors, omissions and misinterpretations are the responsibility of the author.


TABLE OF CONTENTS

I.

Introduction ...................................................................................................... 1 A. Review of the Literature on Pretrial Failure to Appear .................................. 2 B. Review of the Literature on Pretrial Re-Arrest .............................................. 5 1. Literature on the General Defendant Population...................................... 5 2. Literature on Domestic Violence Defendants........................................... 7 C. Summary of Literature on Pretrial Misconduct............................................ 10 D. Research Plan ............................................................................................ 10

II.

Methodology................................................................................................... 13 A. Overview of the CJA Database and the Combined First Quarter 2001 and Third Quarter 2002 Dataset ........................................................................ 13 B. Identifying Domestic Violence Cases.......................................................... 14 C. Identifying Re-Arrests for Domestic Violence Offenses .............................. 16 D. Selection of the Crimes against Persons and Property Subsample ............ 17 E. Using a Defendant-Based Data File............................................................ 19 F. Plan of Analysis .......................................................................................... 19

III.

Pretrial Release Outcomes............................................................................ 21 A. B. C. D. E.

IV.

Pretrial Release Decisions.......................................................................... 21 Pretrial Failure to Appear ............................................................................ 26 Pretrial Re-Arrest ........................................................................................ 27 Combined Rates of Pretrial Failure to Appear and Pretrial Re-Arrest ......... 27 Summary and Discussion of Findings......................................................... 29

Predicting Pretrial Failure to Appear for DV Defendants............................ 31 A. Development of a Model Predicting Pretrial Failure to Appear ................... 31 B. Model Predicting Pretrial Failure to Appear for DV Defendants .................. 34 C. Summary and Discussion of Findings......................................................... 38

V.

Predicting Pretrial Re-Arrest for DV Defendants......................................... 41 A. Development of a Model Predicting Pretrial Re-Arrest for a New DV Offense ................................................................................................. 41 B. Model Predicting Pretrial Re-Arrest for a New DV Offense for DV Defendants ................................................................................................. 42 C. Summary and Discussion of Findings......................................................... 44

Table of Contents Continues on Next Page i


TABLE OF CONTENTS, CONTINUED

VI.

Conclusion...................................................................................................... 49 A. Major Findings ............................................................................................ 49 B. Discussion .................................................................................................. 50 C. Conclusion.................................................................................................. 57

VII. References ..................................................................................................... 61 Appendix A: Statistical Methods.......................................................................... 69 1. Logistic Regression Analysis....................................................................... 69 2. Correcting for Selection Bias....................................................................... 72 Appendix B: Distribution of Variables for Regression Models ......................... 77 Appendix C: Coding of Variables for Regression Models ................................. 81

ii


LIST OF TABLES

Table 4-1

Logistic Regression Model Predicting Likelihood of Failure to Appear........35

Table 5-1

Logistic Regression Model Predicting Likelihood of Pretrial Re-Arrest for a New DV Offense .........................................................................................43

Table A-1

Logistic Regression Model Predicting Likelihood of Pretrial Release......... 73

iii


LIST OF FIGURES

FIGURE 1

Cases Disposed at Arraignment............................................................ 21

FIGURE 2

Release Status at Arraignment.............................................................. 22

FIGURE 3

Bail Making at Arraignment ................................................................... 23

FIGURE 4

Stage of First Release........................................................................... 23

FIGURE 5

Number of Days between Arraignment and First Release .................... 24

FIGURE 6

True Cash Alternatives for Cases with Bail Set at Arraignment............. 24

FIGURE 7

Amount of Bail Set at Arraignment ........................................................ 25

FIGURE 8

Amount of Bail at Time of First Release ................................................ 25

FIGURE 9

Failure-to-Appear Rates for Defendants Who Were Ever Released ..... 26

FIGURE 10A Pretrial Re-Arrest Rates for Any New Offenses for Defendants Who Were Ever Released.................................................. 28 FIGURE 10B Pretrial Re-Arrest Rates for New DV and Non-DV Offenses for Defendants Who Were Ever Released.................................................. 28 FIGURE 11

Combined Rates of FTA and Pretrial Re-Arrest for Defendants Who Were Ever Released ..................................................................... 29

iv


PRETRIAL FAILURE TO APPEAR AND PRETRIAL RE-ARREST AMONG DOMESTIC VIOLENCE DEFENDANTS IN NEW YORK CITY I. INTRODUCTION The processing of domestic violence (DV) cases often presents special problems for the criminal justice system. A DV defendant who is released prior to the disposition of a criminal case may harm or threaten the victim while the case is pending. The DV defendant’s motivation may be to retaliate against the victim for having the defendant arrested and/or to discourage the victim from participating in the prosecution of the case. While defendants in Non-DV cases may also retaliate against or threaten their victims, victims in DV cases are often in greater danger of facing renewed threats or violence. DV defendants have easier access to the victim, since they usually know where the victim lives and works. Because of their emotional ties to the victim, DV defendants may have greater motivation to threaten or retaliate. Emotional and economic ties also may provide defendants with greater leverage against victims in DV cases than in Non-DV cases. Furthermore, domestic violence, more than other types of violence, often occurs in a private location, and is therefore difficult to detect and prevent. Because of concerns that some DV defendants may commit new DV offenses during the pretrial period, many in the criminal justice system are particularly interested in pretrial release decisions in DV cases. How often are DV defendants released? How often do they fail to appear for their scheduled court appearances? How often are they re-arrested for new DV offenses while awaiting a disposition on their original case? How often are they re-arrested for any new offense? Finally, how do the answers to these questions differ for DV and Non-DV defendants? Are the risks greater for DV defendants? This report seeks to answer these questions using a sample of DV and Non-DV defendants in New York City. It examines pretrial release outcomes for domestic violence cases and compares them to the outcomes for Non-DV cases. It then examines models predicting two types of pretrial misconduct: failure to appear for a scheduled court appearance, and re-arrest for a new DV offense during the pretrial period. Concerns about pretrial release of DV defendants are generally focused on the risk of re-offending. Preventive detention, aimed at preventing re-arrest for new offenses, is permitted in the federal courts and in many state courts. However, New York State’s Criminal Procedure Law (CPL) §510.30(2)(a) requires that decisions about releasing the defendant on recognizance or setting bail be based on factors related to insuring that the defendant will make scheduled court appearances. New York State law does not allow these decisions to be based on concerns about the safety of the victim or the community. Nevertheless, the law does allow the court to address concerns about safety through the conditions of release. The court has wide latitude to set conditions of release, including issuing a temporary order of protection limiting the


2 defendant’s contact with the victim. Because New York State law requires criminal courts to take different approaches to address concerns about failure to appear and victim safety, the current study will focus on the differential impact of release decisions on these two outcomes. We begin by reviewing previous research on factors that affect pretrial failure to appear (FTA) and pretrial re-arrest. A. Review of the Literature on Pretrial Failure to Appear Much of the research on pretrial failure to appear, and in particular much of the recent research, has been conducted at the New York City Criminal Justice Agency, Inc. (CJA) (Siddiqi 1999, 2000, 2002, 2003a, 2006). This review draws heavily on a recent report (Siddiqi 2006), which summarizes findings from studies of the factors affecting FTA. Comparing these studies is difficult because they differ in their methods of calculating FTA (Siddiqi 2006). Some studies have used appearance-based FTA rates (Eskridge 1979) while others have used defendant-based FTA rates (Center for Governmental Research, Inc. 1983, Schaffer 1970, Siddiqi 1999, Thomas 1976). Appearance-based rates are calculated by dividing the number of missed appearances by the number of scheduled appearances. Defendant-based rates are calculated by dividing the number of defendants who missed at least one court appearance by the total number of defendants. Unfortunately, many studies did not specify which type of measure of FTA they were using. FTA rates may also change over time. In New York City, defendant-based FTA rates in the Criminal (lower) Court have declined from 30% in 1989 (Siddiqi 1999) to 18% in 1998 and 15% in 2001 (Siddiqi 2002). FTA rates in New York City have remained relatively stable since that time (Siddiqi 2003a). A second problem with reviewing prior research is the inconsistency across studies in measuring time at risk (Siddiqi 2006). While all researchers agree that the effect of other variables should be assessed after controlling for time at risk, they differ in their measurement of time at risk. Some studies calculated the number of days the defendant was free from pretrial detention (Center for governmental Research, Inc. 1983, Toborg 1981), while others counted time from the date of release until the date of disposition (Goodman 1992). Others counted only the time from the date of release until the first missed court appearance prior to case disposition (Toborg 1981). None of these methods is ideal, and we will develop a different measure for use in this study (see Chapter 4). Recognizing that FTA and time at risk were measured differently in different studies, and that FTA rates vary considerably across jurisdictions and over time, it is nevertheless possible to draw some general conclusions about factors that affect FTA. Criminal History. Defendants with more serious criminal histories are more likely to fail to appear (Siddiqi 1999). A record of prior arrests (Center for Governmental Research, Inc. 1983, Goldkamp, Gottfredson and Herzfeld 1981, Ozanne, Wilson and Gedney 1980, Reaves 1993, Reaves and Perez 1994, Toborg 1981) or prior convictions


3 (Goodman 1992, Maxwell 1999, Reaves 1990, Winterfield, Coggeshall and Harrell 2003) was associated with higher rates of failure to appear. Prior research also consistently shows that a record of prior failure to appear is a strong predictor of failure to appear in new criminal cases (Belenko, Mara-Drita and McElroy 1992, Center for Governmental Research, Inc. 1983, Cuvelier and Potts 1993, Eskridge 1979, Goodman 1992, Reaves 1993, 1994, Reaves and Perez 1994, Siddiqi 1999, 2000, 2002, Straus and Golbin 1980, Wilson 1979, Winterfield et al. 2003). Only Maxwell’s (1999) study of felony cases in New York City found no effect of prior failure to appear. Community Ties. Research has also examined whether a defendant’s community ties influence the likelihood of FTA. Some researchers found that strong community ties were associated with lower rates of FTA (Belenko et al. 1992, Siddiqi 1999, 2000, 2002, Toborg 1981, Venezia 1973, Wilson 1975), although some studies have found no relationship (Eskridge 1979, Feeley and McNaughton 1974). Employment, in particular, was identified as a predictor of lower FTA rates in many studies (Boudouris, Thomasgard and Lacsina 1977, Center for Governmental Research, Inc. 1983, Lazarsfeld 1974, Ozanne et al. 1980, Rhodes, Hyatt and Scheiman 1996, Roth and Wice 1980, Siddiqi 1999, 2000, 2002). CJA research has consistently shown that having a telephone, residing in the local area, and expecting someone at arraignment were associated with lower rates of FTA (Siddiqi 1999, 2000, 2002). A defendant’s length of time living at his/her current address had a significant effect on the likelihood of FTA in one CJA study, based on 1989 data (Siddiqi 1999; see also Ozanne et al. 1980), but not in two other CJA studies, based on 1998 and 2001 data (Siddiqi 2000, 2002). Charge Characteristics. Defendants charged with drug offenses generally have higher rates of FTA, while those charged with violent offenses and public order offenses generally have lower rates of FTA (Bureau of Justice Statistics 1992, Cohen and Reaves 2006, Hart and Reaves 1999, Rainville and Reaves 2003, Reaves 1990, 1994, 1998, 2001, Reaves and Perez 1994, Reaves and Smith 1995, Siddiqi 1999, 2000). A few studies have found that defendants charged with burglary and property crimes have higher rates of FTA than defendants facing other types of charges (Kirby 1977, Maxwell 1999, Reaves 1993, Reaves and Perez 1994, Siddiqi 1999, 2000). Interestingly, most studies have found that as charge severity increases, rates of FTA decrease (Belenko et al. 1992, Center for Governmental Research, Inc. 1983, Eskridge 1979, Rhodes et al. 1996, Roth and Wice 1980, Siddiqi 1999, 2000, Straus and Golbin 1980), although one study found no relationship (Landes 1974). Type of Release. Many studies have focused on the impact of the type of release on the likelihood of failure to appear. Researchers, like courts, have been interested to know if defendants released on bail are more or less likely to appear in court than those released on recognizance (ROR). Results are decidedly mixed. Some studies reported that defendants released on bail were more likely to FTA than those released on recognizance (Bell et al. 1974, National Institute of Law Enforcement and Criminal Justice 1976, Reaves 1994), while others reported the opposite—higher rates of FTA for those released on recognizance (Alameda Regional Criminal Justice


4 Planning Board 1976, Helland and Tabarrok 2004, Obert 1973, Reaves 1990, 1993, Reaves and Perez 1994). Still other studies reported no impact of type of release (Conklin and Meagher 1973, Goldkamp et al. 1981, Maxwell 1999, Wheeler and Wheeler 1981). Confirming the mixed nature of the findings, Thomas (1976) examined 20 jurisdictions across the country and found that in some cities, defendants released on bail had higher rates of FTA than those released on recognizance (Chicago, San Francisco), while in other cities the reverse was true (Boston, Detroit, Los Angeles) and in other cities there was no difference. A similar study of 16 jurisdictions produced similar patterns of findings (U.S. Office of Economic Opportunity 1973). Research on felony defendants in state courts also showed variation by jurisdiction (Kennedy and Henry 1996). Research at CJA found that defendants released on recognizance in New York City were more likely to fail to appear than defendants released on bail (Siddiqi 1999). Demographic Characteristics. Some studies reported that younger defendants were more likely to FTA than older defendants (Goldkamp et al. 1981, Illinois Criminal Justice Information Authority 1992, Siddiqi 1999) while others did not (Maxwell 1999, Reaves 1990, 1993, 1994, Reaves and Perez 1994, Roth and Wice 1980). Similarly, some studies found higher FTA rates for Black defendants (Goldkamp et al. 1981, Illinois Criminal Justice Information Authority 1992, Reaves 1993, Reaves and Perez 1994, Siddiqi 1999) and Hispanic defendants (Reaves and Perez 1994), while some found no differences by ethnicity (Maxwell 1999, Reaves 1994, Roth and Wice 1980). Two studies reported that women were more likely than men to fail to appear (Maxwell 1999, Siddiqi 1999), although others reported higher FTA rates for men (Reaves 1990, Reaves and Perez 1994), and others reported no difference (Reaves 1993, 1994). Drug Use. Finally, researchers have found mixed results on the effects of drug use on FTA. Several studies reported that drug users had higher rates of FTA (Goldkamp et al. 1981, Roth and Wice 1980, Smith, Wish and Jarjoura 1989, Winterfield et al. 2003), but others found no effect (Belenko et al. 1992, Britt, Gottfredson and Goldkamp 1992, Rhodes et al. 1996). In summary, research on the general population of defendants found that defendants with more serious criminal histories, especially those with a history of prior FTA, were more likely to fail to appear. Results on the effect of community ties were mixed, with some finding that stronger community ties reduced FTA, while others found no impact. One measure of community ties, employment, was consistently associated with lower rates of failure to appear. Charge type seemed to influence FTA, with drug defendants more likely to FTA and mixed results for burglary and property crimes. Charge severity was inversely related to FTA—the more severe the charge, the lower the likelihood of FTA. The effect of type of release (bail vs. ROR) varied considerably from study to study and place to place, so no definitive conclusion can be drawn. Age, race, sex and drug use may have an impact, but again some studies found no impact. Overall, the main conclusion that can be drawn from prior literature is that criminal history and employment affected the likelihood of FTA, while results for the influence of other factors were not consistent. At best, the literature suggests variables that should


5 be examined and tested in new studies, but provides only limited guidance about what to expect. While there has been extensive research on FTA in the general population of defendants, we are not aware of any published research on FTA among domestic violence defendants. B. Review of the Literature on Pretrial Re-Arrest 1. Literature on the General Defendant Population This review of research on pretrial re-arrest begins by focusing on re-arrest among the general population of defendants, and relies heavily on reviews by Siddiqi (2003b, 2006). Pretrial re-arrest rates varied considerably across studies, depending on the year and the population of defendants studied. Several studies focused on felony defendants. Reaves and Perez (1994) reported a pretrial re-arrest rate of 14% in their detailed analysis of 1992 data on felony defendants in state courts from the National Pretrial Reporting Program (NPRP). NPRP also has published less detailed data from other years, showing rates of re-arrest during the pretrial period fluctuating between 14% and 18% from 1988 to 2002 among felony defendants in state courts (Cohen and Reaves 2006, Hart and Reaves 1999, Rainville and Reaves 2003, Reaves 1990, 1993, 1998, 2001, Reaves and Smith 1995). Reaves (1994) found that the pretrial re-arrest rate for released federal felony defendants in 1990 was much lower—about 3%. Other studies included defendants charged with misdemeanor and lesser offenses, as well as felony defendants, in their analyses. Visher and Linster (1990) reported a 26% pretrial re-arrest rate for defendants released on recognizance in the District of Columbia in 1984. Using a 1999 sample of defendants released on recognizance or bail in the District of Columbia, Winterfield et al. (2003) reported a pretrial re-arrest rate of 19% in 1999. Siddiqi (2003b, 2003c) found that pretrial re-arrest rates for all defendants in New York City declined over time, from 30% in 1989 to 20% in 1998 and 17% in 2001. In their examination of factors affecting pretrial re-arrest, many studies controlled for the effect of time at risk (e.g., Belenko, Fagan and Dumanovsky 1994, Siddiqi 2003b, 2003c, Toborg et al. 1984, Visher and Linster 1990), but some did not (e.g., Reaves 1994, Reaves and Perez 1994, Winterfield et al. 2003). As a result, comparisons of findings may be affected by differences across the studies in defendants’ time at risk. Inconsistent findings reported below may be the result of these differences. Criminal History. As was the case in studies of FTA, studies of pretrial re-arrest generally found that defendants with more serious criminal histories were more likely to be re-arrested. Defendants with prior arrests, prior criminal convictions, or pending charges were more likely to be re-arrested (Landes 1974, Reaves 1990, 1993, 1994, Rhodes et al. 1996, Roth and Wice 1980, Siddiqi 2003b, 2003c, Toborg et al. 1984, Visher and Linster 1990, Winterfield et al. 2003). In some studies, a record of prior


6 failure to appear also affected the likelihood of pretrial re-arrest (Goodman 1992, Reaves 1994, Siddiqi 2003b, 2003c). Community Ties. Only a few studies of pretrial re-arrest examined the effect of community ties. Visher and Linster (1990) found that employment reduced the likelihood of re-arrest, at least early in the pretrial period (see also Goodman 1992, Roth and Wice 1980). They also found that defendants with more years of education were less likely to be re-arrested. Siddiqi’s analyses (2003b, 2003c) of three New York City datasets (1989, 1998, 2001) found consistent effects of two community ties items. Employed defendants and defendants who had a telephone were less likely to be rearrested during the pretrial period than those who were not employed or had no telephone. Charge Characteristics. Several studies have examined the influence of charge characteristics. Most studies generally found that defendants charged with violent crimes were more likely to be re-arrested during the pretrial period (Reaves 1994, Winterfield et al. 2003). However, Siddiqi (2003b, 2003c) found that defendants charged with violent crimes were less likely to have a pretrial re-arrest. Felony charges generally increased the risk of re-arrest, as did drug charges (Siddiqi 2003b, 2003c). Visher and Linster (1990) found that defendants charged with violent crimes were more likely to be re-arrested early in the pretrial period, while those charged with drug or larceny offenses were more likely to be re-arrested as time at risk increased. Winterfield et al. (2003) reported that defendants charged with bail offenses and those charged with obstructing justice were more likely to be re-arrested. Type of Release. Only a few studies explicitly examined the effect of type of release (bail vs. ROR). Reaves (1993) found that felony defendants in state courts who were released on recognizance had a higher rate of pretrial re-arrest than those released on bail. However, Reaves (1994) found little difference in rates of pretrial rearrest between federal felony defendants released on bail and those released on recognizance. Siddiqi (2003b) found that defendants released on recognizance were more likely to be re-arrested during the pretrial period than those released on bail in 1989 and 1998. However, a subsequent analysis in 2001 found that the type of release had no effect on the rate of pretrial re-arrest (Siddiqi 2003c). Demographic Characteristics. Among demographic variables, age was a consistent predictor of pretrial re-arrest in most studies. Younger defendants were more likely to be re-arrested than older defendants (Belenko et al. 1994, Goldkamp et al. 1981, Goodman 1992, Reaves 1990, 1993, 1994, Siddiqi 2003b, 2003c, Toborg et al. 1984, Visher and Linster 1990, Winterfield et al. 2003). Reaves (1993, 1994) reported lower rates of pretrial re-arrest for Whites than for Blacks. Siddiqi (2003b, 2003c) found lower re-arrest rates for Whites in one of her three samples and higher re-arrest rates for Blacks and Hispanics in two of her three samples. Most studies (Reaves 1990, 1993, Siddiqi 2003b, 2003c) reported that males had higher pretrial re-arrest rates than females, although Reaves (1994) found no meaningful differences by sex among federal felony defendants.


7

Drug Use. Defendants who tested positive for drug use had higher rates of pretrial re-arrest (Bureau of Justice Statistics 1992, Visher and Linster 1990, Winterfield et al. 2003). In summary, research on the general population of defendants found that defendants with more serious criminal histories were more likely to be re-arrested during the pretrial period. Community ties, particularly employment, reduced the likelihood of pretrial re-arrest. Charge characteristics were also influential, although the findings varied. Some studies found that defendants charged with violent crimes, felonies and drug offenses were more likely to be re-arrested, while others did not. Findings on the effect of type of release (bail vs. ROR) were also inconsistent. Younger defendants were more likely to be re-arrested, while findings on the effects of ethnicity and gender varied across studies. Defendants who tested positive for drug use were more likely to be re-arrested. There seemed to be a consensus across studies that criminal history and community ties were influential predictors of pretrial re-arrest. Findings on the effects of other factors were inconsistent, except that all the reviewed studies found that pretrial re-arrest rates were lower for older defendants. As was the case in our review of literature on pretrial FTA, the studies reviewed here suggest the types of variables that should be examined in new studies, while providing only limited information about what findings to expect. 2. Literature on Domestic Violence Defendants We have found only two published studies that examined pretrial re-arrest among domestic violence defendants. Newmark et al. (2001) studied defendants who were indicted on felony domestic violence charges in Brooklyn, New York, between 1995 and 1997. The study was designed to compare DV cases processed in 1995-1996 to cases processed by a new Felony Domestic Violence Court, which was established in early 1997. One model in the study examined the likelihood of a pretrial re-arrest for any new offense (data were not available on re-arrests for new DV offenses). The likelihood of a pretrial re-arrest was not affected by the establishment of the court. Defendants were more likely to be re-arrested during the pretrial period if they had any prior criminal convictions and if the victim in the case had attempted to drop the charges. Three other variables were nearly statistically significant and increased the likelihood of re-arrest: whether the defendant had any prior violent felony convictions, whether the defendant had any prior criminal contempt convictions, and whether the defendant was currently charged with criminal contempt. These findings indicate that criminal history variables were strong predictors of pretrial re-arrest in felony DV cases. In particular, defendants with current or past criminal contempt charges, which usually involve the violation of an order of protection, were more likely to commit new offenses and to be re-arrested for them. Lasley (2003) studied defendants who were arrested on felony domestic violence charges and released on bail in Los Angeles and Orange Counties, California between


8 1996 and 1999. Re-arrests for each defendant were tracked for six weeks following arrest. In these jurisdictions, a defendant released on bail does not make his/her first court appearance until five to six weeks after arrest. For each defendant, the six week follow-up period was only part of the pretrial period. While this limited the ability of the study to provide information on all pretrial re-arrests, it did enable the study to focus on the early period following the initial arrest, when a re-arrest is most likely to occur. Almost 12% of the 496 defendants in the study were re-arrested for a new domestic violence offense during the six week follow-up period. Lasley’s (2003) study was designed to evaluate the effect of a program of intensive bail supervision. Defendants were randomly assigned to receive either normal supervision or intensive supervision. Normal supervision rarely involved any contact between the bail bondsman and the defendant. Intensive supervision involved having the bail bondsman make two random telephone or personal contacts with the defendant every week. The study found that defendants on intensive bail supervision were less likely to be re-arrested for a new domestic violence offense than those on normal supervision. Lasley also examined the effects of other factors. Defendants with prior arrests were more likely to be re-arrested, which is consistent with research on pretrial re-arrest among the general population of defendants. Lasley examined one community ties measure: employment. Overall, employed defendants were no less likely to be re-arrested than unemployed defendants. However, when defendants were placed under intensive bail supervision, employed defendants were less likely to be rearrested than unemployed defendants. This suggests the possibility that community ties reduce the likelihood of re-arrest for a new DV offense only when defendants are placed under supervision during the pretrial period. Finally, Lasley (2003) found that the defendant-victim relationship did not affect the likelihood of re-arrest—married defendants were no less likely to be re-arrested than unmarried defendants. While we have identified only two published studies that examined pretrial rearrests among domestic violence defendants, we found several studies that examined rates of post-disposition re-arrests for DV defendants. Most of these studies were designed to assess the effect of the type of disposition on re-arrest rates for new DV offenses in the post-disposition period. None of the studies found that more severe case outcomes and criminal sanctions reduced the likelihood that a DV defendant would be re-arrested for a new DV offense (see Peterson 2003a for a full review of these studies). However, the studies did identify some variables that had an impact on postdisposition re-arrests for new DV offenses. Criminal History. Defendants with more serious criminal histories were more likely to be re-arrested for a new DV offense after case disposition (Davis, Smith and Nickles 1998, Fagan et al. 1984, Peterson 2003a, Sullivan et al. 2000, Tolman and Weisz 1995, Wooldredge and Thistlethwaite 2002). For example, Peterson (2003a) found that defendants who had any prior criminal convictions or open cases and defendants with more prior misdemeanor convictions were all more likely to be re-arrested for a new DV offense. Notably, the study also found that a pretrial re-arrest for a new DV offense was a strong predictor of the likelihood of a post-disposition re-


9 arrest for a new DV offense (Peterson 2003a). This finding highlights the importance of the current study, because it suggests that defendants re-arrested for a new DV offense in the pretrial period are more likely to be re-arrested for a new DV offense after their case is disposed. An early pattern of pretrial re-arrest may signal an ongoing pattern of re-offending that will continue beyond the disposition of the current case. Community Ties. Two studies found that defendants with weaker community ties were more likely to be re-arrested for a new DV offense after case disposition (Peterson 2003a, Wooldredge and Thistlethwaite 2002). Wooldredge and Thistlethwaite (2002) found that defendants who lived at the same residence for 5 years or more were significantly less likely to be re-arrested than those who lived at their residences for less than 5 years. Peterson (2003a) conducted a more detailed analysis of the effect of community ties on re-arrest rates. He found that defendants who were unemployed, who did not expect someone to be at their arraignment, who did not report having a telephone, and who lived at their current address one year or less were all more likely to be re-arrested for a DV offense. These findings suggest that several aspects of community ties affect re-arrest rates. One exception to this finding is that living with a partner (Wooldredge and Thistlethwaite 2002) or living with someone (Peterson 2003a) was associated with higher rates of re-arrest for a new DV offense. Both studies speculated that defendants who live with their partners had greater opportunity for committing new DV offenses. Charge Characteristics. One study found an effect of charge type on postdisposition re-arrest for a new DV offense. Defendants charged with criminal contempt (usually for violating an order of protection) were more likely to be re-arrested than defendants facing other charges in domestic violence cases (Peterson 2003a). Type of Release. No study has examined the effect of type of release on postdisposition re-arrests for a new DV offense. However, Peterson (2003a) examined the effect of whether the defendant was ever released during the pretrial period, and found no impact on the likelihood of post-disposition re-arrest. Demographic Characteristics. Regarding demographic variables, three studies have found that younger defendants were more likely to be re-arrested for a new DV offense after case disposition (Murphy, Musser and Maton 1998, Peterson 2003a, Wooldredge and Thistlethwaite 2002). Two studies found that men were more likely to be re-arrested than women (Peterson 2003a, Wooldredge and Thistlethwaite 2002). Peterson (2003a) reported that Hispanics were less likely than Whites, Blacks and others to be re-arrested for a new DV offense. The defendant-victim relationship had no effect on the likelihood of re-arrest (Peterson 2003a). In summary, the findings from prior research on pretrial and post-disposition rearrest among DV defendants suggest that criminal history, community ties, and age were consistent predictors of re-arrest, while findings for the effect of other variables were either inconsistent or untested. The finding that pretrial re-arrest for a DV offense


10 was a strong predictor of post-disposition re-arrests for DV offenses suggests that for some DV defendants a pattern of repeated re-offending is undeterred by arrest and prosecution. These defendants pose considerable problems for the criminal justice system, and for the processing of domestic violence cases. C. Summary of Literature on Pretrial Misconduct The literature on pretrial misconduct provided limited guidance on what findings to expect in new studies. The most consistent findings were that defendants with more serious criminal histories and weaker community ties were more likely to engage in pretrial misconduct. Some studies found that charge characteristics mattered, though the findings on the specific type of charges associated with pretrial misconduct varied. Findings regarding the effect of type of release, race/ethnicity and sex were inconsistent across studies. Younger defendants and defendants who used drugs were more likely than older defendants to be re-arrested during the pretrial period, but findings on the effect of age and drug use on FTA were inconsistent. Taken together, findings of research on pretrial misconduct provided guidance for the current study in three ways. First, the research provided valuable information on the factors that should be considered. Specifically, measures of criminal history, community ties, charge characteristics, type of release, demographics, and drug use should be included if available. Second, previous research suggested that multiple measures of each of these factors should be used. For example, while studies generally found an effect of community ties, the particular measures of community ties varied from study to study. The use of multiple measures whenever possible maximizes the chance of finding an effect, and of determining which particular variables are most influential. Third, previous research suggested that findings of models predicting failure to appear are likely to differ from those of models predicting pretrial re-arrest. While these two types of pretrial misconduct share some common predictors, other variables are likely to predict one or the other, but not both. Moreover, even when the same predictor affects both FTA and pretrial re-arrest, it may be much stronger in one model than the other. With this general guidance based on previous research, we are now ready to describe the plan of research for the current study. D. Research Plan The current study is designed to address four research questions: 1) What are the pretrial release outcomes for defendants in domestic violence cases? 2) How do these outcomes compare to those for defendants in non-domestic violence cases? 3) What are the factors that influence the likelihood of pretrial failure to appear among DV defendants?


11

4) What are the factors that influence the likelihood of pretrial re-arrest for new DV offenses among DV defendants? These questions will be addressed through an analysis of data on New York City defendants drawn from the New York City Criminal Justice Agency, Inc. database. As described in Chapter 2, the dataset includes defendants arrested in the first quarter of 2001 and the third quarter of 2002. The data included in the study include information from the CJA interview, the arrest report, and case processing information. The first two questions will be addressed in Chapter 3 by providing information on a variety of pretrial release outcomes for both DV and Non-DV defendants. A model predicting pretrial failure to appear among DV defendants is developed and presented in Chapter 4. Chapter 5 discusses a model predicting pretrial re-arrest among DV defendants. The report concludes with a summary of findings, and a discussion of their implications.


12

[This page intentionally left blank]


13 II. METHODOLOGY A. Overview of the CJA Database and the Combined First Quarter 2001 and Third Quarter 2002 Dataset The data for this study were drawn primarily from the CJA database. This database contains information about the arrest, case processing, and case outcomes of most New York City arrestees. The CJA database includes data from three sources: CJA’s pre-arraignment interview,1 the New York City Police Department’s On-Line Booking System (OLBS) Database, and the New York State Office of Court Administration (OCA). Information concerning demographic characteristics and the community ties of the defendants is taken from the CJA pre-arraignment interview. Information about the arrests is based on the OLBS data. Detailed Criminal Court and Supreme Court processing and outcome data on each of the arrests are drawn from the OCA data. This report is based primarily on analyses of two datasets. The First Quarter 2001 Dataset includes data collected on a three-month cohort of arrests made from January 1, 2001 to March 31, 2001. The dataset includes information on 91,729 arrests where the district attorney elected to bring charges and where a docket number was assigned. The Third Quarter 2002 dataset includes data collected on a three-month cohort of arrests made from July 1, 2002 to September 30, 2002. The dataset includes information on 77,427 arrests where the district attorney elected to bring charges and where a docket number was assigned. In addition to information in the CJA database, the datasets also include information provided by the New York State Division of Criminal Justice Services (DCJS).2 DCJS data were used to supplement and check the reliability of criminal history information that was routinely collected by CJA interviewers. For cases that had multiple dockets, case-processing information in this study is based on the docket that had the most severe arraignment charge (based on Penal Law severity) in Criminal Court.3 When the most severe arraignment charges on two or 1

CJA conducts pre-arraignment interviews to measure the defendant’s community ties and to serve as the basis for making a recommendation as to whether or not the defendant should be released on recognizance at his or her first court appearance. Defendants who are arrested on a bench warrant, given a Desk Appearance Ticket (DAT), or who are held for arraignment on prostitution charges in the downtown Manhattan Criminal Court are not interviewed by CJA. CJA collects arrest and Criminal Court information for all arrestees, and arrestees were included in the First Quarter 2001 Dataset or Third Quarter 2002 Dataset whether or not they were interviewed by CJA. 2 DCJS, OCA, and the NYPD are not responsible for the methods or conclusions of this report. 3 New York State Penal Law categorizes most offenses according to their severity. The most serious crimes are A felonies, followed by felonies classified as being of severity B through E. Misdemeanors are less severe than felonies, and are classified as A or B misdemeanors or Unclassified misdemeanors (A misdemeanors are more severe than B misdemeanors, and Unclassified misdemeanors are less severe than B misdemeanors). Violations are less severe than misdemeanors, and are not considered crimes, although they can result in jail sentences. No distinctions of severity are made within the category of violations.


14 more dockets are of equal Penal Law severity, the top charge is determined according to guidelines developed by OCA. These guidelines provide a consistent set of rules for determining which of two arraignment charges of equal severity will be identified as the top arraignment charge. In New York State’s two-tiered court system for handling criminal cases, the Criminal Courts only have trial jurisdiction over cases having a most serious charge of misdemeanor or lesser severity. Most defendants charged with felonies are first arraigned in Criminal Court. Cases sustained at the felony level must be brought for prosecution in Supreme Court. In felony cases where the DA decides not to prosecute the case in Supreme Court (or the Grand Jury fails to return an indictment), the case may be disposed in Criminal Court by dismissal or by a plea to a reduced charge less severe than a felony, or by a transfer to another court’s jurisdiction (e.g., Family Court).4 The cases selected for inclusion in the analyses in this report include only cases that reached a final disposition in Criminal Court. The overwhelming majority (about 98%) of domestic violence cases citywide were disposed in Criminal Court. The First Quarter 2001 Dataset includes case processing information in Criminal Court through final disposition (and sentencing, if there was a conviction), or until November 30, 2001. The Third Quarter 2002 Dataset includes case processing information in Criminal Court through final disposition (and sentencing, if there was a conviction), or until May 6, 2003. Information about any final dispositions in Criminal Court beyond these cutoff dates was not included in the dataset. B. Identifying Domestic Violence Cases Social scientific and legal definitions of domestic violence have changed over the last 30 years (Peterson 2001). In New York State the statutory definition of domestic violence approximates what has come to be known in the social scientific literature as “family violence.” Under New York State’s Criminal Procedure Law (CPL) §530.11 (as amended by the 1994 Family Protection and Domestic Violence Intervention Act), family offenses are defined as offenses committed against a member of the same family or household, where “family or household” are defined as: (1) persons related by consanguinity or affinity; (2) persons legally married to each other; (3) persons who were formerly married, and; (4) persons who have a child in common, whether or not they have ever been married or lived together. New York State’s statutory definition of domestic violence excludes unmarried partners, unless they have a child in common. However, the New York City Police Department (NYPD) operates with an expanded definition of domestic violence that includes individuals who are not married, but who are cohabiting or have previously 4

The Family Courts have concurrent jurisdiction over certain domestic violence cases (Aldrich and Domonkos 2000). Some DV cases are heard only in Criminal Court, some are heard in both Criminal Court and Family Court, and others are heard only in Family Court. We do not have access to data on DV cases that are heard only in Family Court, and our report draws no conclusions about these cases.


15 lived together. This NYPD definition of “family” expands on New York State law by including “common-law” marriages, same-sex couples, and registered New York City domestic partners (NYPD 2000). By citywide agreement, the DA’s offices and the Criminal Courts in all five boroughs also use this expanded definition to identify DV cases, whether or not the relationship between the victim and defendant meets the New York State statutory requirements contained in CPL §530.11. To identify domestic violence cases, assistant district attorneys (ADAs) use information collected by the police about the relationship between the victim and the defendant, if any. The ADAs also often ask victims about their relationship with the defendant. When this information indicates that the defendant-victim relationship meets the NYPD expanded definition of domestic violence, the case is identified as a DV case. In all five boroughs DV case files are then given beige “backs” (special color-coded back sheets) to distinguish them from other case files. At Criminal Court arraignment, court clerks assign an arraignment hearing type of “DV” to domestic violence cases, and this designation is entered in OCA’s computerized court records. At the time the defendants in the first quarter of 2001 and the third quarter of 2002 were arrested, cases identified at arraignment as DV cases were processed in different ways depending on the borough. In all boroughs, most cases with a DV hearing type were sent to a specialized Criminal Court domestic violence part for post-arraignment appearances. However, there were exceptions. In Brooklyn and the Bronx, cases with a DV hearing type that involved physical or sexual abuse of children or other types of non-intimate partner violence (e.g., violence between siblings) were not sent to the specialized domestic violence parts. Finally, some cases that did not have a DV hearing type at arraignment were also sent to the specialized DV parts, presumably because information that these cases involved domestic violence became available only after arraignment. In this study, we identified “domestic violence” cases by relying on the court’s identification of these cases. We used information about both hearing type and court part, since not all DV cases were assigned a DV hearing type. We identified cases as domestic violence cases if the Office of Court Administration reported that: (1) the case had a domestic violence hearing type at Criminal Court arraignment, and/or, (2) the case had one or more appearances in a specialized domestic violence Criminal Court part.5

5

In the first quarter of 2001, the specialized domestic violence Criminal Court parts were AP-12 and AP-15 in Brooklyn (renamed DV1 and DV2 in January 2001), AP-10 and TAP-2 in the Bronx, AP-4 in Queens, D and JP13 (for DV jury trials) in Manhattan and AP2-DV in Staten Island. By the third quarter of 2002, the following specialized DV parts were added: Part IDV in the Bronx and part T-DV in Queens. Although AP2-DV in Staten Island was identified as a separate court part in our data, it was actually a specialized DV calendar. DV cases on this calendar were heard in an all-purpose part two days a week. We identified cases as DV cases if they had one or more appearances on this calendar.


16 Using these criteria, we identified 7,572 domestic violence cases in the First Quarter 2001 Dataset, about 9% of the total sample of 84,953 cases disposed in Criminal Court. There were 7,027 DV cases in the Third Quarter 2002 Dataset, about 10% of the 70,891 cases disposed in Criminal Court. In the combined First Quarter 2001 and Third Quarter 2002 Dataset, about 65% of the 14,599 DV cases had both a DV hearing type at arraignment and at least one appearance in a specialized DV part. An additional 18% of cases had a DV hearing type at arraignment but no appearance in a specialized DV part. These included cases that were disposed at arraignment as well as DV cases that were sent to Non-DV parts. Finally, 17% of the cases had at least one appearance in a specialized domestic violence part, but did not have a domestic violence hearing type at Criminal Court arraignment. As we noted in a previous report (Peterson 2003a), the measure identifying DV cases does have some limitations. First, there may be instances where a DV case was not identified as such in court records (i.e., it did not receive a DV hearing type at Criminal Court arraignment and did not appear in a specialized domestic violence part). The measure did not identify these as DV cases but instead categorized them as Non-DV cases. In the current study, this limitation affected our analyses in two ways. First, it reduced the sample size of DV cases on which we report. Nevertheless, we have an adequate sample size for our analyses, and were able to draw valid conclusions about DV cases from our sample. Second, we slightly overestimated the number of Non-DV cases in the sample. However, the number of DV cases misidentified as Non-DV cases is likely to be a very small proportion of the total number of Non-DV cases. C. Identifying Re-Arrests for Domestic Violence Offenses We measured recidivism in this study by examining re-arrests for new offenses during the pretrial period (i.e., between arrest and case disposition). We classify each re-arrest as a DV re-arrest or a Non-DV re-arrest. Our analyses use two measures of pretrial re-arrest. The first measure indicates whether the defendant had at least one re-arrest during the pretrial period for a new DV offense. The second measure indicates whether the defendant had at least one re-arrest during the pretrial period for any new offense. Unfortunately, re-arrest rates are likely to underestimate recidivism. New DV offenses may not lead to re-arrest, since many victims do not call the police when a new offense occurs, and police may not make an arrest even when they are called. To overcome the limitations of re-arrest, some studies measure recidivism using interviews with the victim. Interviewers can learn about incidents that did not result in calls to the police and re-arrest of the defendant. Rates of recidivism based on victim interviews are generally higher than rates based on re-arrest. Victim interviews also have weaknesses, however. It is often very difficult to reach victims and to complete interviews with them. Furthermore, victim interviews ignore the possibility that the defendant has re-offended with a new victim.


17 Because both types of data have strengths and weaknesses, we would have preferred to measure recidivism using both victim interviews and re-arrest data. We used re-arrest data for practical reasons—it was the only measure available to us. Although re-arrest may underestimate recidivism, it has two advantages over victim interviews. Data are potentially available for all defendants, not just those for whom victim interviews were completed. Also, it measures recidivism against new victims as well as against the same victim. One study of domestic incidents in New York State found that limiting a measure of recidivism to new incidents with the same victim reduced estimates of recidivism by 15% to 20% (Frisch et al. 2001). This suggests that a significant number of domestic violence defendants move on to new victims. Our measures of re-arrest also are affected by the problems described above regarding the identification of DV arrests. To the extent that the courts fail to identify DV arrests as such, our measures of re-arrest underestimated the actual number of DV rearrests and overestimated the actual number of Non-DV re-arrests. To create our re-arrest measures, we collected data on all re-arrests and then determined whether each re-arrest was for a DV case or a Non-DV case. This determination was made using the criteria described in Section II above. That is, we counted a re-arrest as a domestic violence case if either: (1) the case had a DV hearing type at arraignment, or (2) the case had any appearances in a specialized DV part. For re-arrests that were declined for prosecution, we used NYPD’s information about the nature of the defendant-victim relationship to identify DV cases. D. Selection of the Crimes against Persons and Property Subsample The current study used data from the Crimes Against Persons and Property (CAPP) Subsample described in a previous report (Peterson 2003a).6 This subsample was selected so that we could examine domestic violence cases across a wide range of charges. We began by selecting cases for this subsample where there was an alleged attempt to cause injury or where an overt threat of injury was made (Weis 1989). We used the most severe arraignment charge (based on Penal Law severity) to determine the nature of the offense, since this charge determines how the case is handled in the court system. We did not use the most severe arrest charge, which reflects charging decisions made by the police. We initially selected all cases that had a top (i.e., most severe) arraignment charge from any of the following New York State Penal Law articles: PL 120 (Assault), PL 130 (Sex Offense), PL 160 (Robbery), PL 260 (Crimes Against Children), or PL 265 (Weapons). Unfortunately, we were not able to include cases disposed in Criminal Court that had top arraignment charges from PL 125

6

In one of our reports (Peterson 2001), we also discussed results for an Assaults Subsample, which included all cases where the top arraignment charge was assault (PL 120). We used the Assaults Subsample to provide a more focused comparison of DV cases to similar Non-DV cases. In this study, we focus primarily on DV cases, and make only a few comparisons of DV to Non-DV cases. We therefore use only the Crimes Against Persons and Property Subsample for our analyses. This subsample includes information about the full range of DV offenses, including assaults as well as violations of orders of protection, crimes against children, etc.


18 (Homicide), PL 150 (Arson), and PL 135 (Kidnapping).7 Cases charged with offenses in these Penal Law articles were excluded since there were too few domestic violence cases with top arraignment charges in each of these Penal Law articles for reliable multivariate analysis. Recognizing that domestic violence often includes offenses that result in financial and psychological harm, rather than just physical harm, we also selected cases if they had a top arraignment charge from one of the following Penal Law articles: PL 140 (Burglary), PL 145 (Criminal Mischief), PL 155 (Larceny), PL 205 (Escape and Resisting Arrest), PL 215 (Criminal Contempt),8 and PL 240 (Public Order Offenses). Within each Penal Law article, we selected only those cases that had charges that could plausibly include elements comparable to those found in domestic violence cases. For example, cases with a top arraignment charge of Assault in the Third Degree (PL §120.00) were included in the subsample. However, cases of Gang Assault in the First Degree (PL §120.07) or Gang Assault in the Second Degree (PL §120.06) were excluded, since it is unlikely that a domestic violence case would include a gang assault charge. Similarly, in PL 240, prostitution charges (PL §240.37) were excluded. We then narrowed the subsamples further to identify an appropriate group of cases for the analysis. First, we limited the sample to Summary Arrests (i.e., cases in which the defendant was held in custody pending Criminal Court arraignment), excluding cases where the defendant was issued a Desk Appearance Ticket (DAT) and released by the arresting officer. DAT’s are rarely issued in DV cases. We also excluded cases with juvenile defendants (under age 16), cases that did not reach a final disposition by the cutoff date, and cases that were missing data on the defendant’s criminal history or sex. We also excluded cases that were disposed in Criminal Court on Vehicle and Traffic Law (VTL) or Administrative Code (AC) charges. After these exclusions, the Crimes Against Persons and Property Subsample included 12,882 DV cases in the combined First Quarter 2001 and Third Quarter 2002 Dataset. For comparisons to Non-DV cases, we used a sample of Non-DV cases drawn only from the First Quarter 2001 Dataset. The Crimes Against Persons and Property subsample includes 19,195 Non-DV cases in the First Quarter 2001 Dataset. Although we could have included Non-DV cases from the Third Quarter 2002 Dataset as well, we elected not to do that for three reasons. First, time and resource constraints prevented us from collecting re-arrest data for Non-DV cases in that dataset. Second, the sample size of Non-DV cases in the First Quarter 2001 dataset is sufficiently large to allow us to make reliable statistical comparisons with DV cases. Finally, the results for 2002 are likely to be quite similar to the results for 2001.

7 8

Most, but not all, of these cases were sustained as felonies and disposed in Supreme Court. Penal Law article 215 includes violations of orders of protection.


19 E. Using a Defendant-Based Data File The dataset analyzed in this study is a defendant-based data file that includes information on all defendants who were arrested in the first quarter of 2001 and third quarter of 2002. While the vast majority of defendants had only one case in the case-based data file, some defendants were arrested two or more times. For most defendants who had multiple arrests, we included information in the defendant-based data file about the first arrest, i.e., the arrest that occurred earliest. However, for defendants whose first arrest was not for a DV case but who had a second or subsequent DV arrest, we selected the first DV arrest, not the first arrest. This procedure enabled us to identify all the defendants who had at least one DV arrest. We first created two defendant-based files, one for the first quarter of 2001 and one for the third quarter of 2002. We then combined the two files, and took additional steps to eliminate arrests from the third quarter 2002 for defendants who had already been arrested in the first quarter 2001. After eliminating these re-arrests from the Third Quarter 2002 Dataset, the number of DV cases in the combined First Quarter 2001 and Third Quarter 2002 Dataset was reduced by 310 cases, from 12,882 to 12,572. Finally, we restricted the sample to docketed cases, resulting in a sample of 11,938 DV cases. The combined dataset included only one arrest for each defendant. F. Plan of Analysis Chapter 3 uses the combined First Quarter 2001 and Third Quarter 2002 Dataset to describe pretrial release outcomes for DV cases, and to compare those patterns to Non-DV cases. Subsequent chapters focus only on DV defendants, since other research has addressed pretrial misconduct among the larger population of all defendants. Chapter 4 examines the factors that influence the likelihood of pretrial failure to appear among DV defendants. Chapter 5 examines the factors that influence the likelihood of pretrial re-arrest for a new DV offense. Chapter 6 summarizes the results of the study.


20

[This page intentionally left blank]


21 III. PRETRIAL RELEASE OUTCOMES This chapter describes pretrial release outcomes for DV defendants in New York City, and compares them to the outcomes for Non-DV defendants. As explained in Chapter 2, data for DV defendants were drawn from the combined First Quarter 2001 and Third Quarter 2002 Dataset (also referred to below as “the combined dataset”). Data for Non-DV defendants were drawn solely from the First Quarter 2001 dataset. Data for both DV and Non-DV defendants was limited to cases in the Crimes Against Persons and Property Subsample, as described in Chapter 2, Section D. A. Pretrial Release Decisions In New York State, judges have several options when considering the release of a defendant at Criminal Court Arraignment (Marks et al. 1996). In most cases, the judge can release the defendant on recognizance (ROR), or can set bail in any amount and decide on the type of bond or whether a lower cash alternative to posting bond would be accepted. In a small number of felony cases, the judge also has the option of remanding the defendant without bail. In this section, we describe various aspects of the release decision in DV and Non-DV cases. Many criminal cases in New York City are disposed at arraignment, and the question of whether the defendant will be released during the pendency of the case never arises. However, by citywide agreement, domestic violence cases are generally not disposed at arraignment. This allows the District Attorney’s offices and the courts time to learn more about the details of these often complex cases and to transfer many of these cases to the specialized DV court parts (see discussion in Chapter 2, section B). This policy insures that the court must make a release decision at arraignment in almost all DV cases. These different policies for DV and Non-DV cases produce large differences in the rate at which cases are disposed at arraignment. As shown in Figure 1, only about 2% of DV cases were disposed at arraignment, FIGURE 1 CASES DISPOSED AT ARRAIGNMENT Combined First Quarter 2001 and Third Quarter 2002 Dataset

Disposed at Arraignment 2%

DV CASES

NON-DV CASES

(N=11,938)

(N=18,331)

Not Disposed at Arraignment 98%

Disposed at Arraignment 47%

Not Disposed at Arraignment 53%


22 compared to 47% of comparable Non-DV cases. As we discuss pretrial release practices for DV and Non-DV cases, it is important to remember that an arraignment release decision was made in almost all DV cases, but only in about half of Non-DV cases. Any differences we found between DV and Non-DV cases may have been affected by the exclusion of certain types of Non-DV cases from the comparison. Among cases that were not disposed at arraignment, over two thirds of defendants were released on recognizance—67% of DV defendants and 63% of NonDV defendants (see Figure 2). A small proportion was released on bail (4% of DV defendants and 5% of Non-DV defendants). Almost all the defendants who were not released were held on bail; less than 1% were remanded (data not shown). The rarity of remand is not surprising, since this dataset consists of cases that were disposed in the (lower) Criminal Court. Remand is generally reserved for defendants facing felony charges whose cases are likely to lead to indictment and disposition in the (upper) Supreme Court. FIGURE 2 RELEASE STATUS AT ARRAIGNMENT (Defendants with Cases Continued Beyond Arraignment) Combined First Quarter 2001 and Third Quarter 2002 Dataset

ROR 67%

DV CASES

NON-DV CASES

(N=11,737)

(N=9,711)

Not Released at Arraignment 29%

Bail Made 4%

ROR 63%

Not Released at Arraignment 32%

Bail Made 5%

A closer look at the cases in which bail was set indicated that only a small proportion of defendants were able to make bail at arraignment. About 12% of DV defendants and 13% of Non-DV defendants for whom bail was set at arraignment were released on bail at arraignment (See Figure 3). Defendants who were not released at arraignment were sometimes released later. Figure 4 shows that about one sixth of defendants were released after arraignment—17% of defendants in DV cases and 16% of defendants in Non-DV cases. (A few of these defendants made bail on the arraignment day, but after the arraignment [data not shown]. These defendants were first held by the Department of Correction, and then released from custody.) In Non-DV cases, one sixth of defendants were never released (16%), while only 12% of defendants in DV cases were never released. This difference is largely explained by the higher proportion of defendants in DV cases who were released on recognizance at arraignment, as indicated in the discussion of Figure 2.


23 FIGURE 3 BAIL MAKING AT ARRAIGNMENT (Defendants for Whom Bail Was Set at Arraignment) Combined First Quarter 2001 and Third Quarter 2002 Dataset DV CASES

NON-DV CASES

(N=3,862)

(N=3,488)

Bail Set, Not Made 88%

Bail Made 12%

Bail Set, Not Made 87%

Bail Made 13%

FIGURE 4 STAGE OF FIRST RELEASE (Defendants with Cases Continued Beyond Arraignment) Combined First Quarter 2001 and Third Quarter 2002 Dataset DV CASES

NON-DV CASES

(N=11,737)

(N=9,711) Never Released 12%

Released at Arraignment 71%

Released After Arraignment 17%

Released at Arraignment 68%

Never Released 16%

Released After Arraignment 16%

Defendants who were released after arraignment were usually released within a week of the arraignment. The median time to release for these defendants was 5 days, for both DV and Non-DV defendants (see Figure 5). Because some cases took considerably longer, the mean time to release was 11 days for DV defendants and 12 days for Non-DV defendants. When bail is set for a defendant, judges set the amount of a bond that must be posted for the release of the defendant. They also have the option of setting a cash alternative, which is the amount of cash that would have to be posted with the court for release of the defendant. Since bonds can usually be obtained by putting up half the value of the bond in cash with a bail bondsman, a cash alternative is considered a true cash alternative in this study only if it is less than half the amount of the bond. (Bail bondsmen actually vary in the percentage of the bond required in cash, but we use 50% as a typical figure.) If a judge sets bail at $1,000 bond or cash, there is no true cash alternative, since a bond often can be obtained for $500. Similarly if the judge sets bail at $1,000 bond or $500 cash, there is no true cash alternative, since the bond usually


24 FIGURE 5 NUMBER OF DAYS BETWEEN ARRAIGNMENT AND FIRST RELEASE (Defendants Released After Arraignment) Combined First Quarter 2001 and Third Quarter 2002 Dataset 14

12

11

MEAN

5

7

5 MEDIAN

0 DV CASES

NON-DV CASES

(N=1,959)

(N=1,500)

could be obtained for $500 cash. However, if the judge sets bail at $1,000 bond or $250 cash, there is a true cash alternative, since bail can be secured with $250 cash, which is less than the $500 cash that would be needed to secure the $1,000 bond. To calculate whether there was a true cash alternative, we excluded cases with $1 bond (indicating that the defendant was already being held on another pending case). In cases with multiple dockets, we summed the amounts set on each docket. As shown in Figure 6, when judges set bail at arraignment, they offered a true cash alternative about 6% of the time for both DV and Non-DV defendants. FIGURE 6 TRUE CASH ALTERNATIVES FOR CASES WITH BAIL SET AT ARRAIGNMENT (Defendants for Whom Bail Was Set at Arraignment, Excluding Cases with $1 Bond) Combined First Quarter 2001 and Third Quarter 2002 Dataset DV CASES

NON-DV CASES

(N=3,795)

(N=3,295)

No True Cash Alternative 94%

No True Cash Alternative 94% True Cash Alternative 6%

True Cash Alternative 6%

To calculate the amount of bail set at arraignment, we used the amount of the bond required unless there was a lower cash alternative set. In cases with a lower cash alternative, we used the lower cash amount, even if it was not a “true” cash alternative


25 (i.e., even if it was not less than 50% of the bond amount). As before, we summed the amounts of bail set across dockets, and excluded cases with $1 bond. The median bail set at arraignment was $1,000 for both DV and Non-DV defendants (see Figure 7). However, because high bail amounts were more common for Non-DV defendants, the means differed considerably. The mean bail set for DV defendants at arraignment was $2,060, compared to $3,700 for Non-DV defendants. FIGURE 7 AMOUNT OF BAIL SET AT ARRAIGNMENT (Defendants for Whom Bail Was Set at Arraignment) Combined First Quarter 2001 and Third Quarter 2002 Dataset $3,700

$4,000

MEAN

$2,060 $2,000

$1,000

$1,000

DV CASES

NON-DV CASES

(N=3,795)

(N=3,295)

MEDIAN

$0

We also examined the amount of bail posted at the time of their first release for those defendants who were released on bail. The median bail posted by DV defendants was $750 at the time of their first release, compared to $1,000 for Non-DV defendants (see Figure 8). The mean amount of bail posted also differed. DV defendants posted bail of $1,210, on average, compared to an average of $1,940 for Non-DV defendants. FIGURE 8 AMOUNT OF BAIL AT TIME OF FIRST RELEASE (Defendants Released on Bail) Combined First Quarter 2001 and Third Quarter 2002 Dataset $1,940

$2,000 $1,500 $1,000

MEAN

$1,210

$1,000 MEDIAN

$750

$500 $0 DV CASES (N=1,683)

NON-DV CASES (N=1,240)


26

To summarize our findings on arraignment release decisions, we found very few differences in the treatment of DV and Non-DV defendants whose cases were not disposed at arraignment. DV defendants were only slightly more likely to be released at arraignment. Most defendants (DV and Non-DV) were released on recognizance at arraignment. Both DV and Non-DV defendants for whom bail was set at arraignment were unlikely to post bail at arraignment. However, DV defendants who were held at arraignment were more likely to secure their release after arraignment than Non-DV defendants. There was little difference between DV and Non-DV defendants in time to first release. The median amount of bail set was the same for DV and Non-DV defendants, although the mean was higher for Non-DV defendants. DV defendants who were released on bail were released on lower amounts of bail than Non-DV defendants. B. Pretrial Failure to Appear Arraignment release decisions are supposed to insure that the defendant returns for future court appearances. We now examine defendant-based rates of pretrial failure to appear, specifically whether defendants who were released ever failed to appear in court during the pendency of the case. (Note that defendants who missed multiple appearances were counted only once in calculating the FTA rate.) About 12% of NonDV defendants failed to appear, compared to 10% of DV defendants (see Figure 9). This difference is statistically significant. Although DV defendants were slightly more likely to be released than Non-DV defendants, this does not appear to have increased the risk of failure to appear. In fact, DV defendants were slightly more likely to come back for their court appearances than Non-DV defendants. FIGURE 9 FAILURE-TO-APPEAR RATES FOR DEFENDANTS WHO WERE EVER RELEASED Combined First Quarter 2001 and Third Quarter 2002 Dataset 20% 12% 10% 10%

0% DV CASES

NON-DV CASES

(N=10,299)

(N=8,129)

The data presented in Figure 9 do not take into account differences between DV and Non-DV defendants in time at risk. However, time at risk was actually longer for DV defendants than for Non-DV defendants (data not shown). So the higher rate of FTA for Non-DV defendants cannot be attributed to differences in time at risk.


27 C. Pretrial Re-Arrest Although arraignment release decisions in New York State are not supposed to be made on grounds of public safety, safety considerations are of considerable interest. Other states permit judges to take the likelihood of re-offending into consideration in making arraignment release decisions. In New York State, there are concerns that DV victims may be exposed to renewed violence or threats of violence when the defendant is released during the pretrial period. Orders of protection are issued at arraignment in the cases of almost all DV defendants. However, although violations of orders of protection increase the penalties against the defendant, it is not clear how effective these orders are at deterring new DV offenses. To explore these safety issues in DV cases, we now examine rates of re-arrest during the pretrial period. As noted in Chapter 2, re-arrest has limitations as a measure of recidivism, and the results presented here probably underestimate recidivism. DV and Non-DV defendants who were released were equally likely to be rearrested during the pretrial period. About 15% of released defendants were re-arrested for at least one new offense prior to the disposition of their case (see Figure 10A). There were, however, significant differences in the types of crimes for which DV and Non-DV defendants were re-arrested. About 9% of DV defendants were re-arrested for at least one new DV offense during the pretrial period, compared to only 1% of Non-DV defendants (see Figure 10B). About 7% of released DV defendants were re-arrested for at least one new Non-DV offense during the pretrial period, compared to 14% of Non-DV defendants. (Note that some defendants were re-arrested for both DV and Non-DV offenses.) These findings indicate that although DV and Non-DV defendants were equally likely to be re-arrested for a new pretrial offense, DV defendants posed the greatest risk of committing a new DV offense. These findings do not control for time at risk, which was slightly longer for DV defendants (data not shown). However, the difference in time at risk is unlikely to affect the basic pattern of findings reported here. D. Combined Rates of Pretrial Failure to Appear and Pretrial Re-Arrest To examine overall rates of pretrial misconduct, we now combine the information about pretrial failure to appear and pretrial re-arrest. Specifically, we look at two measures of pretrial misconduct. One measure is the combined rate of FTA and pretrial re-arrest for any new offense. This measure indicates the percentage of defendants who failed to appear and/or were re-arrested for any new offense during the pretrial period. The second measure is the combined rate of FTA and pretrial re-arrest for a new DV offense. This measure indicates the percentage of defendants who failed to appear and/or were re-arrested for a new DV offense during the pretrial period. Each measure includes some defendants who engaged in both types of pretrial misconduct. As shown in Figure 11, DV and Non-DV defendants were about equally likely to FTA and/or to be re-arrested for any new offense during the pretrial period. About 21% of DV defendants


28

FIGURE 10A PRETRIAL RE-ARREST RATES FOR ANY NEW OFFENSES FOR DEFENDANTS WHO WERE EVER RELEASED Combined First Quarter 2001 and Third Quarter 2002 Dataset 20% 15%

15%

DV CASES

NON-DV CASES

10%

0% (N=10,299)

(N=8,129)

FIGURE 10B PRETRIAL RE-ARREST RATES FOR NEW DV AND NON-DV OFFENSES FOR DEFENDANTS WHO WERE EVER RELEASED Combined First Quarter 2001 and Third Quarter 2002 Dataset

14%

20% 7% 10%

Pretrial Re-arrest Rates for New Non-DV Offenses

9% 1%

0%

DV CASES

NON-DV CASES

(N=10,299)

(N=8,129)

Pretrial Re-arrest Rates for New DV Offenses


29 and 22% of Non-DV defendants engaged in one or both of these types of pretrial misconduct. There was, however, a significant difference between the two groups when DV re-arrests were taken into account. About 17% of DV defendants failed to appear and/or were re-arrested for a new DV offense during the pretrial period. Reflecting their lower rates of committing a new DV offense (as shown in Figure 10B), about 12% of Non-DV defendants failed to appear and/or were re-arrested for a new DV offense during the pretrial period.

FIGURE 11 COMBINED RATES OF FTA AND PRETRIAL RE-ARREST FOR DEFENDANTS WHO WERE EVER RELEASED Combined First Quarter 2001 and Third Quarter 2002 Dataset

22%

21% 24%

17% 12%

0%

12%

Combined Rates of FTA and Pretrial Re-Arrest for Any New Offenses Combined Rates of FTA and Pretrial Re-Arrest for New DV Offenses

DV CASES

NON-DV CASES

(N=10,299)

(N=8,129)

E. Summary and Discussion of Findings Our review of pretrial release decisions showed that there was one major difference between DV and Non-DV cases. Almost all DV cases (98%) were continued beyond arraignment, requiring the arraignment judge to make a pretrial release decision. By contrast, only 53% of comparable Non-DV cases were continued beyond arraignment. In almost half of Non-DV cases, the arraignment judge did not make a pretrial release decision because the case was disposed at arraignment. The high continuance rate at arraignment for DV cases reflected citywide policy, which generally requires DV cases to be continued while further information about the case is gathered. When analyses were restricted to cases continued beyond arraignment, pretrial release decisions in DV and Non-DV cases were relatively similar. Defendants in DV cases were only slightly more likely to be released on recognizance at arraignment (67% vs.


30 63%). DV and Non-DV defendants were about equally likely to be released on bail at arraignment, and those who were held at arraignment were about equally likely to be released at some time after arraignment. The median time to first release was about the same, 5 days for both groups. This finding was not surprising, since many defendants must be released on recognizance when the DA does not corroborate the complaint within 5 days of the arraignment (6 days if a Sunday intervenes), as required by New York CPL ยง170.70. Both DV and Non-DV defendants were offered a true cash alternative in only 6% of cases in which bail was set, and the median amount of bail set was $1,000. However, DV defendants who were released on bail posted lower amounts of bail than Non-DV defendants. The median was $750 for DV defendants and $1,000 for Non-DV defendants. Among those who were released at some time prior to the disposition of their cases, defendants in DV cases were slightly less likely to fail to appear than defendants in Non-DV cases (10% vs. 12%). This difference is surprising given that many more DV defendants had their cases continued beyond arraignment, and a slightly higher proportion of DV defendants were released prior to the disposition of their case. Although a higher proportion of DV defendants were released, their FTA rate was lower than for Non-DV defendants. Defendants in DV cases and Non-DV cases were equally likely to be re-arrested for a new offense during the pretrial release period. About 15% of each group were rearrested. However, 9% of DV defendants were re-arrested for a new DV offense compared to only 1% of Non-DV defendants. So although the overall re-arrest rate was the same for DV and Non-DV defendants, the risk of committing a new DV offense during the pretrial period was much greater for DV defendants. Combined rates of pretrial misconduct were similar if we consider whether defendants failed to appear and/or were re-arrested for any new offense during the pretrial period. About 21% of DV defendants and 22% of Non-DV defendants engaged in one or both types of pretrial misconduct. However, DV defendants were more likely than Non-DV defendants to engage in pretrial misconduct if we consider only re-arrests for new DV offenses. About 17% of DV defendants, but only 12% of Non-DV defendants, failed to appear and/or were re-arrested for a new DV offense during the pretrial period.


31 IV. PREDICTING PRETRIAL FAILURE TO APPEAR FOR DV DEFENDANTS A. Development of a Model Predicting Pretrial Failure to Appear New York State law directs judges to make release decisions based on the restrictions that are necessary to insure that the defendant returns for future court appearances (New York State Criminal Procedure Law §510.30). Factors that must be considered and taken into account “include the defendant’s character, reputation, habits and mental condition; employment and financial resources; family ties and the length of residence in the community; criminal record; previous adjudication as a juvenile delinquent; previous record of failure to appear for required court appearances; the strength of the evidence and any other factor indicating probability of conviction and the likely sentence if convicted” (Phillips 2004, p. 2). As discussed in Chapter 3, about 10% of DV defendants failed to appear for at least one court appearance during the pretrial period. In this chapter, we develop a logistic regression model to examine which factors increased or decreased the likelihood of pretrial failure to appear (FTA) among DV defendants. (For a description of logistic regression analysis and a discussion of how to interpret regression results, see Appendix A.) A variety of factors that previous research suggests may affect FTA were considered for inclusion in the logistic regression model (see review of literature in Chapter 1). To the extent possible, the model also considered factors that are identified in the New York State Criminal Procedure Law as relevant to the court’s release decision. Drawing on previous literature and state statutes, several categories of variables were considered for inclusion in the model: criminal history, release recommendation, community ties, charge characteristics, release characteristics, and geographic and demographic characteristics. Information on drug use by the defendant was not available. Criminal History. The available measures of criminal history included prior arrests, convictions, sentences and warrants. We were able to determine whether the defendant had any prior arrests, any prior misdemeanor convictions, and/or any prior felony convictions. Information was also available about the number of open cases, the number of prior misdemeanor convictions, and number of prior felony convictions. The sentencing variables included measures of the number of prior misdemeanor jail sentences, whether the defendant had ever been sentenced to prison, was currently on parole from prison, and/or had ever been sentenced as a youthful offender. We expect that defendants with more serious criminal records are more likely to fail to appear. Because prior failure to appear is a strong predictor of future likelihood of failure to appear, we also included a measure which indicated whether the defendant had two or more previous bench warrants on his/her record. Release Recommendation. CJA’s release recommendation indicates whether the defendant received a full or qualified release recommendation, or was not


32 recommended for release because of weak community ties or because of an active bench warrant at the time of arrest. Community Ties. Although CJA’s release recommendation is based on a composite assessment of a variety of community ties, we also considered whether individual measures of community ties independently affected the likelihood of failure to appear. The community ties measures considered were: employment status, length of time at current address, whether the defendant lives with someone, whether the defendant expects someone to attend the arraignment, whether the defendant has a telephone, and whether the defendant lives in the New York City area. In general, we expect that defendants with stronger community ties are less likely to fail to appear. Charge Characteristics. Several charge characteristics were also considered for inclusion in the models. Number of arrest charges and the type and severity of the charge at arraignment were tested in the model. The type of arraignment charge was classified in three categories, based on the most severe arraignment charge: assault, criminal contempt (usually for a violation of an order of protection), and other. Charge severity was based on New York State Penal Law, and ranged from most severe (B felony) through least severe (violation). We expect that defendants with more arrest charges are more likely to fail to appear, however we have no expectations about the type of arraignment charge. Regarding severity of arraignment charge, we expect that defendants facing more serious charges will be less likely to fail to appear. Although this expectation may be counterintuitive, it is consistent with previous research (Siddiqi 2006). Type of Release. The model also tested for the impact of two aspects of the court’s release decision: whether the defendant was released at arraignment or at a later appearance, and whether the defendant was released on recognizance, or on bail. We expected that defendants released at arraignment were less likely to fail to appear. Those who were not released until a later appearance were initially considered by the judge as a poor risk for release. Prior research has produced mixed results regarding the effect of release on bail vs. recognizance. However, based on the results of prior CJA research on failure to appear in New York City (Siddiqi 1999), we expect defendants released on recognizance to have higher rates of FTA than those released on bail. Geographic and Demographic Characteristics. The geographic and demographic variables considered for the model are: borough, age, sex, ethnicity, and defendant-victim relationship. We did not expect any differences in rates of failure to appear by borough, ethnicity, sex, or defendant-victim relationship. We expected older defendants to be less likely to fail to appear. Control Variables. Finally, our model includes two control variables. One is a correction for selection bias associated with the likelihood of release. Since our analyses of failure to appear are limited to defendants who were released prior to case disposition, this correction adjusts for the influence of some variables on both the


33 probability of release and the probability of failure to appear (see Appendix A, Section 2, for a detailed discussion of selection bias). The second control variable included in the model is time at risk. Since defendants whose cases take longer to reach a disposition have more opportunities to fail to appear, it is essential that the model predicting failure to appear adjust for the amount of time each defendant was at risk. To calculate time at risk, we counted the number of days between the defendant’s release and the disposition of the case. We then subtracted any time that the defendant was out on his or her first bench warrant for failing to appear. We were not able to adjust time at risk for time on a second or later bench warrant, but very few defendants had two or more bench warrants on their case. Nor were we able to subtract from time at risk any time the defendant was held (either on bail or remand) after his or her first release. This calculation would have been extremely complicated, and would be difficult to accurately complete. Because few defendants had multiple periods of detention and release, we do not believe that making this adjustment would have changed our results significantly. The two control variables (the correction for selection bias and time at risk) were entered first. Whether they were statistically significant or not, we wanted these two variables included in the final model. To develop the remainder of the logistic regression model and to decide which variables to include in the final model, we used a stepwise procedure. This procedure allowed us to identify and select the strongest predictors of failure to appear. First we tested all the criminal history measures, adding the strongest criminal history variable to the model first, followed by the second strongest criminal history variable, etc. We added variables to the model until all the statistically significant criminal history measures were entered. We then used the stepwise procedure to enter community ties variables, adding the strongest ones first, until all statistically significant community ties variables were entered. Similar procedures were followed in succession for charge characteristics, release characteristics, and geographic/demographic characteristics. Information about the distribution of each set of variables is included in Appendix B. One final point about the development of the model merits discussion. We limited our analysis to DV defendants. We did this for three reasons. First, when we pooled the data for DV and Non-DV defendants, we found that there were no statistically significant differences in their FTA rates after controlling for the other variables in the model (data not shown). DV defendants are no less likely to fail to appear than Non-DV defendants, once the influence of the other variables in the model is taken into account. Second, when we analyzed data for Non-DV defendants separately (data not shown), we found that the model was substantially similar to the model for DV defendants. Although some of the variables included in the final models differed for DV and Non-DV defendants, the basic findings were quite similar. Third, there is extensive prior research on the factors influencing pretrial failure to appear among the general population of defendants in New York City (Siddiqi 1999, 2000, 2002, 2003a). The current study is primarily focused on the issue of pretrial misconduct by DV defendants, and the results of a model for DV defendants are of particular interest.


34 B. Model Predicting Pretrial Failure to Appear for DV Defendants Based on the procedure described above, we developed the final logistic regression model predicting failure to appear among DV defendants (see Table 4-1). The proportion of variance explained by the model was relatively low—about 10% (see Nagelkerke R2), suggesting that the likelihood of failure to appear is primarily a function of factors not considered here. Nevertheless, this model provided information about a wide variety of factors that did have an influence on the likelihood of failure to appear. We now consider each of these in turn. Of the two control variables, only time at risk had a statistically significant effect on the likelihood of failure to appear. As expected, defendants with a longer time at risk were more likely to fail to appear. This variable was actually one of the strongest predictors in the model, as indicated by the size of the standardized beta, .45. Although the correction for selection bias was not statistically significant, its effect was in the expected direction. Defendants who were more likely to be released by the court were less likely to fail to appear. As expected, defendants with more serious criminal histories were more likely to fail to appear. Defendants with any prior arrests, defendants with two or more prior misdemeanor jail sentences, and defendants with two or more prior bench warrants were all more likely to fail to appear. The most important of these variables, as indicated by the large standardized beta, was the measure of any prior arrests. The remainder of our measures of criminal history did not have any effect on the likelihood of failure to appear, and were not included in the model. CJA’s release recommendation was also a statistically significant predictor of FTA rates. As expected, defendants who had an open bench warrant at the time of arrest (hence ineligible to be recommended) were more likely to FTA than those who were recommended or who received a qualified recommendation. However, defendants who received no release recommendation (because of weak New York City ties) were no more likely to FTA than those who were recommended or who received a qualified recommendation. While this may at first seem surprising, the model also includes the individual measures of community ties that were used to determine the recommendation. As we will see in the discussion of the individual community ties items, some of these items are statistically significant. The effects of these individual items have captured the effect of the “No Recommendation” category. A separate analysis (not shown) indicates that when the individual items are excluded from the model, defendants who received “No Recommendation” are significantly more likely to fail to appear. This indicates that the “No Recommendation” category and the three individual items were roughly equivalent substitutes for each other in capturing the effect of community ties on the likelihood of failure to appear. Although the CJA recommendation was included in the model, we also considered whether some of the individual community ties items that are used in the CJA recommendation had an independent effect. Defendants who were unemployed,


35 TABLE 4-1 LOGISTIC REGRESSION MODEL PREDICTING LIKELIHOOD OF FAILURE TO APPEAR 1 DV CASES, CRIMES AGAINST PERSONS AND PROPERTY SUBSAMPLE Combined 2001-2002 Dataset

INDEPENDENT VARIABLES

2

Standardized

Odds

β

Ratio

CONTROL VARIABLES SELECTION BIAS CORRECTION: LIKELIHOOD OF RELEASE TIME AT RISK

-0.03 0.45 ***

0.80 1.01

0.32 ***

1.62

-0.01 0.19 *** 0.15 ***

0.97 1.80 1.61

0.06 0.17 *** 0.02

1.11 1.80 1.09

0.19 *** 0.18 *** 0.13 *

1.36 1.34 1.28

0.04 0.09 *

1.10 1.19

0.17 ** -0.30 ***

1.40 0.54

DEFENDANT'S CRIMINAL HISTORY ANY PRIOR ARRESTS PRIOR MISDEMEANOR JAIL SENTENCES Reference Category: None One Prior Misdemeanor Jail Sentence Two or More Prior Misdemeanor Jail Sentences TWO OR MORE PRIOR BENCH WARRANTS RELEASE RECOMMENDATION Reference Category: Recommended or Qualified Recommendation No Recommendation (Weak NYC Ties) Open Bench Warrant at Time of Arrest Missing DEFENDANT'S COMMUNITY TIES UNEMPLOYED DOES NOT EXPECT ANYONE AT ARRAIGNMENT HAS NO TELEPHONE ARRAIGNMENT CHARGE TYPE Reference Category: Assault Criminal Contempt Other RELEASE CHARACTERISTICS RELEASED AFTER ARRAIGNMENT RELEASED ON BAIL

Table Continues on Next Page


36 TABLE 4-1 (continued)

INDEPENDENT VARIABLES

2

Standardized

Odds

β

Ratio

BOROUGH Reference Category: Brooklyn Manhattan Queens Staten Island Bronx

0.02 0.01 0.14 ** 0.07

1.04 1.02 1.71 1.15

-0.28 *** -0.50 *** -0.48 ***

0.63 0.44 0.43

DEFENDANT'S DEMOGRAPHIC CHARACTERISTIC AGE: Reference Category: Age 16-20 Age 21-29 Age 30-39 Age 40 and over 2

Nagelkerke R (N of cases)

0.10 *** (10,299)

NOTES 1 2

See text for a description of the dataset and the subsample. See Appendix C for information about the measurement and coding of the variables.

* Statistically significant at p < .05 ** Statistically significant at p < .01 *** Statistically significant at p < .001


37

defendants who did not expect anyone at arraignment, and defendants who had no telephone were all more likely to fail to appear. For each of these variables, the odds of FTA were about 1.3 times greater for defendants with weak community ties when compared to defendants with strong community ties. Our findings regarding the effect of arraignment charge type show that defendants were more likely to fail to appear when they were charged with “other” charges, i.e., charges other than assault or criminal contempt. Since the “other” category included a wide variety of charges, this finding is difficult to interpret. Both release characteristics had a statistically significant effect on the likelihood of failure to appear. Defendants released after arraignment (as opposed to those released at arraignment) were more likely to fail to appear. This suggests that defendants who were not released at arraignment generally posed a higher risk of failure to appear if they were subsequently released. Although these defendants were later released, their inability to secure release at arraignment was presumably due to the identification by the arraignment judge of significant reasons to set bail in an amount the defendant could not make. Their subsequent release, either on ROR or reduced bail, or on the original bail, posed greater risks. Our findings also indicated that defendants released on bail had a lower risk of failure to appear than defendants released on recognizance. This finding was consistent with prior CJA research (Siddiqi 1999), although, as noted earlier, other research has produced mixed results. Since our model controlled for many factors that may be correlated with this variable, the effect identified here seems unlikely to be attributable to many of the case or defendant characteristics associated with release on bail vs. ROR. Most importantly, prior criminal history, community ties and type of charge were included in the model and did not account for the lower FTA rate associated with release on bail. This suggests that other, unmeasured, characteristics (e.g., the defendant’s financial resources) accounted for this effect, and/or that bail was more effective than ROR at preventing FTA for these defendants. We found that the borough where the case was docketed had a statistically significant effect on the likelihood of failure to appear. Specifically, DV defendants whose cases were docketed in Staten Island were more likely to fail to appear than defendants whose cases were docketed in other boroughs. The number of defendants in Staten Island was relatively small, and this finding may be a statistical anomaly. In any event, we make no attempt to interpret this finding, as we are not aware of any reason why the FTA rate for DV defendants should be higher in Staten Island. Finally, we found that age had a very powerful effect on the likelihood of FTA among DV defendants. Older defendants were less likely to fail to appear than younger defendants. The odds that a defendant aged 21-29 would make all scheduled appearances were 1.6 times (1/.63) greater than for 16-20 year olds. For defendants aged 30-39 and those aged 40 and over, the odds of making all scheduled appearances were about 2.3 times (1/.44 and 1/.43, respectively) greater than for 16-20 year olds.


38 As indicated by the size of the standardized betas, being in the age 30-39 category or the age 40 and over category were the two strongest predictors of failure to appear in the model. These findings indicate that for DV defendants, age was the most important predictor of failure to appear. C. Summary and Discussion of Findings Predicting pretrial failure to appear is a difficult task. Our model was only able to account for 10% of the variance in the likelihood of failure to appear. This problem is fairly typicalâ&#x20AC;&#x201D;most previous research has found that predictive models of failure to appear have only weak explanatory power. In spite of these limitations, we were able to identify a number of factors that had an influence on failure to appear. After controlling for time at risk, the strongest predictor of failure to appear among DV defendants was the defendantâ&#x20AC;&#x2122;s age. Older defendants, particularly those over 30, were much more likely to make all their scheduled court appearances. The next strongest predictors were whether the defendant had any prior arrests and whether the defendant was released on bail. DV defendants who had any prior arrests were more likely to fail to appear than those with no prior arrests. DV defendants who were released on bail (about 17% of those who were released) were less likely to fail to appear than those released on recognizance. This difference cannot be explained away by any of the factors included in the model. The lower FTA rate for defendants released on bail cannot be attributed to community ties, criminal record, age, or any of the other factors included in the model. This may indicate that releasing DV defendants on bail reduced their likelihood of FTA, or it may indicate that we did not include in our model other relevant differences between defendants released on bail vs. those released on recognizance. Other strong predictors of FTA (as measured by the size of the standardized beta) were additional measures of criminal history, community ties measures, and whether the defendant was released after arraignment. Among the criminal history measures, defendants with two or more prior misdemeanor jail sentences were more likely to fail to appear. Defendants with two or more prior bench warrants, as well as those with an open bench warrant at the time of arrest, were more likely to fail to appear. This finding suggests that defendants with a prior history of failure to appear were more likely to fail to appear on their current DV case. What is somewhat surprising about this finding is not that a history of bench warrants increased the likelihood of failure to appear, but rather that this variable was not the strongest predictor of failure to appear. Prior CJA research has found that a history of bench warrants is one of the strongest predictors of FTA in New York City (Siddiqi 2004). The findings regarding community ties confirmed, as expected, that defendants with weak community ties were less likely to make all their scheduled court appearances. Specifically, defendants who were unemployed, who did not expect anyone at arraignment, and who had no telephone were more likely to fail to appear. Defendants who were released after arraignment (as opposed to being released at


39 arraignment) were more likely to fail to appear. This finding suggests that the courtâ&#x20AC;&#x2122;s initial reluctance to release the defendant on recognizance at arraignment was an accurate indication that the defendant posed a higher than average risk of failure to appear. The remaining variables included in the model had relatively weak effects that were difficult to interpret, so we do not address them further. It is worth identifying the variables that were tested but not included in the model. While some criminal history variables were statistically significant predictors of the likelihood of failure to appear, several other criminal history variables did not add explanatory power to the model: open cases at the time of arrest, prior misdemeanor convictions, prior felony convictions, prior prison sentences, current parole status, and prior sentences as a youthful offender. Three community ties items were not statistically significant in the model: length of time at current address, whether the defendant lived with someone, and whether the defendant lived in the New York City area. Two charge characteristics had no effect on failure to appear, and were not included in the model: number of arrest charges, and severity of arraignment charge. Finally, sex, ethnicity, and defendant-victim relationship had no effect on failure to appear, after controlling for the effects of the other variables in the model. Some of these variables had very little variance (e.g., only 3% of DV defendants lived outside the New York City area, and only 2% were on parole), so it is not surprising that they did not have a statistically significant impact on failure to appear. Some of these variables were similar to other variables that were included in the model, such as the community ties measures, criminal history measures, and charge characteristics. It may be that the effects of community ties, criminal history, and charge characteristics can be captured by a few key variables, and that other measures are redundant. Finally, the remaining variables that were not included in the model (ethnicity and defendant-victim relationship) were not expected to have an effect, and their absence was not surprising. Many of the findings reported here for DV defendants are similar to findings from research on general populations of defendants. The influence of criminal history, particularly prior warrant history, and of community ties on the likelihood of failure to appear is consistent with most previous research. A somewhat surprising finding was that age was the strongest predictor of failure to appear for DV defendants. While prior research has found that FTA rates decrease with age, age has generally not been found to be one of the strongest predictors of FTA. Similarly, the influence of type of release (bail vs. ROR) was consistent with prior research, though its impact in this model appears somewhat stronger than in previous studies. The influence of the timing of release (i.e., whether the defendant was released at or after arraignment) is a new findingâ&#x20AC;&#x201D;this variable has generally not been tested in prior research.


40

[This page intentionally left blank]


41 V. PREDICTING PRETRIAL RE-ARREST FOR DV DEFENDANTS A. Development of a Model Predicting Pretrial Re-Arrest for a New DV Offense While judges are permitted to consider the defendantâ&#x20AC;&#x2122;s likelihood of court attendance when setting bail or deciding whether to release the defendant on recognizance, they are not permitted to take into account public safety factors, such as the likelihood that the defendant will commit a new offense during the pretrial release period (Marks et al. 1996, Phillips 2004). New York State law, unlike federal law, does not allow for preventive detention of defendants based on the potential risk of reoffending. As mentioned in Chapter 3, many jurisdictions outside New York State do consider public safety factors, and some explicitly address public safety in domestic violence cases (e.g., Kentucky; see discussion in Mahoney et al. 2001). In such jurisdictions, domestic violence defendants are seen as posing an enhanced risk of reoffending against the same victim. The social science literature suggests that DV defendants may harm or threaten to harm the victim as a way to discourage her or him from participating in the prosecution of the case (Peterson 2003a). Because of these concerns, questions about the likelihood of pretrial re-arrest in DV cases, and the factors that influence it, are of considerable interest in New York State. From time to time, legislation is proposed to permit judges to make release decisions in DV cases by taking into account the likelihood of re-offending (Pirro 2004, Ross and Gendar 2005). As discussed in Chapter 3, about 9% of DV defendants were re-arrested for at least one new DV offense during the pretrial period. In this chapter, we develop a statistical model to examine which factors increase or decrease the likelihood of pretrial re-arrest for a new DV offense among DV defendants. The analysis excludes Non-DV defendants, who had a very low probability of committing a new DV offense (about 1%). The variables considered for inclusion in the model of pretrial re-arrest included all the variables considered in the FTA model discussed in Chapter 4. One variable, time at risk, is measured differently for the re-arrest model than it was for the FTA model. Specifically, time at risk is measured as the number of days between the defendantâ&#x20AC;&#x2122;s release and the disposition of the case. We did not subtract any time that the defendant was out on a bench warrant for failing to appear, as we did in the FTA model. The reason we did not make this adjustment is that the defendant continues to be at risk of re-arrest while out on a bench warrant. As before, our measure does not subtract from time at risk any time the defendant was held (either on bail or remand) after his or her first release. This adjustment would be very difficult to make, and we believe it affects relatively few defendants and consequently would not affect the results of our model. Our expectations regarding the effects of the independent variables on the likelihood of pretrial re-arrest for a new DV offense are based on prior research on rearrest as discussed in Chapter 1. Specifically, we expect criminal history, community ties, and age to affect the likelihood of pretrial re-arrest. The literature provides no


42 consistent evidence regarding what to expect regarding the effect of charge characteristics, type of release, or geographic or demographic variables. The stepwise procedure used to develop the logistic regression model of pretrial re-arrest for a new DV offense is the same procedure used to develop the model predicting pretrial failure to appear. We entered the two control variables (correction for selection bias and time at risk) first, to insure that the effects of other variables were independent of these two variables. We then entered variables stepwise within each of the major categories: criminal history, community ties, charge characteristics, release characteristics, and geographic/demographic characteristics. B. Model Predicting Pretrial Re-Arrest for a New DV Offense for DV Defendants The final logistic regression model predicting pretrial re-arrest is shown in Table 5-1. The model indicates that a variety of factors influence the likelihood of pretrial rearrest. We discuss each set of factors in turn. Both control variables have a statistically significant effect in this model. As expected, the correction for selection bias had a negative effect on the likelihood of rearrest. This indicates that defendants who had a higher probability of release (based on their background and case characteristics) were less likely to be re-arrested during the pretrial period. Those who had a lower probability of release (but were released anyway) were more likely to be re-arrested. Those who had a lower probability of release presumably had more risk factors for re-arrest than those with a higher probability of release. The second control variable, time at risk, had a statistically significant positive effect on the likelihood of re-arrest. As indicated by the size of the standardized beta (.66) this was by far the strongest predictor of pretrial re-arrest. Not surprisingly, the greater the time at risk, the greater the likelihood that the defendant committed at least one new DV offense during the pretrial period. Two measures of defendantsâ&#x20AC;&#x2122; criminal history had a statistically significant effect. Defendants with any prior arrests were more likely to be re-arrested for a new DV offense during the pretrial period, as were those with two or more open cases at the time of arrest. These findings indicate that a history of previous arrests, particularly a recent history of arrests, is a good predictor of pretrial re-arrest on the current case. Several community ties measures had statistically significant effects on the likelihood of pretrial re-arrest for new DV offenses. Unemployed defendants were more likely to be re-arrested than those who were employed. The odds of re-arrest were about 1.3 times greater for unemployed defendants. Similarly, defendants who lived at their current address for one year or less were more likely to be re-arrested than those who lived at their current address for longer than one year. Defendants who lived with someone at the time of their arrest were more likely to be re-arrested for a new DV offense during the pretrial period. Living with someone is usually considered a sign of stronger community ties and would normally be expected to be associated with lower likelihood of re-arrest. However, for domestic violence defendants, living with someone


43 TABLE 5-1 LOGISTIC REGRESSION MODEL PREDICTING LIKELIHOOD OF PRETRIAL RE-ARREST FOR A NEW DV OFFENSE DV CASES, CRIMES AGAINST PERSONS AND PROPERTY SUBSAMPLE1 Combined 2001-2002 Dataset INDEPENDENT VARIABLES

2

Standardized

Odds

β

Ratio

CONTROL VARIABLES SELECTION BIAS CORRECTION: LIKELIHOOD OF RELEASE TIME AT RISK

-0.10 * 0.66 ***

0.49 1.01

ANY PRIOR ARRESTS

0.25 ***

1.57

OPEN CASES AT TIME OF ARREST Reference Category: None One Open Case at Time of Arrest Two or More Open Cases at Time of Arrest

0.03 0.10 **

1.08 1.49

0.15 0.12 0.09 0.08

*** ** * *

1.30 1.26 1.18 1.20

0.20 *** 0.03

1.70 1.06

DEFENDANT'S CRIMINAL HISTORY

DEFENDANT'S COMMUNITY TIES UNEMPLOYED AT CURRENT ADDRESS 1 YEAR OR LESS LIVES WITH SOMEONE HAS NO TELEPHONE ARRAIGNMENT CHARGE TYPE ARRAIGNMENT CHARGE PENAL LAW ARTICLE: Reference Category: Assault (PL 120) Criminal Contempt (PL 215) Other DEFENDANT'S DEMOGRAPHIC CHARACTERISTICS SEX (Female) ETHNICITY: Reference Category: Non-Hispanic Black Non-Hispanic White Hispanic Other AGE: Reference Category: Age 16-20 Age 21-29 Age 30-39 Age 40 and over

-0.16 **

0.70

0.10 * -0.12 ** 0.03

1.29 0.80 1.11

-0.10 -0.13 -0.27 ***

0.83 0.78 0.58

0.09 0.05 -0.12 * 0.05

1.23 1.13 0.73 1.10

DEFENDANT-VICTIM RELATIONSHIP Reference Category: Married Boyfriend-Girlfriend Common-Law Marriage Other Relationship Missing Nagelkerke R2 (N of cases)

.12 *** (10,299)

NOTES 1 2

See text for a description of the dataset and the subsample. See Appendix C for information about the measurement and coding of the variables. * Statistically significant at p < .05 ** Statistically significant at p < .01 *** Statistically significant at p < .001


44

may indicate that they have easy access to the victim. When released, these defendants may have greater opportunity to commit a new offense against a family member living in their household, and therefore may be at greater risk for re-arrest for a new DV offense. (The measure used here did not indicate specifically who the defendant lived with, nor did it indicate if the defendant lived with the victim in the current case. However, it was probably strongly correlated with living with the victim— strongly enough to show an effect on the likelihood of a new DV offense.) Finally, defendants who had no telephone, as expected, were more likely to be re-arrested for a new DV offense. The arraignment charge type had a statistically significant effect on the likelihood of re-arrest. Specifically, defendants who were charged with criminal contempt were more likely to be charged with a pretrial re-arrest for a new DV offense than defendants who faced other charges. In DV cases, criminal contempt is usually charged when the defendant has violated an order of protection. This finding shows that defendants who had previously shown a willingness to violate a court order and who had been charged with criminal contempt were more likely to commit a new DV offense during the pretrial period. There were three demographic characteristics that had a statistically significant effect on the likelihood of re-arrest. Women were less likely than men to be re-arrested. Ethnicity also had a statistically significant effect on re-arrest. After controlling for the effects of the other variables in the model, Non-Hispanic Whites were more likely to be re-arrested for a new DV offense during the pretrial period than Non-Hispanic Blacks. The odds of re-arrest were 1.29 times greater for Non-Hispanic Whites than for NonHispanic Blacks. Hispanics, on the other hand, were less likely to be re-arrested than Non-Hispanic Blacks. Their odds of re-arrest were only .80 times as large as the odds for Non-Hispanic Blacks. Age was also a statistically significant predictor of re-arrest. Defendants aged 40 and over were significantly less likely than defendants aged 16-20 to be re-arrested for a new DV offense. Finally, defendant-victim relationship had a statistically significant effect on the likelihood of re-arrest. Defendants in an “other” relationship with the victim were less likely to be re-arrested than defendants who were married to the victim. These “other” relationships included parent-child, sibling, and other non-intimate relationships. Since there were no differences among the various categories of intimate partner relationships (married, common-law, and boyfriend-girlfriend), this finding indicates that those defendants arrested for offenses against other types of family members had a lower probability of re-arrest for a new DV offense. C. Summary and Discussion of Findings The explanatory power of our model of pretrial re-arrest for a new DV offense was relatively weak—the model explained only about 12% of the variation in the likelihood of re-arrest. This is similar to the explanatory power of the model of pretrial


45 FTA presented in Chapter 4, and is generally similar to the explanatory power of other models of re-arrest (e.g., Peterson 2003a). After controlling for time at risk and the correction for selection bias, the strongest predictor of the likelihood of pretrial re-arrest for a new DV offense was age. Defendants age 40 and over were significantly less likely to be re-arrested than defendants age 16-20. The next strongest predictors of re-arrest were whether the defendant had any prior arrests and whether the defendant was charged with criminal contempt at arraignment. Defendants who had any prior arrests were more likely to be re-arrested for a new pretrial DV offense than those with no prior arrests. Defendants charged with criminal contempt at arraignment were more likely to be re-arrested than those charged with assault. In DV cases, criminal contempt charges are generally brought when the defendant has violated an order of protection. This finding suggests that defendants who have been charged with violating orders of protection were more likely to commit new DV offenses and to be re-arrested for them during the pretrial period. This finding is consistent with previous research on DV defendants’ predisposition re-arrests for any new offense (Newmark et al. 2001) and on DV defendants’ post-disposition re-arrests for new DV offenses (Peterson 2003a), suggesting that this factor influences both pretrial and post-disposition re-arrests. The community ties measures were also strong predictors of re-arrest (as measured by the size of the standardized beta). In general, defendants with weaker community ties were more likely to be re-arrested. Defendants who were unemployed, who lived at their current address for one year or less, and defendants who had no telephone were all more likely to be re-arrested for a new DV offense during the pretrial period. Living with someone is generally viewed as a strong community tie, but in this model, living with someone increased the likelihood of re-arrest. While this finding appears to be inconsistent with the “community ties” hypothesis, in the context of domestic violence the effect of this variable should be re-interpreted. Domestic violence defendants who lived with someone were probably more likely to be living with the victim of their original offense, and therefore living with someone was a measure of the greater opportunity these defendants had to commit a new offense. As has been noted in other studies (Peterson 2003a, Wooldredge and Thistlethwaite 2002), living with someone is more appropriately interpreted as a measure of access to the victim or opportunity to commit a new DV offense, rather than as a measure of strong community ties. Viewed in this way, the finding reported here is not surprising; moreover it should not be interpreted as evidence that stronger community ties increase, rather than decrease, the likelihood of re-arrest. Other variables in the model that had statistically significant effects included having two or more open cases at the time of arrest, the sex and ethnicity of the defendant, and the defendant-victim relationship. Defendants who had two or more open cases at the time of arrest were more likely to be re-arrested for a new DV offense during the pretrial period. This indicates that these defendants were in a period of frequent criminal activity leading to multiple re-arrests in a relatively short time. Female defendants were less likely to be re-arrested, a finding which is consistent with prior


46 research on DV defendants. Surprisingly, re-arrest rates for new pretrial DV offenses varied by ethnicity. Non-Hispanic Whites were more likely, and Hispanics were less likely, than Non-Hispanic Blacks to be re-arrested. These differences cannot be attributed to any of the measures of criminal history, community ties, charge type, or demographic factors included in the model. These results may be a statistical anomaly, or may be due to unmeasured factors not included in the model. While defendant-victim relationship affected the likelihood of re-arrest, it turns out that only one of the categories of this variable differed from the others. Defendants who had an “other” relationship with the victim were less likely to be re-arrested for a new pretrial DV offense. “Other” relationships included primarily parent-child relationships and sibling relationships. There were no differences in the likelihood of re-arrest among the various categories of intimate partners—married couples, common-law marriages, and boyfriend-girlfriend relationships. Additional criminal history, community ties and charge type variables were tested but were not included in the model because they were not statistically significant. Among the criminal history variables that were not included were: prior misdemeanor convictions, prior felony convictions, prior jail and prison sentences, current parole status and prior sentences as a youthful offender. Three community ties measures were not included in the model: the CJA release recommendation, whether the defendant expected anyone at arraignment, and whether the defendant lived in the New York City area. Two charge characteristics had no effect on re-arrest and were not included in the model: number of arrest charges and severity of arraignment charge. As noted at the end of Chapter 4, some of these variables had very little variance, and it is not surprising that they did not help to predict the likelihood of re-arrest. Some of the variables were similar to other variables already in the model (e.g., criminal history and community ties) and may have been redundant. Among these it is notable that a prior record of bench warrants, or having an open bench warrant, had no effect on the likelihood of re-arrest. Knowing a defendant’s history of one type of misconduct does not necessarily help predict the other. This suggests that failure to appear and pretrial re-arrest for new DV offenses may not have the same underlying causes. To further explore the differences between re-arrest for new DV offenses and FTA, we now consider similarities and differences between the findings from the models predicting each outcome. There were several important similarities between the models predicting re-arrest and FTA. Both explained about the same proportion of variance (12% and 10%, respectively). After controlling for time at risk and the correction for selection bias, age was the most important predictor in both models. Older defendants were less likely to FTA and less likely to be re-arrested for a new DV offense than younger defendants. A record of prior arrests was also an important predictor in both models. Weak community ties, particularly being unemployed and not having a telephone, were important predictors of pretrial misconduct in both models. However, there are several notable differences between the models predicting re-arrest and FTA. Most importantly, release characteristics were important predictors of FTA, but had no effect on the likelihood of re-arrest. After controlling for the effects of


47 the other variables in the model, defendants released on bail were much less likely to fail to appear than defendants released on recognizance. However, there was no difference in the likelihood of re-arrest between defendants released on bail and those released on recognizance. Similarly, defendants released after arraignment were more likely to FTA than those released at arraignment, but this variable had no impact on rearrest. This suggests important differences in the factors affecting these two types of pretrial misconduct. Factors related to release affected failure to appear, but had no impact on pretrial re-arrest. There was also an interesting difference in the impact of charge type. Defendants arraigned on “other” charges were more likely to FTA, whereas defendants arraigned on “criminal contempt” charges (primarily violating an order of protection) were more likely to be re-arrested for a new DV offense. So although charge type affected both outcomes, it did so in different ways. Beyond the effect of age, which had a strong influence in both models, demographic and relationship characteristics had an influence only on re-arrest, not FTA. Specifically, ethnicity, gender and defendant-victim relationship affected the likelihood of re-arrest for a new DV offense, but had no impact on FTA. Finally, there were minor differences in the effect of specific items of criminal history and community ties on the two outcomes. A history of prior bench warrants, and having an open bench warrant, affected the likelihood of FTA on the current case but did not affect the likelihood of pretrial re-arrest for a new DV offense. On the other hand, having two or more open cases at the time of arrest affected the likelihood of rearrest, but not of FTA. With regard to community ties, not expecting someone at arraignment increased the likelihood of FTA but not of re-arrest, whereas living with someone and living at the current address for one year or less increased the likelihood of re-arrest but not of FTA. Taken together, these patterns suggest that the underlying factors influencing FTA and re-arrest share some basic similarities, but also that there are significant differences in the underlying processes.


48

[This page intentionally left blank]


49 VI. CONCLUSION A. Major Findings The current study has examined pretrial outcomes for DV and Non-DV defendants, and developed models to predict pretrial misconduct among DV defendants. The study used data from defendants arrested in New York City in the first quarter of 2001 and the third quarter of 2002. The study addressed four questions: 1) What are the pretrial release outcomes for defendants in domestic violence cases? 2) How do these outcomes compare to those for defendants in non-domestic violence cases? 3) What are the factors that influence the likelihood of pretrial failure to appear among DV defendants? 4) What are the factors that influence the likelihood of pretrial re-arrest for new DV offenses among DV defendants? There are four major findings from the study. First, most defendants in DV cases were released during the pretrial period and their rates of pretrial misconduct are relatively low. About two thirds of DV defendants were released on recognizance at arraignment, about 4% were released on bail at arraignment, about one sixth were released after arraignment, and 12% were never released. The median bail set for DV defendants at arraignment was $1,000; the median bail for DV defendants who were ever released on bail was $750. About 10% of DV defendants failed to appear for at least one scheduled appearance. About 15% of DV defendants were re-arrested during the pretrial period, including 9% who were rearrested for a new DV offense. Second, pretrial release outcomes and rates of pretrial misconduct for DV defendants were generally similar to those for Non-DV defendants. However, there were two noteworthy differences. The first is that almost all DV defendants had their cases continued beyond arraignment, whereas nearly half of comparable Non-DV defendants had their cases disposed at arraignment. This difference reflects citywide policy, which generally requires that DV cases be continued beyond arraignment. As a result of this policy, the court must make a release decision for almost all DV defendants, while no release decision is needed for many Non-DV defendants. The second difference is that DV and Non-DV defendants were re-arrested for different types of offenses. DV defendants were much more likely than Non-DV defendants to be re-arrested for a new DV offense during the pretrial period (9% vs. 1%). Non-DV defendants were much more likely than DV defendants to be re-arrested for a new NonDV offense during the pretrial period (14% vs. 7%). These results suggest DV and Non-


50 DV defendants are each more likely to be re-arrested for the same type of offense as their original arrest. Third, pretrial failure to appear among DV defendants was most strongly affected by age, criminal history, community ties, and release characteristics. Age had the strongest impactâ&#x20AC;&#x201D;younger defendants were more likely than older defendants to fail to appear for at least one scheduled court appearance. Defendants with more serious criminal histories, particularly those with a prior record of failing to appear, were more likely to FTA. Defendants with stronger community ties were generally less likely to FTA. Defendants released after the initial arraignment appearance, and those released on recognizance were more likely to FTA than those released at arraignment or those released on bail. Although all these factors were important predictors, the model only accounted for 10% of the variation in likelihood of failure to appear. Fourth, the strongest predictors of pretrial re-arrest for a new DV offense among DV defendants were age, criminal history, community ties, charge characteristics, and demographic characteristics. As with FTA, age had the strongest impact. Younger defendants were more likely to be re-arrested during the pretrial period than older defendants. Defendants with more serious criminal histories, particularly those with two or more open cases, were more likely to be re-arrested for a new DV offense. Defendants with stronger community ties were generally less likely to be re-arrested, while those charged with violating an order of protection were more likely to be rearrested. Female defendants and those in â&#x20AC;&#x153;otherâ&#x20AC;? family relationships with the victim (primarily parent-child or sibling) were less likely to be re-arrested. Non-Hispanic Whites were more likely to be re-arrested, and Hispanics were less likely to be rearrested, than Non-Hispanic Blacks. Overall, the model predicting pretrial re-arrest accounted for 12% of the variation, which was slightly more than the model predicting FTA. B. Discussion This study has described the extent of pretrial misconduct among DV defendants in New York City. It has considered three types of pretrial misconduct: failure to appear, re-arrest for a new DV offense, and re-arrest for any new offense. The study also has attempted to identify factors that are associated with higher or lower rates of pretrial misconduct. The factors that were considered included criminal history, community ties, charge characteristics, release characteristics, and demographic characteristics. To interpret the results of the study, we evaluate the findings in the context of three issues. First, our understanding of pretrial misconduct among DV defendants in New York City depends, in part, on whether we evaluate each type of misconduct separately, or consider all types of misconduct together. Considered individually, rates of particular types of pretrial misconduct among DV defendants were relatively low. About 10% of defendants in DV cases missed at least one scheduled court appearance, and 9% were re-arrested for a new DV offense during the pretrial period. Even when considering re-


51 arrests for any new offense during the pretrial period, 15% of defendants in DV cases were re-arrested. However, when considering all types of pretrial misconduct together, the rates of misconduct for DV defendants were somewhat higher. About 17% of defendants in DV cases failed to appear and/or were re-arrested for a new DV offense during the pretrial period. If we expand the measure of re-arrest to include re-arrest for any new offense, the rate of pretrial misconduct increases to 21%. While the evidence available in this study indicates that the majority of defendants in DV cases did not engage in pretrial misconduct, it does show that over one fifth failed to appear at least once and/or were re-arrested. Furthermore, as noted in Chapter 2, re-arrest rates are likely to underestimate recidivism since new offenses may go unreported or may not lead to re-arrest. This suggests that rates of pretrial misconduct may be even higher than 21%. From this perspective, the current study suggests that pretrial misconduct is a frequent problem among DV defendants. Second, our understanding of pretrial misconduct among DV defendants in New York City also depends on how their rates of misconduct compare to Non-DV defendants, and on which types of misconduct we compare. Some comparisons suggest little difference between DV defendants and Non-DV defendants. Compared to Non-DV defendants, DV defendants posed no greater risk for FTA or for re-arrest for any new offense. However, DV defendants posed a much greater risk of committing a new DV offense during the pretrial period than comparable Non-DV defendants (9% vs. 1%). From this perspective, the current study suggests that pretrial re-arrest for a new DV offense is a problem that deserves special attention when DV defendants are prosecuted. Finally, our understanding of pretrial misconduct among DV defendants in New York City depends on whether we examine the factors that affected failure to appear or those that affected pretrial re-arrest for a new DV offense. Some factors, such as age, prior arrests and community ties, had similar effects on both FTA and re-arrest. Other factors, such as prior misdemeanor jail sentences, prior bench warrants, and release characteristics, affected FTA but not re-arrest for a new DV offense. Yet open cases at the time of arrest, criminal contempt charges, sex, ethnicity, and defendant-victim relationship affected re-arrest for a new DV offense, but not FTA. These findings suggest that efforts to reduce pretrial misconduct among DV defendants could benefit from a focus on the different factors influencing specific types of misconduct. Reducing Failure to Appear. Taken together, the findings from this study suggest that pretrial misconduct is a problem for a significant minority of DV defendants and they suggest which factors increased the risk of pretrial misconduct. We now consider how the findings of this study could be used to reduce the risk of pretrial misconduct among DV defendants. Failure to appear among DV defendants in New York City appears to be most strongly affected by age, criminal history, community ties, and release characteristics. The bail statutes in New York State do not permit the criminal courts to consider demographic characteristics such as age as a factor in determining the type or


52 conditions of release. However, the court is required to take criminal history and community ties into account, and the court determines the release characteristics. Our findings suggest that courts should pay special attention to criminal history in making release decisions for defendants in DV cases. Specifically, those with any prior arrests, those who have two or more prior bench warrants, those who have an active bench warrant at the time of arrest, and those with two or more prior misdemeanor jail sentences all pose greater risks of failure to appear. Additionally, courts could pay special attention to three aspects of community ties: whether the defendant is employed, expects anyone at arraignment, and has a telephone. Information about these criminal history and community ties items is available to the criminal court at the time of arraignment, and could be used to assist the court in making release decisions. While our research suggests that certain criminal history and community ties items should be given special attention to reduce the likelihood of FTA, the question of how to use this information remains to be addressed. What changes in release decision-making should be made if a defendant has a more serious criminal history or weak community ties? Our research addressed two characteristics of release decisions, and found that both influenced the likelihood of FTA. The first characteristic is whether the defendant is released at arraignment or after arraignment. The odds that a DV defendant released after arraignment will fail to appear are 1.4 times higher than the odds for a DV defendant released at arraignment. This suggests that courts might reduce FTA among DV defendants by reducing the rate at which defendants who have been detained at arraignment are released after arraignment. This might be accomplished by refusing to reduce bail or refusing to ROR certain DV defendants at post-arraignment appearances. However, in practice, it would be difficult for the court to prevent most post-arraignment releases. Many DV defendants released after arraignment have made bail and cannot be detained. Other DV defendants must be released on recognizance because the DA has not corroborated the complaint within five days of the arraignment (six days if a Sunday intervenes), as required by New York CPL ยง170.70. The second release characteristic that affects the likelihood of FTA is whether the DV defendant is released on bail or on recognizance. This suggests that courts might reduce FTA by setting bail rather than releasing the defendant on recognizance. Some defendants will not be able to make bail and will be detained, thus insuring their return to court. Others will be released, but as our research shows, DV defendants released on bail have a lower rate of failure to appear than those who are released on recognizance. The odds that a DV defendant released on recognizance will fail to appear are about twice as high as the odds that a DV defendant released on bail will fail to appear. While setting bail more often at arraignment might appear to be a relatively easy way to reduce FTA, it would raise numerous problems in DV cases. First, our research does not test or provide evidence supporting the assumption that changing release practices would change the FTA rate. Our findings describe the effects of current practices, but do not explicitly examine whether changing those practices would


53 produce changes in the FTA rate. For example, we do not have evidence that setting bail for defendants who are now granted ROR would be effective in reducing FTA among released defendants. There may be differences between defendants who currently are granted ROR and those who currently are released on bail that we have not measured or taken into account in our models. These differences may account for the difference in their FTA rates. If there are unmeasured differences between these two groups of defendants, then setting bail for defendants who are currently granted ROR might not reduce the FTA rate for these defendants. Second, judges are likely to be reluctant to set bail more often, or to set higher bail, in DV cases. Domestic violence victims often want the defendant treated leniently, and do not want the defendant detained. Many DV cases are initiated when a victim calls the police to stop a particular incident of domestic violence. However, these victims are often reluctant to follow up in prosecuting the case, sometimes because they fear threats or retaliation, sometimes because they fear social stigma (e.g., among extended family), sometimes because they remain emotionally attached to the defendant, and sometimes because they are financially dependent on the defendant. When DV defendants are convicted, many victims do not want the defendant sentenced to jail, often expressing a wish that the defendant receive â&#x20AC;&#x153;treatment,â&#x20AC;? e.g., participate in a batterer program (Peterson 2001, 2003b). Although there are some victims who want the criminal justice system to deal with misdemeanor DV defendants harshly, it is important to remember that many do not. When victims do not participate in the prosecution, conviction is unlikely in DV cases (Peterson 2001, 2002, 2003c). Even if convicted of a misdemeanor offense, most DV defendants are not sentenced to jail (Peterson 2001, 2002, 2003c). Judges are generally reluctant to set bail more often, or to set higher bail, for defendants in misdemeanor cases who are unlikely to be convicted, or if convicted, unlikely to be sentenced to jail. Third, setting higher bail or setting bail more often for DV defendants may have a significant negative impact on their families, and may sometimes hurt the victim. An employed defendant who provides financial support to a victim and his/her children may lose a job if detained on a DV case. Or the defendant may be unable to provide adequate financial support if required to post bail. This suggests that in cases where the defendant and victim live together, judges may be reluctant to make release decisions that will financially harm the victim. Finally, setting bail more often at arraignment would cause the detention of numerous defendants who would not fail to appear if released. Furthermore, increasing the number of detained defendants would increase costs to the criminal justice system for detaining and transporting defendants. In New York City, these costs are considerable, and routine detention of larger numbers DV defendants would be problematic. For all of these reasons, efforts to encourage courts to set higher bail or to set bail more often in DV cases are unlikely to produce much change in release practices. Furthermore, even if such changes were implemented, they might not significantly


54 reduce the problem of FTA among DV defendants. This suggests that it would be worthwhile for the courts to consider other changes in release practices. Other possibilities, such as supervised release or additional conditions of release might be effective at reducing FTA. We will consider these possibilities in further detail after next reviewing ways to reduce the likelihood of re-arrest for a new DV offense. Reducing Pretrial Re-Arrest for a New DV Offense. Pretrial re-arrest for a new DV offense among DV defendants in New York City is most strongly affected by criminal history, arraignment charge type, community ties, and demographic characteristics. As noted earlier, decisions about whether to release the defendant on bail or recognizance, and decisions about the amount of bail, must be based only on factors that are related to the likelihood of failure to appear. Furthermore, bail statutes in New York State do not permit the criminal courts to consider demographic characteristics such as age, ethnicity, or gender as factors in determining the type or conditions of release. However, as will be discussed below, there are several ways that courts can address concerns about pretrial re-arrest. Our findings on re-arrest, like those for FTA, suggest that the courts should pay special attention to criminal history, community ties, and arraignment charge type in making decisions about type and conditions of release for defendants in DV cases. However some of the criminal history and community ties factors that are relevant are different for re-arrest than for FTA. To identify defendants with a criminal history associated with a higher risk of pretrial re-arrest for a new DV offense, the courts should focus on whether the defendant had any prior arrests, and whether the defendant had two or more open cases. Having any prior arrests was also associated with FTA, while having two or more open cases is important for predicting re-arrest but not FTA. Courts could also pay special attention to three community ties items related to the likelihood of pretrial re-arrest for a new DV offense: whether the defendant is employed, whether the defendant has a telephone, and whether the defendant has lived at his/her current address for a year or more. The first two of these items were also associated with FTA, while the latter was not. Finally, the courts should consider arraignment charge type as a factor related to the likelihood of pretrial re-arrest for a DV offense. Specifically, defendants charged with criminal contempt are more likely to be re-arrested for a new DV offense during the pretrial period. While our research suggests that criminal history and community ties should be given special attention to reduce the likelihood of pretrial re-arrest for a new DV offense, the question of how to use this information remains to be addressed. What changes in release decisionmaking should be made if a defendant has a more serious criminal history, weak community ties, or is charged with criminal contempt? Our research did not identify any characteristics of release decisions that are associated with likelihood of re-arrest. Whether the defendant is released at or after arraignment, and whether the defendant is released on bail or recognizance have no influence on likelihood of pretrial re-arrest for a new DV offense. Not only did we fail to find an effect of release characteristics, but the courts are also limited in the actions they can take regarding risk


55 of pretrial re-arrest. Given these restrictions, what actions can the courts take to address concerns about risk of pretrial re-arrest in DV cases? First, many of the factors that influence the risk of pretrial re-arrest for a new DV offense are included in the list of factors that courts must take into consideration in assessing risk of FTA. To the extent that courts consider criminal history, community ties, and arraignment charge type9 for reasons related to FTA when making release and bail decisions, they are also coincidentally basing these decisions on factors relating to re-arrest. This would appear to be within the bounds of the statutes. As Marks et al. (1996) note, it is likely that courts frequently consider risk of re-arrest in making release and bail decisions without expressly acknowledging it. Second, while the courts cannot explicitly take risk of pretrial re-arrest into account in making decisions about bail and release, they are authorized to issue a Temporary Order of Protection (TOP) in domestic violence cases. The decision to issue a TOP can be based on the risk of injury or intimidation to the victim (Marks et al. 1996). A TOP can require the defendant to stay away from the victim’s home, school or work place and/or to refrain from harassing, intimidating, threatening, or committing a family offense against the victim. Such orders can prevent the defendant from returning to his or her own home if the defendant shares the home with the victim. Temporary Orders of Protection are routinely ordered in DV cases, and were issued in virtually all the DV cases in our dataset. Nevertheless, our study has shown that re-arrests for new DV offenses occur in spite of the widespread use of TOPs. The only recommendation that could be made regarding TOPs is that they more often include tighter restrictions on the actions of the defendant. Specifically, some TOPs are “full” while others are “limited.” A “full” TOP (also referred to as a “full stay-away”) allows no contact between the defendant and the victim. A “limited” TOP allows contact for specific purposes (often to allow the defendant to visit his/her children). We have no data on whether TOPs were “full” or “limited,” nor are we aware of any evidence suggesting that “full” TOPs are more effective than “limited” TOPs at preventing re-arrest. Courts could presumably try to order more “full” TOPs in an effort to reduce recidivism. However, it is likely that some victims will prefer the “limited” TOP, and it may be difficult in practice for courts to find situations where “full” TOPs can be effective. Finally, a court can set other conditions of release, in addition to a TOP, although the statutes are not specific about these other conditions. Case law suggests that courts are authorized, for example, to require the defendant to report to the police regularly or to enroll in an alcohol treatment program (Marks et al. 1996). Violation of these conditions can result in revocation of bail or ROR; however, in misdemeanor cases, the court is required to issue “another order of bail or recognizance, presumably with more stringent conditions” (Marks et al. 1996, p. 201). Some restrictive conditions, such as a tight curfew that amounts to preventive detention, are not permitted (Marks et al. 1996). Presumably, permissible conditions must be related to insuring the 9

Consideration of charge type is permissible insofar as the court is required to take into account the likely sentence if the defendant is convicted (Kluger and Swern 1998).


56 defendant’s return to court, but there appears to be considerable latitude in determining the relevance of the conditions to this goal. This latitude opens the possibility for supervised release in New York, a practice that is common in other jurisdictions. Supervised release programs are generally designed to monitor defendants’ compliance with the conditions of their release while awaiting the disposition of their cases. A national survey of pretrial programs in 2001 found that supervision services were available in 93% of the jurisdictions surveyed (Clark and Henry 2003). Among the most common services available were having the defendant report by telephone or in person, referring the defendant to substance abuse treatment or mental health treatment, testing the defendant for drug and alcohol use, and electronic monitoring. The 2001 survey also reported that special procedures to supervise DV defendants were used in one third of the jurisdictions (Clark and Henry 2003). The special procedures include referral to counseling, maintaining contact with the victim to monitor compliance with TOPs, electronic monitoring, a specialized DV caseload, and enhanced supervision. To facilitate the use of special procedures for monitoring DV defendants, some pretrial programs use special risk assessment procedures for DV cases. A supervised release program for DV defendants in Milwaukee is run by the Domestic Violence Court Commissioner, who oversees all misdemeanor DV pretrial proceedings (Campbell, Damiani and Mengrahj 2004). DV defendants who have a previous domestic violence record may be ordered to participate in the Pretrial Monitoring Program (PMP) as a condition of bail. Under this program, DV defendants appear before the Commissioner three times during the pretrial period, and must also have three in-person check-in sessions with the bail monitor. The bail monitor also visits the defendant’s address to verify compliance with TOPs and other bail conditions. If violations of the conditions are found, bail can be modified; occasionally, new charges are filed. Victims in PMP cases are offered referrals to services, and are provided information about court procedures and the defendant’s conditions of bail (Vera Institute of Justice 2006). The use of supervised release programs for both the general population of defendants, and for DV defendants in particular, has grown more common over the past decade (Clark and Henry 2003). Studies of supervised release in Philadelphia (Goldkamp and White 1998), Milwaukee, and Portland, Oregon (Austin, Krisberg and Litsky 1984) found that such programs were effective at preventing FTA when compared to unsupervised release.10 Lasley’s (2003) study of DV defendants released on bail suggests that supervision of released DV defendants reduces re-arrests for a new DV offense. Although the study is limited to DV defendants released on bail, the findings suggest the possibility that supervision of DV defendants released on recognizance may also be effective at deterring re-arrests. While it is beyond the scope of the current study to develop a plan for a supervised release program for DV defendants in New York City, we will outline some 10

However, supervised release did not prevent FTA in Miami (Austin et al. 1984).


57 general guidelines for such a program. The brief review of supervised release programs provided here suggests several supervision practices that could be included in a supervised release program. (See Pretrial Services Resource Center (1999) for an extensive discussion of pretrial supervision practices, and Lotze et al. (1999) for descriptions of several pretrial supervision programs for DV defendants.) DV defendants should be interviewed on the day of their arraignment in Criminal Court to provide contact information as well as information for a risk assessment. Each defendant should be provided at that time with a full explanation of the conditions of release, including the TOP. Defendants should be required to make contact with an oversight agency by telephone or in person for a specified number of times per week while the case is pending. The oversight agency should contact the defendant if s/he fails to check in as required. The oversight agency should also notify the defendant of upcoming court appearances, report to the court about the defendantâ&#x20AC;&#x2122;s compliance or noncompliance with the conditions of release, and contact defendants who miss court appearances to encourage them to return to court. The oversight agency could also contact the victim to verify the defendantâ&#x20AC;&#x2122;s compliance with the TOP, however this must be done carefully to avoid endangering the victim. For example, if the oversight agency reports the victimâ&#x20AC;&#x2122;s statements to the court in the presence of the defendant, the defendant may retaliate against the victim for reporting his violation. Notably absent from these recommendations is a requirement that the defendant enroll in a batterer intervention program. While the use of such programs as a condition of pretrial release is commonplace for felony cases in the Supreme Court DV parts in New York City, the volume of misdemeanor cases in the Criminal Court DV parts far exceeds the capacity of existing programs to accommodate them. Widespread use of batterer programs for defendants in misdemeanor DV cases in New York City is impractical. Construction of a specialized risk assessment instrument for supervised release of DV defendants would have to be done carefully, and in consultation with the courts. Factors that affect the likelihood of FTA could be included, but it is not clear whether and how factors that affect the likelihood of pretrial re-arrest could be included. Presumably, such factors could be included if they are used only to determine conditions of release designed to prevent intimidation of or harm to the victim. This use would be consistent with the purposes for which TOPs are issued in DV cases. C. Conclusion Pretrial misconduct, including failure to appear and re-arrest, is a frequent problem among misdemeanor DV defendants in New York City. Furthermore, such misconduct often includes a re-arrest for a new DV offense. Younger DV defendants, those with a more serious criminal history and those with weak community ties are more likely to engage in pretrial misconduct. What can be done to reduce the rates of pretrial misconduct?


58 Statutorily, the courts have more options to reduce failure to appear than to reduce re-arrest. Releasing defendants on bail rather than on recognizance and setting higher bail are permissible methods for insuring the defendantâ&#x20AC;&#x2122;s return to court. However, greater use of bail, either at arraignment or after arraignment, is unlikely and impractical. Many victims do not participate in the prosecution of DV cases and they are unlikely to support greater use of bail. Because most DV cases are unlikely to end in conviction and a jail sentence, judges are reluctant to set bail in an amount that will detain the defendant. If bail were used more often, many DV defendants would be detained who would not engage in pretrial misconduct if released and detention costs would increase significantly. Finally, greater use of bail may not be an effective way to reduce failure to appear. In short, changes in bail-setting practices significant enough to reduce pretrial misconduct among DV defendants are unlikely. While increasing the general reliance on bail in DV cases does not appear to be feasible, there are other changes courts can make to reduce pretrial misconduct. Setting conditions of release, particularly by requiring supervised release for some DV defendants, may be an effective way to address concerns about both FTA and pretrial re-arrest for new DV offenses. The oversight provided by a supervised release program during the pretrial period might also reduce recidivism in the post-disposition period. Prior CJA research has shown that the duration of court oversight is associated with lower rates of post-disposition re-arrest (Peterson 2003a, 2004). Some DAs extend the duration of court oversight as a means of monitoring DV defendants during the pretrial period (Peterson and Dixon 2005). Increasing the intensity of pretrial oversight, through the use of a supervised release program, might provide added benefits in terms of reducing post-disposition re-arrests as well as pretrial re-arrests. Additional changes to reduce pretrial misconduct by DV defendants would require legislative action. Allowing preventive detention for defendants who pose a risk to a victim, a witness, or the community would enable the courts to base release and bail decisions on these risks. This would allow courts to detain, or require supervised release for, high-risk defendants. Such legislation is proposed from time to time, but has not been passed (Ross and Gendar 2005). Another possible legislative action would involve a change in the courtsâ&#x20AC;&#x2122; authority to revoke bail or recognizance when a TOP is violated in the pretrial period. New York State law currently authorizes the court to revoke bail or recognizance and detain defendants charged with intimidating a victim or witness (Marks et al. 1996). This is a situation where the court can legally engage in preventive detention. Under this limited circumstance, defendants can be remanded for 90 days or until the re-arrest charges are reduced to a misdemeanor (whichever is shortest). In our dataset, only a handful of DV defendants were re-arrested and charged with intimidating a victim or witness. However, about 12% of released DV defendants were charged with violating an order of protection, and an additional 5% were re-arrested during the pretrial period and charged with violating an order of protection (data not shown). If courts were permitted to revoke bail or ROR for some of these DV defendants, or to assign them to a supervised release program, pretrial misconduct, and particularly pretrial re-arrest for a new DV offense,


59 might be reduced. This would require an extension of the provision allowing pretrial detention for intimidating a victim or witness to apply to violating an order of protection. Remand would be allowed only for felony violations of an order of protection (PL ยง215.51 or ยง215.52), but higher bail or supervised release could be authorized for misdemeanor violations of an order of protection (PL ยง215.50). This study has demonstrated that pretrial misconduct is a serious problem for a significant minority of defendants in DV cases. It has also identified several factors associated with pretrial misconduct, and suggested some ways to reduce it. However, remedies are likely to be difficult to implement, and legislative change is probably needed to effectively address this problem.


60

[This page intentionally left blank]


61 VII. REFERENCES Alameda Regional Criminal Justice Planning Board. 1976. An Evaluation of the Comprehensive Pretrial Services Delivery System in Alameda County. Oakland, CA: Alameda Regional Criminal Justice Planning Board. Austin, James, Barry Krisberg, and Paul Litsky. 1984. Evaluation of the Field Test of Supervised Pretrial Release. Washington, D.C.: National Institute of Justice. Austin, James, Barry Krisberg, and Paul Litsky. 1985. “The Effectiveness of Supervised Pretrial Release.” Crime and Delinquency 31:519-537. Belenko, Steven, Iona Mara-Drita, and Jerome E. McElroy. 1992. “Drug Tests and the Prediction of Pretrial Misconduct: Findings and Policy Issues.” Crime and Delinquency 38:557-582. Belenko, Steven, Jeffrey A. Fagan, and Tamara Dumanovsky. 1994. “The Effect of Legal Sanctions on Recidivism in Special Drug Courts.” The Justice System Journal 17:53-81. Bell, Marshall, et al. 1974. Bail System Development Study: Final Report. Columbus, Ohio: The Ohio State University Research Foundation. Berk, Richard A. 1983. “An Introduction to Sample Selection Bias in Sociological Data.” American Sociological Review 48:386-398. Boudouris, James, Richard Thomasgard, and Teresa Lacsina. 1977. Pretrial Decision Making in Iowa: Comparisons of Pretrial Programs in Polk County and Linn County. Des Moines, IA: Bureau of Correctional Evaluation. Britt, Chester L., III, Michael R. Gottfredson, and John S. Goldkamp. 1992. “Drug Testing and Pretrial Misconduct: An Experiment on the Specific Deterrent Effects of Drug Monitoring Defendants on Pretrial Release.” Journal of Research in Crime and Delinquency 29:62-78. Bureau of Justice Statistics. 1992. Drugs, Crime, and the Justice System. Washington, D.C.: U.S. Department of Justice. Campbell, Robin, Cristina Damiani and Suzanne Mengrahj. 2004. Enhancing Responses to Domestic Violence: Promising Practices from the Judicial Oversight Demonstration Initiative. New York: Vera Institute of Justice. Center for Governmental Research, Inc. 1983. An Empirical and Policy Examination of the Future of Pretrial Release Services in New York State. Rochester, New York: Center for Governmental Research, Inc.


62 Clark, John and D. Alan Henry. 2003. Pretrial Services Programming at the Start of the 21st Century: A Survey of Pretrial Programs. Washington, D.C.: U.S. Department of Justice. Cohen, Thomas H. and Brian A. Reaves. 2006 Felony Defendants in Large Urban Counties, 2002. Washington, D.C.: U.S. Department of Justice. Conklin, John E. and Dermot Meagher. 1973. “The Percentage Deposit Bail System: An Alternative to the Professional Bondsman.” Journal of Criminal Justice 1:99317. Cuvelier, Steven J. and Dennis W. Potts. 1993. Bail Classification Project: Harris County, Texas. Alexandria, VA: State Justice Institute. Davis, Robert C., Barbara E. Smith and Laura B. Nickles. 1998. “The Deterrent Effect of Prosecuting Domestic Violence Misdemeanors.” Crime and Delinquency 44:434-442. Eskridge, Chris W. 1979. “An Empirical Study of Failure to Appear among Accused Offenders: Construction and Validation of a Prediction Scale.” Annual Journal of Pretrial Services 2:105-17. Fagan, Jeffrey, Elizabeth Friedman, Sandra Wexler, and Virginia Lewis. 1984. National Family Violence Evaluation Final Report. Washington, D.C.: U.S. Department of Justice. Feeley, Malcolm M. and John McNaughton. 1974. A Pretrial Process in the Sixth Circuit: A Quantative and Legal Analysis. New Haven: Yale University. Frisch, Lisa A., Melissa I. Mackey, Donna Hall, and Alissa Pollitz Worden. 2001. Family Protection and Domestic Violence Intervention Act of 1994: Evaluation of the Mandatory Arrest Provisions (Final Report to the Governor and Legislature). Albany, NY: Division of Criminal Justice Services. Goldkamp, John S., Michael R. Gottfredson, and Susan Mitchell Herzfeld. 1981. Bail Decisionmaking: A Study of Policy Guidelines. Washington, DC: U.S. Department of Justice, National Institute of Corrections. Goldkamp, John S. and Michael D. White. 1998. Restoring Accountability in Pretrial Release: The Philadelphia Pretrial Release Supervision Experiments, Final Report. Philadelphia: Crime and Justice Research Institute. Goodman, Rebecca. 1992. Hennepin County Bureau of Community Corrections Pretrial Release Study. Minneapolis, MN: Planning and Evaluation Unit.


63 Hart, Timothy C. and Brian A. Reaves. 1999. Felony Defendants in Large Urban Counties, 1996. Washington, D.C.: U.S. Department of Justice. Helland, Eric and Alexander Tabarrok. 2004. “The Fugitive: Evidence on Public versus Private Law Enforcement from Bail Jumping.” Journal of Law and Economics 47:93122. Illinois Criminal Justice Information Authority. 1992. Pretrial Release and Crime in Cook County. Chicago, Illinois: Illinois Criminal Justice Information Authority. Kennedy, Spurgeon and D. Alan Henry. 1996. Commercial Surety Bail: Assessing Its Role in the Pretrial Release and Detention Decision. Washington, D.C.: Pretrial Services Resource Center. Kirby, Michael P. 1977. The Effectiveness of the Point Scale. Washington, D.C.: Pretrial Services Resource Center. Kluger, Judy Harris and Anne J. Swern. 1998. “Arraignment.” Pp. 53-76 in Lawrence N. Gray (ed.), New York Criminal Practice, 2nd Edition. New York: New York State Bar Association. Landes, William M. 1974. “Legality and Reality: Some Evidence on Criminal Procedure.” Journal of Legal Studies 3:287-300. Lasley, James. 2003. “The Effect of Intensive Bail Supervision on Repeat Domestic Violence Offenders.” The Policy Studies Journal 31:187-207. Lazarsfeld, Paul F. 1974. An Evaluation of the Pretrial Services Agency of the Vera Institute of Justice. New York: Vera Institute of Justice. Lotze, Evie, John Clark, D. Alan Henry and Jolanta Juskiewicz. 1999. The Pretrial Services Reference Book: History, Challenges, Programming. Washington, D.C.: Pretrial Services Resource Center. Mahoney, Barry, Bruce Beaudin, John A. Carver III, Daniel B. Ryan, and Richard B. Hoffman. 2001. Pretrial Services Programs: Responsibilities and Potential. Washington, D.C.: National Institute of Justice. Marks, Lawrence K., Robert S. Dean, Mark Dwyer, Anthony J. Girese, and James A. Yates. 1996. New York Pretrial Criminal Procedure. St. Paul, MN: West Publishing. Maxwell, Sheila Royo. 1999. “Examining the Congruence between Predictors of ROR and Failures to Appear.” Journal of Criminal Justice 27:127-141.


64 Menard, Scott. 1995. Applied Logistic Regression Analysis. Sage Publications: Thousand Oaks, CA. Mohr, Lawrence B. 1990. Understanding Significance Testing. Sage Publications: Thousand Oaks, CA. Murphy, Christopher M., Peter M. Musser, and Kenneth I. Maton. 1998. “Coordinated Community Intervention for Domestic Abusers: Intervention System Involvement and Criminal Recidivism.” Journal of Family Violence 13:263-284. National Institute of Law Enforcement and Criminal Justice. 1976. Community-Based Corrections in Des Moines. Washington, D.C.: U.S. Department of Justice. Newmark, Lisa, Mike Rempel, Kelly Diffily and Kamala Mallik Kane. 2001. Specialized Felony Domestic Violence Courts: Lessons on Implementation and Impacts from the Kings County Experience. Washington, D.C.: The Urban Institute. NYPD. 2000. Patrol Guide. New York: New York City Police Department. Obert, Ronald. 1973. Pre-trial Release Program in an Urban Area: Final Report, Santa Clara Pre-trial Release Program. San Jose, CA: Santa Clara Pre-Trial Release Program. Ozanne, Marq R., Robert A. Wilson and Dewaine L. Gedney, Jr. 1980. “Toward a Theory of Bail Risk.” Criminology 18:147-161. Peterson, Richard R. 2001. Comparing the Processing of Domestic Violence Cases to Non-Domestic Violence Cases in New York City Criminal Courts. New York: NYC Criminal Justice Agency, Inc. Peterson, Richard R. 2002. Cross-Borough Differences in the Processing of Domestic Violence Cases in New York City Criminal Courts. New York: NYC Criminal Justice Agency, Inc. Peterson, Richard R. 2003a. The Impact of Case Processing on Re-Arrests among Domestic Violence Offenders in New York City. New York: NYC Criminal Justice Agency, Inc. Peterson, Richard R. 2003b. Combating Domestic Violence in New York City: A Study of DV Cases in the Criminal Courts. New York: NYC Criminal Justice Agency, Inc. Peterson, Richard R. 2003c. Combating Domestic Violence in New York City, 2001 (Research Brief #4). New York: NYC Criminal Justice Agency, Inc.


65 Peterson, Richard R. 2004. The Impact of Manhattan’s Specialized Domestic Violence Court. New York: NYC Criminal Justice Agency, Inc. Peterson, Richard R. and Jo Dixon. 2005. “Court Oversight and Conviction under Mandatory and Nonmandatory Domestic Violence Case Filing Policies.” Criminology and Public Policy 4:535-558. Phillips, Mary T. 2004. Factors Influencing Release and Bail Decisions in New York City: Part 3. Cross-Borough Analysis. New York: NYC Criminal Justice Agency, Inc. Pirro, Jeanine. 2004. “Giving Victims Their Due.” New York Law Journal, February 11, p. 2. Pretrial Services Resource Center. 1999. The Supervised Release Primer. Washington, D.C.: Pretrial Services Resource Center. Rainville, Gerard and Brian A. Reaves. 2003. Felony Defendants in Large Urban Counties, 2000. Washington, D.C.: U.S. Department of Justice. Reaves, Brian A. 1990. Pretrial Release of Felony Defendants, 1988. Washington, D.C.: U.S. Department of Justice. Reaves, Brian A. 1993. Pretrial Release of Felony Defendants, 1990. Washington, D.C.: U.S. Department of Justice. Reaves, Brian A. 1994. Pretrial Release of Federal Felony Defendants. Washington, D.C.: U.S. Department of Justice. Reaves, Brian A. 1998. Felony Defendants in Large Urban Counties, 1994. Washington, D.C.: U.S. Department of Justice. Reaves, Brian A. 2001. Felony Defendants in Large Urban Counties, 1998. Washington, D.C.: U.S. Department of Justice. Reaves, Brian A. and Jacob Perez. 1994. Pretrial Release of Felony Defendants, 1992. Washington, D.C.: U.S. Department of Justice. Reaves, Brian A. and Pheny Z. Smith. 1995. Felony Defendants in Large Urban Counties, 1992. Washington, D.C.: U.S. Department of Justice. Rhodes, William M., Raymond Hyatt, and Paul Scheiman. 1996. “Predicting Pretrial Misconduct with Drug Tests of Arrestees: Evidence from Eight Settings.” Journal of Quantitative Criminology 12:315-348.


66 Ross, Barbara and Alison Gendar. 2005. “Morgy: Higher Bail for Abusers.” New York Daily News, March 11. Roth, Jeffrey A. and Paul B. Wice. 1980. Pretrial Release and Misconduct in the District of Columbia. Washington, D.C.: Institute for Law and Social Research. Schaffer, Stephen A. 1970. Bail and Parole Jumping in Manhattan in 1967. New York: Vera Institute of Justice. Siddiqi, Qudsia. 1999. Assessing Risk of Pretrial Failure to Appear in New York City: A Research Summary and Implications for Developing Release-Recommendation Schemes. New York, NY: NYC Criminal Justice Agency, Inc. Siddiqi, Qudsia. 2000. Prediction of Pretrial Failure to Appear and Alternative Pretrial Release Risk-Classification Schemes in New York City: A Validation Study. New York, NY: NYC Criminal Justice Agency, Inc. Siddiqi, Qudsia. 2002. Prediction of Pretrial Failure to Appear and an Alternative Pretrial Release Risk-Classification Scheme in New York City: A Reassessment Study. New York, NY: NYC Criminal Justice Agency, Inc. Siddiqi, Qudsia. 2003a. An Examination of the Existing and New Pretrial Release Recommendation Schemes in New York City: A Pre-Implementation Analysis. New York, NY: NYC Criminal Justice Agency, Inc. Siddiqi, Qudsia. 2003b. Predicting the Likelihood of Pretrial Re-Arrest: An Examination of New York City Defendants. New York: NYC Criminal Justice Agency, Inc. Siddiqi, Qudsia. 2003c. Predicting the Likelihood of Pretrial Re-Arrest Among New York City Defendants: An Analysis of the 2001 Dataset. New York: NYC Criminal Justice Agency, Inc. Siddiqi, Qudsia. 2004. CJA’s New Release-Recommendation System (Research Brief #5). New York: NYC Criminal Justice Agency, Inc. Siddiqi, Qudsia. 2006. Predicting the Likelihood of Pretrial Re-Arrest for Violent Felony Offenses and Examining the Risk of Pretrial Failure Among New York City Defendants: An Analysis of the 2001 Dataset. New York: NYC Criminal Justice Agency, Inc. Smith, Douglas A., Eric D. Wish and G. Roger Jarjoura. 1989. “Drug Use and Pretrial Misconduct in New York City.” Journal of Quantitative Criminology 5:101-126. SPSS, Inc. 1999. SPSS Regression Models 9.0. Chicago: SPSS, Inc.


67 Straus, F. and J.J. Golbin. 1980. Suffolk Probation's Pre-trial Services Program, Report #3. Suffolk County, NY: Suffolk County Probation Department. Sullivan, Cris M., Joanne Belknap, Ruth Fleury, Amy Leisenring, Heather Melton, and Meghan Chandek. 2000. “The Relationship between Domestic Violence Case Disposition and Victimization Six Months Later.” Paper presented at the Annual Meeting of the American Society of Criminology, San Francisco, CA, November. Thomas, Wayne. 1976. Bail Reform in America. Berkeley, CA: University of California Press. Toborg, Mary A. 1981. Pretrial Release: A National Evaluation of Practices and Outcomes. Washington, D.C.: The Lazar Institute. Toborg, Mary A., Anthony M. J. Yezer, Philip Tseng, B. Lynn Carpenter. 1984. Pretrial Release Assessment of Danger and Flight: Method Makes A Difference. Washington, DC: National Institute of Justice. Tolman, Richard M. and Arlene Weisz. 1995. “Coordinated Community Intervention for Domestic Violence: The Effects of Arrest and Prosecution on Recidivism of Woman Abuse Perpetrators.” Crime and Delinquency 41:481-495. U.S. Office of Economic Opportunity. 1973. Pretrial Release Program: Working Papers. Washington, D.C.: U.S. Office of Economic Opportunity. Venezia, Peter S. 1973. Pretrial Release with Supportive Services for High Risk Defendants: Three-year Evaluation of the Polk County Department of Court Services Community Corrections Project. Davis, CA: National Council on Crime and Delinquency. Vera Institute of Justice. 2006. Pretrial Innovations: Supporting Safety and Case Integrity. New York: Vera Institute of Justice. Visher, Christy A. and Richard A. Linster. 1990. “A Survival Model of Pretrial Failure.” Journal of Quantitative Criminology 6:153-184. Weis, Joseph G. 1989. “Family Violence Research Methodology and Design.” Pp. 117-162 in Lloyd Ohlin and Michael Tonry (eds.), Crime and Justice: A Review of Research, Volume 11: Family Violence. Chicago: University of Chicago. Wheeler, Gerald R. and Carol L. Wheeler. 1981. “Two Faces of Bail Reform: An Analysis of the Impact of Pretrial Status on Disposition, Pretrial Flight, and Crime in Houston.” Review of Policy Research 1:168-182.


68 Wilson, Robert A. 1975. A Practical Procedure for Developing and Updating Release on Recognizance Criteria. Newark, Delaware: University of Delaware, Division of Urban Affairs. Wilson, Robert A. 1979. Evaluation of the Chester County Bail Agency. Harrisburg, PA: Pennsylvania Commission on Crime and Delinquency. Winterfield, Laura, Mark Coggeshall, and Adele Harrell. 2003. Development of an Empirically-Based Risk Assessment Instrument. Washington, D.C.: Urban Institute. Wooldredge, John and Amy Thistlethwaite. 2002. â&#x20AC;&#x153;Reconsidering Domestic Violence Recidivism: Conditioned Effects of Legal Controls by Individual and Aggregate Levels of Stake in Conformity.â&#x20AC;? Journal of Quantitative Criminology 18:45-70.


69 APPENDIX A: STATISTICAL METHODS 1. Logistic Regression Analysis This report used logistic regression analysis to predict the likelihood of failure to appear and the likelihood of re-arrest. Logistic regression analysis is a statistical technique that is used when the outcome to be explained (i.e., the dependent variable) has two categories. In our analyses in Chapter 4, all cases were coded on our dependent variable in one of two categories: did not fail to appear (coded 0) and failed to appear (coded 1). The models predict the likelihood that defendants failed to appear for a scheduled court appearance during the pendency of their case. In our analyses in Chapter 5, all cases were coded on our dependent variable in one of two categories: not re-arrested (coded 0) and re-arrested (coded 1). The models predict the likelihood that defendants were re-arrested for a DV offense during the pendency of their case. In both chapters, these predictions were made on the basis of information we have about a variety of defendant and case characteristics (i.e., the independent variables). Logistic regression techniques provide several ways of evaluating the effect of these independent variables. The current study examined three statistical measures to evaluate the effect of the independent variables on a dependent variable. First, we report the statistical significance of each independent variable. Statistical significance takes into account the size of the sample as well as the magnitude of the effect of the independent variable. Taking this information into account, statistical significance assesses the probability that the effect observed in the sample could have occurred by chance alone. In this report, following standard convention, significance levels of .05 or less were treated as statistically significant. In other words, when an effect has a 5% or less probability of having occurred by chance, we conclude that the independent variable is a statistically significant predictor of the likelihood of re-arrest. One weakness of using statistical significance to measure the effect of an independent variable is that when sample sizes are large (e.g., more than several thousand cases), many independent variables have statistically significant effects even when the magnitude of their effects is small. For example, in a very large sample, we may be able to say that having a prior arrest has a statistically significant effect on the likelihood of re-arrest, even though the odds of rearrest for those with a prior arrest are only 1.01 times larger than for those without a prior arrest. In this hypothetical example, we can say that the effect of having a prior arrest is unlikely to be due to chance. However, it is also clear that knowing whether or not a defendant had a prior arrest does not explain much of the variation in likelihood of re-arrest. Our second statistical measure used to evaluate the effect of the independent variables is the odds ratio. The odds ratio supplements information about statistical significance by evaluating the magnitude of the effect of the independent variable. Specifically, it tells us how much the odds of an outcome (e.g., re-arrest) change, for each one unit increase in the independent variable. If an independent variable is coded in two categories (e.g., 0 and 1), then the odds ratio tells us how the odds of the


70 outcome change when cases are coded 1 on the independent variable (vs. cases coded 0). An odds ratio greater than one indicates an increase in the likelihood of the outcome occurring, while an odds ratio less than one indicates a decrease in the likelihood of the outcome occurring. An odds ratio of 1 indicates that the odds of an outcome occurring are not affected by the independent variable. To return to our previous example, if the odds ratio for the effect of having a prior arrest on the likelihood of re-arrest was 1.12, this would mean that in cases where the defendant had a prior arrest, the odds of re-arrest are 1.12 times greater than in cases where the defendant did not have a prior arrest. In contrast, if we examined the impact of whether the defendant was released on recognizance while the case was pending, we might find an odds ratio less than 1. For example, if the odds ratio was .83, this would mean that in cases where the defendant was released on recognizance, the odds of re-arrest are only .83 times as large as the odds when the defendant was released on bail. To simplify interpretation of odds ratios less than 1, it is common to examine the inverse of the odds ratio (1 divided by the odds ratio). When this is done, the interpretation of the effect of the independent variable is reversed. For example, if the odds ratio for being released on recognizance is .83, we can take the inverse of the odds ratio, 1.20 (1 divided by .83), and say that in cases where the defendant was released on bail, the odds of re-arrest were 1.20 times greater than in cases where the defendant was released on recognizance. Finally, if the odds ratio was 1.00, this would mean that whether the defendant was released on recognizance or on bail had no impact on the odds of re-arrest. (These examples are hypothetical and do not necessarily reflect our expectations about the findings.) In the analyses presented in this report, results are presented for independent variables coded in three different waysâ&#x20AC;&#x201D;categorical variables that have two categories, categorical variables that have more than two categories, and continuous variables that measure the quantity of a defendant or case characteristic (e.g., the number of prior felony convictions for the defendant). When a categorical independent variable has two categories, the odds ratio measures the change in the odds when cases are in one category vs. the other (e.g., defendant had a prior arrest vs. did not have a prior arrest). When a categorical independent variable has more than two categories, one of the categories is chosen as a reference category, and the odds ratios measure the effect of being in each of the other categories vs. being in the reference category (e.g., defendants age 40 and over are compared to defendants age 16-20, which is used as the reference category). Finally, when the independent variable is continuous, the odds ratio measures the change in the odds associated with an increase of one unit on the scale of the independent variable (e.g., for number of arrest charges, the odds ratio measures the effect of having one additional arrest charge). Our third statistical measure used to assess the effect of the independent variables is the beta coefficient (Menard 1995). The beta (Î&#x2019;) coefficient takes into account not only the change in the likelihood of the outcome associated with a change in the independent variable, but also the distribution of the cases among the categories of the independent variable. Being in one category of an independent variable may


71 have a large effect on the likelihood of an outcome (and therefore the variable may have a large odds ratio), but if there are relatively few cases in that category, the variable will not help to explain much of the variation in the likelihood of the outcome. For example, a defendant charged with a sex offense might have a high probability of re-arrest, and this variable would have a high odds ratio. However, if only a small number of defendants in the sample are charged with a sex offense, this variable would not be able to explain much of the variation in likelihood of re-arrest. Standardized betas measure this overall effect of the independent variable on the dependent variable. Standardized betas vary from -1 to +1; values closer to zero indicate that the effect of the independent variable is relatively small, while values closer to +1 or -1 indicate that the effect of the independent variable is relatively strong. There are no commonly accepted absolute standards to determine whether a standardized beta is strong or weak. Consequently, we will discuss the relative strength of variables, describing some as stronger or weaker than others. In the current study, we used all three of the measures discussed above. We used the statistical significance level to distinguish those independent variables that had a detectable11 effect on the dependent variable from those that did not. We used the odds ratio to evaluate the size of the effect of the independent variable and we used the standardized beta to evaluate the ability of the independent variable to account for variation in the dependent variable. The models we discuss include a large number of predictors of the dependent variable. In these models, the measures of the effect of each independent variable (statistical significance, odds ratio, and standardized beta) evaluate the effect of that independent variable after controlling for the effects of all the other independent variables in the model. These effects represent the net effect of a given independent variable after the effect of all the other independent variables have been taken into account. This net effect differs from the total effect of the independent variable, which is the effect of the independent variable when it is used as the only predictor of the dependent variable. To evaluate the overall ability of all the independent variables in the logistic regression model to predict the dependent variable, we use a statistical measure called Nagelkerke R2 (SPSS, Inc. 1999). This measure varies from 0 to +1. It can be roughly interpreted as indicating what proportion of the variation in the dependent variable is explained by all the independent variables in the model (see Menard 1995 for a full discussion of the R2 statistic in logistic regression models). Low values of R2 (closer to 0) indicate that the model as a whole is relatively weak in accounting for variation in the dependent variable. High values (closer to +1) indicate that the model as a whole is very successful in accounting for variation in the dependent variable. 11

Due to sampling error, and limitations of logistic regression techniques, it is possible that some independent variables that do affect the dependent variable are found to be statistically insignificant in our particular sample of cases. See Mohr (1990) for a further discussion of these issues.


72 2. Correcting for Selection Bias Our models predicting likelihood of pretrial failure to appear and pretrial re-arrest for a new DV offense included a control variable to correct for selection bias. Selection bias is a problem that arises in statistical analysis when the group of cases that could have ended up in the sample is restricted to a selected set of respondents (Berk 1983). When selection bias occurs, the statistical estimates of the effects of the independent variables may be biased. These estimates may overstate, or understate, the influence of an independent variable. If problems of selection bias are not addressed, the interpretation of the results may be misleading. In our analysis, the models predicting failure to appear and re-arrest include only cases that were released at some point between arraignment and disposition. The variables that influence whether a defendant was released also may influence whether the defendant fails to appear and/or is rearrested. For example, number of prior misdemeanor convictions may affect both the likelihood of release and the likelihood of failure to appear. If a model predicting the likelihood of failure to appear among those released does not control for selection bias, the estimate of the effect of number of prior misdemeanor convictions will be overstated. Part of its effect on likelihood of failure to appear will actually be due to its influence on the likelihood of release, i.e., on the likelihood that the case ended up in the sample of released cases. The remainder of its effect, if any, will be due to its influence on the likelihood of failure to appear. To control for this kind of selection bias, we included in the models a control variable that measures the predicted probability of release. This predicted probability of release was created using the logistic regression model shown in Table A-1. To avoid statistical problems12 the predicted probability of release used as a correction for selection bias was created using a somewhat different set of independent variables than was used in the models presented in Chapters 4 and 5. The predicted probability of release can theoretically vary from a low of 0.00 to a high of 1.00. Of course, the model predicting likelihood of failure to appear included only those cases where the defendant actually was released. The predicted probability of release for convicted cases was skewed toward the higher end of the scale. As shown in Appendix B, the mean predicted probability for the released cases in our analyses was .90. Nevertheless, even among cases where the defendant was released, there was significant variation: the predicted probability of release ranged from .17 to .99 (data not shown). It is this variation that enables the predicted probability of release to correct for selection bias. Among released cases, those with a low predicted probability of release are more similar to those who were not released, while those with a high predicted probability of release are more representative of those actually in the sample. The influence of the 12

When the predicted probability of release is included in a model predicting the likelihood of failure to appear, it is important that the predicted probability of release not be highly correlated with other variables in the model. This problem, known as multicollinearity, is particularly likely if the same set of independent variables is used in both the release and failure-to-appear models. For this reason, the model creating the predicted probability of release uses some different independent variables than those used in the analyses predicting failure to appear and re-arrest that were presented in this report.


73

TABLE A-1 LOGISTIC REGRESSION MODEL PREDICTING LIKELIHOOD OF PRETRIAL RELEASE CRIMES AGAINST PERSONS AND PROPERTY SUBSAMPLE1 Combined 2001-2002 Dataset

2

INDEPENDENT VARIABLES

Standardized

Odds

β

Ratio

ARRAIGNMENT CHARGE TYPE Reference Category: Assault (PL 120) Criminal Contempt (PL 215) Harassment (PL 240) Crimes Against Children (PL 260) Other

-0.10 *** -0.03 0.03 -0.08 ***

0.60 0.85 1.29 0.78

ANY PRIOR ARRESTS ANY PRIOR MISDEMEANOR CONVICTIONS ANY PRIOR FELONY CONVICTIONS ANY OPEN CASES AT TIME OF ARREST

-0.40 -0.15 -0.11 -0.04

*** *** *** **

0.32 0.64 0.68 0.88

PRIOR MISDEMEANOR JAIL SENTENCES Reference Category: None One Prior Misdemeanor Jail Sentence Two or More Prior Misdemeanor Jail Sentences ANY PREVIOUS PRISON SENTENCE ON PAROLE FROM PRISON DEFENDANT HAD 2 OR MORE BENCH WARRANTS AT TIME OF ARREST

-0.02 -0.14 -0.03 -0.04 -0.07

*** * ** ***

0.88 0.54 0.86 0.71 0.73

-0.18 *** -0.08 ***

0.77 0.78

0.22 *** -0.13 *** -0.08 ***

1.88 0.53 0.62

DEFENDANT'S CRIMINAL HISTORY

CHARGE CHARACTERISTICS NUMBER OF ARREST CHARGES ARREST CHARGE IS A FELONY RELEASE RECOMMENDATION Reference Category: No Recommendation (Weak NYC Ties) Recommended or Qualified Recommendation Open Bench Warrant at Time of Arrest Missing

Table Continues on Next Page


74

TABLE A-1 (continued)

2

Standardized

Odds

β

Ratio

INDEPENDENT VARIABLES DEFENDANT'S DEMOGRAPHIC CHARACTERISTICS SEX (Female) ETHNICITY: Reference Category: Non-Hispanic Black Non-Hispanic White Hispanic Other AGE: Reference Category: Age 16-20 Age 21-29 Age 30-39 Age 40 and over Nagelkerke R2 (N of cases)

NOTES 1 2

See text for a description of the dataset and the subsample. See Appendix C for information about the measurement and coding of the variables.

* Statistically significant at p < .05 ** Statistically significant at p < .01 *** Statistically significant at p < .001

0.14 ***

1.69

0.08 *** -0.02 0.08 **

1.39 0.93 1.59

0.04 0.04 0.07 **

1.13 1.13 1.27

0.29 *** (21,448)


75 predicted probability of release on the likelihood of failure to appear controls for the influence of variables that affect both release and failure to appear. When the predicted probability of release is included as a control variable in the model predicting the likelihood of failure to appear, the estimates of the effects of the other independent variables in the model are more accurate.13 Similarly, the influence of the predicted probability of release on the likelihood of pretrial re-arrest controls for the influence of variables that affect both release and pretrial re-arrest. Including the predicted probability of release as a control variable in the model predicting pretrial re-arrest makes the estimates of the effects of the other independent variables in the model more accurate. To return to the example discussed above, the number of prior misdemeanor convictions may influence both the likelihood of release and the likelihood of failure to appear. In our model predicting the likelihood of failure to appear, the estimate of the effect of number of prior misdemeanor convictions was more accurate because the model controlled for the influence of number of prior misdemeanor convictions on the likelihood of release.14 As a result, we had greater confidence in our estimates of the effects of this and other independent variables, as well as in our interpretation of the results of the model. Although we included the predicted probability of release as a control variable in the model, we did not discuss the impact of this variable since its primary purpose was to enable us to estimate accurately, and to interpret, the effects of the other variables.

13

See Heckman (1979) and Peterson (1989) for a more detailed discussion of selection bias and corrections for it. 14 The extent to which the estimates are more accurate depends on the ability of the model predicting the probability of release to explain a significant portion of the variation in likelihood of release. Our model was reasonably successful at explaining variation in likelihood of release. The model correcting for selection bias accounted for approximately 29% of the variation in likelihood of release.


76

[This page intentionally left blank]


77 APPENDIX B: DISTRIBUTION OF VARIABLES FOR REGRESSION MODELS

VARIABLES AND DISTRIBUTION DEPENDENT VARIABLES DEFENDANT EVER FAILED TO APPEAR Never failed to appear Ever failed to appear Total, all cases

90% 10% 100%

DEFENDANT RE-ARRESTED FOR A NEW DV OFFENSE PRIOR TO CASE DISPOSITION Not re-arrested Re-arrested Total, all cases

91% 9% 100%

CONTROL VARIABLES MEAN PREDICTED PROBABILITY OF RELEASE MEAN NUMBER OF DAYS AT RISK OF FAILING TO APPEAR MEAN NUMBER OF DAYS AT RISK OF PRETRIAL RE-ARREST

.90 86% 88%

DEFENDANT'S CRIMINAL HISTORY ANY PRIOR ARRESTS No Yes Total

50% 50% 100%

ANY PRIOR MISDEMEANOR CONVICTIONS No Yes Total

76% 24% 100%

ANY PRIOR FELONY CONVICTIONS No Yes Total

83% 17% 100%

NUMBER OF PRIOR MISDEMEANOR CONVICTIONS None 1 2 3 4 5 6 7 or more Total

77% 10% 4% 3% 2% 1% 1% 3% 100%1

NOTE Percentages do not sum to 100% due to rounding.

1


78 APPENDIX B: DISTRIBUTION OF VARIABLES FOR REGRESSION MODELS (continued)

VARIABLES AND DISTRIBUTION NUMBER OF PRIOR FELONY CONVICTIONS None 1 2 3 or more Total

83% 10% 4% 3% 100%

NUMBER OF OPEN CASES None 1 2 or more Total

81% 14% 5% 100%

PRIOR MISDEMEANOR JAIL SENTENCES None 1 2 or more Total

88% 5% 7% 100%

ANY PREVIOUS PRISON SENTENCE No Yes Total

92% 8% 100%

ON PAROLE FROM PRISON No Yes Total

98% 2% 100%

DEFENDANT HAD 2 OR MORE BENCH WARRANTS AT TIME OF ARREST Defendant did not have 2+ bench warrants at time of arrest Defendant had 2+ bench warrants at time of arrest Total

94% 6% 100%

DEFENDANT PREVIOUSLY SENTENCED AS A YOUTHFUL OFFENDER Defendant not previously sentenced as a YO Defendant previously sentenced as a YO Total

94% 6% 100%

RELEASE RECOMMENDATION: Recommended or Qualified Recommendation No recommended: weak NYC area ties Open Bench Warrant At Time of Arrest Other or Missing Total

Table Continues on Next Page

66% 24% 5% 5% 100%


79 APPENDIX B: DISTRIBUTION OF VARIABLES FOR REGRESSION MODELS (continued)

VARIABLES AND DISTRIBUTION DEFENDANT'S COMMUNITY TIES UNEMPLOYED Employed Not employed Total

63% 37% 100%

AT CURRENT ADDRESS 1 YEAR OR LESS Defendant at current address more than 1 year Defendant at current address 1 year or less Total

72% 28% 100%

LIVES WITH SOMEONE No Yes Total

72% 28% 100%

DOES NOT EXPECT ANYONE AT ARRAIGNMENT Expects someone at arraignment Does not expect anyone at arraignment Total

36% 64% 100%

HAS NO TELEPHONE Has a telephone Has no telephone Total

79% 21% 100%

LIVES OUTSIDE NYC AREA Lives in NYC area Lives outside NYC area Total

97% 3% 100%

CHARGE CHARACTERISTICS MEAN NUMBER OF ARREST CHARGES

1.78

ARRAIGNMENT CHARGE TYPE Assault (PL 120) Criminal contempt (PL 215) Other Total

66% 12% 21% 100%1

ARRAIGNMENT CHARGE SEVERITY B Felony C Felony D Felony E Felony A Misdemeanor B Misdemeanor Violation Total

1% 1% 5% 3% 86% 4% 0% 100%

NOTE Percentages do not sum to 100% due to rounding.

1

Table Continues on Next Page


80 APPENDIX B: DISTRIBUTION OF VARIABLES FOR REGRESSION MODELS (continued)

VARIABLES AND DISTRIBUTION RELEASE CHARACTERISTICS STAGE OF RELEASE Released at arraignment Released after arraignment Total

81% 19% 100%

TYPE OF RELEASE Released on recognizance Released on bail Total

83% 17% 100%

DEFENDANT'S DEMOGRAPHIC CHARACTERISTICS BOROUGH Brooklyn Manhattan Queens Staten Island Bronx Total

39% 18% 20% 4% 19% 100%

SEX Male Female Total

82% 18% 100%

ETHNICITY: Non-Hispanic Black Non-Hispanic White Hispanic Other Total

48% 13% 33% 6% 100%

AGE: Age 16-20 Age 21-29 Age 30-39 Age 40 and over Total

10% 31% 33% 26% 100%

DEFENDANT-VICTIM RELATIONSHIP Boyfriend-girlfriend Married Common-law marriage Other relationship Missing Total

18% 20% 19% 13% 30% 100%


81

APPENDIX C: CODING OF VARIABLES FOR REGRESSION MODELS

1

CODING SCHEME

VARIABLES DEPENDENT VARIABLES

DEFENDANT EVER FAILED TO APPEAR DEFENDANT RE-ARRESTED FOR A NEW DV OFFENSE PRIOR TO CASE DISPOSITION

Ever failed to appear = 1, Never failed to appear = 0 Re-arrested = 1, Not re-arrested = 0

CONTROL VARIABLES SELECTION BIAS CORRECTION: LIKELIHOOD OF RELEASE

Continuous, ranges from 0.00 to 1.00

TIME AT RISK OF FAILING TO APPEAR

Number of days

TIME AT RISK OF PRETRIAL RE-ARREST

Number of days

DEFENDANT'S CRIMINAL HISTORY ANY PRIOR ARRESTS

Any prior arrests = 1, All other categories = 0

ANY PRIOR MISDEMEANOR CONVICTIONS

Any prior misdemeanor convictions = 1, All other categories = 0

ANY PRIOR FELONY CONVICTIONS

Any prior felony convictions = 1, All other categories = 0

NUMBER OF PRIOR MISDEMEANOR CONVICTIONS

None = 0, One = 1, Two = 2, Three = 3, Four = 4, Five = 5, Six = 6, Seven = 7 or more

NUMBER OF PRIOR FELONY CONVICTIONS

None = 0, One = 1, Two = 2, Three = 3 or more

NUMBER OF OPEN CASES

None = 0, One = 1, Two = 2 or more

PRIOR MISDEMEANOR JAIL SENTENCES Reference Category: None One Prior Misdemeanor Jail Sentence

No prior misdemeanor jail sentence: Reference Category One prior misdemeanor jail sentence = 1, All other categories = 0 Two or more prior misdemeanor jail sentences = 1, All other categories = 0

Two or More Prior Misdemeanor Jail Sentences ANY PREVIOUS PRISON SENTENCE

Any previous prison sentence = 1, All other categories = 0

ON PAROLE FROM PRISON

On parole from prison = 1, All other categories = 0

DEFENDANT HAD 2 OR MORE BENCH WARRANTS AT TIME OF ARREST

Defendant had 2 or more bench warrants at time of arrest = 1, All other categories = 0

DEFENDANT PREVIOUSLY SENTENCED AS A YOUTHFUL OFFENDER

Defendant previously sentenced as a youthful offender = 1, All other categories = 0

RELEASE RECOMMENDATION: Reference Category: No Recommendation (Weak NYC Ties) Recommended or Qualified Recommendation Open Bench Warrant At Time of Arrest Other or Missing

No recommendation (weak NYC ties): Reference Category Recommended or qualified recommendation = 1, All other categories = 0 Open bench warrant = 1, All other categories = 0 Other or missing = 1, All other categories = 0

DEFENDANT'S COMMUNITY TIES UNEMPLOYED

Unemployed = 1, All other categories = 0

AT CURRENT ADDRESS 1 YEAR OR LESS

At current address 1 year or less = 1, All other categories = 0

LIVES WITH SOMEONE

Lives with someone = 1, All other categories = 0

DOES NOT EXPECT ANYONE AT ARRAIGNMENT

Does not expect anyone at arraignment = 1, All other categories = 0

HAS NO TELEPHONE

Has no telephone = 1, All other categories = 0

LIVES OUTSIDE NYC AREA

Lives outside NYC area = 1, All other categories = 0

Table Continues on Next Page


82

APPENDIX C: CODING OF VARIABLES FOR REGRESSION MODELS (continued)

1

CODING SCHEME

VARIABLES CHARGE CHARACTERISTICS

NUMBER OF ARREST CHARGES ARRAIGNMENT CHARGE TYPE Reference Category: Assault (PL 120) Criminal Contempt (PL 215) Other

Number of charges, ranges from 1 to 4

ARRAIGNMENT CHARGE SEVERITY

A Felony = 1, B Felony = 2, C Felony = 3, D Felony = 4, E Felony = 5, A Misdemeanor = 6, B Misdemeanor = 7, Unclassified Misdemeanor = 8, Violation = 9

Assault: Reference Category Criminal Contempt = 1, All other categories = 0 Other = 1, All other categories = 0

RELEASE CHARACTERISTICS STAGE OF RELEASE

Release at arraignment = 1, Released after arraignment = 0

TYPE OF RELEASE DEFENDANT'S DEMOGRAPHIC CHARACTERISTICS

Bail = 1, ROR = 0

BOROUGH Reference Category: Bronx Brooklyn Manhattan Queens Staten Island

Bronx: Reference Category Brooklyn = 1, All other categories = 0 Manhattan = 1, All other categories = 0 Queens = 1, All other categories = 0 Staten Island = 1, All other categories = 0

SEX (Female)

Female = 1, Male = 0

ETHNICITY: Reference Category: Non-Hispanic Black Non-Hispanic White Hispanic Other

Non-Hispanic Black: Reference Category Non-Hispanic White = 1, All other categories = 0 Hispanic = 1, All other categories = 0 Other = 1, All other categories = 0

AGE: Reference Category: Age 16-20 Age 21-29 Age 30-39 Age 40 and over

Age 16-20: Reference Category Age 21-29 = 1, All other categories = 0 Age 30-39 = 1, All other categories = 0 Age 40 and over = 1, All other categories = 0

DEFENDANT-VICTIM RELATIONSHIP Reference Category: Married Boyfriend-girlfriend Common-law marriage Other relationship Missing

NOTE 1

See text for a description of the variables in the models.

Married: Reference Category Boyfriend-girlfriend = 1, All other categories = 0 Common-law marriage = 1, All other categories = 0 Other relationship = 1, All other categories = 0 Missing = 1, All other categories = 0

FTA and DV Re-Arrests 06  
Read more
Read more
Similar to
Popular now
Just for you