Post-Disposition Re-Arrests of Juvenile Offenders by Administrator

Peter C. Kiers

Acting Executive Director

POST-DISPOSITION RE-ARRESTS OF JUVENILE OFFENDERS

Marian J. Gewirtz Project Director and Senior Research Analyst

Final Report September 2016

52 Duane Street, Third Floor, New York, NY 10007-1231 The mission of the New York City Criminal Justice Agency, Inc., is to assist the courts and the City in reducing unnecessary pretrial detention.

(646) 213-2500

POST-DISPOSITION RE-ARREST OF JUVENILE OFFENDERS

Marian J. Gewirtz Project Director and Senior Research Analyst

Research Assistance: Miles Riemer-Peltz Data Analyst

Administrative Support: Lauren A. Wilson Communications Associate

Information Systems Programming: Wayne Nehwadowich IT Deputy Director for Programming

September 2016 This report can be downloaded from nycja.org/library.php.

ďŁŠ 2016 NYC Criminal Justice Agency, Inc. When citing this report, please include the following elements, adapted to your citation style: Gewirtz, Marian J. 2016. Post Disposition Re-Arrest of Juvenile Offenders. New York: New York City Criminal Justice Agency, Inc.

TABLE OF CONTENTS Acknowledgements ........................................................................................................................ iii INTRODUCTION ............................................................................................................................. 1 I. The Research Population and Time at Risk for Re-Arrest ......................................................... 3 EXHIBIT 1: START OF AT-RISK PERIOD FOR RE-ARREST POST DISPOSITION ...................... 4 EXHIBIT 2: START OF AT-RISK PERIOD FOR RE-ARREST POST CONVICTION ..................... 5 II. The Disposition of the JO Case.................................................................................................. 7 EXHIBIT 3: JO CASE OUTCOME ........................................................................................... 7 III. Post-Disposition Re-Arrest ..................................................................................................... 9 A. Overview of Re-Arrest Rates Post Disposition ................................................................... 9 EXHIBIT 4: RE-ARREST RATES POST DISPOSITION ............................................................. 9 EXHIBIT 5: RE-ARRESTS PER JUVENILE (Re-Arrested Juveniles Only) .............................. 11 B. Re-Arrest Rates Post Disposition by Time at Risk to Re-Arrest ....................................... 11 EXHIBIT 6: RE-ARREST RATES POST DISPOSITION BY TIME AT RISK TO RE-ARREST ........ 12 C. The First Re-Arrest Charge and the First Felony Re-Arrest Charge Post Disposition ....... 13 EXHIBIT 7: CHARGE SEVERITY AND CHARGES FOR RE-ARREST ................................................ 14 D. Time At Risk to Re-Arrest Post Disposition ....................................................................... 14 EXHIBIT 8: TIME AT RISK POST DISPOSITION TO RE-ARREST ........................................... 15 EXHIBIT 9: MEDIAN NUMBER OF DAYS AT RISK POST DISPOSITION TO RE-ARREST BY OUTCOME OF THE INITIAL JO CASE ........................................ 16 E. Re-Arrest Post Disposition By Selected Juvenile Characteristics: Gender, Ethnicity and Prior Arrests ............................................................................................................... 17 EXHIBIT 10: RE-ARREST BY GENDER, ETHNICITY AND PRIOR ARRESTS ........................... 18 F. Re-Arrest Post Disposition By Selected JO Case Characteristics: Charge, Borough of Arrest and Release Status ................................................................................................. 18 EXHIBIT 11: RE-ARREST BY JO CHARGE, BOROUGH AND RELEASE STATUS .................... 20 G. Re-Arrest Post Disposition by the Outcome of the Initial JO Case ................................... 20 EXHIBIT 12: RE-ARREST POST DISPOSITION BY OUTCOME OF THE INITIAL JO CASE....... 21

IV. MULTIVARIATE ANALYSES.....................................................................................................23 A. Multivariate Findings ................................................................................................. 24 EXHIBIT 13: COX PROPORTIONAL HAZARD REGRESSION MODELS OF TIME TO FIRST RE-ARREST POST DISPOSITION ...................................... 25 FIGURE 1: Probability of Surviving Without a Re-Arrest Leading to Conviction, By Gender .......................................................................................... 27 FIGURE 2: Probability of Surviving Without a Re-Arrest Leading to Conviction, By JO Case Disposition ....................................................................... 28 FIGURE 3: Probability of Surviving Without a Re-Arrest Leading to Felony Conviction, By JO Case Disposition ........................................................... 29 B. Discussion of Multivariate Findings ............................................................................ 29 V. CONCLUSION .......................................................................................................................... 33 VI. POLICY IMPLICATIONS .......................................................................................................... 35 APPENDIX A: JUVENILE OFFENSES .............................................................................................. 36 APPENDIX B: STATISTICAL PROCEDURES (By Mary T. Phillips, Ph.D.) ...................................... 37

Acknowledgements This report, to use a common phrase, is the result of the efforts of “a village.” Although it is not possible to acknowledge all by name, I want to briefly highlight the contributions that made this research possible. Staff of the NYC Criminal Justice Agency (CJA) in each borough was responsible for interviewing the juvenile defendants prior to their arraignment in the Criminal Court. CJA technical staff integrated the interview data with arrest data from the NYPD and court outcome data from the Office of Court Administration (OCA). Then Wayne Nehwadowich, now IT Deputy Director for Programming, extracted the data from CJA’s main informational system and created the research files that served as the basis for this project. The remaining contributors to the work that culminated in this report are within the research department at CJA. Miles Riemer-Peltz, Data Analyst, focused on gathering the rearrest data across eight years of research files, files which when combined surpassed two and half million records. As the project matured, Miles returned to the already-enormous master file to supplement the data with the outcomes of the re-arrests. Eoin Healy, Senior Research Analyst, offered technical advice as I built and refined the multivariate models. A special debt of gratitude is owed to Mary Phillips, Deputy Director of Research, who offered editorial assistance but went far beyond expectations as she sought greater clarity of analysis and descriptive language with each revision. In addition, Mary Phillips wrote the discussion of the type of survival analysis (Cox proportional hazard regression) used in this research which appears as Appendix B. Finally, this report benefited from the review and comments of Richard Peterson, Director of Research. While credit is due to many, the methodology, analysis, findings, and conclusions, as well as any errors, are the sole responsibility of the author.

iii

INTRODUCTION In accordance with New York State’s Juvenile Offender (JO) Law, youths under age 16 who face serious violent felony charges are brought directly to the adult rather than juvenile court for prosecution. Most of these youths are charged with robbery in the first or second degree, but the list of eligible charges (see Appendix A) also includes murder, manslaughter, burglary, weapons and other offenses. Previous research conducted by the New York City (NYC) Criminal Justice Agency (CJA) documents extremely high rates of recidivism among the youths processed as JOs. CJA’s 2015 research 1 on recidivism among JOs attempted to assess the relationship between the outcome of cases involving JOs and the likelihood of re-arrest. However, that research did not consistently differentiate between pre-and post-disposition recidivism. The report discussed here is a follow-up endeavor, again exploring the relationship between case outcome and recidivism for fourteen- and fifteen-year-olds processed as JOs in the adult courts in NYC, but the current research examines only post-disposition re-arrests. Are youths convicted in JO cases at greater risk of re-arrest than are their counterparts whose cases reached other outcomes? The current research focuses on JO cases arraigned since January 1, 2007, that reached a final disposition in the adult court by December 31, 2014. Re-arrest is tracked starting at disposition for youths who were at risk as of that date, or from the youth’s date of release after conviction and/or sentencing. In addition to identifying factors predictive of time to the first rearrest and to the first felony-level re-arrest post disposition, this research also looks at the outcome of the re-arrests. That is, this research also examines factors predictive of the first rearrest post disposition that resulted in conviction and the first that resulted in a felony-level conviction. The goal of the research is to determine whether any of these measures of recidivism varies by case outcome. The report begins with a description of the research population, including how time at risk for re-arrest was determined. We then examine post-disposition recidivism in terms of the 1

Gewirtz, Marian J. 2015. Recidivism Among Juvenile Offenders in New York City, 2007-2012: A Comparison by Case Outcome. New York: New York City Criminal Justice Agency, Inc., and Gewirtz, Marian J. 2015. “Juvenile Offenders: Re-Arrest and Court Outcomes.” Research Brief series, no. 38. New York: New York City Criminal Justice Agency, Inc.

rates of re-arrests as well as their severity and timing. We also compare the re-arrest rates by selected characteristics of the juveniles (gender, ethnicity and prior arrests) and by selected characteristics of the initial JO case (charge, borough of arrest and release status). Multivariate analyses using Cox proportional hazards regression is used to explore the factors associated with the risk of post-disposition re-arrest. We assess whether the disposition affects the risk of recidivism by building the strongest Cox models and attempting to include the type of disposition as a predictive variable.

I. The Research Population and Time at Risk for Re-Arrest The research began with a defendant-based file of 3,115 juveniles who were arraigned in the Criminal Court between January 2007 and December 2014 whose cases reached final disposition in the adult court as of December 2014. In the event that there was more than one JO case arraigned during that period for a juvenile, only the earliest case was selected for inclusion. The research file is restricted to juveniles who were at risk for re-arrest post disposition (see Exhibit 1). In cases that were dismissed, acquitted or transferred to the Family Court, the juvenile was considered at risk as of the date of disposition. 2 Among juveniles who were convicted, the at-risk date is the first date after disposition that the juvenile was at risk. That date may be the date of release at disposition, between disposition and sentencing, at sentencing or after sentencing. Juveniles who were sentenced to probation, conditional discharge, or time already served prior to sentencing were considered at risk as of the date of sentencing. It was more difficult to determine the date the juvenile was at risk after disposition when the juvenile was sentenced to imprisonment because the date of release from prison is not available. We planned to estimate the at-risk date based on the length of sentence, adjusted for time in detention prior to sentencing, and we have confidence in our estimates when the juvenile was given a relatively short prison term, specifically, a maximum of one year. However, the estimated at-risk date did not seem reliable when the sentences were longer and when the length of the sentence was a range from a minimum to a maximum. For that reason we excluded 270 juveniles who were not at risk after conviction and who were not at risk at any time prior to sentencing and who were sentenced to a maximum of more than one year. Four

Time at risk is overestimated for the juveniles whose cases were transferred to the Family Court because it is likely that some were actually held in detention for a period of time pending the adjudication of the case in the Family Court. However, there were only 144 juveniles at risk for re-arrest who were not already released before their cases were transferred to the Family Court. The Administration for Childrenâ&#x20AC;&#x2122;s Services, which is responsible for the detention services for children, maintains that juveniles facing delinquency charges in the Family Court are held an average of 15 to 30 days, but that youths who face juvenile offender charges in the adult court spend longer in detention. Based on those figures, we temporarily subtracted 30 days from the days at risk for the juveniles who were not released in the adult court prior to the transfer of their cases to the Family Court and who were not already re-arrested within 30 days of the transfer. Since this change did not affect the variables in the multivariate models presented in Section IV in this report, we did not use the re-estimated time at risk for these cases.

juveniles who were not at risk between conviction and sentencing and whose sentences were a year or less were also excluded because they had no time at risk prior to December 31, 2014. EXHIBIT 1: START OF AT-RISK PERIOD FOR RE-ARREST POST DISPOSITION Number of Cases

TIME AT RISK STARTS: At Disposition: Dismissed, Acquitted, Transferred to Family Court At Disposition - Convicted

Percent

1,674 931

59 33

51 54 131

2 2 5

TOTAL AT-RISK FOR RE-ARREST POST DISPOSITION

2,841

100*

Excluded from At-Risk Group: Time At Risk Unknown-Sentenced To More than One Year Not At Risk Though Sentenced To Up To One Year Total Defendant-Based File

270 4 3,115

After Disposition:

Between Conviction and Sentencing At Sentencing After Serving Sentence Up To One Year

*Does not sum to 100% due to rounding

Most (80%) of the 1,167 youths who were convicted in JO cases were at risk for re-arrest as of the date of their conviction and an additional four percent began their at-risk period between conviction and sentencing (Exhibit 2). Five percent (54) of the convicted youths began their at-risk period at sentencing because they were not at risk between conviction and sentencing and their sentence did not require prison time. Finally, there were 131 (11%) who were at risk for re-arrest only after they served some part of a sentence of less than one year. Some of the juveniles who were at risk at or after disposition were later detained pending sentencing or eventually sentenced to time in prison, but the start of the at-risk period for these juveniles was not affected by later detention or prison time. Time at risk was truncated at the date of sentencing for 29 youths who were at risk for re-arrest prior to sentencing and were not re-arrested but who were sentenced to a maximum prison length greater than one year. For the remaining youths at risk prior to sentencing but not re-arrested then and who 4

were later sentenced to a maximum of less than one year in prison, the time at risk to re-arrest was adjusted by subtracting the estimated prison time. EXHIBIT 2: START OF AT-RISK PERIOD FOR RE-ARREST POST CONVICTION TIME AT RISK STARTS:

At Conviction Between Conviction and Sentencing At Sentencing After Serving Sentence Up To One Year

TOTAL AT-RISK FOR RE-ARREST POST CONVICTION

Number of Cases Percent 931 80 51 4 54 5 131 11 1,167

100

The research file contains 2,841 JO cases arraigned between 2007 and 2014 that reached final disposition in the adult court by the end of 2014 and had some time at risk for rearrest after disposition. Time at risk ranged from only a few days to nearly eight years. If all of the JO cases in the file had the maximum time at risk after disposition, it is likely that re-arrest rates would be higher, as would rates of felony re-arrest and rates of re-arrest that resulted in conviction. This report describes the re-arrests post disposition in terms of their timing and severity, and whether or not the re-arrests resulted in conviction, but the focus of the report is on assessing whether re-arrest rates vary by the type of outcome for the initial JO case. The Cox multivariate proportional hazards analyses do not require that all of the cases have an equal time at risk for re-arrest. Cox survival analysis was selected specifically because cases that vary in time at risk can be included.

II. The Disposition of the JO Case In New York State defendants in JO cases cannot plead guilty or be convicted at trial in the Criminal Court, the misdemeanor court, because juveniles under the age of 16 can be held criminally responsible only for one of the enumerated felony JO charges, and virtually all felony convictions take place in the Supreme Court. The juveniles cannot plead guilty to a misdemeanor in the Criminal Court because the juveniles in these cases would no longer be JOs and would no longer be eligible for prosecution in the adult court. The only final dispositions in the Criminal Court for these JO cases are dismissal or transfer to Family Court. The JO case may be transferred to the Family Court for prosecution there at any point in case processing, but most transfers are from the Criminal Court. Similarly, any case can be dismissed from either the Criminal or Supreme Court, but most JO cases that are dismissed are dismissed in the Criminal Court. As shown in Exhibit 3, below, 43 percent of the JO cases reached their final adult-court disposition in the Criminal Court. More than a quarter were dismissed and 16 percent were transferred to the Family Court from the Criminal Court. Most of the JO cases were disposed in the Supreme Court (57%). The most common disposition was conviction, by plea or at trial, in the Supreme Court (41%), although trials are rare in NYC. Transfers to the Family Court (7%) and dismissals (9%) were uncommon in the Supreme Court. Sixteen cases resulted in acquittal after trial. EXHIBIT 3: JO CASE OUTCOME Criminal Court Disposition Dismissed Transferred to Family Court Subtotal Criminal Court Supreme Court Disposition Convicted Dismissed or Acquitted Transferred to Family Court Subtotal Supreme Court TOTAL Disposed as of December 31, 2014

Number of Cases 764 455

1,219

1,167 261 194 1,622 2,841

Percent 27 16

41 9 7

57 100 7

III. Post-Disposition Re-Arrest A. Overview of Re-Arrest Rates Post Disposition Re-arrest rates are extremely high among JOs prosecuted in the adult courts in NYC (Exhibit 4, below). Nearly three quarters of the juveniles were re-arrested and half of the juveniles were re-arrested for a felony-level offense after their initial JO case reached final disposition in the adult court. These findings are consistent with previous research on the JO population in NYC. Re-arrest is defined here as an arrest in NYC for which the offense date is after the initial JO case reached disposition. An arrest during the time the juvenile is in custody and not at risk is not considered a re-arrest. The re-arrest may be prosecuted in the adult court or the Family Court. As mentioned in the introduction to this report, this research also focuses on re-arrests post disposition that result in conviction and on those that result in conviction for a felony-level offense. The outcome of the re-arrest has not been considered in previous JO recidivism research at CJA. The findings again indicate high rates of re-arrest. More than half (53%) of the juveniles who were at risk for re-arrest were convicted in at least one of the re-arrests after disposition and nearly a quarter (23%) were convicted of a felony in at least one of the rearrests. EXHIBIT 4: RE-ARREST RATES POST DISPOSITION

Number of Cases 2,137

Percent 75%

Any Felony Re-Arrest

1,421

50%

Any Re-Arrest Leading to Conviction

1,496

53%

640

23%

2,841

100%

Any Re-Arrest

Any Re-Arrest Leading to Felony Conviction TOTAL At Risk for Re-Arrest

We were concerned that the 23 percent rate of re-arrest that resulted in a felony-level conviction might understate the actual rate of re-arrest leading to a felony conviction because there was little time for re-arrests for 2014 cases to reach a felony conviction. However, there were only 101 juveniles initially arraigned in 2014 who were at risk for re-arrest prior to the end of that year. Although only two of those juveniles were re-arrested for a case that reached a felony conviction by the close of data collection, exclusion of the 2014 at-risk JO cases affected 9

the percent re-arrested for a case that resulted in a felony conviction by less than a percentage point. Exclusion of juveniles initially arraigned in 2013 increased the rate of re-arrest that resulted in a felony conviction to 25 percent. The rate of re-arrest that resulted in any conviction was also affected by the lack of time for more recent re-arrests to reach felony conviction. Exclusion of 2014 JO cases increased the rate of re-arrest that resulted in any conviction to 54 percent and exclusion of 2013 JO cases increased that rate to 58 percent. The at-risk 2014 JO cases had less time at risk and had a lower rate of re-arrest and of felony re-arrest than did the at-risk juveniles who were arraigned in previous years. However, we continued to include all of the at-risk cases in the research discussed in this report, first because the differences in re-arrest rates are small and second, because these cases can still be included in the multivariate analysis since Cox survival analysis takes account of time at risk. Exhibit 5 displays the number of re-arrests and the number of felony re-arrests per juvenile and also displays the numbers as percentages of all of the juveniles who were rearrested and as cumulative percentages. More than eight of every ten re-arrested JOs were rearrested more than once after disposition, 64 percent were re-arrested more than twice and more than half were re-arrested more than three times. Seven percent of re-arrested JOs were re-arrested more than ten times. The 2,137 youths who were re-arrested account for 10,705 re-arrests (the total number of post-disposition re-arrests is not shown). The median (midpoint) number of re-arrests was four. The median was four re-arrests for each of the case outcome categories except the convicted category, for which the median was only three. Twenty-two of the re-arrests were for murder and 47 were for attempted murder (data not shown). There were fewer felony-level re-arrests. More than four of every ten youths who were re-arrested for a felony were re-arrested for a felony only once and more than two thirds were re-arrested once or twice. Seventeen percent were re-arrested for a felony more than three times. The 1,421 JOs re-arrested for at least one felony-level offense account for 3,264 felony re-arrests. The median number of felony-level re-arrests was two for each of the case outcome

categories except the dismissed or acquitted category, for which the median number of felonylevel re-arrests was only 1.5. 3 EXHIBIT 5: RE-ARRESTS PER JUVENILE (Re-Arrested Juveniles Only) Number of Re-Arrests 1 2 3 4 5 6 7 8-10 11 or more TOTAL RE-ARRESTED Median

All Re-Arrests Frequency

411 357 285 236 208 148 128 220 144

19 17 13 11 10 7 6 8 7

Cumulative % 19 36 49 60 70 77 83 93 100

2,137

100

Felony Re-Arrests Cumulative Frequency % % 607 43 43 363 26 68 204 14 83 101 7 90 68 5 95 33 2 97 20 1 98 17 1 99 8 1 100 1,421

100

B. Re-Arrest Rates Post Disposition by Time at Risk to Re-Arrest The post-disposition re-arrest rates presented above reflect re-arrests for all of the juveniles who were at risk for re-arrest after disposition, regardless of how long they were at risk. Although the multivariate analyses discussed in Section IV of this report do not require that all of the cases have an equal time at risk for re-arrest, re-arrest rates presented in Exhibit 6, below, control for the varying amounts of time juveniles were at risk. Nearly all of the 2,841 juveniles in the research file were at risk for re-arrest for a year after disposition of their initial JO case, or were re-arrested within the year (2,677, 94%). The 2,677 base for the one-year rearrest rate is comprised of the 1,339 juveniles re-arrested within one year and the 1,338 who were at risk at least one year but not re-arrested that quickly. The one year re-arrest rate is 50 percent of the 2,677 who were either re-arrested within one year or were at risk at least that long. The number of cases at risk or re-arrested is largest for the â&#x20AC;&#x153;first re-arrestâ&#x20AC;? category compared to the other re-arrest categories because that number is the sum of the volume of juveniles who were still at risk after one year combined with the number of re-arrests within 3

We did not tally the numbers of re-arrests that resulted in any conviction or in a felony-level conviction.

the year. For example, 646 juveniles were re-arrested for a felony within one year and 1,997 were at risk that long but not re-arrested for a felony. The number re-arrested for a felony within one year combined with the number at risk at least that long is 2,643. That number is smaller than the 2,677 base for the one-year-any re-arrest rate because some of the juveniles with a first re-arrest within a year did not have a felony re-arrest that quickly and were not at risk at least one year. As previously mentioned, half of the juveniles who were at risk for one year or more were already re-arrested within the year. Within one year at risk after disposition, 24 percent were re-arrested for a felony-level offense, 22 percent were re-arrested for any offense that resulted in conviction and eight percent were re-arrested for an offense that resulted in a felony conviction. As shown in Exhibit 6, fewer juveniles were at risk longer than four years. The four-year re-arrest rate reflects all of the re-arrests within four years at risk (2,083) as a percentage of all of the juveniles who were at risk longer than four years or who had already been re-arrested (2,432, because 349 were not re-arrested within four years and were at risk at least four years). In other words, the re-arrests cumulate over time to the cutoff year for the time at risk. Fully eighty six percent of the juveniles who were still at risk more than four years were already rearrested within that time. The four-year re-arrest rate was 63 percent for felony-level rearrests, 68 percent for a re-arrest leading to conviction, and 32 percent for a re-arrest leading to felony conviction. EXHIBIT 6: RE-ARREST RATES POST DISPOSITION BY TIME AT RISK TO RE-ARREST

PERCENT RE-ARRESTED OF JUVENILES AT RISK AT LEAST THAT LONG: Within One Year Within Two Years Within Three Years Within Four Years TOTAL AT RISK POST DISPOSITION

First Re-Arrest % 50 71 80 86

N of Cases (2,677) (2,574) (2,508) (2,432)

(2,841)

First Felony Re-Arrest % 24 41 52 63

N of Cases (2,643) (2,446) (2,293) (2,097)

(2,841)

First Re-Arrest First Re-Arrest Leading to Leading to Felony Conviction Conviction % 22 43 57 68

N of Cases (2,619) (2,411) (2,254) (2,059)

% 8 16 24 32

N of Cases (2,605) (2,328) (2,086) (1,749)

(2,841)

(2,841) 12

C. The First Re-Arrest Charge and the First Felony Re-Arrest Charge Post Disposition The first re-arrest was a felony for three of every ten JOs who were re-arrested (Exhibit 7). Five percent were A or B felonies, including one re-arrest for murder and seven re-arrests for attempted murder (not shown). On the other hand, nearly one in five of the first re-arrests were for theft of services (PL 165.15), commonly turnstile jumping, and fourteen percent were marijuana offenses including eleven percent for possession of marijuana (PL 221.10). The first felony re-arrest was an A or B felony for nearly a fifth of the JOs re-arrested for any felony (not shown). That includes six re-arrested for murder and 20 re-arrested for attempted murder at the first felony-level re-arrest. The most frequent charge at the first felony re-arrest was robbery in the second degree (PL 160.10, 23%). An additional seven percent were charged with robbery in the first degree (PL 160.15) and nearly four of every ten first felony-level re-arrests involved any robbery charge, including attempted robberies. After robbery charges, the next most frequent first felony re-arrest charge was assault in the second degree (PL 120.05, 8%). Assault and assault-related charges, including attempted assaults, account for fifteen percent of first felony-level re-arrests. Twelve percent of the first felony rearrests were for weapon offenses and eight percent of the first felony re-arrests were for narcotics charges. A quarter of the first re-arrests leading to a conviction were felony-level convictions, eleven percent were misdemeanor convictions, and more than six of every ten were violations. Nearly half of the first re-arrests that resulted in convictions were convictions for disorderly conduct (PL 240.20). The remaining convictions varied widely in the type of charge at conviction. Thirteen percent of the first re-arrests leading to a conviction were for a robbery, followed by petit larceny charges which account for nine percent of the convictions. There were 640 youths who were re-arrested for at least one case that resulted in a felony conviction. Obviously, the conviction charge was a felony for each of these re-arrests leading to a felony conviction. Half of these were convictions for a robbery charge. The remaining half are distributed roughly evenly across the assault, narcotics, weapon, burglary and other felony charge categories.

EXHIBIT 7: CHARGE SEVERITY AND CHARGES FOR RE-ARRESTS ARREST CHARGES

CONVICTION CHARGES First Re-Arrest First Re-Arrest Leading To Leading To Felony Conviction Conviction 25 100 11 61 3

First Re-Arrest

First Felony Re-Arrest

30 65 5 -

100

100%

ROBBERY

ASSAULT

NARCOTICS

MARIJUANA

WEAPON

BURGLARY PETIT LARCENY CRIMINAL TRESPASS THEFT OF SERVICES DISORDERLY CONDUCT

100%

(2,137)

(1,421)

(1,496)

(640)

CHARGE SEVERITY Felony Misdemeanor Violation Other TOTAL RE-ARRESTED CHARGES:

OTHER TOTAL RE-ARRESTED NUMBER RE-ARRESTED

D. Time At Risk to Re-Arrest Post Disposition Some juveniles were re-arrested very quickly. Nine juveniles were re-arrested within a day of their at-risk date after the disposition of their JO case and five juveniles were re-arrested for a felony that soon (data not shown). Nearly two percent (35 juveniles) were re-arrested within a week, including thirteen re-arrested for a felony. Exhibit 8 shows the amount of time at risk post disposition to each of the types of re-arrests for the juveniles who were re-arrested. 14

The first re-arrest occurred more quickly than did the first felony re-arrest which in turn occurred somewhat more quickly than the first re-arrest leading to a conviction, and more time at risk elapsed prior to the first re-arrest leading to a felony-level conviction. Eight percent of the first re-arrests occurred within a month of time at risk compared to only two percent of rearrests that resulted in a felony conviction. Nearly a quarter of first re-arrests occurred within three months, compared to only seven percent of the re-arrests leading to a felony conviction. Nearly two thirds of the re-arrests took place within a year as did 45 percent of the felony rearrests, 39 percent of the first re-arrests leading to a conviction and 31 percent of those that resulted in a felony conviction. Nearly all (94%) of the first re-arrests took place within three years, compared to just over three quarters of the re-arrests leading to felony-level convictions. EXHIBIT 8: TIME AT RISK TO RE-ARREST POST DISPOSITION

TIME AT RISK

Up to One Month One to 3 Months Three to 6 Months Six to 12 Months One Year Up to 2 Years Two Years Up to 3 Years More than 3 Years TOTAL RE-ARRESTED NUMBER RE-ARRESTED

First Re-Arrest

8 15 16 24 22 9 6

CUM % 8 23 39 63 85 94 100

100* (2,137)

First Felony Re-Arrest

4 10 13 18 26 13 15

100*

CUM % 4 14 27 45 71 84 100

First Re-Arrest Leading To Conviction %

3 6 10 20 30 17 15

100*

CUM % 3 9 19 39 69 86 100

First Re-Arrest Leading To Felony Conviction %

3 4 9 15 26 21 23

CUM % 3 7 15 31 57 77 100

100*

(1,421) (1,496) (640) *May not total to 100% due to rounding

The median number of days at risk to re-arrest is presented in Exhibit 9. The median is the midpoint of the distribution of days to re-arrest. The median number of days at risk to the first re-arrest was 257, compared to a median of 414 days to the first felony re-arrest, 482 days to the first re-arrest leading to a conviction and 624 days to the first re-arrest that resulted in a felony conviction. The median number of days at risk to the first re-arrest was lowest for the juveniles whose cases were dismissed in the Criminal Court (233) or the Supreme Court (238) 15

and longest for those whose cases were transferred to the Family Court from the Supreme Court (292). It is likely that some of the youths whose cases were transferred to the Family Court were not actually at risk as of their date of disposition but were detained as juvenile delinquents for some time pending the outcome of their cases in the Family Court, so the time at risk for these cases may be somewhat overstated.4 The median time at risk to re-arrest was longest or second longest for the juveniles whose cases were transferred to the Family Court from the Supreme Court for all four types of re-arrest. Similarly, the juveniles whose cases were dismissed in the Criminal Court showed the shortest or second shortest time to re-arrest. EXHIBIT 9: MEDIAN NUMBER OF DAYS AT RISK POST DISPOSITION TO RE-ARREST BY OUTCOME OF THE INITIAL JO CASE

JO CASE OUTCOME

CRIMINAL COURT: Dismissed Transferred to Family Court SUPREME COURT: Convicted: At Risk Before Sentencing At Risk After Sentencing Dismissed or Acquitted Transferred to Family Court

TOTAL RE-ARRESTED

First Re-Arrest Median (N)

First Felony Re-Arrest Median (N)

First Re-Arrest Leading To Conviction Median (N)

First Re-Arrest Leading To Felony Conviction Median (N)

233 266

354 342

493 629

570 653

(578) (340)

(392) (224)

(383) (214)

(191) (101)

263

(878) 468 (584) 415 (666) 621 (245) 284 (731) 487 (471) 425 (553) 638 (200) 236 (147) 388 (113) 328 (113) 472 (45) 238 (192) 424 (122) 534 (130) 667 (51) 292 (149) 481 (99) 711 (103) 707 (52) 257

(2,137) 414

(1,421) 482

(1,496) 624

(640)

Time at risk to re-arrest is often close to the overall median for juveniles whose initial JO case resulted in conviction because they constitute the largest category of cases in this We attempted to estimate how much the undercounted time at risk might affect the median time at risk. When we subtracted 30 days from the time at risk for the 116 juveniles at risk for re-arrest who were not already released before their cases were transferred to the Family Court from the Criminal Court and for 28 at risk for rearrest who were not already released before their cases were transferred to the Family Court from the Supreme Court, the median days at risk to the first re-arrest decreased by six days for the transfers from the lower court and by five days for the transfers from the upper court. As mentioned above, this did not affect the variables in the multivariate models presented in section IV so we did not use the re-estimated time at risk for these cases. 4

research. The re-arrested juveniles who were convicted of their initial JO charge took just a few days longer than average to reach their first re-arrest, about two months longer to reach their first felony re-arrest, more than two months shorter to be re-arrested for a case that resulted in conviction and just a few days shorter than average to be re-arrested in a case that resulted in another felony conviction. We explored the time to re-arrest for the juveniles who were convicted in their initial JO case and found that those who were at risk after conviction but before sentencing (the bulk of the convicted), were at risk about as long as were all of the convicted youths who were re-arrested. However, the small number (147) of re-arrested youths who were at risk only after they were sentenced (either to probation, a conditional discharge, or time already served prior to sentencing), or who served a short sentence before they were at risk for re-arrest, were re-arrested as fast or faster than other youths. Convicted youths who were at risk only after sentencing were rearrested a month and a half faster than other convicted youths. Furthermore, their first felony re-arrests and their first re-arrests that resulted in conviction occurred more than three months sooner and their first re-arrests that resulted in another felony conviction occurred nearly six months sooner. E. Re-Arrest Post Disposition by Selected Juvenile Characteristics: Gender, Ethnicity and Prior Arrests As in previous research and as shown below in Exhibit 10, re-arrest rates were higher among males (76%) than females (61%) and the difference was greater when only felony-level re-arrests (53% compared to 29%), re-arrests leading to a conviction (56% compared to 31%), or re-arrests leading to a felony conviction (25% compared to 7%) were considered. Re-arrest after disposition was also more common among black JOs (77%) than among Hispanic youths (72%) and especially compared to the small numbers of white (69%) or Asian (68%) juveniles. The differences are again wider for felony re-arrests: 53 percent of black JOs were rearrested compared to 45 percent of Hispanic juveniles, and 39 percent among both the white and Asian youths. The rates of re-arrests leading to convictions followed a similar pattern, ranging from a high of 55 percent among black youths to a low of 38 percent among Asian juveniles. A quarter of black juveniles were re-arrested and convicted at the felony level, compared to 19 percent of Hispanic youths, thirteen percent of the white youths and nine percent of the Asian juveniles. 17

Juveniles who had prior arrests in the CJA database were more likely to be re-arrested (80%) than were their counterparts with no prior arrests (70%). The difference was again wider for re-arrests for felony offenses. Fifty seven percent of the juveniles with one or more prior arrests were re-arrested for a felony-level offense compared to 41 percent of those with no prior arrests. The rates of re-arrests leading to conviction followed a similar pattern: 59 percent of juveniles with at least one prior arrest were re-arrested in a case that resulted in a conviction and 28 percent were re-arrested in a case that resulted in a felony conviction, compared to 45 percent and 16 percent, respectively, among juveniles who did not have a prior arrest. EXHIBIT 10: RE-ARREST BY GENDER, ETHNICITY AND PRIOR ARRESTS

ALL CASES GENDER Male Female

ETHNICITY Black White Hispanic Asian PRIOR ARRESTS Prior Arrests No Prior Arrests

Any Re-Arrest Leading To Felony Conviction

Any Re-Arrest

Felony Re-Arrest

Any Re-Arrest Leading To Conviction

2,451 390

76 61

53 29

56 31

25 7

1,914 84 787 56

77 69 72 68

53 39 45 39

55 49 49 38

25 13 19 9

1,548 1,293

80 70

57 41

59 45

28 16

N of Cases At Risk 2,841

F. Re-Arrest Post Disposition by Selected JO Case Characteristics: Charge, Borough of Arrest and Release Status Re-arrests rates also varied by the initial JO arrest charge (Exhibit 11). In each of the four types of re-arrest, re-arrest was less common among the juveniles who faced murder or attempted murder charges or sex offenses, followed by those charged with assault, and was more common among the juveniles charged with robbery or other JO charges. There was an 18 point spread across the charges in rates of any re-arrest post disposition (60 to 78%) and a 19 18

point spread across the charges in rates of any felony-level re-arrest (32 to 51%). The spread was widest for rates of re-arrest leading to a conviction: 35 percent of juveniles initially charged with murder or attempted murder and 37 percent of those initially charged with a sex crime were re-arrested and convicted, compared to more than half of those initially charged with robbery (54%) or other offenses (58%). The spread was nearly as wide for re-arrests leading to felony conviction where thirteen to fifteen percent of the juveniles charged with murder or attempted murder, a sex crime, or assault were re-arrested and convicted of a felony compared to nearly a quarter (23%) percent of those initially charged with robbery and 30 percent of those initially charged with another JO offense. Re-arrest rates varied less by borough of arrest (Exhibit 11). The range in rates of rearrests was only eleven percentage points for any re-arrest after disposition and was only six percentage points across the four largest boroughs. Felony-level re-arrest rates varied by only eight percentage points across the boroughs and rates of re-arrest leading to a conviction varied by only nine percentage points. The twelve point spread in rates of re-arrest resulting in a felony-level conviction was the largest across the boroughs. Seventeen percent of juveniles initially arraigned in Staten Island and 18 percent of those arraigned in the Bronx were rearrested for an offense that resulted in a felony conviction, compared to 20 percent of those in Queens, 24 percent of those in Brooklyn and 29 percent in Manhattan. Re-arrest rates varied widely by the pretrial release status of the juveniles in the initial JO case. The small number of juveniles (92) who posted bail at arraignment showed the lowest rates of re-arrest for all four measures (except for the extremely low rate of re-arrest leading to felony-level conviction among the 27 juveniles for whom no pretrial release status was set because their JO case was disposed at the arraignment appearance). Those who were not released prior to disposition had the highest or second highest rates of re-arrest. The largest category, juveniles who were released on their own recognizance at arraignment, show rearrest rates below the average for all cases, while those who were released after arraignment on recognizance showed re-arrest rates above the average for all cases.

EXHIBIT 11: RE-ARREST BY JO CHARGE, BOROUGH AND RELEASE STATUS

ALL CASES

N of Cases At Risk 2,841

Any Re-Arrest

Any Re-Arrest Felony Leading To Re-Arrest Conviction

Any Re-Arrest Leading To Felony Conviction

JO CHARGE Assault Murder or Attempted Sex Crime Robbery Other

180 92 57 2,378 134

69 60 60 76 78

42 39 32 51 50

45 35 37 54 58

15 13 14 23 30

BOROUGH OF JO ARREST Brooklyn Manhattan Queens Staten Island Bronx

1,019 515 468 70 769

76 78 75 83 72

52 53 49 54 46

54 57 48 50 51

24 29 20 17 18

RELEASE STATUS No Release ROR at Arraignment $M at Arraignment ROR after Arraignment $M after Arraignment Disposed at Arraignment

474 1,603 92 261 384 27

81 73 70 80 76 89

59 46 35 57 52 52

59 50 40 57 57 44

30 19 12 33 22 7

G. Re-Arrest Post Disposition by the Outcome of the Initial JO Case Re-arrest rates generally varied little by the outcome of the initial JO case (Exhibit 12), although there are some exceptions and the multivariate models in Section IV will offer a more nuanced picture. The rates of re-arrest hovered within two percentage points of the 75 percent rate across outcomes, except that the juveniles who were not at risk after conviction until they were sentenced or even served some time (up to one year) showed a slightly higher re-arrest rate (79%). The felony re-arrest rates were all within three percentage points of the 50 percent rate across outcomes, again except for the much higher felony re-arrest rate (61%) among those who were at risk only after sentencing. The rate of re-arrest leading to a conviction was at or below the 53 percent rate across outcomes for all of the outcome categories except conviction. However, the 57 percent rate of re-arrest leading to conviction after the juvenile 20

was convicted for the initial JO case reflects the higher rate (61%) of re-arrest among those who were at risk only after sentencing. Finally, the rate of re-arrest that resulted in a felony-level conviction ranged from a low of 20 percent among juveniles whose initial JO case resulted in dismissal or acquittal and those who were convicted but at risk prior to sentencing, to a high of 27 percent among the juveniles whose initial JO case was transferred to the Family Court from the Supreme Court. EXHIBIT 12: RE-ARREST POST DISPOSITION BY OUTCOME OF THE INITIAL JO CASE

ALL CASES CASE OUTCOME

Criminal Court: Dismissed Transferred to Family Court Supreme Court: Convicted: At Risk Before Sentencing At Risk After Sentencing Dismissed or Acquitted Transferred to Family Court

N of Cases At Risk 2,841

Any Re-Arrest Any Re-Arrest Leading To Any Felony Leading To Felony Re-Arrest Re-Arrest Conviction Conviction %

764 455

76 75

51 49

50 47

25 22

1,167

261 194

982 185

74 77

74 79

47 51

48 61

50 53

56 61

20 27

20 24

IV. MULTIVARIATE ANALYSES In this section, we present multivariate analyses of the factors that are associated with post disposition re-arrest among juveniles processed in the adult courts in NYC in an attempt to measure the independent effect of the type of disposition on the probability and timing of rearrest over time. The rationale for undertaking multivariate analyses is that, unlike the descriptive comparisons presented earlier in this report, multivariate analyses yield information about the impact on recidivism of all of the variables included in the analyses. The multivariate analyses are performed using Cox proportional hazard regression (see Appendix B: Statistical Procedures). This type of survival analysis examines how variables affect the rate of failure over time. In this research, “failure” is the hazard of re-arrest after disposition, and juveniles who are not re-arrested are “survivors.” The hazard of re-arrest is a measure that reflects both the fact of recidivism as well as how quickly the re-arrest took place. Separate models examine the time to the first re-arrest, to the first felony re-arrest, to the first re-arrest leading to a conviction, and to the first re-arrest leading to a felony conviction. Cox modeling does not require that all of the cases have an equal time at risk for re-arrest. All of the cases that were ever at risk for re-arrest can be included and contribute to the analysis. To identify factors associated with re-arrest post disposition, a series of Cox regressions was performed in which many possible explanatory variables were included as potential predictors. None of the community ties factors (length of residence, living with parents or legal guardian, school attendance, below grade-level status, or expecting someone at arraignment) had a significant effect. Five variables were statistically significant predictors in all four models: gender, prior arrests in the CJA database, the arrest charge for the initial JO case, ethnicity (coded dichotomously as black versus other), and the pretrial release status during the initial JO case. One of the arrest boroughs was a significant predictor in each of the models but the specific borough with the strongest effect was not the same in each model. Finally, we added the outcome of the initial JO case to determine if the case outcome was a predictor after taking into account the effects of the other variables. The outcome of the JO case was a statistically significant predictor in two of the models, though the effect of the disposition was not the same in those models. In fact, the effect was not even in the same direction.

The results of the Cox regression models predicting the hazard of the first re-arrest, the first felony re-arrest, the first re-arrest that resulted in conviction and the first re-arrest that resulted in a felony-level conviction are summarized in Exhibit 13. The models include rearrests that occurred more than seven years after the disposition of the initial JO case. A. Multivariate Findings All of the models are very strong and at the highest levels of statistical significance based on the overall chi-square statistic (which tests the hypothesis that all regression coefficients for the variables in the model are identically zero), which was statistically significant at p ≤ .001 in all four models. Indicator contrasts were used for all of the predictor variables in all of the models so that the hazard of re-arrest for juveniles in a particular category could be compared to the hazard for the juveniles in the selected reference category. Exhibit 13 displays the hazard ratio for each value of each predictor. The hazard ratio expresses the relative odds of a juvenile in the group being re-arrested more quickly than a juvenile in the reference group, which is assigned a hazard ratio of 1.000. For example, the hazard ratio for re-arrest leading to felony-level conviction (the fourth model in Exhibit 13) for females is .259, indicating that the odds are nearly 4:1 that female juveniles will be re-arrested for a charge leading to a felony conviction more slowly following disposition than their male counterparts. (The .259 hazard ratio for females means that odds of a female being arrested sooner than a male are about 1:4. The odds of being arrested later are the reverse.) The models are very similar to one another. The “gender” and “prior arrests” variables are statistically significant at the highest level in all four models with females at a reduced hazard of re-arrest compared to males, and juveniles with prior arrests in the CJA database at a greater hazard of re-arrest than those with no prior arrests. Ethnicity was also a significant predictor in all of the models but was not as quite as strong a predictor of a re-arrest leading to conviction. The charge in the initial JO case was a statistically significant predictor in all four models though the effect was weaker when predicting re-arrest leading to a felony-level conviction. Juveniles who were charged with assault or with a sex crime were at a lower hazard of re-arrest in all four models than were juveniles who faced a robbery charge in the initial JO case, and the hazard of re-arrest was also lower for juveniles charged with murder or 24

EXHIBIT 13: COX PROPORTIONAL HAZARD REGRESSION MODELS OF TIME TO FIRST RE-ARREST POST DISPOSITION (N=2,841)

Male 2,451 Female 390

Any First Re-Arrest Hazard B Ratio *** ref 1.000 -.534 .586***

First Re-Arrest First Re-Arrest First Felony Leading To Leading To Felony Re-Arrest Conviction Conviction Hazard Hazard Hazard B Ratio B Ratio B Ratio *** *** *** ref 1.000 ref 1.000 ref 1.000 -.775 .461*** -.827 .259*** .437*** -1.352

PRIOR ARRESTS No Prior Arrests 1,293 Prior Arrests 1,548 ETHNICITY Black 1,914 Other 927

*** ref 1.000 .433 1.541*** *** ref 1.000 -.176 .839***

*** *** ref 1.000 ref 1.000 .574 1.755*** .499 1.647*** *** ** ref 1.000 ref 1.000 -.226 .798*** -.160 .852**

CHARGE IN INITIAL JO CASE Robbery 2,378 Assault 180 Murder or Att. Murder 92 Sex Crimes 57 Weapon 92 Other 42 RELEASE STATUS Released at Arraignment 1,695 861 Released After Arraignment 285 Not Released

*** ref 1.000 -.279 .756** -.442 .643*** -.539 .583** ns ns * ref 1.000 .154 1.166** ns

PREDICTOR VARIABLES: N of Cases GENDER

BOROUGH OF ARREST Brooklyn 1,020 Manhattan 515 Queens 466 Staten Island 70 Bronx 770 CASE OUTCOME Convicted 1,167

ref 1.000 ns -.152 .859* ns ns ref 1.000

Dismissed 1,025 Transferred to Family Court RE-ARREST RATE

* p ≤ .05

** p ≤ .005

649 2,841

*** p ≤ .001

75%

*** 1.000 .750* ns .452*** ns ns *** 1.000 1.253*** 1.352***

*** ref 1.000 -.246 .782* -.584 .558*** -.617 .539** ns ns ** ref 1.000 .160 1.173* .220 1.246*

ns ref 1.000 ns -.176 .839* ns ns ns ref 1.000

*** ref 1.000 ns -.332 .717*** ns ns *** ref 1.000

-.288 -.794

ref .226 .301

-.290

.749***

-.290

.749***

50%

ns = not statistically significant

53%

*** ref 1.000 .645 1.905*** *** ref 1.000 -.294 .746*** * ref 1.000 -.506 .603* -.662 .516* -.775 .461* ns ns *** ref 1.000 .322 1.380*** .447 1.563*** ref 1.000 .221 1.248* ns ns ns ref 1.000

ns .259 1.296* 23%

NOTES: A positive coefficient means that the hazard of re-arrest is higher for juveniles with that value. “Ref” denotes the reference category. Juveniles in the other categories are compared to those in the reference category.

attempted murder in all but the first-felony-re-arrest model. The release status variable compared juveniles released post-arraignment and those with no pretrial release to juveniles who were released at arraignment in the initial JO case. Those released postarraignment were at a significantly greater hazard of re-arrest in all four models, and those who were not released were at a significantly greater hazard of re-arrest in all but the first re-arrest model. The borough-of-arrest variable was statistically significant in only 2 of the models. Although juveniles initially arraigned in Queens had a significantly lower hazard of first re-arrest or first felony re-arrest, compared to juveniles arraigned in Brooklyn, overall the borough of arrest was not a statistically significant predictor in either model. The hazard ratio for those initially arraigned in Queens compared to those in Brooklyn was even lower in the model predicting the first re-arrest leading to conviction. In the fourth model, which predicts the hazard of re-arrest leading to a felony conviction, the juveniles arraigned in Manhattan were at a significantly greater hazard of re-arrest than Brooklyn youths. The Cox multivariate survival analysis indicates that the outcome of the initial JO case is not a statistically significant predictor of the hazard of either the first re-arrest or the first felony re-arrest, after taking into account the effects of the other factors that are in the rearrest models. However, case outcome was a strong predictor of the first re-arrest leading to conviction and a significant predictor of the first re-arrest leading to a felony conviction, although, as mentioned above, the effect was in the opposite direction. For the purpose of the multivariate analysis, case outcome was categorized as “dismissed,” “transferred to the Family Court,” or “convicted,” combining the Criminal and Supreme Court dismissals and combining the Criminal and Supreme Court transfers to the Family Court.5 The findings indicate that juveniles whose cases were dismissed or transferred to the Family Court were at an equally reduced hazard of re-arrest leading to conviction compared to their counterparts who were convicted in their initial JO case (both with a hazard ratio of .749). In contrast, juveniles whose

Cox models run with case outcome coded to reflect both the disposition and the court of disposition showed strong effects in the model predicting the first re-arrest leading to conviction but weaker effects in the model predicting the first re-arrest leading to a felony conviction. However, the separate effects of the disposition in the lower and upper court were consistently in the same direction. 5

cases were transferred to Family Court were at a greater hazard of re-arrest leading to a felonylevel conviction (1.296) than were the convicted juveniles. Figures 1â&#x20AC;&#x2019;3 illustrate survival curves generated by the Cox regression models. The survival function displayed in Figure 1 graphically illustrates the probability of the juveniles in the study surviving without failure (re-arrest) by the number of years at risk, separately by gender. The impact of all of the variables (other than gender) entered in the model is reflected in the survival curves. At time equal to zero, a full 100 percent of both the male and female juveniles survived, that is, had yet to be re-arrested for a case that resulted in a conviction. If there were no re-arrests, the line would remain straight across the years at risk. As time at risk elapsed, the predicted probability of surviving without re-arrest declined to 56 percent for females (the upper line) and declined more steeply to 26 percent for males (the lower line) by the end of the study period. FIGURE 1: Probability of Surviving Without a Re-Arrest Leading to Conviction, By Gender GENDER

100%

____Female ____Male 80%

60%

56%

40%

26% 20%

Figure 2 illustrates the effect of the outcome of the initial JO case on the hazard of rearrest that resulted in a conviction, after taking into account the effects of the other variables included in the model. In this graph, the lower survival curve depicts the reduced survival over time without re-arrest leading to conviction among the juveniles who had already been convicted. The upper line is for both those whose cases were dismissed and those whose cases were transferred to the Family Court; they have the same curve because as we have seen, the hazard ratios for these two groups were identical. These juveniles have a higher level of survival over time without re-arrest leading to conviction. FIGURE 2: Probability of Surviving Without a Re-Arrest Leading to Conviction, By JO Case Outcome JO CASE OUTCOME

100%

____Dismissed or Transferred to Family Court ____Convicted 80%

60%

40%

35% 25%

20%

Figure 3 also compares predicted survival by case outcome, this time based on the hazard of re-arrest leading to felony-level conviction. In this graph, the curve for the convicted juveniles is the highest curve, depicting the greater probability of these youths surviving without re-conviction at the felony level. The middle line is for the juveniles whose cases were 28

dismissed, and the bottom line depicts the lowest survival probability for the juveniles whose cases were transferred to the Family Court.

FIGURE 3: Probability of Surviving Without a Re-Arrest Leading to Felony Conviction, By JO Case Outcome JO CASE OUTCOME

100%

____Convicted ____Dismissed ____Transferred to Family Court 90%

80%

75% 70%

71% 69%

60%

B. Discussion of Multivariate Findings The strongest predictors of the hazard of recidivism after the disposition of the initial JO case were gender, prior arrests, ethnicity, the charge in the initial JO case, and pretrial release status, and each was a statistically significant predictor for each of the four measures of recidivism examined in this research. Borough differences were not consistent, as hazard ratios were lower in Queens (compared to Brooklyn) in the first three models and higher in Manhattan in the fourth model. Both policing and prosecutorial practices vary from borough to borough, so differences in the hazard of re-arrest and reconviction are not surprising, even though we cannot point to specific borough practices that would explain these results. 29

The findings concerning the outcome of the initial JO case are challenging to interpret. It seems possible that youths convicted in JO cases are not at a greater hazard of re-arrest or felony re-arrest than were those who were not convicted because re-arrest is so common among juveniles who are processed in adult court on JO charges. The greater hazard of rearrest leading to conviction among those already convicted may reflect the ways in which those who are convicted continue criminal behavior patterns, patterns that may have already contributed to their initial conviction. The finding that juveniles who were convicted were both more likely to be re-arrested and re-convicted, and more likely to be re-arrested quickly may indicate that conviction in a JO case reflects serious criminal behaviors that persist beyond the initial conviction. Re-arrest leading to re-conviction may also reflect policing policies such as â&#x20AC;&#x153;stop and frisk.â&#x20AC;? It is more puzzling to interpret why juveniles whose cases were transferred to the Family Court were at a greater hazard of re-arrest in the felony conviction model than were those who were convicted. We know that some cases are transferred to the Family Court because the charge was reduced in the adult court to one that is not a JO offense, yet other cases transferred to the Family Court retain the JO charge. We know that some youths are detained for a period of time during case processing in the Family Court and that these youths may be detained in secure or non-secure facilities. We donâ&#x20AC;&#x2122;t know enough about what happened to the cases in the Family Court in terms of case processing and outcomes to hypothesize the meaning of the greater hazard of felony re-arrest and re-conviction for JO cases transferred there. Unfortunately, this research encountered unanticipated problems with the quality of the data on time at risk for re-arrest, problems that potentially affect the strength of the JO case outcome variable in the models. As discussed earlier in this report, we excluded juveniles who were not at risk prior to sentencing and who were sentenced to a maximum of more than one year because we could not reliably estimate how long they served in prison. In addition, we also lacked data on detention ordered by the Family Court and on any sentence that might be ordered by the Family Court. The finding that the juveniles whose initial JO cases were transferred to the Family Court were at a greater hazard of re-arrest leading to a felony conviction is strengthened by the overestimated time at risk for re-arrest. However, the finding 30

that those transferred to the Family Court, like those whose cases were dismissed, were at reduced hazard of re-arrest leading to any conviction is weakened by overestimating time at risk. Furthermore, this research understates the actual rates of re-arrest that resulted in conviction and in felony conviction because there was not enough time for the most recent rearrests to reach conviction at the felony level. Nevertheless, when the recent cases were excluded from the Cox survival analyses the results in the multivariate analyses remained the same.

V. CONCLUSION Juveniles who were convicted in their initial JO case had the highest hazard of re-arrest leading to conviction. However, the hazard ratio for convicted youths was not significantly different from the hazard ratio for other youths in the models predicting the time to the first rearrest or the first felony re-arrest. On the other hand, juveniles whose cases were transferred to Family Court were more likely than the youths who were convicted to have a re-arrest leading to a felony-level conviction. We can only speculate why the re-arrest and felony rearrest models are so similar to one another and show no effect of the initial case outcome or why the role of the disposition is so different in the conviction and felony conviction models. Perhaps re-arrest is so common among youths already processed in the adult court for a serious felony that the initial disposition makes no difference and the distinction between any re-arrest and a felony-level re-arrest is not an important difference. On the other hand, while it is easy to understand the greater hazard of re-arrest leading to conviction among those already convicted, it is not clear why youths whose cases are transferred to the Family Court have a higher probability of re-arrest leading to felony conviction compared to those who were already convicted. We donâ&#x20AC;&#x2122;t know enough about processing and outcomes in the Family Court to understand why the youths whose cases were transferred to the Family Court were at a greater hazard of re-arrest leading to a felony conviction than were convicted juveniles. We have seen that the first re-arrest leading to a conviction was likely to be a conviction for a relatively minor offense at the violation level (Exhibit 7) and this is true for the youths who were convicted (data not shown): 57 percent were re-arrested for a charge that resulted in conviction and nearly two thirds of the re-arrests that resulted in conviction were convictions for violations (usually disorderly conduct). Perhaps the finding that conviction in the initial JO case is associated with higher hazard of re-arrest leading to a conviction, although statistically significant at the highest level, is not very important, since most of the first convictions after disposition were not for very serious offenses. The most important model in this study may be the one predicting time to the first rearrest leading to felony conviction. The juveniles transferred to the Family Court from either adult court showed a significantly higher hazard of re-arrest leading to felony conviction than did the juveniles who were convicted, and, as discussed above, the time at risk for Family Court 33

cases is overestimated, suggesting that this finding could be even stronger if the data were more accurate. The multivariate recidivism models may not be the most salient product of this research. Previous studies did not consider the adult-court outcomes of the re-arrests. This project found a high rate of re-arrest leading to conviction (53%) and, even more importantly, an extremely high rate of re-arrest leading to conviction at the felony level (23%). Furthermore, these measures of recidivism are underestimated since so many of the juveniles in recent JO cases had less than the maximum number of years at risk. As shown in Exhibit 8, it took a median of 640 days (nearly 21 months) to the first re-arrest that resulted in a felony conviction. Exhibit 5 displays the re-arrest rates by time at risk: eight percent who were at risk more than one year were already re-arrested for an arrest that resulted in a felony conviction. At two years at risk, 16 percent were re-arrested, and the rate grew to 24 percent re-arrested of those still at risk at three years and 32 percent for those still at risk four years after the disposition of their initial JO case. It is likely that more of the juveniles will be re-arrested and that more rearrests will result in convictions as more time elapses.

VI. POLICY IMPLICATIONS The findings of this research suggest three kinds of policy implications. The first is the need to learn more about case processing in the Family Court and why the Family Court cases showed such a high hazard of the most serious recidivism, re-arrest leading to a felony conviction. The second concerns the need to reconsider the use of re-arrest as a criterion of success or failure in criminal justice program evaluations. Many re-arrests do not result in convictions and those that do result in conviction are often convictions for violations or infractions, convictions that are not â&#x20AC;&#x153;criminalâ&#x20AC;? convictions. Furthermore, some re-arrests reflect policing policies as much as youth misbehavior. The third kind of policy implication is one that has persisted across the findings of all of the research conducted at CJA on recidivism among youths processed as juvenile offenders. Specifically, these youths who are already charged with serious violent offenses under the age of 16 are at a high risk for re-arrest and for felony-level re-arrest. While re-arrest may reflect over-policing, the current research documents extremely high rates of re-arrest that result in convictions and especially in felony convictions, and conviction at the felony level reflects serious criminal behavior. These youths warrant concerted attention from the NYC criminal justice and child welfare communities to address their recidivism.

APPENDIX A JUVENILE OFFENSES Offense Aggravated sexual abuse in the first degree Arson in the first degree Arson in the second degree Assault in the first degree Burglary in the first degree Burglary in the second degree Criminal sexual act* in the first degree Kidnapping in the first degree Attempted kidnapping in the first degree Possession of a weapon in the second degree Manslaughter in the first degree Murder in the second degree Attempted murder in the second degree Rape in the first degree Robbery in the first degree Robbery in the second degree

Penal Law (Subsection) 130.70 150.20 150.15 120.10 (1) (2) 140.30 140.25 (1) 130.50 (1) (2) 135.25 110/135.25 265.03** 125.20 125.25 (1) (2) 125.25 (3)*** 110/125.25 130.35 (1) (2) 160.15 160.10 (2)

Felony Class B A B B B C B A B C B A A B B B C

Defendant Age 14, 15 14, 15 14, 15 14, 15 14, 15 14, 15 14, 15 14, 15 14, 15 14, 15 14, 15 13, 14, 15 14, 15 14, 15 14, 15 14, 15 14, 15

* Changed from â&#x20AC;&#x153;Sodomyâ&#x20AC;? in November 2003 as part of the NYS Sexual Assault Reform Act. ** Added in November 1998, but only where the weapon is possessed on school grounds. PL 265.02 (4), possession of a weapon in the third degree, a D felony, was repealed effective November, 2006, and is now replaced by 265.03 (3). ***But only where the underlying crime is also a JO offense.

APPENDIX B STATISTICAL PROCEDURES Mary T. Phillips, Ph.D. The multivariate statistical procedure used in this report is Cox regression 6, which is a type of proportional hazards analysis used to investigate the effect of several variables upon the time a specified event takes to happen. This type of analysis is frequently used in medical research in which the outcome is death, where the models provide estimates of the effects of treatment and other factors (e.g., demographics) on how long the patient is predicted to survive before death. In the analyses in this report, the outcome is re-arrest and survival refers to the time to the first re-arrest following the disposition of the sample JO case. The regression models were computed using SPSS 7 to produce the statistics discussed below. In this research, four Cox regression models were computed to analyze the time to the first re-arrest following the disposition of the original arrest. “Re-arrest” was defined in four different ways: the first re-arrest of any kind, the first felony re-arrest, the first re-arrest that led to a conviction, and the first re-arrest that led to a felony conviction. Cox regression, rather than the more frequently used logistic regression, was used in this research because the outcome of interest was not only whether a juvenile was re-arrested (for which logistic regression would be appropriate), but how soon (not addressed by logistic regression). Most of the juveniles in the sample were re-arrested, so it was more interesting to investigate predictors associated with time to re-arrest than simply the fact of re-arrest. The outcome in Cox regression combines the two measures by estimating the relative speed with which the event occurred. It does not calculate an absolute length of time, but rather indicates which conditions are associated with an earlier or later outcome. In medical studies, a Cox model can answer the question of whether the average patient will die more quickly with or without the treatment, controlling for other factors. In the current research, we use Cox models to investigate the factors that are associated with the relative speed of re-arrest after disposition of the sample case. Another reason for selecting Cox regression for these analyses is that juveniles in the sample had varying amounts of time during which they were at risk for re-arrest: their cases did not all reach disposition at the same time, and some did not become at risk until after serving a sentence. This is not a problem in this research because Cox regression is based on hazard rates, which take each individual’s time at risk into account. Finally, Cox regression was selected for these analyses because of its ability to account for censored data, a type of missing data problem common to survival analyses. The study ended without re-arrest for some who may have been re-arrested later—we do not know if they were re-arrested after the observation period ended. These subjects are said to be censored and are counted as surviving for the duration of the study. Unlike ordinary regression models, Cox regression correctly uses information from both censored and uncensored observations. 6

D. R. Cox, 1972. “Regression Models and Life-Tables,” Journal of the Royal Statistical Society. Series B (Methodological), Vol. 34, Issue 2: 187-220. 7 IBM SPSS® Statistics Version 22.0.

Hazard Rate The hazard (or incidence) rate is the number of events that occurred divided by the person-time at risk, which is the sum of each individualâ&#x20AC;&#x2122;s time at risk. In this study, the hazard rate is the number of juveniles who had a post-disposition re-arrest during the study period divided by the sum of each juvenileâ&#x20AC;&#x2122;s time at risk. For example, if five juveniles were at risk for one, two, three, four, and five years, respectively, person-years at risk for the sample of five would equal 15. If three of them were re-arrested while at risk, the hazard rate would be 3/15 = 0.2, or 20%. This is the rate at which individuals were re-arrested. The hazard rate would be much higher if all three re-arrests had occurred after only one year at risk: 3/5=.6, or 60%. In epidemiology, hazard rates are interpreted as the number of new cases (of cancer, for example) per population at risk per unit time. If the outcome is death, then the hazard rate is the mortality rate: the number of deaths per population at risk per unit time. Hazard should not be confused with prevalence. The prevalence rate for this hypothetical sample of five would be 3/5 = .6, or 60%. (Note that prevalence is equivalent to hazard, or incidence, if time at risk is uniformly one unit of time for every case in the sample.) Prevalence rates measure the proportion of individuals in the population who were re-arrested by the end of a given time period, whereas incidence measures the rate of re-arrest per unit of time (the rate of new cases of cancer, in the epidemiological example). The difference lies with the denominator: in calculating hazard (incidence), the denominator is person-years at risk; in calculating prevalence, the denominator is the number of individuals in the sample, with no regard for time at risk. Statistics shown in the models presented in Exhibit 13 Î˛ Coefficient The coefficients in a Cox regression relate to hazard: the coefficients are the loghazard ratios (the log of the hazard ratio). Exponentiating them therefore produces hazard ratios, which are more easily interpreted. In epidemiology, a positive coefficient for a treatment variable indicates a worse prognosis and a negative coefficient indicates a better prognosis associated with treatment, compared to no treatment. In the current research, a positive coefficient indicates a greater hazard (higher incidence) of re-arrest associated with a predictor variable. Another way of thinking about it is that a positive coefficient indicates a shorter time to re-arrest associated with the predictor variable, and a negative coefficient indicates a longer time to re-arrest. For example, the positive coefficient for prior arrest in all four models in Exhibit 13 indicates that juveniles with prior arrests have a greater hazard of re-arrest than those with no priors, meaning that they are likely to be re-arrested sooner. The negative coefficient for females indicates a lower hazard, and a longer time to re-arrest, compared to males. Hazard Ratio The hazard ratio is directly related to the coefficient: it is the exponent of the coefficient. This is the statistic most useful in interpreting regression models because it has an intuitive meaning. When the predictor variable is continuous, the hazard ratio can be interpreted as the predicted change in the hazard for a unit increase in the predictor. When 38

the hazard ratio is a dichotomy, it predicts the change in the hazard from one value of the predictor to the other (e.g., female to male). The hazard ratio for each value in the models compares the hazard of that group to the hazard of a reference group. Cox regression provides for a variety of types of contrast coding, to define the comparison for analytical purposes. In this research, indicator contrast coding was used, which allows the researcher to specify a value as the reference category, against which all other values of the variable are compared. For example, males were selected as the reference category for the gender variable, meaning that the hazard ratio for females was compared to males. For the charge variable, robbery was selected as the reference category, so that hazard ratios for juveniles with other charges were compared to juveniles charged with robbery. By definition, the hazard ratio for the reference group is 1.000, and hazard ratios for other categories of the predictor are expressed relative to 1.000. Positive coefficients produce hazard ratios greater than 1.000 and indicate a greater hazard associated with the predictor. Negative coefficients produce hazard ratios less than 1.000 and indicate a reduced hazard. The hazard ratio for females in each model is less than 1.000, indicating the low odds of a female being re-arrested earlier than a comparable male. Hazard ratios are often misinterpreted. A hazard ratio of 2.000 for Group A does not mean that a juvenile in Group A will be re-arrested twice as quickly as a person in Group B (if Group B is the reference group). It does mean that juveniles in Group A are likely to be rearrested more quickly than juveniles in Group B, but only in this specific sense: a juvenile in Group A who has not yet been re-arrested by a certain time has twice the chance of being rearrested at the next point in time compared to someone in Group B. In the model predicting time to the first re-arrest that led to a conviction, the hazard ratio for females was 0.432. This means that the hazard was reduced by 56.8% (1.00(1.00x0.432) = 0.568) for females, relative to males. The hazard for females was reduced even further when the prediction was for the first re-arrest that led to a felony conviction: the hazard ratio for females in this model was .257, which translates to a 74.3% (1.00-(1.00-.257)) reduction in hazard for females, compared to males. An example of a positive hazard ratio was found in all four models for juveniles with prior arrests. In the model predicting time to the first re-arrest that led to a felony conviction, the hazard ratio for prior arrest was 1.905. This means that juveniles with prior arrests had an increase in their hazard for re-arrest; their hazard was almost double the hazard for those with no prior arrest. Assumption of proportionality. Cox regression was described at the beginning of this Appendix as a type of proportional hazards analysis. This means that it is based on the assumption that the hazard ratio remains the same over time, from the beginning to the end of the study period. Statistical Significance Statistical significance is a measure of the likelihood that the relationship between the predictor variable and the dependent variable could have occurred merely by chance. The level of statistical significance of each hazard ratio included in the model is indicated by asterisks, 39

from one—the least stringent level of statistical significance (p ≤.05)—to three—the most stringent level (p ≤.001). Factors that did not have a statistically significant relationship with the dependent variable are indicated by “ns” for “not significant,” and their coefficients and hazard ratios were omitted from Exhibit 13. It is standard practice to consider a relationship to be statistically significant if the likelihood that the result is merely a chance occurrence is equal to or less than 5% (p ≤.05). An even smaller likelihood — for example, equal to or less than 1% (p ≤.01) — is better. At the most stringent level of significance, p ≤.001, the likelihood of the result occurring by chance is equal to or less than 1 in 1,000. Results that are not statistically significant have an unacceptably high probability (greater than 5%) of being merely the result of sampling error, which means that they may not be representative of the larger population. Both the magnitude of the effect and the size of the sample contribute to the level of statistical significance. The sample of Juvenile Offenders used in the Cox regression analyses in this research was large enough to detect moderately strong effects (2,841), but some of the cell sizes were quite small. For example, only 70 juvenile offenders were brought to court in Staten Island, so only a very strong effect (a large difference in the hazard rate for Staten Island compared to Brooklyn) would attain statistical significance; in fact, Staten Island was not significant in any of the models. On the other hand, there were even fewer juveniles in the sample charged with a sex crime (57), but the hazard of re-arrest was so much lower for them compared to the robbery defendants that being initially charged with a sex crime was statistically significant in all four models. Substantive significance should not be confused with statistical significance, which means only that the effect is real, not that it is important. The importance of a weak — albeit statistically significant — effect may be trivial. This is a problem primarily in very large samples, where even very weak effects are often statistically significant. Survival Curve Survival is the inverse of hazard: as the predicted hazard increases, survival time decreases. The survival curves take into account all of the variables entered in the regression models. In this report, we present three graphs showing survival curves generated by the Cox regression models. The first graph compares the survival curves for males and females produced by the model predicting the first re-arrest that led to conviction. The other two graphs compare survival curves for three JO case outcomes that were produced by two different models. Any point on the survival curve shows the predicted probability that the "average" juvenile in the group will survive without re-arrest past that time. The curve starts at the left with 100% survival at time zero for all groups, and the lines diverge for different groups according to a function of their hazard rates. For example, the survival curve for females is much higher than for males because of females’ higher probability of survival without re-arrest. By the end of the tracking period, the average female who had survived that long (between 7 and 8 years) without a re-arrest leading to conviction had a 56% chance of surviving even longer, compared to a 26% probability for males.