Title: Effects of Judges' Sentencing Decisions on Criminal Careers.
Series: Research in Brief
Author: Don M. Gottfredson
Published: National Institute of Justice, November 1999
Subject: Judges, sentencing
17 pages
40,000 bytes

-------------------------------

Figures, charts, forms, and tables are not included in this ASCII plain-text
file. To view this document in its entirety, download the Adobe Acrobat
graphic file available from this Web site or order a print copy from NCJRS
at 800-851-3420 (877-712-9279 for TTY users).

-------------------------------

U.S. Department of Justice
Office of Justice Programs
National Institute of Justice

National Institute of Justice
Research in Brief

Jeremy Travis, Director
November 1999

Effects of Judges' Sentencing Decisions on Criminal Careers

By Don M. Gottfredson

-------------------------------

Issues and Findings

Discussed in this Brief: This study sought to determine the degree to
which judicial sentencing decisions affect subsequent criminal careers. It
examined the criminal careers of 962 felony offenders in Essex County,
New Jersey, sentenced in 1976 and 1977 variously to confinement and
noncustodial programs. The 18 participating judges exercised considerable
discretion in making sentencing decisions. The data collected included
judicial perceptions, the judges' predictions of the offenders' future
criminal behavior, the judges' sentencing purposes, offender backgrounds,
execution of sentences, and offenders' arrests and charges during the 20
years after sentencing. Also measured were the judges' selection of
different sanctions, the validity of subjective and objective predictions of
future criminal behavior (risks), and the offenders' time in the community
(free of the incapacitating effects of jail or prison).

Key issues: Rigorous tests of sentencing policy changes are rare because
serious obstacles impede assessments of the crime control effects of such
policies. The results of controlled experiments involving different
punishments of equivalent groups of offenders cannot be compared
because such studies, rarely believed to be feasible, are not done. 

Other obstacles to sentencing research include: judges' reluctance to
provide detailed records of their reasons for choosing particular sentences;
fixed or mandatory sentencing laws that limit sentencing alternatives
available to judges; and the demands of careful, long-term followup
analysis of sentencing effects on offenders' subsequent criminal careers. 

This study included careful statistical controls for biasing factors (judicial
selection of sentences, offender risk of reoffense, time in confinement, and
time in the community), detailed recording by judges, availability of broad
sentencing discretion, and study of offenders' criminal careers over the 20
years after sentencing.

Key findings: The judges' subjective risk assessments of offenders'
likelihood of recidivism, although only modestly valid, had a substantial
influence on their sentencing choices. More formal, empirically derived
methods provide better measures of the risk of reoffending.

Available sentencing choices had little effect--other than that of
incapacitation--on recidivism as measured by new arrests and charges:

--Whether the offender was confined or given noncustodial sanctions
made no difference.

--Where the offender was confined (i.e., in jail, a youth facility, or prison)
made little difference.

--The length of the offender's maximum imposed sentence made no
difference.

--The length of time the offender actually was confined made little 
difference.

--A "split" sentence of jail and probation made no difference.

--Fines or restitution made no difference.

Some important lessons for current policy are clear from this study; it 
offers little support, aside from incapacitation, for increased use of
confinement, emphasis on longer terms, or more acceptance of specific
deterrence as a crime control strategy. 

Target audience: Judges, legislators, prosecutors, defense attorneys,
sentencing guidelines commissions, corrections administrators,
researchers, and others concerned with sentencing policy.

-------------------------------

Despite the absence of adequate data on the effects of incarceration and
other sanctions on convicted offenders' subsequent recidivism, sentencing
trends over the past two decades have moved toward increased
determinacy, greater use of mandatory sentences with longer terms, and
reduced judicial discretion. A better understanding of how different felony
sanctions impact the future behavior of offenders is needed to provide a
basis for evaluating the efficacy of current sentencing policies. 

To this end, the National Institute of Justice sponsored research that
examined the crime control effects of sentences, over a 20-year period, on
962 felony offenders sentenced in 1976 or 1977 in Essex County, New
Jersey. This Research in Brief summarizes findings of the study. 

Research issues

Three main factors bias comparisons of outcomes for differently punished
groups of offenders: selection of sentence, risk of reoffense, and time in
confinement. Fair comparisons of sentencing choices must take these
sources of variation in outcomes into account:

--Selection--groups given different sentences are not equivalent because
judges make their sentencing decisions based on case and offender 
characteristics.

--Risk--differently sentenced groups may represent different risks of future
offending at the time their sentences are imposed (usually called the a
priori risk). 

--Incapacitation--offenders are confined for differing periods of time as a
result of the sentence imposed on them for the current offense and crimes
they commit later.

These factors must be considered when assessing the effects of
punishment, treatment, specific deterrence, and incapacitation. The results
of controlled studies of different punishments cannot be compared because
experiments with different sentences for equivalent groups of offenders--
for example, by random assignment of sentences--rarely are believed to be
feasible and are not often done.

Assessment of sentencing policies ordinarily is hindered by several
problems that cast doubt on the relevance of findings about sentencing
choices:

--The most rigorous research methods, in which offenders would be
randomly assigned to various sentencing alternatives, usually are
precluded.

--Judges rarely are willing to record in detail their views on the seriousness
of the offense or the offender's prior record, their assessment of the risk
that the offender will commit future crimes, their ratings of other case
characteristics, or their specific reasons for selecting the sentences they
impose.

--As changes in sentencing policy have taken place over the past two
decades, judicial discretion often has been reduced by greater determinacy
in sentencing structures or stricter mandatory sentencing laws or
guidelines. As a result, judges have not had as much choice in imposing
sentences as they did previously.

--Perhaps most important, researchers must carefully trace the offenders'
records through any periods of confinement resulting from the sentence
(and from later crimes) to determine the sentence as actually executed and
then observe new arrests, convictions, and sentences after their release into
the community. Sufficient time must elapse after sentencing to enable
most offenders to complete any terms of confinement and then to
experience enough time free in the community to permit a fair assessment
of the effects of the sentences on their criminal careers.

These problems were mitigated in this study by using careful statistical
controls for biasing factors, detailed recording by the judges, the
availability of broad sentencing discretion, and the detailed study of
offenders' criminal careers during the 20 years after they were
sentenced.[1]

Study methods

Beginning in May 1976, 18 judges in the Essex County, New Jersey,
Court recorded their judgments when sentencing 962 felony offenders to
confinement and noncustodial programs. They continued to code the cases
they sentenced during the next year. Under New Jersey's then
indeterminate sentencing law, the judges were allowed much discretion in
their decisionmaking. Besides documenting the sentences they imposed,
the judges recorded their perceptions of the seriousness of the crime and
the offender's past criminal record and their assessment of the likelihood
that the offender would commit future crimes. 

The central issue of this study was the effects of different sanctions on the
offenders' subsequent criminal careers. Secondary issues concerned the
measurement of judicial selection of different sanctions, the validity of
subjective and objective predictions of risks for reoffending, and the
amount of time the offenders were in the community free of the
incapacitating effects of jail or prison. The effects of sentences on later
arrests and charges were assessed through statistical or
"quasi-experimental" methods that adjusted for the nonequivalence of
differently sentenced groups.

Data on the imposition and actual execution of sentences and
postconviction arrests, charges, sentences, and confinements over the next
20 years were examined. The followup data collection ended in February
1997. It relied on the New Jersey Department of Corrections
Offender-Based Correctional Information System and record files, the
U.S. Department of Justice Interstate Identification Index, the National
Crime Information Center Wanted Persons File, and the New Jersey
PROMIS/GAVEL Prosecutors Case Tracking System. Staff familiar with
each of these systems coded the data. No records after sentencing could be
found for 37 persons (4 percent) who were counted as having no known
subsequent arrests.[2]

Measures of sentence selection and the offender's a priori risk (risk as
estimated from data available at the time of sentencing) were developed
using multivariate statistical methods. Time variables were calculated
from recorded dates. Differences in later arrests and charges (at 5 years
and 20 years after sentencing) associated with different sentences were
analyzed with statistical controls for measures of the biasing factors.[3]
The variability in new arrests and charges was analyzed to assess sentence
effects, and survival without arrest over time was examined for differently
sentenced groups.

Crimes, demographics, results

The offenders had been convicted of crimes typically considered to be
felonies. These crimes were classified by legal definitions and groupings
based on a multidimensional approach to scaling offense seriousness.[4]
By legal offense classifications, 37 percent were convicted of violent
crimes, 24 percent of drug offenses, 23 percent of property crimes, 10
percent of weapons offenses, and 6 percent of other crimes. By the
behavioral classification based on judgments of seriousness, one in six was
convicted of an offense involving interpersonal confrontation or physical
violence, although property crimes were most common. Fifteen percent
were serious drug offenders whose crimes included the sale of drugs other
than marijuana.

On other measures, the study sample was fairly typical of sentenced
offender populations. The average age was 29. Most were black males;
one-fourth were white; one-eighth were women. About half had served
prior jail terms; 16 percent had been in prison; and 80 percent had prior
placements on probation. On average, they were first arrested at age 22,
had 2.4 prior arrests and convictions, and were currently charged with two
counts and convicted of one. One in 10 was reported to have used heroin
or barbiturates as a juvenile, and 29 percent had a recorded use of these
drugs in the 2 years prior to their arrest. One-third had histories of drug
offenses; nearly half had a record of property offenses; and one-third had a
history of violent crimes. One in 10 had been committed to a juvenile
facility.

Research results described the sentences as imposed (announced) and as
executed (carried out), the sentencing purposes of the judges, and later
arrests and charges. Next, the validity of the judges' predictions and the
measures of a priori risk and sentence selection needed for the analyses of
effects were considered, as described below.

Sentences

Most offenders were sentenced to incarceration with terms ranging from 1
month to life (typically 6, 12, or 18 months in jail or 5, 7, or 10 years in
prison). Sentences were suspended in whole or in part for many offenders,
who then usually were placed on probation. The main choices, in terms of
executed sentences, were a noncustodial sentence (42 percent), jail (29
percent), prison (19 percent), or a youth facility (10 percent). About 10
percent received "split" sentences of probation with jail. Others received
special conditions, including fines and restitution.

Sentencing purposes. Judges most often reported a crime control aim as
the main reason they imposed the sentences they did. Rehabilitation and
specific deterrence were prominent considerations. Other sentencing goals
included incapacitation and general deterrence. Retribution was the
primary purpose in about one in five cases. Judges typically had more than
one purpose. The distribution of ratings is shown in exhibit 1. The cases
were assigned sequentially according to the availability of the judge, and
all judges participated. It cannot be assumed that the judges received
entirely equivalent cases.

Offenses after sentencing

Thirty percent of the offenders were not arrested in the 20 years after
sentencing. More than half of all, however, were rearrested in the first 5
years after they were sentenced. The 962 offenders were arrested 5.3 times
on average in the 20 years after sentencing. The arrest rate for all offenders
participating in the study, not including the time spent incarcerated, was
0.28 arrests per year. The rate for offenders who were rearrested at least
once was 0.36. For those rearrested, the average time to a new arrest was
3.5 years; half of those with new arrests were rearrested within 2.2 years.

Charges classified by legal offense groups. In terms of the legal offense
classifications, in the 20 years after their initial sentencing, the offenders
were charged with 40 criminal homicides (murders or manslaughters), 455
robberies, 752 assaults, 928 burglaries, 18 rapes, 507 weapons offenses,
682 receiving stolen property offenses, and 16 kidnapings--to list some
charges out of 9,346 allegations of new crimes. The median number of
charges was 5, while the mean was 9.7.

Charges classified by behavioral offense groups. The offenders in this
sample were arrested 5,117 times during the 20 years after sentencing.
Most often, by far, they were charged with property crimes or nuisance
offenses, but they also were charged with a substantial number of serious
personal confrontation crimes and drug offenses.[5] The charges at the
first arrest after sentencing, classified into kinds of offenses as used in this
study, are shown in exhibit 2. The numbers and percentages are those for
the most serious charges among any made after the first arrest in the
followup period. The total numbers of offenses charged during the
followup period, grouped into these general classes, are shown in exhibit
3.

Judges' predictions

Judges' subjective predictions of whether the offenders would commit any
new property, violent, or any type of crimes were valid but modestly so.
Predictions of "any new arrest" accounted for about 6 percent of the
differences in new arrests. The validity of some judges' predictions
compared favorably with some empirically derived, formal prediction
methods. Yet, the predictions of 4 of the 18 judges were invalid as to
whether the offenders would commit any new property, violent, or any
crimes; would be charged with new offenses; or would be rearrested. The
predictions of seven judges as to whether the offenders would commit any
new crimes had no validity.

The validity of the judges' predictions of property crimes was similar to
that for any new offense. Predicting whether the offenders would commit
crimes against other persons proved to be a more difficult problem
because these crimes are relatively infrequent. The judges' predictions
accounted for less than 2 percent of the differences in new arrests for
offenses involving interpersonal confrontation. The predictions of violent
crimes by 12 of the judges had no statistically significant relationship to
new violent offense arrests or charges.

The judges who would be better predictors, of course, could not be
identified beforehand. The judges' predictions appeared to be influenced
mainly by their assessments of the seriousness of the offense, their
judgments of the offender's social stability, and the length of the offender's
arrest and conviction records. The factors that explain the judges' risk
predictions for any new crime are shown in exhibit 4.[6] The validity of
the judges' predictions is illustrated in exhibit 5.[7]

Empirical measures of risk

Two more formal risk measures were developed for use in statistically
controlling for the offenders' a priori risk of new arrests. Each was based
on information available at the time of sentencing. Risk measure 1 was
based on the whole sample. Risk measure 2 was based on probationers
only.[8]

Risk measure 1 included, as predictors, measures of age, the judges'
ratings of the offender's arrest record, race, heroin or barbiturate use in the
2 years prior to arrest, alcohol abuse, and the type of crime (property or
serious drug offense). In combination, these and other factors shown in
exhibit 6 accounted for 23 percent of the differences in new arrests (see
exhibit 7). The same measure substantially correlated with other outcomes,
such as the total number of arrests and charges.

Risk measure 2 included most of the measures in risk measure 1, plus such
factors as the offender's sex, number of prior probation sentences, and
prior incarceration for probation or parole violations (see exhibit 8). The
scores on this measure accounted for 31 percent of the variation in the any
new arrest criterion (see exhibit 9).

Selection of sentences by judges

Measures were devised to control for judicial selection of sentences in the
statistically designed studies of effects.[9] These included the decision
about whether to order confinement ("in-out"), selection of the four main
sentencing alternatives (noncustodial, jail, youth facility, or prison),
selection of the three main custodial sanctions, and selection of "split"
sentences.

Selection for any confinement (in-out). In deciding whether to order
confinement, judges appeared to be influenced mainly by their assessment
of rehabilitation as an important aim of sentencing in the particular case,
the recommendation of the probation officer for confinement (unless the
probation officer and the judge used the same information in arriving at
the decision), their own prediction of whether the offender would commit
future crimes, and the seriousness of the offense (see exhibit 10). These
and other items enabled correct classifications ("predictions") of
confinement in 88 percent of the sample.

Selection of type of sanction. 

Selection of a noncustodial sentence, jail, youth facility, or prison
appeared to be influenced mainly by the judges' predictions of whether the
offender would commit any future crime, the perceived importance of
rehabilitation for the offender, the seriousness of the charge, the offender's
criminal record, the number of counts in the conviction, and the
recommendation of the probation officer. The sentence type also appeared
to be influenced by the age of the offender and whether the felony was a
property crime.

Selection of split sentences. In ordering split sentences, the judges seemed
to be influenced by their own predictions of whether the offender would
commit future crimes, aggravating factors, and the relative importance of
retributive and rehabilitative aims.

Effects of sentences

The effects of different sentences were studied using statistical designs
that controlled for the measures of sentence selection, the a priori risk,
how long the offender was confined as a result of the sentence, and how
long the offender subsequently had an opportunity to reoffend in the
community (20 years minus time served on the sentence and
incapacitation time for later confinements).[10] Typically, a naive
interpretation of the observed outcomes without consideration of the
effects of selection, risk, and incapacitation was compared with adjusted
values taking into account the potentially biasing factors.

Effects of the in-out decision. Seventy percent of offenders who were not
sentenced to confinement and 82 percent of those who were sentenced to
confinement were rearrested during the 20-year followup period. When 
adjustments were made for sentence selection, a priori risk, and the time
spent in the community (not incapacitated), however, there was no
statistically significant difference in the percentage of offenders with new 
arrests between the two groups during that period. Similarly, there was no
difference due to the type of confinement when only the first 5 years after
sentencing were considered (see exhibit 11). When selection, risk, and
incapacitation were taken into account, confinement had no effect on the
offenders' arrest rates, the total number of arrests or charges, or charges for
specific crime categories.

Effects of the type of sentence. The type of sentence--noncustodial, jail,
youth facility, or prison--had a small but statistically significant effect on
new arrests during the first 5 years after the offender was sentenced. After
adjusting for selection, a priori risk, and incapacitation, those offenders
sent to the youth facility had the highest adjusted percentage of new
arrests. 

When persons given noncustodial sanctions were excluded from the
analysis, a statistically significant effect of placement in a jail, youth
facility, or prison was found in the 20-year followup. The adjusted values
for jail and prison were the same, but those offenders sent to the youth
facility had higher adjusted values for new arrests (see exhibit 12).

The choice among the four sentence alternatives had a small effect, mainly
due to higher adjusted values for new arrests for those sent to a youth
facility. Age, selection, a priori risk, or incapacitation could not explain
this effect. Placement in noncustodial programs, jail, or prison had no
effect.

Effects of the time served in confinement. Factors helping to explain the
time actually served in jail or prison are listed in exhibit 13. After
adjusting for selection, a priori risk, and incapacitation, the amount of time
the offender actually served in confinement had a small but statistically
significant effect on new arrests, accounting for less than 2 percent of the
differences in new arrests. A comparison of the actual and adjusted values
for new arrests is shown in exhibit 14.

Survival in the community without arrest. The offender's survival in the
community over time (after serving any incapacitating sentences) without
arrest did not differ between those who were confined and those who were
not confined. This result is reflected in exhibit 15, which shows the
"hazard rates" for offenders who were sentenced to confinement and those
who were not.[11] The curve shows new arrests from the time of release
from the relevant confinement (if the sentence called for incarceration) or
the date of the sentence (if the offender was not incarcerated). Whether the
offender was confined or not makes little or no difference to survival
without arrest when the survival or hazard function is calculated on the
basis of the a priori risk, total time incapacitated, and length of
incapacitation on the present sentence. The hazard function shown
indicates the rates of arrests at particular times, given no arrest until that
time. That is, it is an arrest rate per unit of time. Analyses of survival,
however, indicated that the effects of confinement in the youth facility or 
in prison may change over time and should be further investigated.[12]

Other effects of sentences. No effect on new arrests, beyond
incapacitation, was found for the following sentencing choices: the
maximum sentence imposed, "split" sentences of jail and probation, fines,
restitution, and compliance with fine or restitution orders.

Implications

The judges' diverse selection purposes for sentencing individuals support
the need for greater clarity and consistency in sentencing aims. The
conflict between utilitarian and retributive perspectives was apparent in
this study, despite a general preference on the part of judges for utilitarian
crime control. Clarity could be increased if there were an internally
consistent sentencing theory and if it were consistently applied.

Despite their modest validity, the judges' subjective risk judgments
substantially influenced their sentencing choices. The use of more formal,
empirically derived methods would enhance sentencing rationality when
sentencing theory incorporates risk as a relevant and justifiable
consideration.

The main sentencing choices available to these judges had little effect on
crime control aims:

--Except for the effect of incapacitation, whether the offender was
sentenced to confinement made no difference.

--Where the offender was confined made little difference--except perhaps
for the unfavorable effect of placement in a youth facility. 

--The length of the maximum sentence imposed made no difference.

--The length of time actually confined made a slight difference.

--When jail was imposed along with probation, it made no difference.

--Fines or restitution made no difference.

Aside from general deterrence (not studied) and incapacitation (in this
study it mainly provided a correction for the investigation of other crime
control effects), little justification for differences in sentences was found
from a crime control perspective. The different sanctions, varying in
severity of punishment and incapacitating effect, may have served as a
warning to others or as deserved punishment, but there was little or no
evidence of other effects on crime control objectives.

This study found no evidence to justify the belief that the addition of jail
time to a probation sentence has a specific deterrent effect. Unless it is
believed that jail time is required for punishment, or the hope of an
effective warning to others is maintained, this study would support
abandoning the use of split sentences.

Similarly, if considerations of general deterrence and deserved punishment
are set aside, it must be concluded that confinement or increased length of 
incarceration served the crime control purpose of incapacitation but had
little or no effect as a "treatment" with rehabilitative or specific deterrent
effects. Exhibit 16 illustrates that there is little or no effect of sentences on
crime control objectives--other than by incapacitation--as measured in this
study. It also portrays the large area of unexplained variability in new
offenses by the persons who were sentenced.

Policy implications

How is a study of 20-year-old sentencing practices relevant to current
debates? Some important lessons for present day policy are clear. These 
results offer little support for the policy trends, prominent since this
project began, that have supported increased use of confinement as a
sentencing choice, emphasized longer terms, or accepted specific
deterrence to reduce offenders' recidivism. Another lesson concerns
specific procedures for the sentencing decision process: When the risk of
new offenses is seen as an appropriate element in sentencing policy,
empirically derived methods are equal or superior to subjective
assessments. 

The societal stake in crime control effects of sentences, the limitations of
statistical designs, and the results of this study suggest the importance of
investigating these and similar questions by rigorous experimental
methods whenever they are determined to be ethically justifiable.

Study limitations

Some sentencing concerns were beyond the scope of the project. This
study did not examine sentencing for purposes of retribution or general
deterrence--whether punishments were deserved or whether they served as
an effective warning to others. The relative costs of sentencing choices, in 
either human or monetary terms, were not assessed. 

Conclusions from this study cannot be generalized to other jurisdictions or
other times. The study does not show, for example, that well-designed
interventions implemented with fidelity to a clear, coherent theory cannot
be effective in meeting crime control objectives. No measures of the
quality of rehabilitative programs or the severity of sanctions beyond the
length of sentences were available. The analyses reported here could not
show whether different kinds of sanctions are differentially effective for
different kinds of offenders. If either objectives or sanctions have changed,
for example, to emphasize aims of restorative justice or elaborated
"intermediate sanctions," this report must remain silent about the newer
goals and alternatives.[13]

Nonjudicial decisions may have affected the results. For persons sentenced
to prison, the time actually served in prison reflected decisions by a parole
board as well as the judges.

Statistical designs such as those used in this study always provide less
definitive results than do true experiments, which ensure that offenders
compared after subjection to different sanctions may be considered
equivalent in all respects at the outset. Perhaps other, as yet unknown,
selection factors affect either judicial decisions or risk and therefore
influence the outcomes. 

Other limitations are due to an often used but nevertheless crude
measurement of outcomes. Counts of arrests and charges are not adequate
to assess the full spectrum of costs and harms associated with either the
sanctions imposed or the new crimes committed in the community.

Similarly, a lack of information limited measurement of the main
independent variables of the study: Adequate descriptions of community
treatment programs and jail and prison programs were not available, nor
were measures of the quality of rehabilitative programs and the severity of
sanctions within the main types of sentences imposed.

Related research results ameliorate some of these limitations. The case of
the judges' predictions provides an example. In the past 70 years, 136
studies have documented 617 distinct comparisons of the two basic
measures of prediction (subjective or "in the head" versus more formal or
statistical measures). Some of these studies were conducted in the criminal
justice arena. In criminal justice decisions and in all other areas of
behavioral prediction, the evidence clearly shows that more formal
objective measures are superior to subjective evaluations.[14] Similarly,
related research since the 1960s does not support the expectation that the
length of time someone is incarcerated in prison is related to repeated
offending, particularly when relevant offender characteristics are taken
into account.[15]

Notes

1. This study started more than 20 years ago in collaboration with the
judges of the Essex County, New Jersey, Court, particularly with Judge
John A. Marzulli, then the assignment judge. The support of Robert D.
Lipscher, then director of the New Jersey Administrative Office of the
Courts, and John McCarthy, deputy director for criminal practice for the
New Jersey Administrative Office of the Courts, was essential. The
followup phase of the project described in this Research in Brief relied on
the expertise of an advisory committee composed of McCarthy; Stan
Repko, deputy commissioner for research and planning, New Jersey
Department of Corrections; and Wayne Fisher, deputy director, New
Jersey Division of Criminal Justice. The full report of the study, including
references, is available from the National Criminal Justice Reference
Service. See Gottfredson, Don M., Choosing Punishments: Crime Control
Effects of Sentences, report to the National Institute of Justice,
Sacramento, CA: Justice Policy Research Corporation, 1998.

2. Resources did not permit a search for death records, and the total
number of deaths during the followup period is unknown. Twelve persons
are known to have died during the followup. Two had received
noncustodial sentences; four received jail sentences; four were sent to a
youth facility; and two went to prison. Before their deaths, none of these
offenders had been arrested after sentencing.

3. Typically, these were analyses of covariance with the measures to be
controlled for as covariates. Analyses of "survival" (the time that had
elapsed when arrest occurred) also were used.

4. Gottfredson, S.D., and R.B. Taylor, "Community Context and Criminal
Offenders," in Communities and Crime Prevention, ed. T. Hope and M.
Shaw, London: Her Majesty's Stationery Office, 1988; and Gottfredson,
S.D., and D.M. Gottfredson, "Behavioral Prediction and the Problem of
Incapacitation," Criminology 32 (3) (1994): 450-451.

5. "Nuisance" offenses include, for example, parole and probation rules
violations, possession or use of drugs, marijuana offenses, disorderly
conduct, prostitution, and gambling.

6. The items listed are from regression analysis results. In combination,
these items resulted in a multiple correlation of 0.79 (N=931) with the
judges' ratings, which were scores of 1 to 9. The weights of each item (i.e.,
the standardized regression coefficients) are not reflected in the exhibit but
are given in the full report cited. 

7. The point biserial correlation of these scores with any new arrests was
0.22 (N=960).

8. The items listed are from regression analyses. The point biserial
correlation of risk measure 1 scores with any new arrest was 0.48
(N=960); for risk measure 2, it was 0.56 (N=466). (For standardized
coefficients, see the full report.) The results of logistic regression analyses,
theoretically more appropriate for the dichotomous dependent variable, did
not differ substantively. This is not atypical; see, for reviews and reasons,
Copas, J.B., and R. Tarling, "Some Methodological Issues in Making
Predictions," in Criminal Careers and "Career Criminals," ed. A.
Blumstein, J. Cohen, J. Roth, and C. Visher, Washington, DC: National
Academy of Sciences, 1986; Farrington, D., and R. Tarling, Predicting
Crime and Delinquency, Albany, NY: State University of New York
Press, 1985; Gottfredson, D.M., and M.R. Gottfredson, "Data for Criminal
Justice Evaluation: Some Resources and Pitfalls," in Handbook of
Criminal Justice Evaluation, ed. M.W. Klein and D.S. Teilman, Beverly
Hills, CA: Sage Publications, 1980. In both the ordinary least squares
regression and the logistic regression, the items most helpful in prediction,
in the context of the other variables included, were age, arrest record, race,
and heroin or barbiturate use in the past 2 years. The race variable
("white") was included and retained because the prediction method was
not developed for any operational classification use but only for the
purpose of statistical control in the other analyses described. More
complete specifications of these and other analytical results discussed here
are included in the full report of the study.

9. Selection measures relied on discriminant functions with n-1 equations
where n=the number of groups to be discriminated. The equations,
including the standardized coefficients, are given in the full report of the
study. 

10. Most analyses relied on analyses of covariance, others on regression
methods. The "adjusted values" shown in the exhibits are the adjusted
means from the analyses of covariance.

11. This result is based on a Cox regression analysis.

12. A Cox regression analysis based on the groups given noncustodial,
jail, youth facility, and prison sentences did not meet assumptions of
proportionality required for the use of the model, and there was a
significant effect from a time x treatment interaction. Further analysis 
indicated the hazard rate increased more rapidly over time for those
sentenced to the youth institution, less so for those imprisoned. Survival in
the community without arrest after release from confinement in the youth
facility decreased more rapidly than did survival without arrest for persons
who had been imprisoned.

13. For recent reviews, see Cullen, F.T., J.P. Wright, and B.K. Applegate,
"Control in the Community: The Limits of Reform?" in Choosing
Correctional Options That Work, ed. A. Harland, Thousand Oaks, CA:
Sage Publications, 1996: 113-115; Petersilia, J., "Understanding Probation
and Parole: Theory and Practice," in Oxford Criminology Handbook, ed.
M. Tonry, London: Oxford University Press, 1998; Petersilia, J.,
Community Corrections: Probation, Parole, and Intermediate Sanctions,
London: Oxford University Press, 1998.

14. Meehl, P.E., and W.M. Grove, "Comparative Efficiency of Informal
(Subjective, Impressionistic) and Formal (Mechanical, Algorithmic) 
Prediction Procedures: The Clinical-Statistical Controversy," Psychology,
Public Policy, and Law 2 (2) (1996): 293-323.

15. Glaser, D., The Effectiveness of a Prison and Parole System, New
York: Bobbs-Merrill, 1964: 302-303; Department of Institutions, State of
Washington, Research Monograph No. 27, Olympia, WA: Department of
Institutions, State of Washington, 1967; Department of Corrections, State
of California, Long Jail Terms and Parole Outcomes, Sacramento, CA:
Department of Corrections, State of California, 1967; Assembly
Committee on Criminal Procedure, (Committee Report). Sacramento, CA:
Assembly of the State of California, 1968; Kolodny, S., Parole Board
Reform in California: Order Out of Chaos, Report of the Select Committee
on the Administration of Justice, Sacramento, CA: Assembly of the State
of California, 1970; Gottfredson, D.M., M. Neithercutt, J. Nuffield, and V.
O'Leary, Four Thousand Lifetimes: A Study of Time Served in Prison and
Parole Outcomes, Davis, CA: National Council on Crime and
Delinquency, 1973; Gottfredson, D.M., M.R. Gottfredson, and J. Garafalo,
"Time Served in Prison and Parole Outcomes Among Parolee Risk
Categories," Journal of Criminal Justice 5 (1) (1977).

-------------------------------

Don M. Gottfredson is Richard J. Hughes Professor Emeritus at 
the School of Criminal Justice, Rutgers University, and President of the
Justice Policy Research Corporation. This study was funded in part by
grants 82-IJ-CX-0054 and 95-IJ-CX-0118 by the National Institute of
Justice. 

-------------------------------

Findings and conclusions of the research reported here are those of the
author and do not necessarily reflect the views or endorsement of the
Essex County Court, the judges who participated in the study, the New
Jersey Administrative Office of the Courts, the U.S. Department of
Justice, or any other individuals or agencies.

-------------------------------

The National Institute of Justice is a component of the Office of Justice
Programs, which also includes the Bureau of Justice Assistance, the
Bureau of Justice Statistics, the Office of Juvenile Justice and Delinquency
Prevention, and the Office for Victims of Crime.

-------------------------------

This and other NIJ publications can be found at and downloaded from the
NIJ Web site (http://www.ojp.usdoj.gov/nij).

NCJ 178889