United States of America v. State of California et al

Filing 171

REPLY by United States of America to RESPONSE to 2 Motion for Preliminary Injunction. (Attachments: # 1 Exhibit A, # 2 Exhibit B, # 3 Exhibit C, # 4 Exhibit D, # 5 Exhibit E, # 6 Exhibit F, # 7 Exhibit G, # 8 Exhibit H, # 9 Exhibit I, # 10 Exhibit J, # 11 Exhibit K, # 12 Exhibit L, # 13 Exhibit M, # 14 Exhibit N, # 15 Exhibit O, # 16 Exhibit P, # 17 Exhibit Q, # 18 Exhibit R, # 19 Exhibit S, # 20 Exhibit T, # 21 Exhibit U, # 22 Exhibit V, # 23 Exhibit W, # 24 Exhibit X, # 25 Exhibit Y)(Reuveni, Erez)

EXHIBIT M

Measuring recidivism
Recidivism measures require three characteristics:
1. a starting event, such as a release from prison
2. a measure of failure following the starting event,
such as a subsequent arrest, conviction, or return to
prison
3. an observation or follow-up period that generally
extends from the date of the starting event to
a predefined end date (e.g., 6 months, 1 year,
3 years, 5 years, or 9 years).
This study used four outcome measures to examine the
recidivism patterns of former state prisoners. Arrest data
were used because they provided the offense details
needed to produce these four measures for prisoners
from all 30 states in the study.
1. Cumulative arrest percentage is the percentage of
prisoners who had been arrested at least once at
various points in the follow-up period. For example,
the cumulative arrest percentage for year-5 is
the percentage of all released prisoners who had
at least one arrest during the 5-year period. BJS
previously examined the cumulative percentage
of prisoners who had a subsequent conviction
or returned to prison within 5 years following
release.2 The return-to-prison analysis for the 5-year
follow-up study was limited to 23 of the study’s 30
states with the data needed to identify returns to
prison during the entire observation period.
2. Annual percentage of first arrests is the percentage
of prisoners who had their first arrest following
release during a specific year in the follow-up
period. The denominator for each annual first-arrest
percentage from years 1 through 9 is the total
number of prisoners released in the 30 states
during 2005. The numerators are the number of
prisoners arrested for the first time during each of
those years (i.e., they had not been arrested during
a prior year in the follow-up period). The sum of the
annual first-arrest percentages during a follow-up
period equals the cumulative arrest percentage for
the same period.
3. Annual arrest percentage of released prisoners
includes those who were arrested at least once
during a particular year within the follow-up
2See Recidivism of Prisoners Released in 30 States in 2005: Patterns

from 2005 to 2010, NCJ 244205, BJS web, April 2014.

period. The denominator for each percentage
from years 1 through 9 is the total number of
prisoners released in the 30 states during 2005. The
numerators are the number of prisoners arrested
during the particular year, regardless of whether
they had been arrested during a prior year.
4. Annual volume of arrests is the total number of
arrests of released prisoners during a particular
year in the follow-up period. The total volume of
arrests is the sum of each annual volume of arrests
during the entire follow-up period. A prisoner may
have had multiple arrests during a year or in the
follow-up period, and a single arrest may have
involved charges for more than one crime.

Measuring desistance
Desistance is measured as the percentage of prisoners
who, after a particular year, had no subsequent arrests
during the remainder of the 9-year follow-up period.
For example, if a prisoner was arrested during year-3
but was not arrested during years 4 through 9, the
prisoner would be classified as having desisted during
year-3. While recidivism is a measure of arrest at any
point during the follow-up period, desistance is a
measure of the absence of arrest between a particular
point within the follow-up period and the end of the
follow-up period.

Importance of recidivism and
desistance measures
Measures of recidivism and desistance provide
information relevant to a deeper understanding of
criminal behavior and the administration of justice
in a wide range of policy areas. For example, law
enforcement officials interested in the amount of
crime committed by released prisoners can turn to
statistics on the annual volume of arrests. Parole and
probation agencies interested in the involvement of
various types of former prisoners in criminal activities
after release may focus on variations in cumulative
arrest percentages. Treatment providers looking for
measures of program effectiveness will be interested
in desistance patterns. Additionally, task forces and
policymakers examining the movement of criminals
across state borders will be interested in the types of
released prisoners most likely to commit new crimes
(i.e., recidivate) in other states.

2018 UPDATE ON PRISONER RECIDIVISM: A 9-YEAR FOLLOW-UP PERIOD (2005-2014) | MAY 2018

3

Forty-four percent of prisoners released in 2005
were arrested during the first year following release
(figure 3). Sixteen percent of released prisoners were
arrested for the first time during the second year after
release, and 8% were arrested for the first time during
the third year. Fifteen percent of released prisoners
were arrested for the first time during years 4 through
9, including 11% arrested for the first time during years
4 through 6 and 4% arrested for the first time during
years 7 through 9.
Of the released prisoners who were arrested for a new
crime during the 9-year follow-up period, the majority
of the prisoners’ first post-release arrests occurred
during the first 3 years of the follow-up period. More
than half (53%) of all prisoners released in the 30 states
in 2005 who were arrested during the 9-year follow-up
period were arrested for the first time during the first
year (not shown). Among all released prisoners arrested
within 9 years, about 5 in 6 prisoners (82%) were
arrested within the 3-year follow-up period (not shown).
Figure 3
Percent of prisoners released in 30 states in 2005 who
were arrested after release, by year of first arrest
Percent of released prisoners
50

Longer follow-up periods show substantial
declines in apparent desistance
This study examined the extent to which released
prisoners appeared to have desisted from criminal
activity using various follow-up periods. Thirty-two
percent of released prisoners had not been arrested
within 3 years, compared to 21% within 6 years
(figure 4). Within 9 years following release in 2005, the
percentage of prisoners without a new arrest following
release declined to 17%. That is, almost half (47%) of
prisoners with no arrest within 3 years of release had
an arrest during years 4 through 9.
Figure 4
Percent of prisoners released in 30 states in 2005 who
were not arrested since release, by year following release
Percent of released prisoners
60
50
40
30
20
10

40

0
30

1st

2nd

3rd

4th 5th 6th
Year after release

7th

8th

9th

Note: See table 2 for estimates and appendix table 3 for standard errors.
Source: Bureau of Justice Statistics, Recidivism of State Prisoners
Released in 2005 data collection, 2005–2014.

20
10
0

1st

2nd

3rd

4th 5th 6th
Year of first arrest

7th

8th

9th

Note: The denominator for the annual percentage was 401,288 (total
state prisoners released in 30 states in 2005). See appendix table 4 for
estimates and standard errors.
Source: Bureau of Justice Statistics, Recidivism of State Prisoners
Released in 2005 data collection, 2005–2014.

2018 UPDATE ON PRISONER RECIDIVISM: A 9-YEAR FOLLOW-UP PERIOD (2005-2014) | MAY 2018

5

points narrowed from 10% during year-1 to 3%
during year-9, the difference between male and female
prisoners decreased proportionally. During year-1,
the percentage of female prisoners who were arrested
following release was 78% of that for male prisoners,
while during year-9 the percentage of female prisoners
was 87% of that for male prisoners (not shown).
Younger prisoners (those age 24 or younger) were
more likely to be arrested than older prisoners
(those age 40 or older) during each year following
release. For example, 28% of prisoners released at age
24 or younger were arrested during year-9, compared
to 19% of those age 40 or older (figure 8).

FIGURE 8
Annual arrest percentage of prisoners released in
30 states in 2005, by age of prisoner at release

During the first year following release, the percentage
of prisoners released for a property offense who were
arrested for any type of offense (including violent,
property, drug, or public order offenses) was higher
than the percentage of prisoners released for a drug or
violent offense. This general pattern was maintained
across the 9-year follow-up period. It should be noted
that persons could have been serving time in prison
for more than one offense and were categorized for
this report by the most serious offense for which they
were imprisoned: a violent, property, drug, or public
order crime.
FIGURE 9
Annual percentage of prisoners released in 30 states
in 2005 who were arrested for any type of offense, by
most serious commitment offense

Percent of released prisoners
60

Percent of released prisoners arrested*
60

50

50

40

24 or younger

30
40 or older

20

25–39

Most serious offense for
which they were in prison

Property

40

Drug

30

Public order

20

10
0

During the 9 years after release, prisoners
released for a property offense were most likely
to be arrested

Violent

10

1st

2nd

3rd

4th 5th 6th
Year of arrest

7th

8th

9th

Note: See table 5 for estimates and appendix table 7 for standard errors.
Source: Bureau of Justice Statistics, Recidivism of State Prisoners
Released in 2005 data collection, 2005–2014.

0

1st

2nd

3rd

4th 5th 6th
Year of arrest

7th

8th

9th

Note: Public order includes 0.8% of cases in which prisoners’ most
serious offense was unspecified. See table 6 for estimates and
appendix table 8 for standard errors.
*Persons could have been in prison for more than one offense; the
most serious one is reported in this figure.
Source: Bureau of Justice Statistics, Recidivism of State Prisoners
Released in 2005 data collection, 2005–2014.

2018 UPDATE ON PRISONER RECIDIVISM: A 9-YEAR FOLLOW-UP PERIOD (2005-2014) | MAY 2018

9

Five percent of prisoners were arrested during
year-1 following release and not arrested again
during years 2 through 9
Thirty-nine percent of all released prisoners were
arrested during the first year after release and were
also subsequently arrested at least once during years
2 through 9 (figure 11). Five percent of prisoners were
arrested during the first year after release and were
not arrested again during years 2 though 9. Among
prisoners arrested during the first year following
release, nearly 9 in 10 (89%) were arrested again during
the next 8 years (not shown).
During the second year after release, 38% of prisoners
were arrested. A third (33%) of all released prisoners
were arrested during the second year and also arrested
again at least once during years 3 through 9. The
remaining 5% were not arrested again during the
follow-up period.
The percentage who were not arrested during a
subsequent year increased during the later years of the
follow-up period. However, there are fewer observable
years in which to capture a subsequent arrest during
later years of the follow-up period. (For example, in
year-8 there is only one subsequent year.)
FIGURE 11

Percent of prisoners released in 30 states in 2005 who
were arrested after release, by year after arrest and
whether arrested during subsequent years

40

Arrested again during a subsequent year
Not arrested during a subsequent year

30
38.9

33.0

29.1

26.0

23.2

20.1

17.2

5.2

5.9

6.9

7.8

10.2

3rd

4th
5th
6th
Year after arrest

7th

10
0

4.9
1st

4.7
2nd

Twenty-two percent of released prisoners were not
arrested after year-1 of the follow-up period (figure 12).
In other words, when measured by a new arrest, 22% of
prisoners appeared to desist by year-1 because they
were not arrested during years 2 through 9. Some
released prisoners may not have been arrested because
they were incarcerated at certain times during the
follow-up period. Thirty-one percent of prisoners
appeared to have desisted by year-3 and 44% by year-5.
This percentage eclipsed 52% in year-6, at which point
two-thirds of the observable years had elapsed.
FIGURE 12

Prisoners released in 30 states in 2005 who were not
arrested in the remainder of the follow-up period, by
year after release
Percent of released prisoners
80
70
60
50
40
30
20
10
0

Percent of released prisoners
50

20

Forty-four percent of released prisoners were not
arrested after year-5

1st

2nd

3rd
4th
Year after release

5th

6th

Note: Estimates after year-6 are not presented as 3 years of subsequent
arrests could not be measured. See appendix table 12 for estimates and
standard errors.
Source: Bureau of Justice Statistics, Recidivism of State Prisoners
Released in 2005 data collection, 2005–2014.

12.2
13.7
8th

Note: The denominator for the annual percentage was 401,288 (total
state prisoners released in 30 states in 2005). See appendix table 11 for
estimates and standard errors.
Source: Bureau of Justice Statistics, Recidivism of State Prisoners
Released in 2005 data collection, 2005–2014.

2018 UPDATE ON PRISONER RECIDIVISM: A 9-YEAR FOLLOW-UP PERIOD (2005-2014) | MAY 2018

13

Longer follow-up periods provide additional data on the nature of criminal
careers, but take time
Research on the criminal activities of persons released
from prison has employed different follow-up periods.
A 3-year follow-up period has been common, but other
time periods (e.g., 1-year, 5-year, and 9-year, as in this
report) have been used. Which follow-up period is
employed is often driven by the availability of data or
the need to study a specific cohort (e.g., those released
3 years ago). Independent of these constraints, deciding
which follow-up period to use is closely linked to the
competing concerns of accuracy and immediacy.
This study provides empirical evidence that may be
used to inform which follow-up period is preferred for
various research efforts and policy applications. For
example, comparing the 3-year and 9-year follow-up
periods showed that the basic recidivism percentage
(defined as the cumulative arrest percentage
following release) was underestimated by an average
of 15 percentage points using the 3-year window.
Similarly, the 9-year follow-up period showed that
the percentage of released prisoners arrested in
states outside the state that released them was twice
as high as that observed in a study with a 3-year
follow-up period.
With a follow-up period of 3 years, researchers and
policymakers would not have observed more than half
of the arrests of prisoners after their release. This study’s
9-year follow-up period showed that 60% of all arrests
of released prisoners occurred more than 3 years after
their release.
A longer follow-up period enables researchers and
policymakers to better explore the attributes of
desistance. This study found that 32% of released
prisoners had no arrests following release during the

3-year follow-up period (and appeared to have desisted
from criminal activity), but almost half of those (15%)
were arrested during the subsequent 6 years, leaving
17% who had no arrests during the 9-year follow-up
period (see table 2). In addition, the study found that
24% of released prisoners were still actively involved in
criminal activity and were arrested during year-9, which
could be viewed as inviting an even longer period of
review. The longer period also enables researchers
to understand more complex patterns of desistance.
For example, of the 44% of released prisoners who
were arrested during their first year after release, 1 in 9
(5% of all released prisoners) had no additional arrests
during the 9-year follow-up period.
Counterbalancing the value of a longer follow-up
period is the need for up-to-date information.
Offending patterns may change with time and the
offending patterns of prisoners released 10 years ago
may be different than those of prisoners released in
recent years. In addition, policymakers and practitioners
have a need for timely information and may not have
time for a recidivism study with a long observation
period to be completed to assess the value of a
rehabilitation program for released prisoners or a policy
change affecting sentencing.
There is no standard length for follow-up periods used
in studies of the criminal careers of released prisoners
or any other cohort of offenders. This study shows how
recidivism and desistance measures change when
longer or shorter follow-up periods are used. With these
additional data, designers and users of recidivism and
desistance studies have more information to determine
which follow-up period is best for their needs.

2018 UPDATE ON PRISONER RECIDIVISM: A 9-YEAR FOLLOW-UP PERIOD (2005-2014) | MAY 2018

14

maintains criminal history records for which it has sole
responsibility for disseminating, such as information
on federal arrests. The identification bureaus that
operate the central repositories in each state respond
automatically to requests over the Nlets network with
an individual’s criminal history record. Put together,
these requests represent the individual’s national
criminal history record.
Once BJS received approval from the FBI’s Institutional
Review Board to conduct this recidivism study on
prisoners released in 2005, Nlets transmitted the
state and FBI identification numbers on the sampled
prisoners to the FBI’s III system to collect the criminal
history records on behalf of BJS. The criminal history
records include information from the state of release
and all other states in which the sampled prisoners
had been arrested both prior to the release in 2005
and afterwards.
Nlets parsed the fields from individual criminal
history records into a relational database consisting
of state- and federal-specific numeric codes and
text descriptions (e.g., criminal statutes and case
outcome information) into a uniform record layout.
NORC at the University of Chicago assisted BJS
with standardizing the content of the relational
database into a uniform coding structure to support
national-level recidivism research.
BJS conducted a series of data-quality checks on
the criminal history records to assess the accuracy
and completeness of the information, including
an examination of the response messages and the
identification numbers that failed to match a record
in III. To ensure that the correct records were received
on the released prisoners using their fingerprint-based
identification numbers, BJS compared other individual
identifiers in the NCRP data to those reported in the
criminal history records. For 98% of cases, a released
prisoner’s date of birth in the NCRP data exactly
matched the prisoner’s birthdate in the criminal
history records. Nearly 100% (99.9%) of the NCRP and
criminal history records matched prisoner sex, race,
and Hispanic origin.
BJS reviewed the criminal history records for
differences and inconsistencies in reporting practices
and noticed some variations across states. During
data processing and analysis, steps were taken to
standardize the information and to minimize the
impact these variations had on the overall recidivism
and desistance estimates. For example, administrative
(e.g., a criminal registration or the issuance of a

warrant) and procedural (e.g., transferring a suspect to
another jurisdiction) records embedded in the criminal
history data that did not refer to an actual arrest were
identified and removed. Traffic offenses (except for
vehicular manslaughter, driving while intoxicated, and
hit-and-run) were also excluded because the reporting
of these events in the criminal history records varied
widely by state.
Deaths during the follow-up period
BJS documented that 2,173 of the 70,878 sampled
prisoners died during the 9-year follow-up period,
and BJS removed these cases from the recidivism and
desistance analysis along with four additional cases
that were determined to be invalid release records. The
fingerprint-verified death notices obtained through
the FBI’s III system were used to identify some of
the sampled prisoners who died within the 9 years
following release in 2005. Additional deaths were
identified through the Social Security Administration’s
(SSA) public Death Master File (DMF). While the
public DMF provided a more complete source of death
information than the FBI’s III system, the public DMF
provided death information only for the years 2005 to
2011. Therefore, the identification of those who died
between 2012 and 2014 was limited to the FBI’s III
data, which included only fingerprint-verified deaths.
The number of released prisoners who were identified
as dead between 2005 and 2011 in the public DMF
is an undercount of the actual number of deaths
within the sample. Due to state disclosure laws, the
public DMF does not include information on certain
protected state death records received via SSA’s
contracts with the states. Beginning in 2011, the SSA
removed more than 4 million state-reported death
records from the public DMF and began adding fewer
records to the public DMF. As a result, the public DMF
contains an undercount of annual deaths.
The extent to which the public DMF undercounts
the annual number of deaths is not exactly known.
Analyses of deaths in the public DMF compared to
those reported by the Centers for Disease Control
and Prevention’s (CDC) mortality counts suggest that
the public DMF undercounted the overall number of
deaths in the United States by about 10% in 2005. The
undercount increased during succeeding years, and as
of 2010, the public DMF contained less than half (45%)
of the deaths reported by the CDC. If the number
of released prisoners who died during the follow-up

2018 UPDATE ON PRISONER RECIDIVISM: A 9-YEAR FOLLOW-UP PERIOD (2005-2014) | MAY 2018

16

period and were removed from the recidivism and
desistance analysis were adjusted to account for this
undercount, the estimated cumulative recidivism rate
would likely increase by about one percentage point.
Missing criminal history records
Among the 68,701 sampled prisoners not identified
as deceased during the follow-up period, BJS did not
receive criminal history records on 735 prisoners,
either because the state departments of correction
were unable to provide their FBI or state identification
number or because the prisoner had an identification
number that did not link to a criminal history record
either in the FBI or state record repositories. To
account for the missing criminal history records and
to ensure the recidivism and desistance statistics
were representative of all 68,701 prisoners in the
analysis, BJS developed weighting class adjustments to
account for those prisoners without criminal history
information to reduce nonresponse bias.
To create the statistical adjustments, the 68,701
sampled prisoners were stratified into groups by
crossing the two categories of sex (male or female),
five categories of age at release (24 or younger, 25 to
29, 30 to 34, 35 to 39, or 40 or older), four categories
of race/Hispanic origin (non-Hispanic white,
non-Hispanic black, Hispanic, or other race), and four
categories of the most serious commitment offense
(violent, property, drug, or public order). Within each
of the subgroups, statistical weights were applied to
the data of the 67,966 prisoners with criminal history
information to allow their data to represent the
735 prisoners without criminal history information.
Conducting tests of statistical significance
This study was based on a sample, not a complete
enumeration, so the estimates are subject to sampling
error. One measure of the sampling error associated
with an estimate is the standard error. The standard
error can vary from one estimate to the next. In
general, an estimate with a smaller standard error
provides a more reliable approximation of the true
value than an estimate with a larger standard error.
Estimates with relatively large standard errors should
be interpreted with caution. BJS conducted tests to
determine whether differences in the estimates were
statistically significant once sampling error was taken
into account.

All differences discussed in this report are statistically
significant at the 95% confidence interval level.
Standard errors were generated using Stata, a statistical
software package that calculates sampling errors for
data from complex sample surveys.
Offense definitions
Violent offenses include homicide, rape or sexual
assault, robbery, assault, and other miscellaneous or
unspecified violent offenses.
Property offenses include burglary, fraud or forgery,
larceny, motor vehicle theft, and other miscellaneous or
unspecified property offenses.
Drug offenses include possession, trafficking, and
other miscellaneous or unspecified drug offenses.
Public order offenses include violations of the peace or
order of the community or threats to the public health
or safety through unacceptable conduct, interference
with a governmental authority, or the violation of civil
rights or liberties. This category includes weapons
offenses, driving under the influence, probation and
parole violation, obstruction of justice, commercialized
vice, disorderly conduct, and other miscellaneous or
unspecified offenses.
Arrests for probation and parole violations
In this report, arrests for probation and parole
violations were included as public order offenses.
Excluding arrests for probation and parole violations
from the analysis would have had only a small impact
on the recidivism rates. Excluding probation and
parole violations from the annual arrest percentages,
39.5% of prisoners released in 30 states in 2005 were
arrested in year-1, 34.3% were arrested in year-2, 31.5%
in year-3, 29.7% in year-4, 28.2% in year-5, 25.9% in
year-6, 25.9% in year-7, 24.6% in year-8, and 23.0%
in year-9. Overall, excluding probation and parole
violations, 82.4% of prisoners released in 30 states
in 2005 were arrested within 9 years. In other words,
99% of prisoners who were arrested during the 9-year
follow-up period were arrested for an offense other
than a probation or parole violation.

2018 UPDATE ON PRISONER RECIDIVISM: A 9-YEAR FOLLOW-UP PERIOD (2005-2014) | MAY 2018

17