Skip navigation

Redemption in the Presence of Widespread Criminal Background Checks Carnegie Mellon Univ

Download original document:
Brief thumbnail
This text is machine-read, and may contain errors. Check the original document to verify accuracy.
\\server05\productn\C\CRY\47-2\CRY210.txt

unknown

Seq: 1

1-MAY-09

10:08

REDEMPTION IN THE PRESENCE OF
WIDESPREAD CRIMINAL BACKGROUND
CHECKS*
ALFRED BLUMSTEIN
KIMINORI NAKAMURA
The H. John Heinz III College
Carnegie Mellon University
KEYWORDS: redemption, criminal background checks, criminal-history
records, employment
Criminal background checks have now become ubiquitous because
of advances in information technology and growing concerns about
employer liability. Also, a large number of individual criminal records
have accumulated and have been computerized in state repositories and
commercial databases. As a result, many ex-offenders seeking employment could be haunted by a stale record. Recidivism probability
declines with time “clean,” so some point in time is reached when a
person with a criminal record, who remained free of further contact
with the criminal justice system, is of no greater risk than a counterpart
of the same age—an indication of redemption from the mark of crime.
Very little information exists on this measure of time until redemption
and on how its value varies with the crime type and the offender’s age
at the time of the earlier event. Using data from a state criminal-history
repository, we estimate the declining hazard of rearrest with time clean.
We first estimate a point of redemption as the time when the hazard
intersects the age–crime curve, which represents the arrest risk for the
general population of the same age. We also estimate another similar
*

The authors are most appreciative of the support of David J. van Alstyne of New
York’s Division of Criminal Justice Services for providing considerable help in
accessing the data. Partial funding for this work has been provided by the
National Institute of Justice under Grant 2007-IJ-CX-0041. We also thank Daniel
Nagin, Melvin Stephens, James Jacobs, and three anonymous reviewers for their
helpful comments and suggestions. A previous version of this article was
presented at the 2007 annual meeting of the American Society of Criminology
(Atlanta, Georgia). Direct correspondence to Alfred Blumstein, the H. John
Heinz III College, Carnegie Mellon University, Pittsburgh, PA 15213 (email:
ab0q@andrew.cmu.edu).

 2009 American Society of Criminology

CRIMINOLOGY

VOLUME 47

NUMBER 2

2009

327

\\server05\productn\C\CRY\47-2\CRY210.txt

328

unknown

Seq: 2

1-MAY-09

10:08

BLUMSTEIN & NAKAMURA
redemption point when the declining hazard comes “sufficiently close”
to the hazard of those who have never been arrested. We estimate both
measures of redemption as a function of the age and the crime type of
the earlier arrest. These findings aid in the development of guidelines
for the users of background checking and in developing regulations to
enhance employment opportunities for ex-offenders.
For 30 years I’ve lived a good life—so why should I have to tell a
potential employer about my past? (Scanlon, 2000: 10)

THE BASIC PROBLEM OF REDEMPTION
People like Scanlon quoted above are not rare. Many people have made
mistakes in their youthful past but have since turned themselves around
and now live a respectful life. We define redemption, which is a term
rooted in the religious concept that refers to forgiveness of past sins, as the
process of “going straight” and being released from bearing the mark of
crime. Until recently, society had a natural redemption process at work in
the sense that a person who committed a crime could prove to be
redeemed by leading a life as a productive member of society. In recent
years, the opportunity for redemption has been in serious question. The
following two important trends make the problem of redemption a growing public concern: 1) there has been an increasing demand for background checks for a wide variety of purposes, most importantly for
employment assessment, and 2) a growing number of individual criminal
records have accumulated and are becoming easily accessible electronically. With the rapid advancement in information technology, individuals
with a criminal record are haunted not only by the question about their
criminal background on job applications but also are faced with computerized criminal background checks, which are increasingly relied on by
employers.1 Criminal background checks reveal the individual’s old criminal record and highlight that fact, which overshadows a law-abiding life led
since. Computerized criminal records can have long memories, and this
article is intended to provide guidance for imposing some limits to that
memory.
Employers conduct background checks on job applicants for several different reasons. One reason may be to verify their moral character.
Another reason, which is more directly related to the context of criminalhistory background checks, may be the desire to assess their risk of committing crimes that could cause physical, financial, and reputational damage to the organization. We focus on this risk of reoffending by those
1.

The concern has been raised at least since the 1970s (Maltz, 1976; Westin and
Baker, 1972).

\\server05\productn\C\CRY\47-2\CRY210.txt

unknown

Seq: 3

CRIMINAL BACKGROUND CHECKS

1-MAY-09

10:08

329

individuals who have prior criminal records. Considerable evidence exists
that, after an initial period, the probability of recidivism declines monotonically with time free and clear of further contact with the criminal justice system. The current article addresses the following questions: How
long does it take for an individual with a prior criminal record and no
subsequent criminal involvement to be of no greater risk than persons of
the same age in the general population?2 How does an individual with a
prior record compare with individuals with no prior record? How do those
risks vary with the characteristics of the prior record, such as the crime
type and age at the prior arrest?

PREVALENCE OF CRIMINAL
BACKGROUND CHECKING
With the advancement in information technology and the Internet, individuals’ criminal records have never been more easily accessible. The
background-check industry is burgeoning. Numerous companies exist that
acquire and compile criminal justice information obtained from the police
and the courts and assemble a database for commercial purposes (Barada,
1998; Munro, 2002). SEARCH (the National Consortium for Justice Information and Statistics) reports that, “in addition to a few large industry
players, there are hundreds, perhaps even thousands, of regional and local
companies” that compile and/or sell criminal justice information to the
end users (SEARCH, 2005: 7). They provide background-check services to
private employers at their convenience in a timely manner at decreasing
costs (SEARCH, 2005). A recent survey of firms from multiple cities in
the United States reveals that about 50 percent check the criminal background of job applicants (Holzer, Raphael, and Stoll, 2004). Another survey finds that 80 percent of the large employers in the United States now
run criminal background checks on their prospective employees (Society
for Human Resources Management, 2004).
Some employers may conduct criminal background checks on job applicants voluntarily to identify those individuals who may commit criminal
acts in the workplace to minimize loss and legal liability of negligent hiring
that could result from such acts (Bushway, 1998). For some job positions
that involve vulnerable populations, such as children and the elderly, laws
require employers to conduct such background checks (Hahn, 1991). In
addition, employers may use criminal history records to assess character
flaws, such as lack of honesty and trustworthiness (Kurlychek, Brame, and
2.

We know that recidivism risk declines with age, and so it is important to make
the comparison with age-comparable individuals.

\\server05\productn\C\CRY\47-2\CRY210.txt

330

unknown

Seq: 4

1-MAY-09

10:08

BLUMSTEIN & NAKAMURA

Bushway, 2007; Pager, 2007). Also, occupational licensing laws could disqualify many individuals based on the requirement of “good moral character” (Harris and Keller, 2005; May, 1995).3 As the use of criminal
background checks by employers has become widespread, criminal
records could have lingering effects on employment prospects as “invisible
punishment” or collateral consequences of contact with the criminal justice system (Travis, 2002).4 Many employers show considerable reluctance
to hire individuals with criminal records (Holzer, Raphael, and Stoll, 2003;
Pager, 2003; Schwartz and Skolnick, 1962; Holzer, Raphael, and Stoll,
2004;5 others have shown the relationship between criminal records and
poorer employment prospects (Bushway, 1998; Grogger, 1995; Nagin and
Waldfogel, 1995; Western, Kling, and Weiman, 2001).

PREVALENCE OF CRIMINAL RECORDS
In 2007, according to the Uniform Crime Report, law-enforcement
agencies across the United States made over 14 million arrests (Federal
Bureau of Investigation, 2008). On December 31, 2003, over 71 million
criminal-history records were in the state criminal-history repositories
(Bureau of Justice Statistics, 2006).6 The increasing automation of criminal
history records in the repositories has increased the number of records
that are electronically accessible. At the end of 2003, about 90 percent of
the records were automated, and the level of automation increased 57 percent from 1995 (Bureau of Justice Statistics, 2006).
Prior research suggests that the general public’s chance of being
arrested in their lifetime is high. Over 40 years ago, it was estimated that
50 percent of the U.S. male population would be arrested for a nontraffic
offense in their lifetime (Christensen, 1967). Among those who have an
3.

4.

5.

6.

We do not elaborate more on employers’ concern over whether a criminal record
signals a lack of good character. The investigation of such considerations and its
relationship with time clean warrant future research on employer judgments.
Collateral consequences of contact with the criminal justice system occur mostly
outside the public view and affect ex-offenders beyond the imposed sentences
(Travis, 2002: 16). They include restrictions on professional and occupational
licensing, which are possibly important means for ex-offenders to increase their
employment opportunities. The occupations that are affected by the restrictions
range from health care, nursing, and education, to plumbing and barbering. Collateral consequences could also include denial of governmental benefits, such as
welfare and public housing, termination of parental rights, and revocation or suspension of driver’s licenses (Kethineni and Falcone, 2007; May, 1995; Petersilia,
2003; Samuels and Mukamal, 2004; Wheelock, 2005).
Some evidence suggests that the negative effect of criminal background checks
on the hiring of ex-offenders is strongest for employers who are legally required
to conduct such background checks (Stoll and Bushway, 2008).
An individual offender may have had records in multiple states.

\\server05\productn\C\CRY\47-2\CRY210.txt

unknown

Seq: 5

1-MAY-09

CRIMINAL BACKGROUND CHECKS

10:08

331

arrest record, some have an isolated record that was acquired years ago
and have maintained a clean record since then, but the evidence of contact
with the criminal justice system, even if it was in the distant past, could
remain in the repositories forever.

RELEVANCE OF CRIMINAL HISTORY
One rationale behind the practice of checking the criminal background
of job applicants is that the employers recognize the strong positive relationship between past and future criminal offending. The continuity in
criminal behavior has been validated by many studies (Blumstein, Farrington, and Moitra, 1985; Brame, Bushway, and Paternoster, 2003; Farrington, 1987; Piquero, Farrington, and Blumstein, 2003). Although these
studies lend support to employers who would avoid any potential employees with a criminal-history record, these employers would also be well
advised by some interlinked lines of research in criminology, which present equally strong evidence of desistance from crime in a subpopulation
of those with past offenses. One line of research argues that changes in the
life course of offenders affect their risk of future involvement in crime. For
example, it is well established that a stable marriage and employment are
powerful predictors of such desistance (Sampson and Laub, 1993; Sampson, Laub, and Wimer, 2006; Uggen, 1999; Wallman and Blumstein, 2006;
Warr, 1998). Also, in another line of research, the age–crime curve demonstrates a steady decline in criminal activity after a peak in the late teens
and young-adult period, and aging is one of the most powerful explanations of desistance (Farrington, 1986; Hirschi and Gottfredson, 1983;
Sampson and Laub, 1993, 2003).
Most importantly for the current study, time clean since the last offense
strongly affects the relationship between past and future offending behavior. Studies on recidivism consistently demonstrate that those who have
offended in the past will have the highest probability of reoffending within
several years, and the probability will decline steadily afterward (Maltz,
1984; Schmidt and Witte, 1988; Visher, Lattimore, and Linster, 1991). Two
studies that tracked released U.S. prisoners show that of all those who
were rearrested in the first 3 years, approximately two thirds were arrested
in the first year, which indicates the declining recidivism rate over time
(Beck and Shipley, 1997; Langan and Levin, 2002). Another study
examined the effects of sentences on 962 felons convicted between 1976
and 1977 in Essex County, New Jersey, by following their recidivism (measured by rearrest) for over 20 years (Gottfredson, 1999). This study shows
that although half of those rearrested were arrested within 2.2 years, 30
percent of the offenders remained arrest-free after the original sentence.
The calculation based on the Essex data reveals that among those felons

\\server05\productn\C\CRY\47-2\CRY210.txt

332

unknown

Seq: 6

1-MAY-09

10:08

BLUMSTEIN & NAKAMURA

who stayed free of crime for 10 years after the original conviction, only 3.3
percent were reconvicted within the next 10 years (Community Legal Services, Inc., 2005).
Numerous other studies have shown that recidivism occurs relatively
quickly. However, little attention has been paid to the smaller population
of ex-offenders who stay crime free for an extended period of time.
Recent papers by Kurlychek and her colleagues have shed some light on
the population characterized by long-time avoidance of crime (Kurlychek,
Brame, and Bushway, 2006, 2007). Examining the hazard rate, they show
that the risk of offending for those with criminal records converges toward
the risk for those without a record as substantial time passes.
Kurlychek, Brame, and Bushway (2006) used the longitudinal data from
the Second Philadelphia Birth Cohort Study (Tracy, Wolfgang, and Figlio,
1990). The major advantage of such longitudinal samples is that they have
a representative population of nonoffenders, which makes it possible to
compare the hazard of those with records to those without. However, longitudinal samples are often limited in size, and the follow-up may not be as
complete as one desires.

MEASURES OF REDEMPTION
Although past wrongdoings are a useful sign of future trouble, this
information has decreasing value over time because the risk of recidivism
decreases monotonically with time clean. Thus, there can be a point at
which we can be confident that redemption has occurred, where the risk of
reoffending has subsided to the level of a reasonable comparison group.
The problem here is that little empirical information exists that can help to
establish that point. The absence of reliable empirical guidelines leaves
employers no choice but to set their own arbitrarily selected cutoff points
based on some intuitive sense of how long is long enough—inevitably with
a conservative bias.7 Given the importance of this issue, particularly for
those individuals with other employment vulnerabilities, it becomes
7.

For example, the Transportation Security Administration requires maritime
workers to obtain a Transportation Worker Identification Credential (TWIC) to
access secure areas of port facilities. Individuals are disqualified from getting a
TWIC if they have been convicted for certain disqualifying criminal offenses
within 7 years of the TWIC application (Transportation Security Administration,
n.d.). To the best of our knowledge, the choice of the cutoff points is arbitrary
and not based on any empirical analysis. Although 7 years seems to be a common
restorative period, perhaps based on a view that 5 years is too short and 10 years
is too long, some evidence exists that the cutoff points set by users of criminal
records could be much larger or could be “indefinite” (Carey, 2004: 50). The Fair
Credit Reporting Act states that a vendor of criminal history records may not
report arrest information that is older than 7 years (Hinton, 2004).

\\server05\productn\C\CRY\47-2\CRY210.txt

unknown

Seq: 7

CRIMINAL BACKGROUND CHECKS

1-MAY-09

10:08

333

important to develop empirical estimates of a reasonable point of
redemption.
One such point, which we denote as T*, is where the recidivism risk
declines and crosses the level of the general population of the same age,
and so it can serve as a point of redemption. These data can help an
employer who has selected a job applicant for a position and wants to
compare that individual’s risk of arrest with someone of the same age from
the general population. The crossover occurs because the general population includes people who have no criminal records as well as people who
have multiple arrests.
Now, suppose an employer has multiple job applicants for a position,
and a background check is run on all applicants. Those with no prior
record (whom we designate as the “never arrested”) are inherently less
risky than those with a prior record, but that difference can diminish with
the amount of time the individual with a prior arrest stays clean. This provides another point of redemption, when the recidivism risk of an individual with a criminal record is “sufficiently close” to one without, and we
designate that point as T**. T** should be larger than T* because the comparison group (the never arrested) are less risky than the general
population.
It is reasonable to expect that T* and T** will vary with the crime type
of the earlier arrest, which is denoted as C1. Recidivism studies have
shown that the crime type for which state prisoners were released was
related to recidivism rates (Beck and Shipley, 1997; Langan and Levin,
2002). Prisoners who were released for “crimes for money,” such as burglary, robbery, larceny, and motor vehicle theft, had the highest recidivism
rates in both studies. T* and T** could also vary with the age of the prior
arrest, which is denoted as A1, and criminological research consistently
indicates that an earlier onset age is a good predictor of a serious criminal
career, which is characterized by a larger number of offenses and a longer
career duration (Blumstein et al., 1986; Farrington et al., 1990; Farrington
et al., 2003; Piquero, Farrington, and Blumstein, 2007). Because a prior
record of violence, especially at younger ages, predicts more serious and
chronic offending (Elliott, 1994; Farrington, 1991; Piquero, Farrington, and
Blumstein, 2007), recidivism risk is expected to be higher for those whose
early arrest was for violence (Piper, 1985).
Age and crime type of the prior arrest also should be taken into account
in estimating T* and T** because the information about these factors usually appears in the criminal background reports that employers obtain, and
so the information is available to be used in the hiring decision.
We are interested in developing estimates of T* and T** as a function of
these characteristics of the earlier record. This approach is related to the
more familiar approach of estimating recidivism probability. It is more

\\server05\productn\C\CRY\47-2\CRY210.txt

334

unknown

Seq: 8

1-MAY-09

10:08

BLUMSTEIN & NAKAMURA

complicated, however, because one must examine the record over an
appreciably longer period of time. In recidivism studies, it is usually sufficient to track individuals for as short as 5 years, because the large majority
of individuals who will recidivate will do so within the first several years
(e.g., Beck and Shipley, 1997; Langan and Levin, 2002). However, the estimation of T* and T**, particularly as a function of A1 and C1, requires
observation over a much longer interval, long enough for the recidivism
probability to become small enough. This process requires larger initial
samples than those used in past studies (Kurlychek, Brame, and Bushway,
2006, 2007) so that we can estimate the recidivism probability with sufficient precision after most of any initial cohort has already recidivated.

RESEARCH APPROACHES AND RESULTS
This section first introduces the data used in the analysis to estimate
hazard. It then describes the hazard estimation procedure. Next, an
approach to comparing redemption candidates with the general population and the resulting estimates of T* are discussed. Then, an approach to
comparing redemption candidates with those who have never been
arrested and the resulting estimates of T** are discussed.
DATA
Our research approach requires starting with criminal-history records
initiated long enough ago that we can be confident that after having been
free and clean of arrests, the individuals with those records have a low
residual risk of recidivism. On the other hand, we would like records from
a time when the computerization of rap-sheet information was sufficiently
advanced so that the computer records would provide an appropriate sample. Thus, we contacted the criminal-history repository in New York State
asking for a sample of individuals arrested for the first time as adults in
1980. This information provided an interval of 27 years to follow the individuals and assess their recidivism probabilities. It also provided a large
enough population to disaggregate into a reasonable number of interesting
crime types and age at first arrest and still have an adequate number of
individuals who have remained clean of crime 10, 20, and even 25 years
later.

\\server05\productn\C\CRY\47-2\CRY210.txt

unknown

Seq: 9

CRIMINAL BACKGROUND CHECKS

1-MAY-09

10:08

335

Over 88,000 individuals were recorded as experiencing their first arrest
in 1980 in New York State.8 From this total population, we focus on individuals whose age at first arrest was 16, 18, and 20 years.9 The crime types
we focus on here have to be sufficiently numerous, reasonably serious
[e.g., we avoid driving under the influence (DUI)], and less than permanently damaging (e.g., murder). Our analysis of T* focuses on three
offenses: robbery, burglary, and aggravated assault. This selection of
offenses is based on their prevalence in arrest records and on the fact that
a first adult arrest for one of them is relatively unlikely to lead to incarceration, and especially to a long period of incarceration, which would complicate our analysis.10 The analysis of T** uses two broader categories of
C1, violent and property, to have the larger sample sizes that are needed to
generate the desired precision in estimating hazard as t becomes large.
Table 1 provides the distribution of the sample by age and crime type at
first arrest.
ESTIMATION MODELS OF REARREST RISK
We model time to recidivism using survival analysis, which is a statistical
method developed to study the occurrence and the timing of events.
Because the methods are flexible and generic, they have been used for
studying a wide variety of events, such as deaths, marriages, cancer cures,
unemployment, militarized disputes, earthquakes, equipment failures, and
so on. Criminologists have long used the methods to study recidivism (e.g.,
Maltz, 1984; Schmidt and Witte, 1988; for review, see Chung, Schmidt, and
Witte, 1991). Our analysis uses hazard (or hazard rate) to examine the
8.

9.
10.

The data received include all individuals with an arrest recorded in the New York
State Division of Criminal Justice Services repository of criminal-history records.
There are other individuals with one or more arrests that were sealed but with no
unsealed arrests; these individuals were not included in the files we examined. In
a background check, these individuals would presumably appear as never
arrested. It is also possible that individuals with an initial arrest in 1980 that was
sealed before they had an opportunity for a second arrest after 1980, and then
appeared at a later time with an arrest that was not sealed; in that case, their
second arrest would have been recorded with a different ID number and would
not have been included in our 1980 sample. We were unable to link the two
components of such an individual’s records. This selection process dropped people whose arrest frequency (l) may have been relatively low from our sample;
thus, our hazard estimates may be somewhat higher than if they were included,
and that would also have made our T* and T** estimates somewhat higher.
In contrast to most other jurisdictions, New York considers 16-year-olds to be
“adults.”
Incarceration after the first arrest is very unlikely for our three crime types
except for robbery. About 10 percent of those who were arrested for robbery in
1980 went to prison. Excluding them does not change our findings in any important way.

\\server05\productn\C\CRY\47-2\CRY210.txt

336

unknown

Seq: 10

1-MAY-09

10:08

BLUMSTEIN & NAKAMURA

Table 1. Initial Sample Size, n by First Offense (C1) and
Age at First Arrest (A1)
First Offense

16

Age at First Arrest
18

20

Robbery
Burglary
Aggravated assault

937
1,956
230

382
763
205

197
387
169

Violent
Property

1,861
5,238

1,151
2,986

794
1,833

timing of redemption and how it varies with offense type and age at first
arrest.
Let T be the time until a new arrest. Hazard, h(t), is the conditional
probability of a new arrest given surviving without an arrest up to time t,
which can be written as follows:
h(t)=Pr(T=tT ≥ t) =

# of the 1980 sample who have a new arrest time period t
# of the 1980 sample who have not had a new arrest before t

(1)

This measure is precisely the quantity employers and others would use to
evaluate the offending risk of a person who has been revealed by the background check to have committed a crime t years ago and none since (Kurlychek, Brame, and Bushway, 2006).
In calculating h(t), we count a new arrest (after their initial arrest in
1980) for any offense type except DUI.11 Thus, for example, a new arrest
is marked when a person whose first arrest in 1980 was for burglary is
rearrested for burglary or for any nonburglary offense, other than DUI.
We estimate the hazard, given conditions at first arrest, namely the age
A1 and the crime type C1 of the first arrest. Figure 1a displays h(t) for A1 of
16, 18, and 20 years for C1 of burglary.12 Figure 1b shows h(t) for A1 = 18
years for C1 = robbery, burglary, and aggravated assault. As expected, h(t)
varies with A1 and C1. The hazard curves differ primarily in the first 10
years, with robbery tending to have the highest conditional rearrest
probability, whereas burglary and aggravated assault follow a similar,
11.

12.

In some cases, we find that an arrest is followed quickly by another arrest. We are
concerned that what seems to be a new “arrest” might be related to the same
crime event as the prior arrest (e.g., transfer to a different jurisdiction), so we
count an arrest as a new arrest only if it occurs at least 30 days after the prior
arrest.
To reduce random fluctuations, all hazard curves (h(t)) for t = 2 in figures 1 to 4
are smoothed using five-point smoothing, which is also known as a running mean
or a moving average with a window width of five (i.e., h’(t) = [h(t-2) + h(t-1) +
h(t) + h(t+1) + h(t+2)] / 5), and the horizontal axis begins at t = 2.

\\server05\productn\C\CRY\47-2\CRY210.txt

unknown

Seq: 11

1-MAY-09

CRIMINAL BACKGROUND CHECKS

10:08

337

Figure 1a. Hazard Rate h(t): Age 16, 18, 20 Burglary

Probability of Rearrest

.25
.20
.15
Age 16 Burglary
.10

Age 18 Burglary
Age 20 Burglary

.05
.00
2

4

6

8

10

12

14

16

18

20

Years Since First Arrest

Figure 1b. Hazard Rate h(t): Age 18 Robbery, Burglary,
and Aggravated Assault

Probability of Rearrest

.25
.20
.15
Age 18 Robbery
.10

Age 18 Burglary
Age 18 Aggravated
Assault

.05
.00
2

4

6

8

10

12

14

16

18

Years Since First Arrest

lower trend. Also, a younger A1 is associated with a higher hazard. These
patterns have important implications in estimating T*, to which we turn
next.
COMPARISON WITH THE GENERAL POPULATION
APPROACH
We are interested in finding T*, which is the value of t where the risk of
a new arrest matches the risk of arrest for the general population of the
same age. We estimate the risk of arrest for the general population by the
age–crime curve whose horizontal axis is age (A) and whose vertical axis is
the age-specific arrest rate of people of age A, which is the ratio of the
number of arrests of age A to the population of age A from a particular

\\server05\productn\C\CRY\47-2\CRY210.txt

338

unknown

Seq: 12

1-MAY-09

10:08

BLUMSTEIN & NAKAMURA

year. Instead of using the conventional age–crime curve, we construct a
progressive age–crime curve, in which the age-specific arrest probability
for those who were of age A1 in 1980 is calculated from the number of
arrests and the population of age A1 in 1980, the number of arrests and the
population of age (A1 + 1) in 1981, those of age (A1 + 2) in 1982, and so
on.13 As a result of the way this progressive age–crime curve is constructed, it takes into account the period effect, which is not observed in
the traditional age–crime curve.14 Here, we count arrests for any offense
except DUI, suspicion, and “other” offenses so that the range of offenses
for which an arrest is made is comparable with the range of offenses considered for a new arrest for redemption candidates.
The two curves, the hazard and the age–crime curve, are expected to
cross at T* years for two reasons. First, the age–crime curve includes,
among the larger population, those who were never arrested as well as
those who recently offended and thus have a reasonably high risk of reoffending. In contrast, the redemption candidates have been arrest free for
T* years, during which time the risk, or hazard rate, should have fallen
substantially.
RESULTS
Table 2 shows the values of T* by offense type at first arrest (C1) and
age at first arrest (A1).15 In general, reasonable differences in values of T*
are observed across offense types and ages at first arrest. Overall, those
who were arrested for robbery take the longest time, about 9 years for 16year-olds, about 8 years for 18-year-olds, and about 4 years for 20-yearolds, to be similar to their age cohorts from the general population in
13.

14.

15.

More generally, the value of the age–crime curve in year t after the first arrest of
persons of A1 in 1980 is given by the number of arrests of people of age (A1 + t)
in year (1980 + t) divided by the population of that age in that year. The sample
cohort is from New York, so the age–crime curve as a comparison is also from
New York. The number of arrests by age in New York is from the Uniform Crime
Reports (Federal Bureau of Investigation, 1981–2001; National Consortium on
Violence Research, April 10, 2008), and the population of New York State is
from the census (U.S. Census Bureau, 1996, 2000, 2007). Similar to how the hazard curves are smoothed, the age–crime curve is smoothed using three-point
smoothing.
The period effect could be of special importance in estimating T* for the redemption candidates, who were first arrested in 1980 and were followed thereafter.
The late 1980s through the early 1990s witnessed a significant increase in the
violent crime rate (Blumstein and Wallman, 2006). Because the redemption candidates arrested in 1980 were experiencing an anomalously high crime period,
their progressive age–crime curve incorporates the period effect, which accounts
for the nonmonotonic decline with age.
The values of T* are calculated as the intersection of the smoothed age–crime
curve and the smoothed hazard by linear interpolation.

\\server05\productn\C\CRY\47-2\CRY210.txt

unknown

Seq: 13

1-MAY-09

CRIMINAL BACKGROUND CHECKS

10:08

339

terms of the probability of an arrest. Interestingly, the probability of a new
arrest at T* is relatively consistent across different ages at first arrest and
offense types, being close to .10.

Table 2. Values of T* by C1 and A1 (Arrest Probability at
T* in Brackets)

16

Age at First Arrest
18

20

8.5 (.103)
4.9 (.105)
4.9 (.105)

7.7 (.096)
3.8 (.097)
4.3 (.098)

4.4 (.086)
3.2 (.086)
3.3 (.086)

First Offense
Robbery
Burglary
Aggravated assault

Across each of the crime types, the youngest A1 is associated with the
largest value of T*. This result is consistent with general findings in criminology that younger starters persist longer in their criminal careers
(Piquero, Farrington, and Blumstein, 2007). Also, the magnitude of T* is
consistently largest for robbery but varies with A1 for burglary and aggravated assault.
For illustrative purposes, figures 2a and 2b show hazard curves for three
conditions: (figure 2a: C1 = Robbery, A1 = 18) and (figure 2b: Burglary, 16)
and the corresponding progressive age–crime curves (i.e., for the t years
after the first arrest) and the resulting intersection, T*.

Figure 2a. Comparison with Age–Crime Curve: Age 18
Robbery
T* = 7.7, h(T*) = .096

Probability of Rearrest

.25
.20
.15
Age 18 Robbery

.10

General Population
(Age 18 in 1980)

.05
.00
2

4

6

8

10

12

Years Since First Arrest

14

16

18

\\server05\productn\C\CRY\47-2\CRY210.txt

340

unknown

Seq: 14

1-MAY-09

10:08

BLUMSTEIN & NAKAMURA

Figure 2b. Comparison with Age–Crime Curve: Age 16
Burglary
T* = 4.9, h(T*) = .105

Probability of Rearrest

.25
.20
.15
Age 16 Burglary

.10

General Population
(Age 16 in 1980)

.05
.00
2

4

6

8

10

12

14

16

18

20

Years Since First Arrest

COMPARISON WITH THE “NEVER ARRESTED”
Our previous analysis estimated T* as a point of redemption by comparing people with a prior record who have stayed clean with members of the
general population of the same age. In contrast to T*, which can be calculated as an intersection of two curves, a comparison with the never
arrested inherently involves more complex choices. Because the risk of
rearrest for a redemption candidate might be expected to approach, but
not cross, the risk of arrest for the never arrested, it becomes a matter of
having to assess when the two curves are “close enough.”
APPROACH
Approximating the hazard of the never arrested. Information about the
never arrested is not directly available in any repository-based data set
that contains records of only those who have been arrested.16 One
approach to estimating the hazard of the never arrested involves using the
1980 age distribution of New York and the age distribution of 1980 firsttime arrestees. Assuming stationarity as in estimation of life tables, we can
approximate the population of the never arrested at age A (Pna(A)) as
follows:17
Pna(A) = Population of NY of age A in 1980 – S (first-time arrestees in
1980 for all A1 < A).
16.

17.

Kurlychek, Brame, and Bushway (2006, 2007) pursued this issue using cohort
data sets, but such data sets are often too limited for estimating hazard rates for
the small fraction of individuals with a prior arrest who remain clean for a reasonable time.
We only consider arrests at adult ages in NY (A1 = 16).

\\server05\productn\C\CRY\47-2\CRY210.txt

unknown

Seq: 15

1-MAY-09

CRIMINAL BACKGROUND CHECKS

10:08

341

As a result, the hazard of the never arrested at age A, (hna(A)), is calculated as follows:18
hna(A) =

Number of first-time arrestees for A1 = A

(2)

Pna(A)

Figure 3 displays our estimate of hna(A). It is evident that the younger
ages are associated with higher risk of arrest, but even at age 16, the hazard is less than .03, which is clearly much lower than the risk of rearrest of
those with a prior arrest. We can now compare the hazard of redemption
candidates whose first arrest occurs at age A1, h(t), with the hazard of the
never arrested, hna(t=A−A1).

Figure 3. Hazard of the Never Arrested, hna(A)

Probability of Arrest

.035
.030
.025
.020
.015
.010
.005
.000
16

18

20

22

24

26

28

30

32

34

36

38

Age

Determining “close enough.” We designate as T** the point when the
hazard of an individual with a criminal record h(t) is sufficiently close to
that of one without. Figure 4 shows h(t) for A1 = 18 for C1 = property
crimes and violent crimes, as well as hna(t).19 We first note that h(t)
declines considerably faster than hna(t). However, aside from random fluctuations, h(t) comes very close to hna(t) but remains above it, even at t >
18.

19.

For the same reason as discussed in footnote 8, hna(A) may be higher if the individuals who have one or more arrests that were sealed but with no unsealed
arrests were included.
The comparison with the risk of the never arrested is particularly sensitive to the
diminished sample size, so we use two broad categories of C1, violent and property crimes. Here, violent crimes are designated to include robbery, aggravated
assault, forcible rape, and simple assault. Murder and non-negligent manslaughter are not included as C1 because special conditions are likely to apply to their
redemption. Property crimes are designated to include burglary, larceny, motor
vehicle theft, stolen property, fraud, forgery, and embezzlement.

\\server05\productn\C\CRY\47-2\CRY210.txt

342

unknown

Seq: 16

1-MAY-09

10:08

BLUMSTEIN & NAKAMURA

20. Given these observations, our question is when the redemption candidate’s risk is deemed “close enough” to that of the never arrested.
Our approach invokes the use of confidence intervals. Using the estimated risk of the never arrested hna(t), we estimate T** as the lowest value
of t such that the upper bound of the confidence interval of h(t) becomes
smaller than or equal to (hna(t)+δ), where δ represents a risk difference
that an employer is willing to tolerate.20, 21
20.

This approach of estimating T** is equivalent to carrying out a hypothesis test
where the null hypothesis states that the difference between h(t) and hna(t) is
greater than δ. The alternative hypothesis is that the difference is less than or
equal to δ. Thus, it is of the following form:
H0:h(t)>d+hna(t) versus H1:h(t)≤d+hna(t).
We would reject the null hypothesis at t = T**, where the upper bound of the
confidence interval of h(t) first intersects (hna(t) + δ). This approach is motivated
by the literature on (bio)equivalence tests where the studies want to show that
the effectiveness of new treatments (drugs, vaccines, diagnoses, etc.) is no worse
than the standard, existing treatment by a specified margin (e.g., Barker et al.,
2001; Westlake, 1976).
Our approach is different from the more familiar approach of determining
whether h(t) is “close enough” to hna(t) by carrying out a hypothesis test with the
null hypothesis stating h(t) is equal to hna(t) and concluding that h(t) is “statistically indistinguishable” from hna(t) when we fail to reject the null hypothesis. This
corresponds to constructing confidence intervals around h(t) and denoting T** as
the intersection of the lower bound of the confidence interval of h(t) with hna(t).
However, smaller sample sizes inevitably make confidence intervals wider, which
reflects the larger uncertainty of the estimates. If T** were estimated using the
lower bound of the confidence interval of h(t), then wider confidence intervals
would lead inappropriately to smaller values of T**, possibly producing unreasonable values of T** less than T*. By introducing δ and using the upper bound
of the confidence interval, our approach circumvents this shortcoming.
The conventional standard error of h(t) can be calculated by the formula
[( h ( t ) ∗ (1 − h ( t )] / n ( t ) . However, this formula relies on the asymptotic normality of the estimate of h(t). Because the sample sizes defining h(t) become small
when t is large, the standard errors calculated by the formula above are questionable. Moreover, in this case, the symmetric confidence intervals can include negative lower endpoints, which are a problem of “overshoot” (Newcombe, 1998);
because we are trying to estimate the proportion of those who are rearrested at t,
those estimates have to be bounded between 0 and 1, and so cannot go negative.
The standard confidence interval of a proportion (often referred to as the Wald
interval) is also known to show erratic behaviors in terms of the coverage
probability, regardless of sample sizes and the values of h(t) (Brown, Cai, and
DasGupta, 2001). Given the limitations of the Wald interval for h(t), we use the
statistical method of “bootstrap.” The bootstrap provides a reliable method to
estimate the uncertainty of an estimator via resampling, without relying on the
asymptotic properties of the estimator. We used the bias-corrected and accelerated (BCa) bootstrap intervals for h(t), with the number of bootstrap samples, B
= 2001 (Efron, 1987; Efron and Tibshirani, 1993; Wu, 1989). Confidence intervals
can be constructed using methods other than the bootstrap (Brown, Cai, and
DasGupta, 2001; Newcombe, 1998).

\\server05\productn\C\CRY\47-2\CRY210.txt

unknown

Seq: 17

1-MAY-09

CRIMINAL BACKGROUND CHECKS

10:08

343

Figure 4. Comparison with the Never Arrested (Age 18
Violent, Property)

RESULTS
Suppose that an employer can accept δ = .05, whereby a redemption
candidate’s hazard can be as much as .05 higher than the hazard of a never
arrested person of the same age. Then we estimate T** = 4.8 for C1 =
property and T** = 8.0 for C1 = violent (both for A1 = 18) using the 95
percent confidence interval. The more tolerant an employer is (larger
value of δ), the shorter the redemption time (smaller value of T**). Figure
5 shows this tradeoff between δ and T** for three different conditions of
the first arrest. Violent offenders have consistently higher values of T**
than property offenders, indicating that violent offenders need to stay
clean longer for the same risk-tolerance difference. It also demonstrates
that a younger A1 is associated with a longer time necessary for property
offenders to be comparable with the never arrested of the same age at a
given tolerance level δ.22
For the employer who is more accepting of risk and willing to focus on
the intersection of the two hazard curves (h(t) and hna(t)), the values of
21.
22.

Alternatively, an employer can formulate the risk tolerance as a risk ratio (or a
relative risk) of h(t) to hna(t).
Another approach to comparing redemption candidates with the never arrested
is to recognize that the comparison need not be of two candidates of the same
age. Because the hazard declines with age, younger never arrested individuals
may exist whose hazard is no less than that of an older individual with a prior
arrest but who has stayed clean for a long period. It could also be the case that,
based on some existing base rates for workplace deviant behaviors (e.g., Bachman, 1994; Slora, 1989), some employers might have a specific risk level, g, below
which the risk is tolerable or acceptable for the purpose at hand (e.g., a particular
job position in a particular industry).

\\server05\productn\C\CRY\47-2\CRY210.txt

344

unknown

Seq: 18

1-MAY-09

10:08

BLUMSTEIN & NAKAMURA

T** at the same value of δ (.05) and A1 (18) would be 4.2 years for property and 7.0 years for violence compared with 4.8 years for property and
8.0 years for violent using the above approach that employs the upper
bound of the confidence interval. The values of T** based on the intersection of the hazards are lower than those for the conservative employer
who wants high confidence that the candidate represents a low risk.

Figure 5. Tradeoff between δ and T** (Based on the Upper
Bound of the Confidence Interval of h(t))

ISSUES STILL TO BE ADDRESSED
We believe that our results represent a significant step forward in an
area where so little is known empirically about the redemption process. As
usual, however, some important efforts remain. We have identified T* as
the minimum duration of time clean in New York State for the recidivism
probability to drop below the norm for New Yorkers of the same age. We
have also identified approaches to estimating T**, when the recidivism
probability falls below any specified level compared with people never
arrested. It is possible, however, that an individual who stayed clean in
New York was arrested in another state. Thus, our estimates are lower
bounds on T* (and T**) and the associated recidivism probability. One
study on the recidivism of prisoners estimated that 7.6 percent of the
released prisoners were rearrested out of state (Langan and Levin, 2002).
Another finds that, among the prisoners who were released from 11 state
prisons in 1983, roughly 10 percent of them have out-of-state arrests within
3 years of their release (Orsagh, 1992). To address this concern about
mobility, we have approached the Federal Bureau of Investigation, which
maintains a national index of rap-sheet records in the Interstate Identification Index. We can present them with identification information of the

\\server05\productn\C\CRY\47-2\CRY210.txt

unknown

Seq: 19

CRIMINAL BACKGROUND CHECKS

1-MAY-09

10:08

345

individuals who have stayed clean in New York and should be able to
obtain information on their arrests elsewhere in the nation. That method
will raise the h(t) curve somewhat and so increase the value of T* and T**.
The correction could be reasonably large for a state like New York, where
the large fraction of offenders from New York City could easily commit
other offenses in a neighboring state.23 We would anticipate that the correction would be appreciably less in a state like California, where the
major metropolitan areas are much more remote from neighboring states.
A second issue that warrants additional analysis is the distinction
between arrest information and conviction information. In many settings,
it is considered either inappropriate or illegal to ask about an arrest record
in the absence of a following conviction.24 We intend to pursue this analysis using only conviction information. Of course, the initial sample will
become smaller because many of our arrests were not followed by convictions. But of those convicted, we would anticipate that T* and T** would
be larger, because people who were convicted (“true” offenders) would be
more likely to have subsequent arrests than those who were acquitted or
whose cases were dismissed (“ambiguous” offenders).
Another small correction should be made for time in custody. The estimation of hazard assumes that the entire initial sample of arrestees is at
risk of an additional arrest shortly after their prior arrest. However, those
who are incarcerated as a result of the first arrest are at risk of a new
arrest only after their release from incarceration. Thus, the estimation
needs to be adjusted for the incarceration time. Unfortunately, identifying
correction information from rap sheets is not easy. However, considering
that the 1980 arrest is the first arrest for the sample of arrestees, it is not
likely that any lengthy period of incarceration follows that first arrest.
It is possible that conditions in New York are distinctively different from
other states or that offenders first arrested in 1980 were different from
those arrested more recently, so it is important that we generate robustness tests of the findings presented here. That approach will include collecting data from multiple states to examine how patterns of redemption
vary across the states. This analysis will provide an opportunity to look
across the states to determine whether their offending patterns or their
arrest patterns differ. We also intend to take subsequent draws of people
whose first arrest occurred in 1985, 1990, and 1995. These samples will
have a shorter observation period, especially for the 1995 sample, but we
23.
24.

We anticipate that the younger arrestees we focus on are less mobile than older
counterparts and thus are less likely to have out-of-state records.
According to the guidelines published by the Equal Employment Opportunity
Commission (EEOC), employers may not deny employment based on arrests
that did not lead to convictions unless there is a business justification (EEOC,
1990).

\\server05\productn\C\CRY\47-2\CRY210.txt

346

unknown

Seq: 20

1-MAY-09

10:08

BLUMSTEIN & NAKAMURA

anticipate that what we lose in observation time will be more than compensated with the richer quality of the records as we move into more contemporary computerization of records. Since the National Criminal
History Information Program was initiated in 1995 to improve the quality
of criminal-history records in state repositories, it is expected that we will
observe increased accuracy and completeness in the criminal history of the
1995 or even the 1990 sample. Also, if our estimates of T* roughly persist,
then a 10-year observation interval should be adequate.25 Examining multiple cohorts of arrestees will also allow us to generate information on time
trends in arrest patterns and in recidivism patterns as well as information
on any period effect.

POLICY IMPLICATIONS
The information and approaches we have generated here should be of
considerable value in enhancing redemption opportunities and consequent
employment opportunities for individuals who made a mistake in the past
but have since lived a lawful life. The knowledge of T* and T** could be
used in many ways by various pertinent parties to facilitate the redemption
process.
USERS OF CRIMINAL RECORDS
EMPLOYERS
Employers who run background checks on job applicants could be given
a brief document informing them of the diminished value of records older
than T* or T** years for risk assessment purposes.26 Because employers
have a strong concern about liability suits, a statute could protect them
from such due-diligence vulnerability in case they hire someone whose last
arrest was longer ago than T* or T**.27 This would be a relief for employers who are otherwise willing to hire individuals with criminal records, and
it would add to the existing incentives such as Work Opportunity Tax
Credit (WOTC) and Federal Bonding Program (FBP).28
25.

26.

27.

28.

Depending on the approach to estimating T** and the choice of δ and g, the
values of T** could be as large as 25 years. Thus, the desired estimation of T**
might not be possible from a 1990 sample, especially from a 1995 sample.
Users of background checks should base their decision not only on the information about criminal history but also on information about other important factors
(such as employment history, marriage, and educational attainment).
Although such legal protections would most likely be welcomed by employers,
their concern over possible damage to the organization’s reputation would not be
eliminated (Fahey, Roberts, and Engel, 2006).
For more details about WOTC, see http://www.doleta.gov/business/incentives/
opptax/. More details about FBP are available at http://www.bonds4jobs.com/
index.html.

\\server05\productn\C\CRY\47-2\CRY210.txt

unknown

Seq: 21

CRIMINAL BACKGROUND CHECKS

1-MAY-09

10:08

347

Such liability-protection statutes could also be applicable to employers
that ask applicants about their criminal background, but it would limit
their inquiries to criminal involvements that occur within the last T* or
T** years. This statute would be relevant to the concerns of the “ban the
box” movement.29
PARDON BOARDS
The governor of each state is empowered to grant a pardon as an act of
clemency and forgiveness. Most typically, a pardon board reviews relevant
information about the individual seeking clemency and makes a recommendation to the governor. Although the length of the law-abiding period
is considered one of the most important factors in pardon applications, it is
not clear whether pardon boards have reliable guidelines as to how long a
law-abiding period should be for the individual to be deemed appropriate
for pardon.30 Although pardons are hard to obtain, especially for the poor,
pardons have a significant restorative effect that signals that the pardoned
individual is rehabilitated (Love, 2003).
DISTRIBUTORS OF CRIMINAL RECORDS
REPOSITORIES
State record repositories could adopt a policy not to disseminate criminal record information older than T* or T** years. This regulation could
apply specifically to the states that make their criminal-history information
publicly available on the Internet.31 States are clearly moving in the direction of making individual criminal records more publicly accessible
(Jacobs, 2006). However, given the lasting consequence of disseminated
records on a large number of individuals, finding means to limit the dissemination would be a realistic approach to the problem.32 The state could
29.

30.

31.

32.

The “box” refers to a question on job applications that asks prospective
employee whether they have ever been convicted of a crime. So far, the movements to “ban the box” have been largely limited to employment for city governments (Henry and Jacobs, 2007; National Employment Law Project, 2008).
For example, in Pennsylvania, the Board of Pardons (2005: 1) publicly states that
the length of time free of crime after the offense is one of the best indicators of
rehabilitation that the applicant can demonstrate.
In 2001, 13 states (of the 38 that responded to the survey) provide public access
to criminal history records through the Internet (SEARCH, 2001). [Samuels and
Mukamal (2004) report that 28 states allow Internet access to criminal records.]
Some employers might “statistically discriminate” based on correlating individual
characteristics of a job applicant with generic covariates of criminal activity such
as race and ethnicity. As a result, limiting employers’ access to criminal records
could possibly have an adverse consequence for those without criminal records
(Bushway, 2004; Finlay, in press; Freeman, 2008; Holzer, Raphael, and Stoll,
2006; Pager, 2003; Raphael, 2006).

\\server05\productn\C\CRY\47-2\CRY210.txt

348

unknown

Seq: 22

1-MAY-09

10:08

BLUMSTEIN & NAKAMURA

adopt a policy to seal repository records of events older than T* or T**
years in response to a request from a non-criminal justice agency. Such
sealed records could still be accessible for criminal justice purposes. A
more aggressive approach would be to expunge records older than T* or
T** years.
Even though these judicial procedures tend to be more accessible and
reliable than pardon, the popularity of sealing and expungement peaked in
the 1970s and has severely declined since then in most jurisdictions (Love,
2003, 2006). Moreover, Love (2003, 2006) reports that no one standard
exists in terms of what it means to have a record sealed, expunged, set
aside, vacated, or annulled. A record being expunged does not necessarily
mean that the record is literally destroyed; rather, the expunged records
“almost always remain available for use by law enforcement agencies and
the courts, and in some states they may be accessible to other public agencies and even to private investigative services hired to perform criminal
background checks for employers” (Love, 2003: 121). Furthermore, critics
of sealing and expungement argue that the concealment of records and the
denying of past wrongdoing are institutionalized deception and are not
compatible with the pursuit of truth, the foundation of a legal system
(Franklin and Johnsen, 1981; Kogon and Loughery, 1970).
Despite these criticisms, concealment and denial of criminal records
after some “rehabilitation period” are common in many countries. For
instance, in the United Kingdom, according to the Rehabilitation of
Offenders Act 1974, those who are convicted of certain crimes, after specified rehabilitation periods, are treated as though the crime never happened, and are not obligated to reveal the record when asked at
employment settings.33, 34
COMMERCIAL VENDORS
Because many employers rely on background-check services provided
by commercial vendors of criminal records, if states seal or expunge
records older than T* or T** years, this policy should be accompanied by a
process of requiring those old records also to be erased from commercial
databases.35
33.

34.
35.

The Rehabilitation of Offenders Act of 1974 followed a report called Living It
Down: The Problem of Old Convictions, which is a report of a committee chaired
by Lord Gardiner (1972). The report shows that the longer a convicted person
remains crime free, the less likely that the person will commit another crime.
For more on the sealing and expungement of criminal records in the European
Union, see Loucks, Lyner, and Sullivan (1998).
Given the discrepancy between the records from official sources (state repositories) and the records from commercial databases (Bushway et al., 2007), it is
important that any update (i.e., sealing or expungement) that takes place on the

\\server05\productn\C\CRY\47-2\CRY210.txt

unknown

Seq: 23

CRIMINAL BACKGROUND CHECKS

1-MAY-09

10:08

349

CERTIFICATES OF REHABILITATION
The main criticisms of sealing and expungement include the compromise of governmental transparency as well as the possible adverse effect
on nonoffenders because of statistical discrimination.36 Certificates of
rehabilitation and other similar means can circumvent the problem. Certificates of rehabilitation are designed to remove certain collateral consequences for eligible ex-offenders and can potentially enhance their
employment prospects.37 The certificates reward good behavior of exoffenders by explicitly acknowledging them as being rehabilitated rather
than erasing the record of their contact with the criminal justice system.
Thus, these certificates are similar to pardons in spirit but are relatively
more accessible than pardons. Currently, only a handful of states issue
such certificates (Love and Frazier, 2006; Samuels and Mukamal, 2004),
but they could be used more widely by taking advantage of the empirical
evidence of T* and T**.38

SUMMARY
As background checking has become a routine practice for many
employers, and an increasing number of criminal records have become
electronically accessible, those who made a mistake many years ago but
have since lived a law-abiding life face hardships in finding employment.
The risk of recidivism declines with time clean, so we know that a person
who has stayed clean for an extended period of time must be of low risk.
The question is the extent to which the risk drops over time, and at what
point in time the risk is deemed low enough. This article addresses these
questions by examining the hazard of those who were first arrested in 1980

36.
37.

38.

official records is reflected on the records in the commercial sources. Jacobs and
Crepet (2008) highlight the difficulty in forcing vendors to make such changes
because their right to access the criminal records would be protected by the First
Amendment of the Constitution.
See footnote 32.
Criminal history records are regarded as “negative credentials” that signify
“social stigma and generalized assumptions of untrustworthiness or undesirability” (Pager, 2007: 33; see also Jacobs, 2006 and Jacobs and Crepet, 2008), whereas
certificates of rehabilitation attempt to emphasize the progress made by the exoffender. Regarding more fair representation of riskiness by taking into account
the positive factors, Bushway et al. (2007) mention that it is conceivable for the
government to devise some score (like a credit score) that indicates the risk of
offending, which can be affected by positive factors such as the length of crimefree time, completion of a drug treatment program, and vocational training, as
well as negative factors such as committing another crime (for a similar
approach, see Freeman, 2008).
Bushway and Sweeten (2007) discuss policy implications regarding the diminished value of old criminal records in the context of collateral consequences.

\\server05\productn\C\CRY\47-2\CRY210.txt

350

unknown

Seq: 24

1-MAY-09

10:08

BLUMSTEIN & NAKAMURA

and by estimating a point of redemption by comparing the hazard of
redemption candidates with 1) the risk of an arrest for individuals of the
same age in the general population and 2) with the risk of an arrest for
those who have never been arrested.
The results indicate that the risk indeed declines monotonically over
time and, after some point T*, becomes lower than the risk of arrest of
someone of the same age in the general public represented by the
age–crime curve. The article also produced reasonable approaches to estimating another measure of redemption, T**, which is the number of years
that those who have a prior arrest need to stay clean to be considered
“close enough” to those who have never been arrested. The results also
demonstrate that T* and T** vary with age and crime type of the earlier
arrest. Younger starting age generally points to a longer time necessary to
become comparable with a person of the same age from the general population. We find that violent offenders have to wait longer than property
offenders to meet the same criterion of redemption. Because the information about age and crime type of the earlier arrest is usually available on
criminal background reports that employers and other users of criminal
records obtain, it is important that T* and T** are estimated as a function
of these two factors.
The findings have several important policy implications; they are helpful
in informing two broad categories of entities as follows: those who are in a
position to disseminate criminal-history information (i.e., state repositories
and commercial vendors of criminal records) and those who are responsible for determining the relevance of criminal records (i.e., judges, pardon
boards, and employers). All of the policy approaches discussed could be
considered by the respective entities, but using any of them requires information and judgment about the relevant values of T* or T**.
As we outlined in the section on future research plans, this research is
clearly ongoing. Because currently no empirical basis exists for knowing
about the variability of T* and T** across states or across time, we will be
conducting similar analyses on data from other states and on data from
other arrest cohorts. This approach will allow us to be in a better position
to provide more complete and robust information about T* and T**. In
the meantime, even the preliminary estimates in this article should be
helpful in moving the policy process forward.

REFERENCES
Bachman, Ronet. 1994. Violence and Theft in the Workplace. Washington,
DC: U.S. Department of Justice, Bureau of Justice Statistics.

\\server05\productn\C\CRY\47-2\CRY210.txt

unknown

Seq: 25

CRIMINAL BACKGROUND CHECKS

1-MAY-09

10:08

351

Barada, Paul W. 1998. Exploding the court check myth. HR Magazine
43:16–9.
Barker, Lawrence, Henry Rolka, Deborah Rolka, and Cedric Brown.
2001. Equivalence testing for binomial random variables: Which test to
use? The American Statistician 55:279–87.
Beck, Allen J., and Bernard E. Shipley. 1997. Recidivism of Prisoners
Released in 1983. Washington, DC: U.S. Department of Justice, Bureau
of Justice Statistics.
Blumstein, Alfred, Jacqueline Cohen, Jeffrey A. Roth, and Christy A.
Visher. 1986. Criminal Careers and “Career Criminals.” Washington,
DC: National Academy Press.
Blumstein, Alfred, David P. Farrington, and Soumyo Moitra. 1985. Delinquency careers: Innocents, desisters, and persisters. In Crime and Justice: An Annual Review of Research, vol. 6, eds. Michael H. Tonry and
Norval Morris. Chicago, IL: University of Chicago Press.
Blumstein, Alfred, and Joel Wallman. 2006. The recent rise and fall of
American violence. In The Crime Drop in America, eds. Alfred Blumstein and Joel Wallman. New York: Cambridge University Press.
Brame, Robert, Shawn D. Bushway, and Raymond Paternoster. 2003.
Examining the prevalence of criminal desistance. Criminology
41:423–48.
Brown, Lawrence D., Tony Cai, and Anirban DasGupta. 2001. Interval
estimation for a binomial proportion. Statistical Science 16:101–33.
Bureau of Justice Statistics. 2006. Improving Criminal History Records for
Background Checks, 2005. Washington, DC: U.S. Department of Justice, Bureau of Justice Statistics.
Bushway, Shawn D. 1998. The impact of an arrest on the job stability of
young white American men. Journal of Research in Crime and Delinquency 35:454–79.
Bushway, Shawn D. 2004. Labor market effects of permitting employer
access to criminal history records. Journal of Contemporary Criminal
Justice 20:276–91.
Bushway, Shawn D., Shauna Briggs, Faye S. Taxman, Meridith Thanner,
and Mischelle Van Brakle. 2007. Private providers of criminal history
records: Do you get what you pay for? In Barriers to Reentry? The
Labor Market for Released Prisoners in Post-Industrial America, eds.

\\server05\productn\C\CRY\47-2\CRY210.txt

352

unknown

Seq: 26

1-MAY-09

10:08

BLUMSTEIN & NAKAMURA

Shawn Bushway, Michael A. Stoll, and David F. Weiman. New York:
Russell Sage Foundation.
Bushway, Shawn D., and Gary Sweeten. 2007. Abolish lifetime bans for
ex-felons. Criminology & Public Policy 6:697–706.
Carey, Corinne. 2004. No Second Chance: People With Criminal Records
Denied Access to Public Housing. New York: Human Rights Watch.
Christensen, Ronald. 1967. Projected percentage of U.S. population with
criminal arrest and conviction records. In The President’s Commission
on Law Enforcement and Administration of Justice, Task Force Report:
Science and Technology, Appendix J. Washington, DC: U.S. Government Printing Office.
Chung, Ching-Fan, Peter Schmidt, and Ann D. Witte. 1991. Survival analysis: A survey. Journal of Quantitative Criminology 7:59–98.
Community Legal Services, Inc. 2005. Declaration of Jeffrey Fagan, Ph.D.
http://www.clsphila.org/fagan_declaration.htm.
Efron, Bradley. 1987. Better bootstrap confidence intervals. Journal of the
American Statistical Association 82:171–85.
Efron, Bradley, and Robert J. Tibshirani. 1993. An Introduction to the
Bootstrap. New York: Chapman and Hall.
Elliott, Delbert S. 1994. Serious violent offenders: Onset, developmental
course, and termination—The American Society of Criminology 1993
presidential address. Criminology 32:1–21.
Equal Employment Opportunity Commission. 1990. Notice No. N-915061. Policy guideline on the consideration of arrest records in employment decisions under title VII of the Civil Rights Act of 1964, as
amended, 42 U.S.C. Sec. 2000e et. seq. (1982).
Fahey, Jennifer, Cheryl Roberts, and Len Engel. 2006. Employment of ExOffenders: Employer Perspectives. Boston, MA: Crime and Justice
Institute. http://www.crjustice.org/cji/ex_offenders_employers_12-1506.pdf.
Farrington, David P. 1986. Age and crime. In Crime and Justice: An
Annual Review of Research, vol. 7, eds. Michael H. Tonry and Norval
Morris. Chicago, IL: University of Chicago Press.
Farrington, David P. 1987. Predicting individual crime rates. In Crime and
Justice: An Annual Review of Research, vol. 9, eds. Don M. Gottfredson and Michael H. Tonry. Chicago, IL: University of Chicago Press.

\\server05\productn\C\CRY\47-2\CRY210.txt

unknown

Seq: 27

CRIMINAL BACKGROUND CHECKS

1-MAY-09

10:08

353

Farrington, David P. 1991. Childhood aggression and adult violence: Early
precursors and later-life outcomes. In The Development and Treatment
of Childhood Aggression, eds. by Debra J. Pepler and Kenneth H.
Rubin. Hillsdale, NJ: Lawrence Erlbaum.
Farrington, David P., Darrick Jolliffe, J. David Hawkins, Richard F. Catalano, Karl G. Hill, and Rick Kosterman. 2003. Comparing delinquency
careers in court records and self-reports. Criminology 41:933–58.
Farrington, David P., Rolf Loeber, Delbert S. Elliott, J. David Hawkins,
Denise B. Kandel, Malcolm W. Klein, Joan McCord, David C. Rowe,
and Richard E. Tremblay. 1990. Advancing knowledge about the onset
of delinquency and crime. In Advances in Clinical Child Psychology,
vol. 13, eds. Benjamin B. Lahey and Alan E. Kazdin. New York:
Plenum.
Federal Bureau of Investigation. 1981–2001, 2008. Crime in the United
States, 1980-2000, 2007. Washington, DC: U.S. Department of Justice,
Federal Bureau of Investigation.
Finlay, Keith. In press. Effect of employer access to criminal history data
on the labor market outcomes of ex-offenders and non-offenders. In
Studies of Labor Market Intermediation, ed. David Autor. Chicago, IL:
University of Chicago Press.
Franklin, Marc A., and Diane Johnsen. 1981. Expunging criminal records:
Concealment and dishonesty in an open society. Hofstra Law Review
9:733–74.
Freeman, Richard. 2008. Incarceration, criminal background checks, and
employment in a low(er) crime society. Criminology & Public Policy
7:405–12.
Gardiner, Lord (Chairman). 1972. Living It Down: The Problem of Old
Convictions, The Report of a Committee set up by JUSTICE, The Howard League for Penal Reform, and the National Association for the Care
and Resettlement of Offenders. London, UK: Stevens.
Gottfredson, Don M. 1999. Effects of Judges’ Sentencing Decisions on
Criminal Careers. National Institute of Justice: Research in Brief.
Washington, DC: U.S. Department of Justice.
Grogger, Jeffrey. 1995. The effect of arrests on the employment and earnings of young men. Quarterly Journal of Economics 110:51–71.
Hahn, Jeffrey M. 1991. Pre-employment information services: Employers
beware? Employee Relations Law Journal 17:45–69.

\\server05\productn\C\CRY\47-2\CRY210.txt

354

unknown

Seq: 28

1-MAY-09

10:08

BLUMSTEIN & NAKAMURA

Harris, Patricia M., and Kimberly S. Keller. 2005. Ex-offenders need not
apply: The criminal background check in hiring decisions. Journal of
Contemporary Criminal Justice 21:6–30.
Henry, Jessica S., and James B. Jacobs. 2007. Ban the box to promote exoffender employment. Criminology & Public Policy 6:755–62.
Hinton, Derek. 2004. The Criminal Records Manual: The Complete
National Reference for the Legal Access and Use of Criminal Records.
Tempe, AZ: Facts on Demand Press.
Hirschi, Travis, and Michael R. Gottfredson. 1983. Age and the explanation of crime. American Journal of Sociology 89:552–84.
Holzer, Harry J., Steven Raphael, and Michael A. Stoll. 2003. Employer
demand for ex-offenders: Recent evidence from Los Angeles. Paper
presented at the Urban Institute Roundtable on Offender Re-Entry,
New York.
Holzer, Harry J., Steven Raphael, and Michael A. Stoll. 2004. Will
employers hire former offenders?: Employer preferences, background
checks, and their determinants. In Imprisoning America: The Social
Effects of Mass Incarceration, eds. Mary Patillo, David F. Weiman, and
Bruce Western. New York: Russell Sage Foundation.
Holzer, Harry J., Steven Raphael, and Michael A. Stoll. 2006. Perceived
criminality, criminal background checks, and the racial hiring practices
of employers. Journal of Law and Economics 49:451–80.
Jacobs, James B. 2006. Mass incarceration and the proliferation of criminal
records. University of St. Thomas Law Journal 3:387–420.
Jacobs, James B., and Tamara Crepet. 2008. The expanding scope, use, and
availability of criminal records. Journal of Legislation and Public Policy
11:177–213.
Kethineni, Sesha, and David N. Falcone. 2007. Employment and exoffenders in the United States: Effects of legal and extra legal factors.
Probation Journal 54:36–51.
Kogon, Bernard, and Donald L. Loughery, Jr. 1970. Sealing and expungement of criminal records—the big lie. The Journal of Criminal Law,
Criminology, and Police Science 61:378–92.
Kurlychek Megan C., Robert Brame, and Shawn D. Bushway. 2006. Scarlet letters and recidivism: Does an old criminal record predict future
offending? Criminology & Public Policy 5:483–504.

\\server05\productn\C\CRY\47-2\CRY210.txt

unknown

Seq: 29

1-MAY-09

CRIMINAL BACKGROUND CHECKS

10:08

355

Kurlychek Megan C., Robert Brame, and Shawn D. Bushway. 2007.
Enduring risk? Old criminal records and predictions of future criminal
involvement. Crime & Delinquency 53:64–83.
Langan, Patrck A., and David J. Levin. 2002. Recidivism of Prisoners
Released in 1994. Washington, DC: U.S. Department of Justice, Bureau
of Justice Statistics.
Loucks, Nancy, Olwen Lyner, and Tom Sullivan. 1998. The employment of
people with criminal records in the European Union. European Journal of Criminal Policy and Research 6:195–210.
Love, Margaret C. 2003. Starting over with a clean slate: In praise of a
forgotten section of the model penal code. Fordham Urban Law Journal 30:1705–41.
Love, Margaret C. 2006. Relief from the Collateral Consequences of a
Criminal Conviction: A State-by-State Resource Guide. Buffalo, NY:
Williams S. Hein & Co., Inc.
Love, Margaret C., and April Frazier. 2006. Certificates of rehabilitation
and other forms of relief from the collateral consequences of conviction: A survey of state laws. In Second Chances in the Criminal Justice
System: Alternatives to Incarceration and Reentry Strategies. Washington, DC: American Bar Association.
Maltz, Michael D. 1976. Privacy, criminal records, and information systems. In Operations Research in Law Enforcement, Justice, and Societal
Security, eds. Sidney H. Brounstein and Murray Kamrass. Lexington,
MA: D.C. Heath and Company.
Maltz, Michael D. 1984. Recidivism. Orlando, FL: Academic Press.
May, Bruce E. 1995. The character component of occupational licensing
laws: A continuing barrier to the ex-felon’s employment opportunities.
North Dakota Law Review 71:187–210.
Munro, Neil. 2002. The ever-expanding network of local and federal
databases. Communication of the ACM 45:17–9.
Nagin, Daniel S., and Joel Waldfogel. 1995. The effects of criminality and
conviction on the labor market status of young British offenders. International Review of Law and Economics 15:109–26.
National Consortium on Violence Research. n.d. FBI Uniform Crime
Reports Arrests Data Cube. http://saturn.heinz.cmu.edu/cognos/cgi-bin/
upfcgi.exe 9accessed April 10, 2008).

\\server05\productn\C\CRY\47-2\CRY210.txt

356

unknown

Seq: 30

1-MAY-09

10:08

BLUMSTEIN & NAKAMURA

National Employment Law Project. 2008. Major U.S. Cities Adopt New
Hiring Policies Removing Unfair Barriers to Employment of People
with Criminal Records. http://www.nelp.org/nwp/second_chance_labor
_project/citypolicies.cfm.
Newcombe, Robert G. 1998. Two-sided confidence intervals for the single
proportion: Comparison of seven methods. Statistics in Medicine
17:857–72.
Orsagh, Thomas. 1992. Multi-State Offender: A Report Concerning State
Prisoners Who Were Criminally Active in More than One State. Washington, DC: Bureau of Justice Statistics.
Pager, Devah. 2003. The mark of a criminal record. American Journal of
Sociology 108:937–75.
Pager, Devah. 2007. Marked: Race, Crime, and Finding Work in an Era of
Mass Incarceration. Chicago, IL: University of Chicago Press.
Pennsylvania Board of Pardons. 2005. Factors Considered by the Board of
Pardons in Evaluating Pardon/Commutation Requests. Pennsylvania
Board of Pardons, Commonwealth of Pennsylvania. http://sites.state.
pa.us/PA_Exec/BOP/ref_lib/FactorsPardon.doc.
Petersilia, Joan. 2003. When Prisoners Come Home: Parole and Prisoner
Reentry. New York: Oxford University Press.
Piper, Elizabeth S. 1985. Violent recidivism and chronicity in the 1958
Philadelphia cohort. Journal of Quantitative Criminology 1:319–44.
Piquero, Alex R., David P. Farrington, and Alfred Blumstein. 2003. The
criminal career paradigm: Background and recent developments. In
Crime and Justice: A review of research, vol. 30, ed. Michael H. Tonry.
Chicago, IL: University of Chicago Press.
Piquero, Alex R., David P. Farrington, and Alfred Blumstein. 2007. Key
Issues in Criminal Career Research: New Analyses of the Cambridge
Study in Delinquent Development. Cambridge, UK: Cambridge University Press.
Raphael, Steven. 2006. Should criminal history records be universally
available? Criminology & Public Policy 5:515–22.
Sampson, Robert J., and John H. Laub. 1993. Crime in the Making: Pathways and Turning Points Through Life. Cambridge, MA: Harvard University Press.

\\server05\productn\C\CRY\47-2\CRY210.txt

unknown

Seq: 31

CRIMINAL BACKGROUND CHECKS

1-MAY-09

10:08

357

Sampson, Robert J., and John H. Laub. 2003. Life-course desisters? Trajectories of crime among delinquent boys followed to age 70. Criminology 41:301–40.
Sampson, Robert J., John H. Laub, and Christopher Wimer. 2006. Does
marriage reduce crime? A counterfactual approach to within-individual causal effects. Criminology 44:465–508.
Samuels, Paul, and Debbie Mukamal. 2004. After Prison: Roadblocks to
Reentry. New York: Legal Action Center. http://www.lac.org/lac/
upload/lacreport/LAC_PrintReport.pdf.
Scanlon, Walter. 2000. It’s time I shed my ex-convict status. Newsweek
135:10.
Schmidt, Peter, and Ann D. Witte. 1988. Predicting Recidivism Using Survival Models. New York: Springer-Verlag.
Schwartz, Richard D., and Jerome H. Skolnick. 1962. Two studies of legal
stigma. Social Problems 10:133–42.
SEARCH Group, Incorporated. 2001. Survey of which states provide public access to their criminal history records through the Internet. Sacramento, CA: SEARCH Group Inc. http://www.search.org/files/pdf/
internet_cch_survey_II.pdf .
SEARCH Group, Incorporated. 2005. Report of the National Task Force
on the Commercial Sale of Criminal Justice Information. Sacramento,
CA: SEARCH Group, Inc.
Slora, Karen B. 1989. An empirical approach to determining employee
deviance base rates. Journal of Business and Psychology 4:199–219.
Society for Human Resources Management. 2004. SHRM finds employers
are increasingly conducting background checks to ensure workplace
safety—Press Release. http://www.shrm.org/press_published/CMS_
007126.asp.
Stoll, Michael A., and Shawn D. Bushway. 2008. The effect of criminal
background checks on hiring ex-offenders. Criminology & Public Policy 7:371–404.
Tracy, Paul E., Marvin E. Wolfgang, and Robert M. Figlio. 1990. Delinquency Careers in Two Birth Cohorts. New York: Plenum.
Transportation Security Administration. n.d. Disqualifiers—HAZMAT
Endorsement Threat Assessment Program. Washington, DC: U.S.

\\server05\productn\C\CRY\47-2\CRY210.txt

358

unknown

Seq: 32

1-MAY-09

10:08

BLUMSTEIN & NAKAMURA

Department of Homeland Security, Transportation Security Administration. http://www.tsa.gov/what_we_do/layers/hazmat/disqualifiers.
shtm.
Travis, Jeremy. 2002. Invisible punishment: An instrument of social exclusion. In Invisible Punishment: The Collateral Consequences of Mass
Imprisonment, eds. Marc Mauer and Meda Chesney-Lind. New York:
The New Press.
Uggen, Christopher. 1999. Ex-offenders and the conformist alternative: A
job quality model of work and crime. Social Problems 46:127–51.
U.S. Census Bureau. 1996. Historical Annual Time Series of State Population Estimates and Demographic Components of Change 1980–1990, by
Single Year of Age and Sex. Washington, DC: U.S. Census Bureau.
U.S. Census Bureau. 2000. 1990–1999 Annual Time Series of State Population Estimates by Single Year of Age and Sex. Washington, DC: U.S.
Census Bureau.
U.S. Census Bureau. 2007. Annual Estimates of the Resident Population by
Single-Year of Age and Sex for the United States and States: April 1,
2000 to July 1, 2006. Washington, DC: U.S. Census Bureau.
Visher, Christy A., Pamela K. Lattimore, and Richard L. Linster. 1991.
Predicting the recidivism of serious youthful offenders using survival
models. Criminology 29:329–66.
Wallman, Joel, and Alfred Blumstein. 2006. After the crime drop. In The
Crime Drop in America, eds. Alfred Blumstein and Joel Wallman. New
York: Cambridge University Press.
Warr, Mark. 1998. Life-course transitions and desistance from crime.
Criminology 36:183–216.
Western, Bruce, Jeffrey R. Kling, and David F. Weiman. 2001. The labor
market consequences of incarceration. Crime & Delinquency
47:410–27.
Westin, Alan F., and Michael A. Baker. 1972. DATABANKS in a Free
Society: Computers, Record-Keeping and Privacy. New York: Quadrangle Books.
Westlake, Wilfred J. 1976. Symmetrical confidence intervals for bioequivalence trials. Biometrics 32:741–4.
Wheelock, Darren. 2005. Collateral consequences and racial inequality:
Felon status restrictions as a system of disadvantage. Journal of Contemporary Criminal Justice 21:82–90.

\\server05\productn\C\CRY\47-2\CRY210.txt

unknown

Seq: 33

CRIMINAL BACKGROUND CHECKS

1-MAY-09

10:08

359

Wu, Lawrence L. 1989. Issues in smoothing empirical hazard rates. Sociological Methodology 19:127–59.

Alfred Blumstein is the J. Erik Jonsson University Professor of Urban
Systems and Operations Research and former dean at the Heinz College
of Carnegie Mellon University. He is a Fellow of the American Society of
Criminology, was the 1987 recipient of the Society’s Sutherland Award for
“contributions to research,” and was the president of the Society in
1991–1992. He was awarded the Stockholm Prize in Criminology in 2007.
His research has covered many aspects of criminal-justice phenomena and
policy, which includes crime measurement, criminal careers, sentencing,
deterrence and incapacitation, prison populations, demographic trends,
juvenile violence, as well as drug-enforcement policy.
Kiminori Nakamura is a doctoral student at the Heinz College, Carnegie Mellon University. His research interests include the dimensions of
a criminal career, life-course/developmental criminology, recidivism, collateral consequences of criminal-history records, and quantitative methods, including social network analysis. He received his MA in
demographic & social analysis in 2005 and his BA in criminology, law &
society in 2004 both from the University of California, Irvine.