ArticlePDF Available

The Absorbing Status of Incarceration and its Relationship with Wealth Accumulation

June 2015
Journal of Quantitative Criminology 31(2):207-236

Authors:

University of Alberta

Objectives This study extends our knowledge on the negative effects of incarceration to the accumulation of wealth by examining whether, how, and how much incarceration affects home ownership and net worth. It also investigates how these outcomes vary with the time since a person was incarcerated and the number of incarceration periods, along with addressing potential mechanisms behind this relationship. Methods I apply hybrid mixed effects models that disaggregate within- and between person variation to investigate incarceration’s relationship with home ownership and net worth, using National Longitudinal Study of Youth data from 1985 to 2008. I also incorporate a set of mediation models in order to test for indirect effects of incarceration on wealth through earnings, health, and family formation. Results My results show that incarceration limits wealth accumulation. Compared to never-incarcerated persons, ex-offenders are less likely to own their homes by an average of 5 percentage points, and their probability of home ownership decreases by an additional 28 percentage points after incarceration. Ex-offenders’ net worth also decreases by an average of $42,000 in the years after incarceration. Conclusions When combined with previous research on incarceration, my findings show that incarceration acts as an absorbing status, potentially leading to the accumulation of disadvantage. Although incarceration’s negative effects on wealth accumulation were partially mediated by its relationship with earnings and family formation, incarceration directly affected home ownership and net worth. In most cases, former inmates began with flatter wealth trajectories and experienced additional losses after incarceration.

Content uploaded by Michelle L. Maroto

Content may be subject to copyright.

ORIGINAL PAPER

The Absorbing Status of Incarceration and its

Relationship with Wealth Accumulation

Michelle Lee Maroto

Published online: 22 August 2014

!Springer Science+Business Media New York 2014

Abstract

Objectives This study extends our knowledge on the negative effects of incarceration to

the accumulation of wealth by examining whether, how, and how much incarceration

affects home ownership and net worth. It also investigates how these outcomes vary with

the time since a person was incarcerated and the number of incarceration periods, along

with addressing potential mechanisms behind this relationship.

Methods I apply hybrid mixed effects models that disaggregate within- and between

person variation to investigate incarceration’s relationship with home ownership and net

worth, using National Longitudinal Study of Youth data from 1985 to 2008. I also

incorporate a set of mediation models in order to test for indirect effects of incarceration on

wealth through earnings, health, and family formation.

Results My results show that incarceration limits wealth accumulation. Compared to

never-incarcerated persons, ex-offenders are less likely to own their homes by an average

of 5 percentage points, and their probability of home ownership decreases by an additional

28 percentage points after incarceration. Ex-offenders’ net worth also decreases by an

average of $42,000 in the years after incarceration.

Conclusions When combined with previous research on incarceration, my ﬁndings show

that incarceration acts as an absorbing status, potentially leading to the accumulation of

disadvantage. Although incarceration’s negative effects on wealth accumulation were

partially mediated by its relationship with earnings and family formation, incarceration

directly affected home ownership and net worth. In most cases, former inmates began with

ﬂatter wealth trajectories and experienced additional losses after incarceration.

Electronic supplementary material The online version of this article (doi:10.1007/s10940-014-9231-8)

contains supplementary material, which is available to authorized users.

M. L. Maroto (&)

Department of Sociology, University of Alberta, 6-23 Tory Building, Edmonton, AB T6G 2H4,

Canada

e-mail: maroto@ualberta.ca

123

J Quant Criminol (2015) 31:207–236

DOI 10.1007/s10940-014-9231-8

Keywords Incarceration !Stratiﬁcation !Cumulative disadvantage !Wealth !Home
ownership
Introduction
Criminologists and sociologists have highlighted the role of incarceration in reproducing
inequality in employment, education, voting, and health (Wakeﬁeld and Uggen 2010). As a
stigmatized legal status, incarceration hinders an offender’s re-integration into society and
often becomes a primary status for the evaluation of that person. In addition, state and
federal laws limit access to voting rights, employment, and social services for previously
incarcerated persons (Manza and Uggen 2006; Samuels and Mukamel 2004). Due to the
incarceration of numerous young, minority, and low-income men in the United States, the
lingering negative effects of incarceration disproportionately harm members of these
groups (Western 2006). As a result, incarceration often exacerbates broader societal
inequalities through its negative social status.
The status of an ‘‘ex-offender’’ or ‘‘former prisoner’’ results not only from individual
involvement in criminal activity, but also from the criminal justice system’s varying
enforcement efforts and responses to charges, which often depend on the crime, its loca-
tion, and the characteristics of the defendant and the victim (Pettit and Western 2004;
Wacquant 2001; Western 2006). Once the status of ex-offender is achieved, however, this
status, often referred to as a criminal credential, can lead to long lasting negative conse-
quences for former offenders (Pager 2003,2007; Wakeﬁeld and Uggen 2010). With such
negative effects, a previous incarceration can become an overarching absorbing status and
the basis for the accumulation of disadvantage over time.
In view of the growing body of research on the contribution of incarceration to
inequality, this paper examines whether, how, and how much incarceration affects two
indicators of wealth: home ownership and net worth. The persistent negative effects of
incarceration in multiple spheres likely spread to areas of wealth accumulation. This can
then inﬂuence personal wellbeing because wealth confers a variety of advantages,
including better neighborhoods, social and cultural capital, and political sway (Bricker
et al. 2012; Bucks 2012; Keister 2000a; Keister and Moller 2000). Even a modest amount
of wealth can create a safety net for households in times of ﬁnancial distress (Spilerman
2000).
The many advantages of wealth, along with its unequal distribution in the United States,
make it a key site for inequality where wealth disparities remain across many groups today
(Bricker et al. 2012; Bucks 2012; Keister 2000a). With their limited access to employment,
former prisoners may also face similar barriers to wealth accumulation. Even though it is
likely that former prisoners will hold little wealth due to their multiple disadvantaged
statuses, we have no research that demonstrates the presence of a relationship between
incarceration and wealth accumulation. I seek to remedy this omission by analyzing the
association between incarceration and the outcomes of home ownership and net worth
using data from the National Longitudinal Study of Youth 1979 cohort.
I begin this paper by drawing on theories of ascription, stigma, and cumulative
advantage/disadvantage as well as Pager’s (2003,2007) concept of the criminal credential
in order to introduce incarceration as an absorbing status. I continue by discussing the
negative effects of incarceration and my reasons for expecting a link between incarceration
208 J Quant Criminol (2015) 31:207–236
123

and wealth. After describing my measures and methods, which include hybrid mixed
effects models, I present ﬁndings showing that incarceration is associated with a reduced
probability of home ownership and lower net worth. Because I investigate change over
time and the potential mechanisms that contribute to incarceration’s relationship with
wealth accumulation, a feat that few incarceration studies have accomplished, these
ﬁndings also demonstrate how the negative effects of incarceration extend into this new
area through multiple pathways.
The Status of Incarceration and its Negative Effects
Wakeﬁeld and Uggen (2010, 388) described current and former prisoners as a ‘‘Weberian
status group sharing similar life chances determined by a common and consequential mark
of [dis]honor.’’ This mark of dishonor potentially lasts indeﬁnitely and can outweigh other
statuses that a person may prefer to identify with. Incarceration, therefore, acts as a
stigmatized status that is formalized through a criminal record or ‘‘credential’’ (Pager
2007). As a result, incarceration can become part of a process of ascription, where roles are
assigned and resources are allocated based upon categorical group membership (Kemper
1974; Mayhew 1968).
Incarceration as a Criminal Credential
In deﬁning characteristics as reference points for ascription, researchers often refer to the
difference between distributing resources based on personal attributes acquired at birth
versus individual performance that can change over time. In reality, this ascription-
achievement dichotomy is not always so straightforward where multiple characteristics
have been conceptualized as both achieved and ascribed (Cadge and Davidman 2006;
Jacobsen and Kendrick 1973; Lorber 1994). The status of an ‘‘ex-convict,’’ ‘‘ex-felon,’’ or
‘‘formerly incarcerated person’’ falls somewhere in between this dichotomy of status
acquisition categories, which is encompassed by Pager’s (2007) concept of a criminal
credential that limits access to opportunity.
According to Pager (2007), credentials represent formalized status distinctions that can
be used to deﬁne legal rights or barriers, which legitimizes their use for the distribution of
resources. Like a college degree, a criminal credential certiﬁes an individual’s position in
society, but unlike a degree, this ‘‘earned’’ credential comes with legal restrictions and a
stigma that brands ex-offenders as untrustworthy (Pager 2007; Pettit and Lyons 2007). A
criminal credential is achieved once a person is incarcerated for his or her criminal
behavior and then released. However, after a person acquires this status, it becomes a
lasting marker and a normatively acceptable basis for unequal treatment in a society that
purports to treat people as equals. A previous incarceration then becomes a status that
works ascriptively by determining the distribution of resources.
Incarceration as a Stigmatized Status
Of course, incarceration is more than an achieved status; it is a stigmatized status. As a
stigma, or according to Goffman (1963, 3) ‘‘an attribute that is deeply discrediting,’’
incarceration acts differently than a credential. The term ‘‘credential’’ generally implies the
receipt of a positive status, one that a person would choose to share with others. As
undesirable characteristics, stigmas are not markers that people choose; stigmas are
J Quant Criminol (2015) 31:207–236 209
123

imposed upon people, usually on people with less power. Stigmas can play a critical role
within the process of stratiﬁcation, particularly when they are used to justify the differ-
ential treatment of groups (Link and Phelan 2001).
Like ascription, stigma encompasses a power component. As Link and Phelan (2001,
377) noted: ‘‘stigma exists when elements of labeling, stereotyping, separation, status loss,
and discrimination occur together in a power situation that allows them.’’ Current and
former prisoners typically come from groups who already lack power, which facilitates
passing along this mark of dishonor. Members of less powerful groups are at higher risk of
incarceration than others with far reaching effects, which is reﬂected by the composition of
the prison population in the United States (Western 2002,2006). Prisoners in the United
States tend to come from the most disadvantaged groups—groups whose members lack
social, economic, and political power (Wakeﬁeld and Uggen 2010; Western 2006). They
are young, from low-income families, and average less schooling than a high school
degree. The prison population also disproportionately comprises racial and ethnic minor-
ities.
1
It is likely then that the continuing negative effects of incarceration will compound
these race and class inequalities, exacerbating the disadvantage of former prisoners.
Absorbing Statuses and the Accumulation of Disadvantage
In order to better describe the status of former prisoners, I borrow the term ‘‘absorbing
state’’ from the literature on social mobility, which uses Markov chain models to represent
the relationship between social structure and social mobility (Henry et al. 1971; Matras
1967; McGinnis 1968). In these models, the axiom of cumulative inertia predicts that the
probability of remaining in a state will increase with the time spent in that state, to the
point that some states become ﬁnite absorbing states (Henry et al. 1971; McGinnis 1968).
2
When applied to certain characteristics, this perspective highlights how even achieved
statuses can become lasting markers that act as if they were ascribed. The negative effects
of incarceration show that it can become an absorbing status that is used ascriptively across
multiple spheres of life, leading to a process of accumulating disadvantage.
According to cumulative advantage/disadvantage theory, achieved and ascribed statuses
can have persisting effects in returns to resources, leading to diverging outcomes in which
some people continually build their resources relative to others. This theory originated with
Merton’s (1968,1988) work on recognition in the scientiﬁc community. Since then
researchers have applied aspects of cumulative advantage theory to various arenas,
including crime and delinquency (DiPrete and Eirich 2006; Sampson and Laub 1997). I use
the term accumulating disadvantage broadly to explain the persisting effects of a previous
incarceration on a person’s credit market outcomes, similar to Lyons and Pettit’s (2011)
use of compounded disadvantage. The negative status of ‘‘ex-convict’’ reﬂects incarcera-
tion’s lingering effects that compound any prior disadvantaged status.
1
In 2011 the incarceration rate for black non-Hispanic males ranged between 5 and 9 times that of white
non-Hispanic males, depending on the age group, and the rate for Hispanic males was two to three times that
of white non-Hispanic males (Carson and Sabol 2012). The largest disparities occurred for younger age
groups.
2
In the vacancy chain literature, a vacancy is created when a new resource unit enters a population and an
individual leaves his or her unit behind to take that new position (Chase 1991; White 1970). This move
initiates a sequence of moves as other individuals in the chain transfer into the vacant units. Within a
vacancy chain an absorbing state acts as the end state or the termination of a chain.
210 J Quant Criminol (2015) 31:207–236
123

Incarceration’s Far Reaching Effects
The many negative effects of incarceration support a process of cumulative disadvantage
that continues after prisoners, who already come from disadvantaged groups, are released.
Labor market barriers disadvantage ex-convicts and often lead to segmented labor market
access for those with a previous incarceration on their record (Piore 1970; Western and
Beckett 1999). Research on labor market outcomes using longitudinal data and adminis-
trative records has consistently shown lower employment rates and earnings for previously
incarcerated persons, particularly black men (Apel and Sweeten 2010; Freeman 1992,
1996; Kling 2006; Lott 1990; Nagin and Waldfogel 1998; Waldfogel 1994; Western 2002;
Western and Beckett 1999; Western and Pettit 2000). Even though wage trajectories begin
to recover over time, many years after they are released, previously incarcerated persons
earn less and spend more time without employment than persons who have never been to
prison (Pettit and Lyons 2007; Western 2002; Western and Beckett 1999). Moreover, the
effects of incarceration on labor market outcomes are much greater than the penalties for
more limited interactions with the criminal justice system, such as arrests and convictions
(Western et al. 2001).
In addition to affecting employability and earnings, incarceration reduces men’s like-
lihood of marriage and places those who do marry at a higher risk of divorce while they are
incarcerated, but not always after incarceration (Apel et al. 2010; Lopoo and Western
2005; Wildeman and Muller 2012). Although they often depend on a family’s situation
prior to a paternal incarceration, the consequences of incarceration extend to other family
members as well; they often lead to psychological issues for children, in addition to family
poverty (Turney and Wildeman 2013; Wildeman 2009,2010). Inmates have also been
shown to experience illness, depression, stress, and other psychological problems at dis-
proportionately higher rates than the larger population (Massoglia 2008a,b). These effects
continue and can even worsen long after incarceration (Schnittker and John 2007;
Schnittker et al. 2011). Despite a pressing need for social services and aid, former prisoners
in many states ﬁnd that their criminal credential restricts their ability to obtain state support
(Samuels and Mukamel 2004). In view of these negative effects and added restrictions, it
seems likely that incarceration could further disadvantage former prisoners in their ability
to accumulate wealth.
Incarceration and Wealth
I investigate a potential consequence of incarceration that has not been studied: whether
incarceration limits wealth accumulation. With its basis in various types of property
ownership, wealth shapes people’s economic and personal wellbeing, creates more stability
than income, and provides beneﬁts that extend to many areas (Keister 2000a; Spilerman
2000). For example, home ownership can provide access to better neighborhoods and
school systems, assets can increase social networks, and the ownership of various goods
heightens social status (Keister 2000a; Spilerman 2000).
Due to these advantages, I focus on wealth accumulation and examine the effects of a
previous incarceration on home ownership and net worth. Speciﬁcally, my research
addresses the following questions: How does incarceration affect the accumulation of
wealth for formerly incarcerated individuals? Does the time since a person was incarcer-
ated inﬂuence these outcomes, as a process of accumulating disadvantage implies? Do
J Quant Criminol (2015) 31:207–236 211
123

these outcomes vary by the length of an incarceration or the number of incarceration
periods? Finally, what are the potential mechanisms behind this relationship? This mark
has been shown to shape labor market outcomes, and I expect that a previous incarceration
will present a negative association with a person’s probability of home ownership and
accumulation of net worth through multiple related pathways.
Disadvantage for previously incarcerated persons operates through numerous
mechanisms. These include selection barriers, such as ex-offenders’ limited educa-
tion, skills, and work experience; ex-offenders’ physical, mental, drug, and motiva-
tional problems; discrimination by employers and other gatekeepers; and the
stigmatization of formerly incarcerated persons (Holzer et al. 2003;Pager2007;
Western et al. 2001). Although secondary data analyses seldom address which of
these mechanisms furthers disadvantage, audit studies emphasize the barriers gen-
erated through employer discrimination that ex-offenders face in the labor market.
Audit studies of employers in Milwaukee and New York showed that employers were
one-half to one-third less likely to consider an ex-offender for an open position than
an equally qualiﬁed person without a criminal credential, controlling for race (Pager
2003,2007;Pageretal.2009).
The individual and structural factors that limit the employability of previously incar-
cerated persons could also affect their ability to accumulate wealth, particularly if, like
employers, lenders interpret a previous incarceration as a signal of untrustworthiness or
instability (Holzer 1996; Holzer et al. 2003; Pager and Quillian 2005). In this case, a
previous incarceration would limit access to lending, a general requirement for wealth
building, but incarceration affects wealth in other ways as well. Recent research on legal
ﬁnancial obligations (LFOs) demonstrates how the criminal justice system imposes added
debt burdens on offenders through the use of heavy pre- and post-conviction ﬁnes and fees
that ex-offenders often cannot afford to pay (Harris et al. 2010,2011). Furthermore,
incarceration limits an individual’s ability to make payments, which could lead to debt
delinquency, negative reports from collection agencies, and limitations on future lending.
Thus, I expect that the stigmatized status of incarceration will directly limit wealth
accumulation for previously incarcerated persons. I also expect that these effects will
worsen over time and with additional periods of incarceration, as former prisoners fall
farther behind. With these expectations, I expand on Harris et al. (2010) research to
demonstrate the broader implications of incarceration for wealth accumulation beyond fees
imposed by the legal system.
Because it acts as an absorbing status that affects multiple areas of life, incarceration
should also limit wealth accumulation through multiple pathways or mechanisms. Sev-
eral of these mechanisms operate with respect to a person’s absence from the labor
market. For instance, increased earnings are positively associated with home ownership
and wealth accumulation (Keister and Moller 2000). The lost work experience that
incarceration imposes and its consequences for earnings later on can also impede a
person’s ability to accumulate assets and develop a credit history. Because family for-
mation often leads to wealth accumulation and home ownership (Bricker et al. 2012),
incarceration should affect wealth through its relationship with marriage. Finally, the
added problems created by the health limitations of formerly incarcerated persons should
also affect wealth accumulation. Overall, I expect that ex-offenders’ stigmatization, lost
earnings, limited marriage prospects, and added health limitations will all lead to post-
incarceration wealth reductions.
212 J Quant Criminol (2015) 31:207–236
123

Data
I use the 1979 cohort of the National Longitudinal Study of Youth (NLSY79) to estimate
the effects of incarceration on credit market outcomes.
3
The NLSY79 cohort is a stratiﬁed
multistage sample of 12,686 men and women who were between 14 and 22 years old when
ﬁrst surveyed in 1979.
4
Respondents were interviewed annually until 1994, after which
they were interviewed every 2 years. The survey is ongoing with debt and asset data most
recently collected in 2008.
My data take the form of an unbalanced panel sample in which the number of time
periods may differ across individuals. I restrict my sample to 1985 through 2008 and I
exclude data from 1991, 2002, and 2006 because the NLSY did not consistently collect
data on assets and debt over time. After removing years when wealth questions were not
asked, individuals from the military oversample, observations with missing data, and
extreme outliers on net worth, a sample of 10,274 individuals (or cases) and 96,180
observations (or person-years) remained. My sample therefore covers over 90 percent of
respondents from the original 11,406 individuals who were not a part of the military
subsample.
Methods
For my analyses, I use hybrid mixed effects regression models that include ﬁxed effects for
time-varying covariates and random effects for time-invariant covariates, which allow me
to disaggregate within- and between-person variation in the same models (Allison 2009).
These models take the form of random effects models—also known as multilevel varying-
intercept, mixed, and hierarchical models— that account for correlated disturbance terms
for the same person over time by assigning each person a separate intercept (Allison 2009;
Gelman and Hill 2007). I incorporate ﬁxed effects for time-varying covariates within each
model by expressing these variables as deviations from their person-speciﬁc means, which
represent the within-person variation across time periods (Allison 2009). I also include the
person-speciﬁc means in order to provide estimates based on the average between-person
variation.
By incorporating the deviations from person-speciﬁc means for time-varying covariates
along with the person-speciﬁc means into each model, I am able to discuss the variation of
a particular score for an individual in the sample (or the average of the average change in a
score for an individual over time) as well as the variation in the average score across
individuals. This allows me to control for unobserved, stable, time-invariant, individual-
level characteristics while assessing the effects of time-varying and time-invariant factors
(Allison 2009). Hybrid mixed effects models, therefore, help to overcome the limitations of
3
In using this longitudinal survey I expand upon the research of Freeman (1992), Western and Beckett
(1999), Western (2002), and the more recent work of Massoglia et al. (2013). These studies applied ﬁxed
effects models to NLSY data to investigate the effects of incarceration on labor market outcomes.
4
The original sample comprised a cross-sectional sample of 6,111 respondents, a supplemental sample of
5,295 respondents that oversampled civilian Hispanic, black, and economically disadvantaged non-black/
non-Hispanic youth, and a military sample of 1,280 respondents (NLSY79 User’s Guide). I use the full
dataset except for the military sample in my analyses to observe as many incarcerations as possible. I also
include a variable in all models to indicate whether the respondent was a member of the cross-sectional
sample.
J Quant Criminol (2015) 31:207–236 213
123

separate ﬁxed and random effects models, yet they provide estimates consistent with both
procedures.
5
Equation 1represents the general random effects model for continuous data, which I
use to estimate an individual’s net worth at time, t,
6
yit ¼ltþbXit þcZiþaiþeit ð1Þ
where iindexes the individual respondent and tindexes yearly observations per individual.
In this equation, ltrepresents the time-varying intercept, cZirepresents vectors of the time-
invariant coefﬁcients and predictor variables, bXit represents vectors of the time-varying
coefﬁcients and predictor variables, and eit is the error term that represents random vari-
ation at each point in time. These models assume that ai, which is treated as a set of
random variables with a speciﬁed probability distribution, is independent of all other
variables in the model.
I then partition the within- and between-person variation through person-speciﬁc mean-
centering of the time-varying predictors in Eq. 1(Allison 2009; Curran and Bauer 2011).
The process of person-speciﬁc mean-centering controls for unobserved heterogeneity in
time-varying covariates by decomposing these variables into their within- and between-
person variation. Equation 2illustrates this process,
x&
it ¼xit 'xið2Þ
where x&
it represents the person-mean centered time-varying covariate, xit is the original
score for individual iat time t, and xirepresents the person-speciﬁc mean for individual i. I
include both components in the models, and report the coefﬁcients for within-person and
between-person variation separately. I also report random effects coefﬁcients for time-
invariant covariates in the model. Although I report coefﬁcients for both levels, I primarily
discuss the within-person effects in my results. Equation 3summarizes this full model,
which now includes coefﬁcients and vectors for X&
it and Xi.
yit ¼ltþbX&
it þðc1Xiþc2ZiÞþaiþeit ð3Þ
I also incorporate an autoregressive disturbance term to account for the correlation of
error terms within persons over time. The error term is described by Eq. 4:
eit ¼qeit;t'1þgit ð4Þ
where q
jj
\1 and git is independent and identically distributed with mean 0 and variance
h2
g.
I include a set of mediation models in order to also test for indirect relationships
between incarceration and wealth at both the within- and between-person levels. The most
common method for calculating indirect effects involves estimating two equations for the
outcome variable, one that includes the mediator and one that does not, and then ﬁnding
the difference in coefﬁcients for the initial variable across the two equations (Krull and
5
Although the coefﬁcient estimates are consistent with both procedures, they are not identical. In particular,
the ﬁxed effects coefﬁcients vary from the within-person coefﬁcients because my data are unbalanced.
Random effects coefﬁcients differ from the between-person coefﬁcients because random effects models do
not purely rely on between-person variation. They also consider some within-person variation.
6
I use a logit form of this equation to estimate an individual’s probability of home ownership at time, t.
214 J Quant Criminol (2015) 31:207–236
123

MacKinnon 2001).
7
However, this method often produces biased estimates for non-linear
models, including logit models with binary outcomes, because the coefﬁcients and error
variance are not separately identiﬁed (MacKinnon and Dwyer 1993). In order to estimate
indirect effects and overcome these biases, I apply Karlson et al. (2011) KHB-method that
accounts for model rescaling (Breen et al. 2013; Karlson and Holm 2011). This allows me
to then discuss both the direct and indirect effects of incarceration on wealth accumulation.
Measures
Outcome Variables
My ﬁrst outcome variable is home ownership, that is, whether the respondent owns or
makes payments on his or her dwelling. Most families’ major sources of wealth are their
homes, and home ownership provides beneﬁts that include residential stability, tax breaks,
and access to schooling (Keister 2000a; Shapiro 2004; Spilerman 2000). Moreover, home
ownership entails access to mortgage lending, which generally requires both an income and
a strong credit history. As illustrated by Table 1, the gap in home ownership rates by
incarceration status is obvious; 24 percent of ever-incarcerated respondents owned their
homes in 2008 compared to 67 percent of never-incarcerated respondents.
I use a measure of net worth in 2010 dollars as my second outcome variable. The survey
calculates net worth by subtracting the respondent’s total debts from the total value of all
assets.
8
Using wealth data as my outcome variable creates certain limitations. As Spiler-
man (2000) noted, wealth ﬁgures in representative surveys can be inconsistent due to the
complexity of wealth, a lack of standardization across surveys, and the difﬁculty many
respondents have in estimating their wealth. In order to address these limitations, I exclude
extreme outliers on these variables and remove cases with missing data.
9
I also limit my
analysis to variables consistently collected in the survey. Similar to home ownership,
Table 1shows a gap of approximately $173,000 in net worth by incarceration status. The
disparity is smaller in terms of median net worth; the median net worth of ever-incar-
cerated respondents in 2008 was $507 compared to $92,000 for never-incarcerated
respondents.
7
The basic ‘‘product-of-coefﬁcient’’ method is another popular way to estimate indirect effects (Bauer et al.
2006; Krull and MacKinnon 2001; Zhang et al. 2009). Estimates obtained using the additive and the
product-of-coefﬁcient methods are usually equivalent for linear outcomes in single-level models, but they
often diverge in multilevel and nonlinear models (Krull and MacKinnon 2001; MacKinnon and Dwyer
1993).
8
The NLSY calculates net worth with the following equation: NET WORTH =HOME VALUE -
MORTGAGE -PROPERTY DEBT ?CASH SAVING ?STOCKS/BONDS ?TRUSTS ?BUSINESS
ASSETS -BUSINESS DEBT ?CAR VALUE -CAR DEBT ?POSSESSIONS -OTHER
DEBT ?IRAs ?401Ks ?CDs. Net worth also includes the respondent’s spouse’s assets and debt. In
order to account for this, I include marital status as a control variable in all full models. I also estimated
additional models for a subset of respondents who were never married, and I tested for interactions between
incarceration and marital status. In these models, the effects of incarceration remained statistically signif-
icant and negative.
9
I removed cases with a total net worth greater than $2 million or less than -$2 million 2010 USD. This
removed 313 cases.
J Quant Criminol (2015) 31:207–236 215
123

Table 1 Descriptive statistics of NLSY79 variables for most recent survey year (2008)

Full sample Never incarcerated Ever incarcerated

N individuals 6,033 5,678 355

Home ownership

64.51 67.07 23.66

Net worth (dollars) $203,824 $213,991 $41,205

Ever incarcerated 5.88 * *

Currently incarcerated 0.51 * *

Incarcerated in past year 0.86 * *

Incarcerated 1–5 years ago 0.50 * *

Incarcerated 6–10 years ago 1.66 * *

Incarcerated more than 10 years ago 2.87 * *

Incarcerated for 1 survey year 2.45 * *

Incarcerated for 2–4 survey years 1.99 * *

Incarcerated for 5 or more survey years 1.44 * *

Age (years) 46.63 46.63 46.54

Employment status

Employed full-time 66.83 68.25 44.23

Employed part-time 12.71 12.79 11.55

Unemployed 3.63 3.17 10.99

Out of the labor force 16.82 15.80 33.24

Job gaps since age 18

No job gaps 27.18 28.11 12.39

Single job gap 21.68 22.14 14.37

Two?job gaps 51.14 49.75 73.24

Self-employed 9.56 9.30 13.80

Government employee 16.77 17.45 5.92

Individual earnings (dollars) $39,713 $41,080 $17,842

Years of schooling completed (years) 13.31 13.39 11.90

Marital status

Married 54.14 56.02 23.94

Never married 17.77 16.43 39.15

Formerly married 28.10 27.54 36.90

Any children 56.46 58.42 25.07

Health limitation 14.88 14.11 27.32

Rural 22.61 23.05 15.49

Female 52.13 54.63 12.11

Black

31.03 29.46 56.06

Hispanic origin

18.47 18.39 19.72

AFQT Score (percentile) 38.91 40.16 18.95

Any drug use in teens 70.13 68.95 89.01

Marijuana use in teens 69.45 68.28 88.17

Cocaine use in teens 17.79 16.77 34.08

Other drug use in teens 16.72 16.13 26.20

Source: NLSY 1979 Cohort, full sample

Estimates refer to percentages, unless otherwise noted

Because I analyze the full NLSY79 sample, blacks and Hispanics are overrepresented in these data

216 J Quant Criminol (2015) 31:207–236

123

Predictor Variables

My primary predictor variable measures the respondent’s incarceration status as whether

the respondent was previously incarcerated prior to the current time period. The NLSY

provides information on incarceration through two measures. The 1980 survey, which

included an extensive set of questions related to illegal activity, asked respondents if and

when they were incarcerated. The NLSY also records a respondent’s residence each year,

including whether the person resided in a jail or prison at the time of the interview.

I use

these two measures to create a variable that indicates whether the respondent was ever

incarcerated as an adult over 18 years of age at each time point. To ensure that my

estimates of the effects of incarceration apply only to people who were previously

incarcerated, I control for whether the respondent was incarcerated at the current survey

wave. As Table 1shows, approximately six percent of NLSY respondents had been

incarcerated by 2008.

I also test two categorical variables that encompass different aspects of incarceration

because the effects of incarceration on wealth accumulation can vary based upon the

timing of incarceration, as well as the length and number of incarceration spells. Therefore,

in order to estimate the accumulating effects of incarceration over time, I created a cat-

egorical variable that measures a respondent’s time since incarceration. This variable has

six categories: never incarcerated, currently incarcerated, incarcerated in the past year,

incarcerated one to 5 years ago, incarcerated 6–10 years ago, and incarcerated more than

10 years ago. In addition, I created a categorical variable to measure the effects of multiple

incarcerations and the length of an incarceration. This variable has ﬁve categories: never

incarcerated, currently incarcerated, previously incarcerated for one survey-year, incar-

cerated for 2–4 survey-years, or incarcerated for ﬁve or more survey-years.

Time-Varying Covariates

I include time-varying covariates for demographic, employment, health, family, and

regional variables commonly used in studies that predict wealth and earnings (Keister and

Moller 2000; Kenworthy 2007). I control for an individual’s labor market situation by

including the respondent’s employment status, cumulative job gaps, and earnings. I cate-

gorize employment status as employed full-time (35 or more hours per week), employed

part-time (less than 35 h per week), unemployed, and out of the labor force, with indi-

viduals employed full-time as the referent category. I measure a respondent’s cumulative

number of job gaps lasting 8 weeks or more using a categorical variable with the following

categories: no job gaps, a single job gap, and multiple job gaps. I use the respondent’s

earnings in thousands of 2010 dollars, unless noted otherwise. I also include measures that

indicate whether the respondent was employed in a government job or was self-employed.

Because interviews record whether the respondent was incarcerated only at the time of the interview, my

measure of incarceration misses persons who were not imprisoned at the time of the interview, but spent

some time in prison during that year.

Because the survey reports whether the respondent was incarcerated only at the time of each interview, I

am not able to determine whether being incarcerated for multiple survey-years refers to a single multi-year

incarceration spell or multiple separate incarceration spells. For example, a respondent who was incarcer-

ated for two survey-years could have been incarcerated two separate times for a few months that coincided

with when the interview occurred, or the respondent could have been incarcerated for two full years. Either

situation would place this respondent in the category of incarcerated for two to four survey-years.

J Quant Criminol (2015) 31:207–236 217

123

I control for the respondent’s age, education, marital status, presence of children, and
health limitations. Because the age of this sample is truncated at 51 years, I expect a
positive effect of age on wealth, but I also include a quadratic age-squared term in order to
account for any non-linear relationships (Dynan and Kohn 2007; Modigliani 1986).
12
I
measure education as the number of completed years of schooling because that is how the
NLSY collected education information. Marital status indicates whether the respondent
was married, formerly married (separated, divorced, or widowed), or never married. I treat
any children and health limitations as binary variables. Marriage, the presence of children,
and added education should increase wealth, but marriage dissolution and health limita-
tions should have the opposite effect (Bricker et al. 2012; Dynan and Kohn 2007; Smith
1999).
Time-Invariant Covariates
I include time-invariant covariates for the respondent’s sex, race, AFQT score, and teenage
drug use, covariates that do not change over time and that could affect a respondent’s
likelihood of incarceration and his or her ability to accumulate wealth. Incarceration rates
differ greatly by a person’s race and sex (BJS, 2010; Carson and Sabol 2012; Western
2006), as does wealth accumulation (Keister 2000b; Krivo and Kaufman 2004). I therefore
measure sex as a dichotomous variable of male or female and race as two dichotomous
variables, black and Hispanic origin.
AFQT score refers to the respondent’s Armed Forces Qualiﬁcation Test percentile
score, which the survey collected for each respondent in 1980. Although AFQT score has
been a controversial measure of intelligence, it offers a standardized measure of cognitive
aptitude and school-based knowledge (Farkas 2003; Maume et al. 1996; U.S. Department
of Defense 1982). I also include three binary variables that indicate whether the respondent
reportedly used marijuana,cocaine, or other drugs as a teenager, which can relate to a
person’s level of self control that inﬂuences employment and potentially wealth accu-
mulation (Gottfredson and Hirschi 1990). The category of other drugs includes heroin,
psychedelics, inhalants, and any other drugs.
In addition to these time-invariant covariates, I also include indicator variables to
designate the cross-sectional sample and survey wave year. The cross-sectional sample
variable indicates whether the respondent was a member of the original cross-sectional
sample, where the referent category refers to members of this sample. The survey-year
variable is an indicator variable with 14 categories where the referent is the most recent
wave (2008). I include this variable to account for any unobserved period effects.
Findings
Incarceration was negatively associated with both measures of wealth accumulation across
analyses. Formerly incarcerated individuals generally had lower average wealth, as mea-
sured in terms of home ownership and net worth, than individuals who had never been to
prison. Moreover, formerly incarcerated persons averaged less wealth in the years after
incarceration compared to the years before. In most cases, former inmates began with
ﬂatter wealth trajectories, partly due to their already disadvantaged statuses, lower levels of
education, and limited earnings, and they experienced additional losses after incarceration.
12
Age also acts as a proxy for time or year because the NLSY is a longitudinal cohort sample.
218 J Quant Criminol (2015) 31:207–236
123

My ﬁndings therefore emphasize how prior disadvantages are then compounded by

incarceration.

Home Ownership

Net of time-invariant individual characteristics accounted for by within-person ﬁxed

effects coefﬁcients, formerly incarcerated persons were less likely to be homeowners in the

years after incarceration compared to the years before. As illustrated by the between-

person random effects coefﬁcients that controlled for earnings, employment, family, and

education, ever-incarcerated persons were also less likely than similar never-incarcerated

persons to own a home at any point between 1985 and 2008. These results appear in

Table 2, which uses whether the respondent was previously incarcerated to predict home

ownership.

All models in Table 2include time-invariant covariates, but in Models 1–3, I sequen-

tially added certain time-varying covariates in order to observe changes in the effects of a

previous incarceration on home ownership. Model 1 controls for only age and its quadratic

term, Model 2 includes family and health covariates, and Model 3 (the full model) includes

controls for employment and earnings. Adding controls decreased the magnitude of the

coefﬁcients for incarceration, but across all models a previous incarceration showed a

signiﬁcant negative association with home ownership, as evidenced by the coefﬁcients

representing within- and between-person variation. This means that, on average, the

probability of home ownership was lower for people who experienced incarceration, and,

among those who were incarcerated at one point, the average probability of home own-

ership decreased in the years after being incarcerated.

Net of control variables, the average probability of home ownership was only about 5

percentage points lower for respondents with a previous incarceration when compared to

similar never-incarcerated individuals. Moreover, for those who were incarcerated, the

average probability of home ownership decreased by approximately 28 percentage points

in the years after incarceration, an estimate that corresponds to the within-person coefﬁ-

cient for incarceration in Model 3.

It is also important to note that these estimates do not

include currently incarcerated individuals, who experienced even larger declines in the

probability of home ownership.

As seen in Model 3, most time-varying and time-invariant control variables were sta-

tistically signiﬁcant and associated with home ownership in the expected direction. Mar-

riage, the presence of children, employment, and earnings were all positively associated

with home ownership. Age showed a curvilinear relationship with home ownership, in

which its effects decreased over time. As expected from the literature on race and wealth,

blacks and Hispanics had a lower average probability of home ownership than other racial

groups. Additionally, most teenage drug use variables were not signiﬁcantly associated

with later home ownership, except for cocaine use, which showed a negative association.

Finally, AFQT scores had a positive association, albeit a small one.

These estimates represent the effects of covariates at the mean of the data as determined by the intercept,

which gives the predicted probability of home ownership when all variables are held at their means for

continuous variables and referent categories for categorical variables. This estimate speciﬁcally applies to

the 2008 survey wave because that is the referent category for the survey wave year variable. To obtain the

upper bound of the predictive difference we can also divide the coefﬁcient by four to approximate the

difference at which the slope of the logistic curve is maximized (Gelman and Hill, 2007). Doing so provides

the values of -0.14 and -0.34 for the between- and within-person coefﬁcients for previous incarceration.

J Quant Criminol (2015) 31:207–236 219

123

Table 2 Results from hybrid logistic mixed models predicting home ownership

Model 1 Model 2 Model 3

b SE b SE b SE

Intercept 2.135*** (.326) 3.006*** (.316) 2.505*** (.317)

Time-varying predictors

Within-person coefﬁcients

Currently incarcerated -1.865*** (.218) -1.440*** (.227) -1.092*** (.230)

Previously incarcerated -1.887*** (.178) -1.611*** (.189) -1.375*** (.193)

Age .156*** (.018) .151*** (.017) .156*** (.017)

Age squared -.009*** (.000) -.006*** (.000) -.005*** (.000)

Marital status (ref: married)

Never married -2.358*** (.061) -2.297*** (.061)

Formerly married -2.251*** (.046) -2.270*** (.046)

Any children .860*** (.042) .883*** (.042)

Any health limitation -.321*** (.057) -.144* (.059)

Rural .354*** (.048) .368*** (.049)

Employment status (ref:

employed full-time)

Employed part-time -.137*** (.039)

Unemployed -.170** (.066)

Out of the labor force -.207*** (.052)

Job gaps (ref: no job gaps)

Single job gap -.417*** (.089)

2?job gaps -.859*** (.110)

Total earnings (in 1,000s) .017*** (.001)

Self-employed .168** (.066)

Government employee .151** (.056)

Years of schooling .010 (.026)

Between-person coefﬁcients

Currently incarcerated -4.005*** (.658) -1.918*** (.574) -.567 (.571)

Previously incarcerated -1.585*** (.248) -1.016*** (.223) -.571** (.221)

Age .205*** (.014) .149*** (.013) .131*** (.013)

Age squared -.008*** (.002) -.008*** (.001) -.007*** (.001)

Marital status (ref: married)

Never married -4.445*** (.105) -4.047*** (.103)

Formerly married -4.427*** (.119) -4.064*** (.117)

Any children .285** (.094) .565*** (.094)

Any health limitation -1.399*** (.177) -.324 (.184)

Rural 1.177*** (.088) 1.299*** (.087)

Employment status (ref:

employed full-time)

Employed part-time -.475** (.164)

Unemployed -1.729*** (.292)

Out of the labor force -.429** (.153)

Job gaps (ref: no job gaps)

220 J Quant Criminol (2015) 31:207–236

123

To better understand the effects of incarceration on home ownership, Table 3displays

the results from models that considered the time since a person was incarcerated and the

length of incarceration. The timing and length of incarceration showed a stronger and

more consistent association with home ownership when comparing individuals before

and after incarceration, as opposed to comparing average outcomes for previously and

never-incarcerated persons. In addition, the negative effects of incarceration on home

ownership seem to increase over time, as a theory of accumulating disadvantage would

predict.

As shown in Table 3Model 1, the effects of incarceration on home ownership increased

over time, but not by much. Most of the between-person coefﬁcients were not signiﬁcantly

different from zero. In terms of the within-person coefﬁcients, the gap in the predicted

probability of home ownership was greatest for currently incarcerated individuals and

persons incarcerated over ten years ago, at 21 and 25 percentage points. This indicates that

the negative effects of incarceration continue well after a person is released. Model 2,

which partitions incarceration based on the number of survey-years for which the

respondent had been incarcerated, shows fairly consistent disparities in home ownership,

except for respondents who were incarcerated for more than 4 survey-years. However, the

currently incarcerated population likely inﬂuences this outcome because over half of all

currently incarcerated observations at time twere incarcerated for more than four survey-

years.

Table 2 continued Model 1 Model 2 Model 3

b SE b SE b SE

Single job gap -.442*** (.080)

2?job gaps -.741*** (.072)

Total earnings (in 1,000s) .024*** (.002)

Self-employed .556** (.188)

Government employee .306** (.114)

Years of schooling .035* (.017)

Time-invariant predictors

Female .345*** (.064) .128* (.062) .656*** (.067)

Black -2.143*** (.099) -.639*** (.094) -.696*** (.094)

Hispanic origin -1.124*** (.106) -.451*** (.099) -.510*** (.097)

AFQT Score .022*** (.001) .019*** (.001) .008*** (.001)

Used marijuana in teens -.243*** (.072) -.029 (.065) .018 (.064)

Used cocaine in teens -.549*** (.093) -.216** (.084) -.224** (.082)

Used other drugs in teens -.350*** (.092) -.149 (.083) -.052 (.081)

AIC 77,344.38 67,680.80 66,413.14

BIC 77,638.08 68,069.23 66,972.10

Rho .70 .63 .62

Analyses are for the sample of all individuals. Continuous variables (earnings, age, work experience, grade,

and AFQT score) are grand mean centered. All models include covariates for the cross sectional sample and

survey wave year. Within-person coefﬁcients approximate ﬁxed effects. Between-person coefﬁcients

approximate random effects

Source: NLSY 1979 Cohort, 1985–2008. N =10,274 individuals and 96,180 observations

*** p\.001, ** p\.01, * p\.05

J Quant Criminol (2015) 31:207–236 221

123

Net Worth

Table 4, which presents the results from hybrid mixed effects regression models predicting

total net worth, shows that the effects of incarceration on wealth accumulation extend

beyond home ownership. Although the interpretation varied when comparing between- and

within-person coefﬁcients, net of stable unobserved individual characteristics and of

Table 3 Results from hybrid logistic mixed models predicting home ownership based on the time since

incarceration and the length of incarceration

Model 1 Model 2

b SE b SE

Intercept 2.522*** (.317) 2.510*** (.317)

Time-varying predictors

Within-person coefﬁcients

Incarceration status (ref: never incarcerated)

Currently incarcerated -1.618*** (.267)

Incarcerated in past year -1.074*** (.312)

Incarcerated 1–5 years ago -1.201*** (.246)

Incarcerated 6–10 years ago -1.234*** (.249)

Incarcerated 11 or more years ago -1.849*** (.282)

Incarceration length/times (ref: never incarcerated)

Currently incarcerated -1.592*** (.279)

Incarcerated for 1 survey year -1.348*** (.246)

Incarcerated for 2–4 survey years -1.265*** (.297)

Incarcerated for 5 or more survey years -1.095* (.429)

Between-person coefﬁcients

Incarceration status (ref: never incarcerated)

Currently incarcerated -1.011 (.531)

Incarcerated in past year .011 (.404)

Incarcerated 1–5 years ago -.656 (.708)

Incarcerated 6–10 years ago -1.255 (.783)

Incarcerated 11 or more years ago -.234 (.703)

Incarceration length/times (ref: never incarcerated)

Currently incarcerated -1.296* (.586)

Incarcerated for 1 survey year -.561* (.260)

Incarcerated for 2–4 survey years -.838 (.460)

Incarcerated for 5 or more survey years .914 (.916)

AIC 66,422.42 66,426.31

BIC 67,038.23 67,023.17

Rho .62 .61

Analyses are for the sample of all individuals. Models include all covariates from Model 3 in Table 2.

Continuous variables (earnings, age, grade, and AFQT score) are grand mean centered. Within-person

coefﬁcients approximate ﬁxed effects. Between-person coefﬁcients approximate random effects

Source: NLSY 1979 Cohort, 1985–2008. N =10,274 individuals and 96,180 observations

*** p\.001, ** p\.01, * p\.05

222 J Quant Criminol (2015) 31:207–236

123

observed control variables that change over time, incarceration was also associated with a

decrease in net worth.

As seen in Model 1, when controlling for only age and time-invariant covariates,

previously incarcerated individuals had an average net worth that was $28,000 less than

that of never-incarcerated individuals, and, for those with an incarceration, their net worth

was about $67,000 less in the years after incarceration compared to the years before. This

disparity declined, but still remained statistically signiﬁcant, after controlling for family

and health aspects in Model 2. When I incorporated additional controls in Model 3, the

between-person coefﬁcient was no longer associated with net worth, indicating that dif-

ferential rates of employment, earnings, and home ownership explain much of the average

disparity in net worth across never- and formerly-incarcerated individuals. However, the

within-person coefﬁcient for incarceration showed that, even after controlling for labor

market aspects and home ownership—by far the largest contributor to wealth—net worth

decreased by approximately $42,000 in the years after incarceration.

Most time-varying and time-invariant covariates were signiﬁcantly related to the out-

come variable in the expected direction. For example, marriage, earnings, education, and

home ownership were positively associated with between- and within-respondent net worth

disparities over time. In terms of time-invariant covariates, blacks and Hispanics held

lower net worth than non-blacks and non-Hispanics, which demonstrates the continuing

gaps in racial wealth. However, there were several variables that did not affect net worth in

the expected direction. In this model, use of cocaine as a teenager was positively associated

with net worth, although this variable was negatively related to home ownership in

Table 2. Unemployment also showed a positive relationship with net worth, but this

association was positive only in models that controlled for income and education, as Model

3 does.

Replicating the analysis for home ownership, Table 5presents models that consider the

time since a person was incarcerated (Model 1) and the length of incarceration (Model 2).

Interestingly, the between-person coefﬁcients in this table indicate that the timing and

length of incarceration were not associated with average net worth between ever- and

never-incarcerated persons, but the within-person coefﬁcients show that, net of stable

unobserved individual-level characteristics, the negative effects of incarceration grew over

time and with the length of incarceration. For instance, formerly incarcerated persons had a

net worth of approximately $105,000 less, on average, in the years where they were

incarcerated 11 or more years ago, when compared to the years prior to their incarceration.

However, the gap for those incarcerated 1–5 years ago was about $45,000. In terms of the

length of incarceration, having spent 5 or more survey-years in prison was associated with

a $71,000 decrease in net worth, compared to the years prior to incarceration, a value more

than double that of those who were incarcerated for only one survey-year. These results

therefore partially support my expectation that incarceration’s negative effects on net

worth would increase over time and with the length of incarceration.

Potential Mechanisms

Isolating the effects of any single variable is a complex process, particularly in the case of

incarceration because it affects so many aspects of a person’s life. It decreases education

and employment opportunities, impedes family formation, and leads to physical and

psychological health limitations (Wakeﬁeld and Uggen 2010). Incarceration is also

implicated in the broader U.S. race discrimination system (Reskin 2012). Incarceration,

therefore, acts a cause and consequence of multiple aspects within my models, which

J Quant Criminol (2015) 31:207–236 223

123

Table 4 Results from hybrid linear mixed models predicting net worth

Model 1 Model 2 Model 3

b SE b SE b SE

Intercept 171,681.9*** (13,278.1) 217,913.2*** (13,346.0) 50,529.0*** (12,581.1)

Time-varying predictors

Within-person coefﬁcients

Currently incarcerated -18,048.7*** (5,432.1) -13,892.3** (5,417.8) -8,955.9 (5,439.2)

Previously incarcerated -67,473.6*** (7,691.8) -60,082.8*** (7,586.4) -41,505.4*** (7,343.7)

Age 3,840.2*** (731.0) 3,327.4*** (715.6) 3,027.3*** (653.7)

Age squared 40.2 (24.2) 110.8*** (24.0) 158.1*** (23.0)

Marital status (ref: married)

Never married -36,316.4*** (2,474.6) -19,700.2*** (2,452.5)

Formerly married -39,561.8*** (2,163.4) -26,214.1*** (2,148.2)

Any children 22,991.0*** (1,850.5) 17,317.5*** (1,820.1)

Any health limitation -10,991.6*** (2,127.8) -8,487.4*** (2,120.7)

Rural 12,715.5*** (2,101.7) 10,561.7*** (2,060.1)

Employment (ref: employed full-time)

Employed part-time 14,164.0*** (1,344.0)

Unemployed 14,400.1*** (2,033.7)

Out of the labor force 18,807.7*** (1,891.4)

Job gaps (ref: no job gaps)

Single job gap -5,882.0 (3,971.0)

2?job gaps -19,404.1*** (4,959.4)

Total earnings (in 1,000s) 951.2*** (30.3)

Self-employed 25,200.7*** (2,560.5)

Government employee -1,628.9 (2,142.5)

Years of schooling 4,376.8*** (1,187.4)

224 J Quant Criminol (2015) 31:207–236

123

Table 4 continued

Model 1 Model 2 Model 3

b SE b SE b SE

Home ownership 58,083.4*** (1,572.2)

Between-person coefﬁcients

Currently incarcerated 6,418.0 (20,383.0) 18,778.8 (19,780.4) -3,494.5 (17,987.9)

Previously incarcerated -28,011.3** (9,061.9) -17,072.8* (8,791.0) 974.6 (7,734.8)

Age 2,203.4*** (548.1) 2,192.9*** (547.7) 94.6 (492.9)

Age squared 30.3 (66.2) 28.9 (64.3) 60.4 (57.7)

Marital status (ref: married)

Never married -79,404.7*** (4,402.8) -15,194.8*** (4,318.8)

Formerly married -101,487.8*** (5,064.9) -26,967.4*** (4,913.1)

Any children -24,653.5*** (4,105.4) -28,338.1*** (3,696.8)

Any health limitation -44,046.5*** (7,351.8) -27,534.0*** (6,971.2)

Rural -15,461.6*** (3,873.3) -10,318.1** (3,457.6)

Employment (ref: employed full-time)

Employed part-time 59,447.3*** (6,436.7)

Unemployed 63,720.0*** (10,413.6)

Out of the labor force 120,419.1*** (5,934.6)

Job gaps (ref: no job gaps)

Single job gap -3,408.8 (3,173.4)

2?job gaps -5,361.1 (2,864.9)

Total earnings (in 1,000s) 2,558.7*** (75.4)

Self-employed 111,233.1*** (7,459.4)

Government employee -13,859.1** (4,548.4)

Years of schooling 5,495.8*** (658.8)

Home ownership 121,660.6*** (4,377.3)

J Quant Criminol (2015) 31:207–236 225

123

Table 4 continued

Model 1 Model 2 Model 3

b SE b SE b SE

Time-invariant predictors

Female -3,631.2 (2,611.2) 2,425.1 (2,724.3) 19,904.8*** (2,648.6)

Black -37,235.5*** (3,949.3) -22,057.9*** (4,040.5) -16,817.7*** (3,640.4)

Hispanic origin -10,399.8* (4,294.2) -9,729.8* (4,310.8) -6,250.1 (3,789.4)

AFQT Score 1,576.7*** (51.9) 1,288.1*** (52.4) 174.7** (57.6)

Used marijuana in teens -3,084.7 (2,921.7) 787.5 (2,845.2) 1,074.1 (2,494.8)

Used cocaine in teens 5,490.2 (3,732.7) 8,117.9* (3,622.7) 7,304.5* (3,161.0)

Used other drugs in teens -20,428.3*** (3,754.6) -15,516.9*** (3,637.1) -8,206.4** (3,177.6)

AIC 2,562,063 2,560,636 2,555,441

BIC 2,562,375 2,561,044 2,556,038

Rho .69 .67 .65

Analyses are for the sample of all individuals. Continuous variables (earnings, age, work experience, grade, and AFQT score) are grand mean centered. All models include

covariates for the cross sectional sample and survey wave year. Within-person coefﬁcients approximate ﬁxed effects. Between-person coefﬁcients approximate random effects

Source: NLSY 1979 Cohort, 1985–2008. N =10,274 individuals and 96,180 observations

*** p\.001, ** p\.01, * p\.05

226 J Quant Criminol (2015) 31:207–236

123

complicates my ability to extract speciﬁc mechanisms, but it also supports incarceration’s
role as an absorbing stigmatized status.
In the case of wealth accumulation, other areas, such as the family or labor market, can
act as mechanisms that inﬂuence home ownership and net worth for former prisoners.
Tables 2and 4show that multiple covariates inﬂuence the relationship between incar-
ceration and wealth accumulation, as the coefﬁcients for incarceration decreased with the
addition of controls across Models 1–3. Due to its absorbing status, it is likely that
incarceration affects other family, employment, and health variables, which in turn
inﬂuence net worth and home ownership. Therefore, in order to parse out some of these
potential pathways, I also tested for mediating effects with marital status, earnings, health,
and home ownership for a previous incarceration.
14
The model results that appear in Tables 6and 7show that the total effect of incar-
ceration on wealth outcomes is much greater than just its direct effect, once indirect effects
are also considered. These tables present coefﬁcients for a previous incarceration’s total
and direct effects on home ownership (Table 6) and net worth (Table 7), as well as the
indirect effects of a previous incarceration on the outcome variables through the listed
mediating variables. In addition to coefﬁcients, I also include estimates of the relative
percentage of the total effect explained by the decomposed relationships.
Except for the between-person effects for net worth, the direct effects of incarceration
accounted for the majority of its effects on wealth, but they played a larger role in
explaining home ownership than net worth. The indirect effects through marriage, earn-
ings, and health limitations explained 14 percent of the within-person effects and 23
percent of the between-person effects for home ownership. These variables, along with
home ownership, accounted for 35 percent of the within-person effects and 99 percent of
the between-person effects for net worth.
Marriage was a strong and consistent predictor of home ownership and net worth in all
models (Tables 2,4). Marriage also mediated the relationship between incarceration and
home ownership (Table 6), but not between incarceration and net worth (Table 7).
Incarceration presented additional indirect effects on home ownership and net worth via
earnings, which was positively associated with home ownership and net worth in all cases
(Tables 2,4). However, earnings was a much stronger mediating variable for the rela-
tionship between incarceration and net worth, than for home ownership, accounting for
17–40 percent of the total effects in terms of net worth. Finally, although health limitations
generally showed a negative association with wealth outcomes, the indirect effects related
to a previous incarceration were negligible. Overall, though, health limitations may not be
the best measure of health status because they often refer to certain disabilities, not speciﬁc
illnesses or diseases.
Home ownership, as one of the major sources of wealth for most people, was by far the
strongest mediator between incarceration and net worth (Table 7). This relationship likely
explains many of the non-signiﬁcant effects of incarceration on net worth in Model 3 in
Table 4, which controls for home ownership. In addition, the indirect effects of incar-
ceration on net worth through home ownership were very large, in terms of both within-
and between-person variation.
In summary, the mechanisms behind these incarceration effects resemble those implied
by labor market research, which include ex-offenders’ limited education, skills, and work
experience and their physical, mental, drug, and motivational problems (Holzer 1996;
14
To simplify this set of analyses, I coded marital status as currently married or not currently married for
the mediation models.
J Quant Criminol (2015) 31:207–236 227
123

Holzer et al. 2003; Pager 2007). I control for many of these mechanisms by including

certain covariates in my models and by focusing on within-person variation over time. I

also test for indirect relationships between marriage, earnings, health limitations, and

wealth outcomes. I ﬁnd that, by reducing ex-offenders’ labor market experience and

Table 5 Results from hybrid linear mixed models predicting net worth based on the time since incarcer-

ation and the length of incarceration

Model 1 Model 2

b SE b SE

Intercept 51,982.5*** (12,583.2) 50,511.9*** (12,580.1)

Time-varying predictors

Within-person coefﬁcients

Incarceration status (ref: never incarcerated)

Currently incarcerated -22,964.2** (7,751.9)

Incarcerated in past year -18,948.5 (9,703.1)

Incarcerated 1–5 years ago -45,296.5*** (9,492.1)

Incarcerated 6–10 years ago -75,595.6*** (10,561.7)

Incarcerated 11 or more years ago -104,552.8*** (12,455.0)

Incarceration length/times (ref: never

incarcerated)

Currently incarcerated -35,612.3*** (7,691.9)

Incarcerated for 1 survey year -33,041.1*** (9,389.2)

Incarcerated for 2–4 survey years -52,949.9*** (9,908.6)

Incarcerated for 5 or more survey years -71,173.9*** (12,956.9)

Between-person coefﬁcients

Incarceration status (ref: never incarcerated)

Currently incarcerated -1,517.3 (16,309.7)

Incarcerated in past year -2,383.1 (14,132.5)

Incarcerated 1–5 years ago 47,740.5 (24,518.1)

Incarcerated 6–10 years ago -18,299.0 (29,256.6)

Incarcerated 11 or more years ago 485.7 (25,520.3)

Incarceration length/times (ref: never

incarcerated)

Currently incarcerated 2,732.9 (17,814.4)

Incarcerated for 1 survey year 1,245.8 (9,471.4)

Incarcerated for 2–4 survey years 6,677.3 (14,810.4)

Incarcerated for 5 or more survey years -18,902.2 (30,508.9)

AIC 2,555,401 2,555,437

BIC 2,556,055 2,556,071

Rho .65 .65

Analyses are for the sample of all individuals. Models include all covariates from Model 3 in Table 4.

Continuous variables (earnings, age, work experience, grade, and AFQT score) are grand mean centered.

Within-person coefﬁcients approximate ﬁxed effects. Between-person coefﬁcients approximate random

effects

Source: NLSY 1979 Cohort, 1985–2008. N =10,274 individuals and 96,180 observations

*** p\.001, ** p\.01, * p\.05

228 J Quant Criminol (2015) 31:207–236

123

earnings, as well as their marriage prospects, incarceration limits their opportunities to

accumulate wealth. Incarceration’s effects on wealth are also related, where home own-

ership mediates its relationship with net worth.

Beyond these mechanisms and mediating relationships, current and previous incarcer-

ations were directly associated with wealth outcomes. This occurs in part because time in

prison reduces a person’s opportunity to accumulate wealth and debt by removing him or

her from society. In this situation, growing legal debt, hiatuses, and periods without

spending money, concomitants of incarceration, can negatively affect future wealth

accumulation. In addition, the fact that conviction and incarceration leave a ‘‘record’’

means that the absorbing status of incarceration confers a stigma that restricts a person’s

ability to re-enter society and start over. In view of evidence that discrimination and the

general stigmatization of formerly incarcerated persons limit their labor market outcomes

(Pager 2003,2007), it follows that incarceration could directly diminish wealth accumu-

lation through similar pathways.

Discussion

My ﬁndings further conﬁrm incarceration’s continuing negative effects on people’s post-

incarceration lives. In addition to employment disadvantages, previously incarcerated

Table 6 Results from hybrid mixed effects mediation models predicting home ownership

Within-person Between-person

Estimate SE Estimate SE

Coefﬁcients

Total effect -1.600*** (.193) -.759*** (.220)

Direct effect -1.376*** (.193) -.571*** (.221)

Indirect effects -.225*** (.031) -.187*** (.032)

Via marriage -.084*** (.027) -.128*** (.029)

Via earnings -.138*** (.013) -.047*** (.009)

Via health limitation -.002 (.002) -.012 (.007)

Relative percentage

Direct effect 85.97 76.92

Indirect effects 14.03 23.08

Via marriage 5.24 16.83

Via earnings 8.64 6.25

Via health limitation .15 1.61

Analyses are for the sample of all individuals. Models include all covariates from Model 3 in Table 2.

Within-person coefﬁcients approximate ﬁxed effects. Between-person coefﬁcients approximate random

effects. The total effect refers to the effect of the initial variable on the outcome controlling for everything

except the mediator variables. The direct effect refers to the effect of the initial variable on the outcome

variable controlling for the mediator variables. The indirect effect refers to the effect of the initial variable

on the outcome variable mediated by the control variables. Indirect effects were calculated using the KHB-

method (Breen et al. 2013). Standard errors and signiﬁcance tests for the indirect effects were calculated

according to Sobel’s (1982) method. Relative percentage refers to the percentage of the total effect

accounted for by the coefﬁcient

Source: NLSY 1979 Cohort, 1985–2008. N =10,274 individuals and 96,180 observations

J Quant Criminol (2015) 31:207–236 229

123

persons also face disadvantages in accumulating wealth, an area of growing interest for

researchers (Turney and Schneider 2014). My ﬁndings show that, in most models, people

who had spent some time in prison were less likely to own their homes and often accu-

mulated less wealth than similar individuals who had never been incarcerated, net of

unobserved individual aspects and other time-varying and time-invariant indicators. More

importantly, formerly incarcerated persons were less likely to be homeowners and held less

wealth on average in the years after incarceration than in the years prior to their

incarceration.

Beyond illustrating this basic relationship, my ﬁndings highlight many of the nuances in

incarceration’s association with wealth accumulation. This relationship was largely med-

iated by the effects of incarceration on marriage, and earnings, as well as by home own-

ership in the case of net worth. Across models, I also allowed the effects of incarceration

on home ownership and net worth to vary based on the person’s time since incarceration

and length of incarceration. These ﬁndings showed that the effects of the timing and length

of incarceration did not differ between never- and ever-incarcerated persons, net of control

variables. In terms of within-person variation, persons incarcerated longer ago and for

more survey-years had lower rates of home ownership and levels of net worth, which is

consistent with the accumulation of disadvantage.

Table 7 Results from hybrid mixed effects mediation models predicting net worth

Within-person Between-person

Estimate SE Estimate SE

Coefﬁcients

Total effect -64,923.36*** (5,683.50) -11,323.05 (6,968.31)

Direct effect -41,902.01*** (5,693.84) -87.69 (6,975.08)

Indirect effects -23,021.36*** (1,336.75) -11,235.36*** (1,202.46)

Via marriage -886.58** (291.31) -364.29* (142.43)

Via earnings -10,822.37*** (887.50) -4,524.54*** (773.94)

Via health limitation -262.82* (129.72) -935.52*** (254.90)

Via home ownership -11,049.58*** (809.27) -5,411.01*** (760.00)

Relative percentage

Direct effect 64.54 .77

Indirect effects 35.46 99.23

via marriage 1.37 3.22

via earnings 16.67 39.96

Via health limitation .40 8.26

Via home ownership 17.02 47.79