.V

3
0.9003 / ’,
54/? ‘12 0 (‘0’!)

 

This is to certify that the
dissertation entitled

LIBRARY
Michigan State '
UniverSIty

ncarceration, Social Bonds, and the Lifecourse

 

 

 

presented by

Beth Marie Huebner

has been accepted towards fulfillment
of the requirements for the

Ph.D. _ degree in _ Criminal Justice

/

J

 

 

Maj Professo sSignature

7 [2103 __

Date

 

MSU is an Affirmative Action/Equal Opportunity Institution

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

MRI 0560mm

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJCIRC/DatoDuopfiS-sz

INCARCERATION, SOCIAL BONDS, AND THE LIFECOURSE

Beth Marie Huebner

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
College of Social Science

2003

ABSTRACT

INCARCERATION, SOCIAL BONDS, AND THE LIFECOURSE

By

Beth Marie Huebner

In the current study, the lifecourse perspective, as posited by Sampson and
Laub (1993), was used to examine the relative effect of incarceration on social bond
attainment. It was hypothesized that individuals who have been incarcerated would be
less likely to attain bonds to marriage and work and the nature of bonds attained would be
further diminished by the event. The hypotheses were tested using data fiom the
National Longitudinal Survey of Youth.

Results from regression and growth curve models confirm the lifecourse
perspective. Across all models estimated, incarceration was negatively associated with
both the likelihood of attainment and the nature of the social bond. A number of
significant relationships were found between static-individual predictors and social bond
attainment; however, individual demographic factors were found only to be moderately
related to the initial status of the individual and had little effect on the nature of change
over time. The findings from this study reinforce the importance of adult social bonds in
determining life trajectories. Implications of these findings are discussed in terms of their

relevance to the study of prisoner reentry.

Copyright by
BETH MARIE HUEBNER
2003

To J.R.S. with all my love

iv

ACKNOWLEDGEMENTS

I have been very fortunate to work with and get to know a number of wonderful
people during my time at Michigan State. Words cannot express the gratitude I have for
each of you. I would like to take this opportunity to thank a few special people. I would
first like to convey my appreciation to my dissertation committee. Dr. Timothy Bynum,
Dr. Christina DeJong, Dr. Merry Morash, and Dr. Kenneth Frank have dedicated
countless hours to assisting me with this work. Thank you for always challenging me to
be a better scholar. I would like to extend special appreciation to my dissertation
chairperson, mentor, and friend Tim Bynum. I am a better person to have worked with
you.

I would also like to thank the many friends that I have made while at Michigan
State. I consider Joe Schafer to be one of my best friends. I look forward to working
closely with you for years to come. I want to thank John McCluskey and Sean Varano,
the original members of Team Bynum, for their constant support and encouragement over
the last five years. I couldn’t have done it without both of you. I would also like to
extend special thanks to Justin Patchin who always made sure that I pulled hard both in
rowing and in academics. Sameer Hinduja also deserves recognition for his enduring
positive outlook on life and his constant willingness to lend a helping hand. I also
enjoyed the long talks about life that I have had with Amanda Burgess-Proctor, and I am
indebted to Karen Ream for the editing assistance and friendship that she has provided. I

am also lucky to have Jeff Cancino, Cathy White, Cynthia Perez McCluskey, Cedrick

Hereaux, Brandon Kooi, Cheryl Reid, Ryan Martz, and Alison Culpepper as colleagues
and fi‘iends.

I would also like to extend special thanks to my family. My parents, Ruth and
Herb Huebner, have always supported me in whatever I have done. Just like at all the
regattas you have attended, your cheers have always been the loudest in the crowd and
have kept me going. My grandmother Barbara Sherman has been an example of strength
and has been an inspiration to all that I do. I would also like to thank my in-laws, Robert
and Merrilee Sachtj en and Bonnie and Jerry Tomczak. I am lucky to have had you
embrace me as family. To the rest of my family including my brother, aunts, uncles,
cousins, and grandparents — thank you.

Finally, I would like to thank my husband and son for their love and support. My
husband Jason Sniff has been my staunchest supporter throughout the long process and
has never questioned the countless hours that I spent in the office or ever doubted that I
would accomplish my goals. Thank you for loving me for better and for worse. I would
also like to acknowledge my son Jackson. Although I know you will never remember

this experience, you have brought me more joy than I ever thought was possible.

vi

TABLE OF CONTENTS

 

 

 

 

 

LIST OF TABLES - - _- IX
LIST OF FIGURES -- - - - - - - - -- - - - -- X
INTRODUCTION: CHAPTER 1 -- - - - - - l
LIMITATIONS OF CORRECTIONAL RESEARCH .............................................................. 4
CURRENT RESEARCH .................................................................................................... 5
SUMMARY ...................................................................................................................... 5
CHAPTER 2: THEORETICAL PERSPECTIVES ...... - -- - - --7
CRIME AND THE LIFECOURSE ....................................................................................... 7
Theoretical Development ........................................................................................... 7
Crime in the Making ................................................................................................ I 7
INCARCERATION, RECIDIVISM, AND REENTRY .......................................................... 24
Individual Level Research ........................................................................................ 25
Recidivism and Reentry ............................................................................................ 26
RESILIENCY FOLLOWING INCARCERATION ................................................................ 28
Employment ............................................................................................................. 29
Attachment to Spouse ............................................................................................... 34
SUMMARY .................................................................................................................... 35
CHAPTER 3: METHODOLOGY ................................................................................ 37
CURRENT RESEARCH .................................................................................................. 37
OVERVIEW OF THE N LSY ........................................................................................... 40
DESCRIPTION OF DATA COLLECTION ......................................................................... 41
SAMPLE ........................................................................................................................ 43
MATCHING .................................................................................................................. 45
Propensity Score Model Specification ..................................................................... 49
Propensity Score Model Statistical Estimation ........................................................ 56
Matched and Comparison Group Estimation ........................................................... 66
Sample Characteristics ............................................................................................ 69
Attainment of Social Bonds ...................................................................................... 71
Nature of the Social Bond ........................................................................................ 72
MEASUREMENT OF VARIABLES ................................................................................... 74
Measurement of Dependent Variables ..................................................................... 74
Measurement of Independent Variables ................................................................... 77
Dynamic Influences .................................................................................................. 77
Static Influences ....................................................................................................... 80
ANALYTIC TECHNIQUE ............................................................................................... 82
RESEARCH IMPLICATIONS .......................................................................................... 86
CHAPTER 4: RESULTS OF ANALYSES- - - - -- - - - -- 88
FIXED-EFFECT MODELS OF SOCIAL BOND DEVELOPMENT ............................................ 89

Vii

Marital Bonds .......................................................................................................... 89

 

 

Marital Attainment by Group ................................................................................... 93
BONDS To EMPLOYMENT .......................................................................................... 104
Employment by Group ........................................................................................... 109
SUMMARY .................................................................................................................. 115
MODELS OF INDIVIDUALS CHANGE .......................................................................... 1 16
The HLM Model ..................................................................................................... I l 7
SUMMARY .................................................................................................................. 147
CHAPTER 5: CONCLUSIONS- - - - - - - -- -- - 152
LIMITATIONS ............................................................................................................. 158
THEORETICAL IMPLICATIONS ........................................................................................ 164
POLICY IMPLICA TIONS .................................................................................................. 166
APPENDICES - - - - - - - - - - - -- _ - -- -170
APPENDIX A: DESCRIPTION OF VARIABLES ................................................................ 171
APPENDIX B: ANALYTICAL MODEL ........................................................................... 173
APPENDIX C. CORRELATION MATRIX FOR SOCIAL BOND To MARRIAGE MODEL ....... 174

APPENDIX D. COLLINEARITY DIAGNOSTICS FOR THE BOND To MARRIAGE MODEL 175
APPENDIX E. CORRELATION MATRIX FOR SOCIAL BOND TO EMPLOYMENT MODEL . 176
APPENDIX F. COLLINEARITY DIAGNOSTICS FOR THE BOND To EMPLOYMENT MODEL
................................................................................................................................... 177
APPENDIX G. FINAL STATISTICAL MODELS FOR THE LIKELIHOOD OF WORK AND
MARRIAGE DEPENDENT VARIABLES — PROPENSITY SCORE ONLY ANALYSES (N =1 ,466)
................................................................................................................................... 178

BIBLIOGRAPHY-m- - - -- - - - - -- -------- ‘--l80

 

viii

LIST OF TABLES

Table 1: Descriptive Statistics for the Variables used in the Development of the

Propensity Scores ........................................................................ 56
Table 2: Logit Model for the Development of Propensity Scores .......................... 58
Table 3: Distribution of the Propensity Scores for Incarceration ........................... 60

Table 4: Distribution of the Propensity Scores for Incarceration — Gender Contrasts. . .60

Table 5: Distribution of the Propensity Scores for Incarceration — Gender and
Incarceration ............................................................................. 61

Table 6: Distribution of the Propensity Scores for Incarceration — Gender and Race 61

Table 7: Logit Models for the Development of Propensity Scores by Gender ............. 65
Table 8: Number of Exact Propensity Score Matches ......................................... 66
Table 9: Descriptive Statistics for the Variables Used in the Development of Propensity
Scores by Group .......................................................................... 68
Table 10: Attainment of Marital Bond ........................................................... 90
Table 11: Marital Attainment in 2000 for Total Sample ..................................... 92
Table 12: Marital Attainment in 2000 by Group .............................................. 96
Table 13: Z Score Coefficients for Sub-Group Comparisons — Marital Attainment Mod9el
............................................................................................ 7
Table 14: Nature of Incarceration and Marriage — Incarcerated Sample Only ........... 100
Table 15: Nature of the Marital Bonds in 2000 for Married Sample ...................... 103
Table 16: Attainment of Employment by Group ............................................ 104
Table 17: Fulltime Employment in 2000 ...................................................... 106
Table 18: Employment in 2000 for Total Sample ........................................... 108
Table 19: Employment in 2000 by Sample Group .......................................... 110

ix

Table 20:

Table 21:

Table 22:

Table 23:

Table 24:

Table 25:

Table 26:

Table 27:

Table 28:

Table 29:

Z Score Coefficients for Sub Group Comparisons — Employment Attainment
Model ................................................................................. 112

Nature of Incarceration and Employment —— Incarcerated Sample Only ...... 1 l4

Variance Components for Random Effects — Likelihood of Marriage and Work
Models ................................................................................. 128

Fixed Effects of Occasion of Measurement Variables on Likelihood of
Attainment of Social Bonds ......................................................... 131

The Effects of Individual-Level Variables on Likelihood of Marriage and

Employment ........................................................................... 134

Fixed Effects of Occasion of Measurement Variables on Likelihood of
Attainment of Social Bonds —- Level II ............................................ 136

Variance Components for Random Effects for Nature of Employment Models
.......................................................................................... 140

Fixed Effects of Occasion of Measurement Variables on Nature of

Work Measures ........................................................................ 137

The Effects of Individual-Level Variables on Nature of Employment ....... 145

Fixed Effects of Occasion of Measurement Variables on Nature of
Work Dependent Variables — Level II .............................................. 137

Figure 1:
Figure 2:
Figure 3:
Figure 4:
Figure 5:

Figure 6:

LIST OF FIGURES

Sampson and Iaub’s Theoretical Model ............................................ 18
Distribution of Propensity Scores for the Total Sample .......................... 59
Likelihood of Employment 1983-2000 ............................................ 126
Likelihood of Marriage 1983-2000 ................................................ 127
Tenure with Employer 1983-2000 .................................................. 138
Cumulative Number of Jobs Worked 1983-2000 ................................. 139

xi

Introduction: Chapter 1

In 2000, about 571,000 prison inmates were released from state prisons back into
the community. At year end 2001, there were over 700,000 persons on parole; this
population represents a three-fold increase Since 1980 (Glaze, 2002). Most of the
individuals returned back to the community will not succeed in abstaining from crime.
The recidivism statistics for incarcerated persons are striking. Of prisoners released from
state institutions in 1994, 68 percent were rearrested for a new offense and 52 percent
were returned back to prison or jail on conviction for a new offense or a technical
violation within three years of release (Langan & Levin, 2002). Among State parole
discharges in 1999, only 42% successfiilly completed their term of supervision (Glaze,
2002)

The increase in prisoner reentry is even more dramatic in light of rising
incarceration rates. In the last twenty-five years the imprisonment rate has increased
nearly fourfold and currently stands at 476 per 100,000 persons. Approximately 1.2
million people are in state prisons on any given day and an additional 600,000 are
incarcerated in local jails (Beck, Karberg, & Harrison, 2002). The ‘imprisonment binge’
(Irwin & Austin, 1994) has taken a substantial toll on both institutional populations and
the community. Few prisoners receive programming while in prison. Less than one third
of all state prisoners in 1997 reported participating in educational or vocational programs
(Lynch & Sabol, 2001). The parole systems have also been inundated with individuals.
Despite this increase, per capita spending on parole services has decreased about $1,500
over the past thirteen years (Lynch & Sabol, 2001). An increasing number of people are

being released into the community ill prepared and with little or no supervision. In light

of the organizational constraints faced by local correctional facilities and the lack of
resources aimed at programming, it is not surprising that prison remains a revolving door.

Despite the dire recidivism statistics, many people do succeed following
incarceration and remain free from involvement with the criminal justice system and
crime. In fact, approximately 40 percent of those individuals released back into the
community are not re-incarcerated following release. Many people do well. Even with
the relative successes, few researchers have examined the factors associated with
resiliency and desistance from crime following release from prison. Most research
conducted to date on incarceration outcomes has been designed to estimate recidivism
(Gendreau, Little, & Goggin, 1996); however, there is great utility in examining
predictors associated with desistance.

Integrating the study of desistance and resiliency into the examination of crime
enhances the research conducted within criminology. Theories of criminology have
traditionally been centered on discovering the factors associated with the etiology of
crime. More recently, lifecourse perspectives have brought to light the importance of
examining different phases of the criminal career. Predictors related to the termination of
a criminal career can have important theoretical implications for the study of the onset of
criminality. For example, if researchers discover that positive social bonds to marriage
increase the likelihood of desistance in adulthood, then it would be important for
researchers to consider if individuals who develop Similar bonds are able to escape
involvement in criminality altogether. Desistance is in many ways the opposite of the
inception of the criminal career where activation refers to the gradual involvement in

criminality, and desistance is signaled by a reduction in criminal behavior and gradual

and eventual conclusion of the career (Loeber & Le Blanc, 1990). Failure to examine all
phases of the criminal career limits the validity of lifecourse perspectives and
criminology in general.

The examination of factors associated with desistance in adulthood also has
important implications for criminal justice policy. Although researchers have begun to
develop comprehensive models of criminal career initiation, it remains difficult to
identify those individuals who will eventually become involved in serious delinquency.
Moreover, there are substantial limitations to conducting research on and intervening in
the lives of youth. Conversely, researchers have greater freedom to become involved
with those individuals who have been formally processed within the criminal justice
system (see Uggen & Piliavin, 1998). Stronger treatments can also be administered to
individuals under legal supervision of the criminal justice system. For example, it is
possible to require participation in a rigorous work treatment program for those
individuals on parole; whereas, this type of intervention would be difficult to conduct
with one who had not been officially sanctioned.

Apart from utilitarian concerns, developing appropriate theories of desistance is
also cost-effective. Institutional placement is extremely costly, especially in light of
current economic crisis and the rising incarceration rate. The reentry point represents an
opportune time in the lifecourse for intervention. Incarceration research has traditionally
been centered on developing appropriate risk prediction instruments. Estimates of risk
are very important for parole decisionmaking and institutional placement, but are of little

assistance in understanding and predicting success following incarceration. The

development of predictive models Specific to desistance will aid researchers and

practitioners alike in crafting appropriate interventions.

Limitations of Correctional Research

Research conducted to date on corrections and reentry has been limited by a
number of predictors. First, much of the literature is centered on the examination of
recidivism following a discrete incarceration event, but more than half of all inmates will
be sentenced to more than one term of incarceration in their lifetime. It is important to
understand how multiple stays of incarceration affect both the relative risk of recidivism
and the likelihood of desistance.

Second, the breadth of the data available to study recidivism has limited current
research. Most research has been centered on the examination of a geographically
limited sample of offenders for a brief time period. Desistance is most ofien considered
a gradual process (Laub, Nagin, & Sampson, 1998). Restricting the time allotted for
subject follow-up truncates the lifecourse experience and inhibits the valid consideration
of the desistance phenomenon. Expanding the nature of the sample and the time frame
for data collection enhances the generalizability and validity of the study.

Finally, the incarceration research has been limited by the nearly singular
emphasis on negative transitions following incarceration. As discussed above, the study
of factors associated with desistance has utility for the expansion of lifecourse research

and is also important for developing appropriate intervention and rehabilitation

programming.

Current Research

The current research serves to address many of the limitations associated with the
existing literature on correctional outcomes and the lifecourse. This research is unique to
corrections research in that the incarceration event is modeled as a lifecourse event. AS
such, a sub sample of a nationally representative, longitudinal dataset was used to
examine how incarceration interrupts and changes life trajectories. Specifically, the
research focused on understanding social bond development following incarceration.
The research has been designed to examine the likelihood of developing social bonds to
work and marriage following incarceration, and how the incarceration event affects the
nature of such bonds. Social bond development has also been operationalized as a
correlate and precursor of desistance; hence, the current work contributes to this body of
literature.

This research also expands current correctional literature by including multiple
measures of incarceration. Incarceration measures were developed to reflect the
frequency and length of incarceration events. In addition, the use of longitudinal data
facilitates the examination of the relationship between timing of incarceration and social

bond development.

Summary

The incarceration and subsequent reentry rates have reached remarkable heights.
The existing corrections research has largely examined the reentry phenomenon in
relationship to recidivism and has ignored factors associated with resilience following

incarceration. The current study adopts a lifecourse perspective to examine incarceration

as an important turning point. The primary goal of this research is to understand social
bond development following incarceration. This research also has important implications

for both theoretical development and correctional policy.

Chapter 2: Theoretical Perspectives

This chapter presents a review of the literature on lifecourse criminology and its
relationship to the study of corrections. First, this chapter examines the general paradigm
of lifecourse criminology. Special attention is paid to the research conducted by
Sampson and Laub (1993). The theory of social capital as posited by Coleman (1988;
1990) is also discussed in relationship to social bond development. Second,
contemporary research on recidivism and prisoner reentry is reviewed. Finally, the
existing research on development of social bonds to marriage and work following
incarceration is discussed. In addition, the review examines the quality of research
conducted to date and draws some conclusions as to what is known about the relationship
between incarceration and successful social bond development and what questions

remain unanswered.

Crime and the Lifecourse
Theoretical Development

The lifecourse perspective is relatively new to the study of criminology. Prior to
the 1970’s, the majority of criminological research was based on static theories of crime
using aggregate, cross-sectional data. Research that was conducted with individual-level
data focused largely on adolescence and ignored early childhood development. This
research failed to provide insight into the heterogeneity among offenders, how
individuals interact with the community, and how criminal involvement changes over

time.

During the past thirty years, a number of seminal longitudinal research studies
have examined crime over the lifecourse (e.g. Elliott, Ageton, & Canter, 1979; West &
Farrington, 1973; Wolfgang, Figlio, & Sellin, 1972). These studies were unique because
they used individual-level measures to examine change in criminal trajectories.
Researchers fi'om this perspective were also the first to incorporate theories and
methodologies from developmental psychology. Much of the prior research in the social
sciences had only examined delinquency in late adolescence and early adulthood;
however, research from the developmental perspective examined how deficits in early
childhood extend to criminality in both adolescence and adulthood (T. E. Mof’fitt, 1993;
T. E. Moffitt, Caspi, Dickson, Silva, & Stanton, 1996; Patterson & Yoerger, 1993). This
research shifted the focus of criminology away fiom the examination of late-adolescent
persistent, high-rate offenders to the study of the etiology of crime.

A central focus within developmental psychology has been the examination of
heterogeneity in lifecourse offending patterns between individuals over time. For
example, Moffitt (1993) argues that individuals who become involved in criminality at a
younger age are qualitatively different from those that begin their criminal involvement
in adolescence. Youth who initiate criminal careers at an early age are substantially more
likely to be involved in criminality later in adulthood when compared to adolescent-onset
offenders. The development of offender typologies also became important as part of this
research. An implicit assumption of this research was that not all offenders experience
criminality and other lifecourse events in the same manner. From this research it became
important to examine individual life trajectories instead of treating youth involved in

criminality as a homogenous class of persons.

Two central findings emerge from early research on lifecourse criminology. First,
there appears to be continuity in offending across time and a strong association between
past and future delinquency at the individual level. Individuals who are involved in
delinquency at a young age are also more likely to be involved in criminality as an adult
(see Blumstein, Cohen, Roth, & Visher, 1986; Tracy & Kemph-Leonard, 1996;
Wolfgang, Thomberry, & Figlio, 1987). Using data from the 1958 Philadelphia Birth
Cohort, Tracy and Kemph-Leonard (1996) found that the majority of sample members
had no record of adult or juvenile crime. Of those who reported having criminal
involvement, the majority had a record of delinquency but had not been involved in
criminality as an adult. One third (32%) of individuals who had been involved in
offending as a juvenile went on to adult criminal careers. In comprehensive reviews of
the literature Olweus ( 1979) and Loeber (1982) also confirm the stability of criminality
across time.

Another consistent finding in the literature iS that aggregate delinquency rates
peak between late adolescence and early adulthood and decline into adulthood. Using
data obtained from the Philadelphia Birth Cohort study of 1967, Wolfgang, Thornberry,
and Figlio (1987) followed a sample of individuals until age thirty and discovered that
delinquency peaks at age sixteen and declines into adulthood (see also Wolfgang et al.,
1972). This finding has been confirmed by a number of researchers using data from
Official arrest reports (F.B.I, 1990), self-reported data (Rowe & Tittle, 1977), and from
populations outside the United States (Farrington, 1986).

Despite the commonality in findings from the early delinquency research, a

number of different theoretical models have been developed to explain the results.

Research within this area can be classified into three groups. The population
heterogeneity explanation sees adult criminality as a function of criminal propensity, or
latent traits, developed in early childhood (e.g. M. R. Gottfredson & Hirschi, 1990).
Conversely, theorists within the state dependence framework argue that criminalin is a
direct result of interaction with the social environment (Blumstein et al., 1986). The final
group includes integrated theories that merge both change and continuity arguments. A
review of the relevant research conducted within each of the three perspectives is

presented below.

Population Heterogeneigr

The stability of crime and antisocial behavior over the lifecourse has been well
documented (Caspi & Bem, 1990; M. R. Gottfiedson & Hirschi, 1990; Loeber, 1982;
Olweus, 1979). A dominant explanation for this stability has been presented by theorists
within the population heterogeneity perspective (see Bushway, Brame, & Paternoster,
1999; Nagin & Paternoster, 1991). Researchers of this tradition have argued that
differences in predisposition to engage in criminal activities explain all differences in
offending over the lifecourse. Latent criminal propensities develop early in life and are
time stable. This theoretical framework is most often associated with the work of
Gottfredson and Hirschi (1990); however, a number of researchers have conducted work
within this framework (Fishbein, 1990; Wilson & Hermstein, 1985). For example,
Wilson and Hermstein (1985) argue that offending can be explained in relation to
criminal propensity, which can include personality traits such as impulsivity, conscience,

and conditionability. Propensity is developed early in life and influences criminality,

10

school performance, employment, and substance abuse. Biological theories of crime also
can be classified as population heterogeneity models because of their emphasis on the
relationship between genetic factors (e. g., chemical imbalances) and criminality (see
Fishbein, 1990).

The General Theory of Crime by Gottfredson and Hirschi (1990) has been one of
the most influential and controversial criminological works. Their population
heterogeneity argument is centered on self-control. Self-control is developed by age four
or five, largely as a result of parental interaction, and remains stable over the lifecourse
(M. R. Gottfredson & Hirschi, 1990:97). Consistent with other theories in the population
heterogeneity framework, they argue that criminal propensity is constant across the
lifecourse; therefore, the best predictor of crime is past criminal behavior (M. R.
Gottfredson & Hirschi, 1990: 107). Based on this theoretical framework, any person who
has inadequate restraint to resist the temptations and gratifications of criminality, and has
an opporttmity to commit a deviant act, will do so. Because criminal propensity is
established at an early age, Gottfredson and Hirschi argue that adult social bonds
established through marriage, meaningful employment, or other control mechanisms will
have little effect on adult trajectories. Any positive correlation found between criminality
and social bonds is spurious. Instead, both processes can be explained in relationship to
self-control.

A number of policies have been proposed based on this theoretical perspective.
Researchers within this population heterogeneity fi'amework have argued that public
agencies should focus on cultivating self-control through social service programming for

youth and their families in early childhood. According to population heterogeneity

11

theorists, the use of any kind of rehabilitation program would be ineffectual. Instead,
programming should focus on early childhood when self-control and other personality
traits are believed to develop. In addition, researchers within the framework contend that
modifications to the criminal justice system in any manner will not reduce involvement in
criminal activities (Gottfredson and Hirschi, 1990). Individuals with low self—control will
not be deterred by traditional criminal justice sanctions. Criminal justice institutions can
be used to protect community safety by separating the offender from society, but the
provision of any form of rehabilitation or prevention services, after the individual has

reached late childhood is futile.

_St_ate Dependence

The second lifecourse perspective has been deemed the state dependence
framework (Brame, Bushway, & Paternoster, 1999; Nagin & Paternoster, 1991). The
state dependence argument has been most often associated with the strain, control, and
labeling theoretical fiameworks. Theorists of this tradition contend that involvement in
crime alters life chances (e. g., ability to vote or attain employment); thereby, increasing
the likelihood of future offending (Nagin & Paternoster, 1991 :166). Involvement in
criminality or deviance at any age changes the offender, making her or him more or less
likely to commit future acts. Most state dependence theorists argue that prior
participation in criminality reduces barriers to fiiture participation making subsequent
deviance more likely. For example, Hirschi (1972) proposed that by committing a crime,
the strength of the offender’s social bond to society is weakened. In damaging the social

bond, the offender limits future legitimate social avenues; therefore, subsequent

12

criminality becomes more attractive. This state dependence framework has also been
identified as the criminal career perspective because theorists in this tradition argue that
early criminal propensity is not necessarily deterministic of adult criminality (see
Blumstein et al., 1986). Instead, the state dependence paradigm takes a dynamic
approach by considering criminality exclusively in relationship to the offenders’

interactions with society.

Age and Offending

The introduction of developmental and lifecourse theoretical perspectives has
fostered a rift in criminology. This divergence in perspectives has grown largely from
the study of the association between age and crime. As stated above, aggregate research
on offending patterns over the lifecourse has found that involvement in delinquency rises
quickly through adolescence, peaks in late adolescence or early adulthood, and declines
steadily into adulthood. The traditional View, associated with the population
heterogeneity framework, has been that the majority of juveniles who participate in
delinquency at a young age will continue to offend into adulthood; although, the
frequency of offending by these individuals will decline markedly over the lifecourse (M.
R. Gottfredson & Hirchi, 1986). Age is correlated with crime. The effects of age do not
depend on demographic influences and cannot be explained in reference to them (Hirschi
& Gottfredson, 19832581). Criminal propensity, developed in childhood, is deterministic
of offending patterns throughout the lifecourse.

Conversely, theorists within the state dependence paradigm make no assumptions

about the fi'equency or duration of offending. Researchers from this perspective also do

13

not deny the evidence of a general age crime curve, but they do argue that there is
variation in criminal trajectories among subgroups when the data are disaggregated
(Blumstein, Cohen, & Farrington, 1988a, 1988b). For this perspective, the key to
understanding crime lies in the identification of sources of variation. This includes
examining individual variation over time and understanding differences between
offenders and non—offenders (Blumstein et al., 1988bz64).

A number of methodological implications follow fi'om this rift. Population
heterogeneity theorists (M. R. Gottfredson & Hirchi, 1986; 1990) have argued that
because criminal behavior is related to stable, underlying group differences, there is little
utility in using longitudinal methodologies (see also Hirschi & Gottfredson, 1983; Hirschi
& Gottfi'edson, 1985). They suggest that researchers pose retrospective questions in a
cross-sectional research design if data are needed on events that occurred prior to the
study period (M. R. Gottfredson & Hirchi, 1986). Criminologists from the state
dependence framework contend that cross-sectional designs are inferior because they
only allow researchers to infer correlation and not causation. Retrospective panel designs
are also not ideal to collect historical data, namely because of problems with imperfect
participant recall (Blumstein eta1., 1988b; Greenberg, 1985). Researchers from the state
dependence perspective agree that prospective longitudinal designs, although not without
problems, allow researchers to obtain data proximate to the event and to make valid
generalizations from the data.

Although the theoretical debate between continuity and change over the lifecourse
remains an important topic of research in criminology to date (see Geis, 2000; Hirschi &

Gottfredson, 2000), a number of researchers have now embraced integrated models. In

14

fact, most researchers widely agree that it is important to take into account stable
individual differences when estimating time-variant models (see Brame et al., 1999;
Nagin & Farrington, 1992a, 1992b; Paternoster, Dean, Piquero, Mazerolle, & Brame,
1997). Research from this debate has spawned a new class of work that melds the central
tenets fi'om the population heterogeneity and state dependence frameworks. A summary

of the lifecourse perspective is presented below.

Lifecourse Perspectives

It was also during this time of great debate within criminology that a new class of
research began to emerge. Hagan & Palloni (1988) argued that criminological research
from both the population and state dependence traditions had been too narrowly focused
on the peak years of criminality (e.g., late adolescence and early adulthood) and high-rate
offenders. The authors encouraged researchers to broaden their examination of
criminality and focus on crime as an element in the larger context of the lifecourse.

Researchers also introduced the concept of a trajectory during this time. A
trajectory represents the long-term development and behavior of an individual over the
lifecourse. Using data from the lifecourse allows researchers to examine criminality in a
larger context of other life events (see Elder, 1985). Crime can represent a turning point
in the lifecourse, but it is not an isolated event. Involvement in crime cannot be separated
from the individual and must be examined in light of other life transitions (e.g., marriage,
educational attainment). The criminal event signals a transition in the lifecourse, and
research should seek to understand how individuals adapt and change following the

transition. Different adaptations to the same event can lead to very different trajectories

15

(Elder, 1985:35). The trajectory of an individual who responds to a narcotics arrest by
seeking treatment could be very different from the trajectory of another individual who
chooses to return to the same community and maintain involvement with crime. Said
another way, crime and incarceration are but events in the larger trajectory; it is the
individual’s response to these events that influences future outcomes.

The timing and the nature of an event can also have a substantial effect on
lifecourse trajectories. Researchers have stressed that both delays in achieving transitions
or milestones and early or precocious transitions can have detrimental effects on firture
trajectories (Caspi & Bem, 1990; Rindfuss, Swicegood, & Rosenfeld, 1987). The time
between transitions has also been linked to differential outcomes (Hogan, 1978, 1980).
For example, it is important to examine not only the attainment of the marital bond, but
also the timing and the nature of the bond. In general, bonds resulting fiom marriage
have been associated with desistance; however, marriage that occurs close to the birth of
a child does not have the same deterrent effect (Laub et al., 1998). Again, social events
do not exist in isolation; the social context and timing of the lifecourse event and the
individual adaptation to the event are as important as the event itself.

The lifecourse perspective is particularly relevant to the study of incarceration and
prisoner reentry. Researchers within the lifecourse perspective include incarceration in a
class of general social events, like marriage and illness, which can impact life trajectories
(Sampson & Laub, 1993). Incarceration is not an isolated event. Instead it is a part of a
larger life trajectory that can be characterized by a number of life changes.

Sampson and Laub (1993) have developed one of the most preeminent lifecourse

perspectives on criminology. Their model rests on the assumption that social bond

16

development over the lifecourse can have profound effects on both conformity and
criminality. Details of the theoretical model posited by Sampson and Laub are presented
below. Specifics on the application of their model to the study of incarceration are also

discussed.

Crime in the Making

Building on the work of the state dependence and offender heterogeneity
perspectives, Sampson and Laub (1993) developed an age-graded integrated theory of
offending (see also Laub & Sampson, 1993; Sampson & Laub, 1990, 1992). Their
theoretical model is a unique contribution to criminological theory in that it incorporates
both continuity and change. They understand the importance that criminal propensity can
have on individual inclination toward criminal behavior; however, they also argue that
social bonds both as a juvenile and as an adult can insulate offenders from criminality
and influence criminal trajectories. Prior criminality is also included in the model in that
the researchers argue that prior criminality can have an influence on future trajectories by
‘knifing of certain life avenues.

In order to test their hypothesis, Sampson and Laub reanalyze Sheldon and
Eleanor Gluecks’ data presented in Unraveling Juvenile Delinquency (1950b). The
sample included 500 White males between the ages often and seventeen from Boston
who had been committed to one of two Massachusetts’s juvenile correctional facilities.
The Gluecks’ also carefiilly constructed a matched sample of 500 non-delinquents
recruited from Boston area public schools. A graphical depiction of Sampson and Laub’s

theoretical framework is presented in Figure l.

17

 

Childhood Adolescence Adulthood

Criminal Offending——> Criminal Offending -———+Criminal Offending
A

Family Relationships, Employment and
School, Peers Marriage

 

Criminal Propensity

 

 

 

Figure 1. Sampson and Laub’s Theoretical Modell

Child Social Bonds and Continuity across the Lifecourse

 

Sampson and Laub present their theoretical model in three parts. The first phase
of the analysis seeks generally to examine why only some individuals, raised in
neighborhoods with structural and economic disadvantages, go on to be involved with
delinquency as an adolescent. The authors posit a control-centered model and contend
that weakened ties to social institutions in childhood and adolescence increase the

likelihood of involvement in delinquency. As part of their test of the control model, the

 

' Figure l was adapted from a figure developed by Bushway, Brame, and Paternoster (1999).

18

authors also include measures of the adolescent’s criminal history and family structure.
In their model criminal propensity, is operationalized using measures of early childhood
behavioral problems. Antisocial behavior also serves as proxy for self—control as
developed by Gottfiedson and Hirschi.

The authors’ analyses establish the importance of early childhood bonds in
mitigating future delinquency. Attachment to family and school are found to be the most
powerful predictors of delinquency in adolescence. Net of early childhood propensities,
individuals who were part of a family characterized by consistent parental discipline,
strong attachment, and parental supervision were significantly less likely to be involved
in delinquency (Laub & Sampson, 1988b; Sampson & Laub, 1993:93). Attachment to
school also had a large negative effect on delinquency net of family controls (Sampson &
Iaub, 199321 19). More importantly, Sampson and Laub discovered that informal social
controls consistently mediate the effect of both individual propensity and structural
background variables. It does appear that there is a small direct effect of childhood
behavioral problems on delinquency; however, the relationship is not strong enough to
predict a substantial amount of variation in delinquency, especially in light of family
factors. This research confirms the importance of early social bonds in predicting
adolescent delinquency.

Not only do weak social bonds increase the likelihood of involvement in
delinquency for adolescents, diminished social bonds continue to shape criminal
trajectories into adulthood. The results presented by Sampson and Laub (1993) support

much of the criminological research that suggests that there is continuity in childhood

19

behaviors that extend through adulthood (see also Laub & Sampson, 1993). Individuals
who participate in delinquency as youth are more likely to have adult criminal histories.
Sampson and Laub’s theoretical argument explaining continuity over the
lifecourse diverges from that of traditional criminology. Continuity does not develop as a
result of criminal propensity. Instead, individuals who have weakened bonds to social
institutions in adolescence can be Shut out of opportunities as adults. This argument,
most often associated with the cumulative continuity or cumulative disadvantage
perspective, posits that criminal behavior has, “a systematic, attenuating effect on the
social and institutional bonds linking adults to society” (Laub & Sampson, 1993:306).
For example, juvenile delinquency can preclude individuals from participating in
educational and training programs limiting the possibilities for meaningful attachment to
work in the future. In short, cumulative disadvantage links investments made in youth to

their life trajectories towards or away from delinquent involvement.

Prosocial Bonds in Adulthood

The finding that individual delinquency inhibits development of prosocial bonds
in adulthood conversely suggests that the attainment of strong bonds in adulthood may
attenuate the effect of delinquency on adult outcomes. Sampson and Laub (1993)
introduce change as an essential part of understanding how, in adulthood, adolescent
offenders are able to realign their criminal trajectory in adulthood toward conformity.
They argue that trajectories of both crime and conformity can be interrupted by social
institutions in the transition to adulthood (see also Laub & Sampson, 1993). Specifically,

strong, quality social bonds in adulthood that create interdependent systems of obligation

20

and restraint are the best predictor of desistance in adulthood (Sampson & Laub,
1993:141). Among members of the sample, adult social bonds to work and marriage
explained significant variation in adult crime, independent of childhood deviance (Laub

et al., 1998; Sampson & Laub, 1993).

Social Ca ital

An integral part of the lifecourse model as posited by Sampson and Laub is the
development of social capital in adulthood. Most lifecourse research has only examined
the attainment of the bond itself in relationship to trajectory change; however, Sampson
and Laub (1990; 1993) argue that the quality or strength of social ties are more important
than the development of the bond per se. The social investment in a relationship dictates
the salience of the bond. It is in relationship to social bond development that Sampson
and Laub incorporate the concept of social capital. The more social capital the individual
acquires from a social relationship, the stronger the social bond.

It is important to consider social capital in the larger purview of the theoretical
construct of capital. The concept of capital was initially posited by Marx (1887) who
argued that economic capital provides individuals with the opportunity to purchase
resources that may aid achievement. This construct was based solely on monetary
exchange and did not incorporate individual differences in the ability to generate capital.
More recently Becker (1964) and Schutz (1961) have expanded the concept of capital to
include human capital. Human capital refers to the capacity of individuals to attain
within themselves, through education or other training, the ability to facilitate social

outcomes (Paxton, 1999:92). The concept of human capital is unique in that it moves the

21

concept of capital from a characteristic of a monetary exchange to an entity that can be
attained by individuals.

Social capital expands the theory of capital to include the examination of
relationships. The concept of social capital was first developed by Pierre Bourdieu
(1986) and then expanded by Coleman (1988; 1990). Bourdieu defined social capital as,
“the aggregate of the actual or potential resources which are linked to . . . membership in
a group” (p.248). Social capital can be viewed generally as a resource present within
communities that facilitates collective action; however, membership in a group, per se, is
not enough to facilitate social capital. Instead, social capital involves transforming daily
relationships into those that duly feel and imply durable obligations (pp. 249-250). The
level of social capital possessed by a group, according to Bourdieu, is in direct
relationship to the networks of relationships that can be mobilized and the amount of
capital possessed by members of that network.

Social capital as posited by Bourdieu was a strictly organizational construct.
Coleman (1988; 1990) has expanded this work to examine how individuals glean capital
from their membership in a group. Coleman argues that social capital does not rest in the
individual; instead social capital develops out of interactions between actors. Both the
group and its members can use the capital produced as a result of the networks for the
achievement of goals. Social capital is productive, making achievement of certain ends
possible (Coleman, 1988:898). Like Bourdieu, Coleman argues that not all network ties
will facilitate capital. Capital develops from relationships that are positive and based on

trust.

22

In short, when examining social capital or social bonds it is important to consider
both the network relations linking people together in a social space, in addition to the
quality of the relationships that tie the group members together. Similarly, Sampson and
Laub (1993) stress the necessity of examining the nature and quality of social bonds. Not
all bonds are characterized by high levels of capital. It is not the bond itself that
attenuates criminal careers, it is the development of social capital that strengthens the

relationship and mitigates involvement in future crime.

Sogial Bonds, Social Capital and the Lifecourse

Criminologists have begun to incorporate the concept of social capital into the
study of criminological problems; however, the theory has yet to be systematically
applied in the research. To date, the theory has been applied largely to the study of
communities, but has also been used to examine the effectiveness of community policing
(e.g. Manning, 1994), general crime rate (e.g. Rose & Clear, 1998), and homicide
(Rosenfeld, Messner, & Baumer, 2001). Macro-level examination of social capital has
been important for the study of community capital and societal outcomes, but this
research provides little insight into how individual capital can facilitate positive outcomes
on the individual level.

Most recently, social capital theory has been used to study individual level
delinquency. Researchers have tied low social capital in families to youth homelessness,
high rates of juvenile crime and violence, and adult criminality (Hagan & McCarthy,
1997; Wright, Cullen, & Miller, 2001). In delinquency research, social capital has been

studied as a buffer developed through a strong relationship with a family. Individuals

23

with strong family bonds have less involvement in delinquent acts themselves, but are
also less likely to participate in secondary correlates of delinquency like association with
delinquent peers and drug use (Wright et al., 2001). To date, the concept of social capital
has not been Specifically applied to the study of the relationship between incarceration
and reentry outcomes, but initial research has documented the low availability of capital
for offenders, especially for incarcerated women (see Reisig, Holtfreter, & Morash,
2002). Low social capital compounded with removal from the community only further
reduces social ties increasing the likelihood of subsequent involvement in criminality.

Although the application of the social capital framework to the study of
criminality is still in its infancy, the initial results reinforce the work conducted by
Sampson and Laub (1993). Social bond development can assist in the development of
social resources that aid in achievement of desistance from crime. Below, a general
review of the existing research on incarceration is discussed. This research provides
firrther evidence of the salience of incarceration as a lifecourse event. The final section
of this chapter addresses the link between incarceration, social capital, and bond

development.

Incarceration, Recidivism, and Reentry

As noted above, the goal of the current research is to examine incarceration as a
lifecourse event. The existing correctional research provides a context for the current
study. Research on individual correctional decisionmaking and recidivism is presented

below.

24

Individual Level Research

Research that has examined imprisonment at the individual level has been
primarily limited to the study of judicial decisionmaking. The extant literature has
focused primarily on two models of decision making, the rational legal model and the
extra-legal model. The extra-legal or conflict model rests on the assumption that certain
groups of people are more likely to be punished harshly. Specifically, researchers have
found that poor, young, minority males are most likely to be incarcerated (Spohn &
Holleran, 2000; Darrell Steffensmeier, Ulmer, & Kramer, 1998). Initial research
discovered inconsistent effects of race on the sentencing decision; however, the
preponderance of research conducted on gender and sentencing decisions indicates that
women are less likely to receive imprisonment when compared to their male counterparts
(Darrell Steffensmeier, Kramer, & Streifel, 1993).

More recently research has identified models that estimate the interaction of
gender, race, and age on sentencing decisions. Overall, young Black men are most likely
to receive a term of incarceration with the impact of race declining as defendants age
(Darrell Steffensmeier et al., 1998). Steffensmeier et a1. (1998) found that independent
of age, Black women are more likely to be imprisoned than White women. The
intersections research provides important insight into the study of incarceration.

Despite the strides that researchers have made in studying individual
decisionmaking, there are many methodological limitations to individual-level sentencing
research. Specifically, much of research suffers from sample selection bias in that data
are most often obtained from one courtroom at one time period. This research is also

imperfect because only the subset of the offender population that reaches the sentencing

25

decision point is considered. The current study improves on the extant decision making
literature in that it examines the incarceration event as a continuous opportunity over the
lifecourse. This modeling strategy allows the researchers to understand how different
individuals reach incarceration, specifically those of different social and racial groups,

and to compare the trajectories of those individuals over time.

Recidivism and Reentry
Research on the outcomes of imprisonment has focused primarily on recidivism.
The development of appropriate predictors of recidivism is an important goal of the
criminal justice system in that findings from this research can be used to develop and
enhance offender treatment programs and appropriate risk models for sentencing and
parole. Researchers have identified two general classes of predictors of recidivism: static
factors which are part of an offender’s past that are immutable (e. g., race) and dynamic
factors which refer to the variable components (e.g., attitudes, cognitions, and values) of
an individual (Andrews & Bonta, 1994; Andrews, Bonta, & Hoge, 1990; Gendreau et al.,
1996). Researchers have found that both static and dynamic factors aid in the prediction
of recidivism. Overall, age, criminal history, gender and family factors are the most
potent static predictors; whereas, the dynamic factors which includes criminogenic needs
(e.g., antisocial behavior), is Significantly associated with recidivism (see Gendreau et al.,
1996 for a review).
In addition to including individual, offender specific characteristics in models of
recidivism, researchers have also considered the nature of the incarceration experience as

a predictor of re-arrest and reentry. The results of the research on time served and

26

recidivism has been mixed. Some researchers have found that time incarcerated is
directly and positively related to incarceration (D. M. Gottfredson, Gottfredson, &
Garafolo, 1977; Schmidt & Witte, 1988), whereas, others were unable to replicate the
relationship (DeJong, 1997; Jones & Sims, 1997; Orsagh & Chen, 1988). A recent three-
year follow up of a national sample of prisoners released in 1994 also found no
relationship between length of incarceration and recidivism (Langan & Levin, 2002).

Researchers have also examined the possibility of a non-linear relationship
between time served and incarceration. Gottfredson et a1. (1977) and Gainey, Payne, and
O'Toole (2000) found evidence of an inverted U shape relationship between time served
and recidivism; whereas, there is a positive effect for shorter sentences and evidence of a
deterrent effect following long stays of incarceration. Orsagh and Chen (1988) also posit
a curvilinear relationship between length of incarceration and recidivism; however, they
argue that time served has a negative impact in the short term.

Research on time to failure has discovered that the first year following release is
the most critical period in relationship to recidivism. For example, Langan & Levin
(2002), found that over Sixty percent of the total re-arrests came in the first year
following release; nearly half of all new convictions also came within the first year.
DeJong (1997) also observed that time to failure was Shorter for first time offenders;
whereas; length of incarceration was positively associated with time to failure for
offenders with prior criminal histories (see also Gainey et al., 2000).

The extant recidivism research provides important insight into individual factors
associated with recidivism; yet, this research is not without limitations. The

methodologies used by the researchers are inconsistent; hence, it is difficult to compare

27

findings across studies. Moreover, data for this research are oflen culled from official
records and do not include sufficient data on dynamic risk predictors (e.g., antisocial
behavior) that may temper the relationships found.

Most importantly, the recidivism research has primarily examined negative
transitions, and not considered factors associated with positive community reentry. The
recidivism literature was used in this paper to inform research on positive lifecourse
transitions into. Although it is not the primary aim of the paper, this study also tests if the

factors associated with negative transitions are also related to successfirl reentry.

Resiliency Following Incarceration

Traditionally, research has focused on the etiology of crime; however, little is
known about why people desist fi'om crime (Uggen & Piliavin, 1998). The research that
has examined factors associated with desistance is mixed (see Laub & Sampson, 2001 for
a review). Initial results indicate that risk factors predicting offending may also have
utility in desistance models (LeBlanc & Loeber, 1993). That said, researchers have also
argued that it is important to develop separate models of desistance because the
functional causes of desistance from crime are different from that of causes of crime
(Uggen & Piliavin, 1998).

There is preliminary evidence that desistance can be linked to development of
social bonds. Family formation and employment have been the most salient predictors of
desistance from crime and other problem behaviors (e.g., alcoholism) in adulthood (Laub
& Sampson, 2001). In the same light, researchers have provided evidence that formal

contact with the criminal justice system reduces the likelihood that an individual will

28

develop positive social bonds and consequently desist from crime (Freeman, 1991;
Sampson & Laub, 1993; Western & Beckett, 1999). Although there is some
inconsistency in the results, researchers have argued that incarceration, especially at a
young age, closes the doors for the development of positive social bonds; hence, limiting
legitimate opportunities for change (see Freeman, 1991; Western & Beckett, 1999).

The nature of the causal relationship between social bond development and
desistance is still unclear. Researchers have argued that the internalized commitment to
the relationship or institution associated with the social bond itself (Sampson & Laub,
1993) reduces the likelihood of future criminality; whereas, others contend that the
relationship between social bond attainment and desistance is more indirect. For
example, Warr (1998) stated that desistance is often correlated with marriage because the
process of being involved in a relationship reduces opportunities to interact with
delinquent others and situations. Because of the nature of the data available for the
current research, it is not possible to examine the change in peer relationships and social
contexts as a result of social bond development; however, this line of inquiry is important
for future analyses. This research is centered on the examination of the attainment and

the quality of the social bond itself.

Employment

As discussed, both negative and positive life events can affect adult trajectories.
The timing of an event is also significant, especially for employment. Early adulthood is
an important time when individuals enjoy strong earnings grth and can attain firm-

specific human capital (Western, Kling, & Weiman, 2001). This phase in the lifecourse

29

is especially fimdamental in that individuals gain Skills and lay the groundwork for future
careers. It follows that changes in employment status, especially in late adolescence and
early adulthood, will have a significant effect on desistance and other outcomes.
Research on employment and desistance has been limited; however, initial research does
reveal the importance of the quality and nature of employment in the examination of
outcomes. The following section outlines the research conducted on employment and
desistance. Specific attention is given to work on the quality of social bonds. In
addition, the research on the attainment of employment following incarceration is
reviewed.

The relationship between unemployment and crime is still being debated by
theorists (e.g. Paternoster & Bushway, 2001). At the aggregate level, the connection
between labor market variables and crime has been well documented (see Chiricos, 1987
for a review); however, the work on individual level outcomes has been inconsistent.
Much Of the research suggests that the relationship between unemployment and crime is
reciprocal, but definite conclusions have yet to be made (Baron & Hartnagel, 1997;
Good, Pirog-Good, & Sickles, 1986; Thornberry & Christenson, 1984). Current work on
criminality and labor force outcomes has shifted away fiom the examination of
unemployment to consider the relationship between attainment of employment and
desistance.

There is initial evidence that the quality of work can have a differential effect on
criminality. Researchers have found that individuals employed in higher quality jobs are
less likely to recidivate (Allan & Steffensmeier, 1989; Duster, 1987). The conclusions

that can be drawn from this research are limited. This research has been mired by self-

30

selection problems in that it is difficult to separate the pre-existing characteristics of
workers from job effects. Many offenders do not have the human capital to attain a high
quality position; hence, it is difficult to separate the attainment of quality work from the
characteristics of the offender.

Researchers have also operationalized participation in the workforce in
relationship to both attachment to work and consistent involvement in the labor force.
Initial evidence suggests that both predictors are important in examining desistance. For
example, Sampson and Laub (1993) found that job stability was central in explaining
desistance from crime. Subjects with low job stability at ages 17-25 were four times
more likely to be arrested in the future (179). Sampson and Laub posit that work per se
does not lead to desistance. Instead, work leads to internalized social controls through
commitment and stability thereby reducing the likelihood that the individual will sever
the bond by engaging in criminality (Sampson & Laub, 1993:304). Uggen (2000) also
found that employment was Significantly related to desistance. This research is unique in
that it examines data fiom the National Supported Work Demonstration Project. Project
participants were randomly assigned to groups where the experimental group was given
career counseling and given a minimum-wage job. The experimental design reduces
many of the methodological concerns (e. g., sample selection bias) associated with prior
research. Uggen found that the effect of participation on recidivism varied with the age
of the offender. Older participants were more amenable to the impact of the work
program. Participation in the program was significantly associated with desistance from
crime for those offenders over the age of twenty-six. The findings from this research

support an age-graded model of employment.

31

Overall, individuals who have developed an attachment to work are less likely to
be involved in criminality. Employment represents an important social bond on the road
to desistance; therefore, it is important to understand how people attain social bonds to
work following incarceration. A number of researchers have examined the relative effect
of incarceration on labor force outcomes (see Western et al., 2001 for a review). When
compared with other criminal justice outcomes (e. g., arrest or conviction), researchers
have found that individuals with incarceration histories are less likely to be employed and
to earn a high wage (Freeman, 1991; Grogger, 1995; Western & Beckett, 1999).
Incarceration is both significantly and negatively associated with labor force outcomes;
however, there is inconsistent evidence on how, or if, the relationship between
incarceration and employment varies over the lifecourse.

Juvenile incarceration has also been consistently linked with reduced earning
potential in adulthood. Researchers who have used the National Longitudinal Survey of
Youth (N LSY), have consistently found that imprisonment as a juvenile had a substantial
long term effect on earnings (Fagan & Freeman, 1999; Freeman, 1991; Western &
Beckett, 1999). Western and Beckett (1999) found that fifteen years after release, the
earning potential of individuals who were incarcerated as youth was still diminished. The
coefficient for incarceration far exceeded those for dropping out of high school and living
in an area with high unemployment ( 1048). These findings reinforce the cumulative
disadvantage perspective presented by Sampson and Laub (1993). Youth who are
incarcerated at a young age are less likely to attain the human and social capital needed to
procure a successful job. Incarceration at a young age can mortgage the firture of youth

by limiting avenues for meaningful employment.

32

The relationship between adult incarceration and employment outcomes has been
less clear. For example, Western (2002) found that incarceration had an initial negative
effect on employment, but the effect of imprisonment on earnings is reduced three to four
years after returning to the workforce. Conversely, Waldfogel (1994) found that
incarceration was negatively associated with both the employment rate and earnings
potential (see also Kling, 1999).

The relationship between adult incarceration and employment also varies by the
characteristics of the offense. Both Kling (1999) and Waldfogel (1994) found that
White-collar offenders were differentially affected by incarceration. White collar
offenders earned ten to thirty percent less following incarceration; whereas, individuals
convicted of Violent or drug offences actually reported an increase in their earning
potential five to eight years following release from prison (Kling, 1999).

Based on the cumulative disadvantage proposition, one would expect that the
length of incarceration would be negatively related to work force outcomes. While
incarcerated, individuals would be less likely to participate in training programs and to
maintain and develop other employment-specific skills. Individuals are also removed
fiom social networks that may facilitate employment and reinforce social bonds to
employment. Using unemployment insurance data, both Kling (1999) and Needels
(1996) found that employment rates did not vary in relationship to time served. Needels
(1996) did find that the length of incarceration was both negatively and significantly
associated with earnings. Increasing incarceration length by one year reduced earnings

by 12 percent over an eight-year period.

33

Overall, adult incarceration does not appear to preclude employment following
incarceration; but wage data does suggest that the same quality or prestige of position
may not be able to be maintained following incarceration. The extant research has
provided important insight into the relationship between work, incarceration, and
desistance; however, additional research is needed. Specifically, the relationship between
the length, nature, and timing of the incarceration is in need of further examination. The
current research serves to further expand the existing research on incarceration and

employment.

Attachment to Spouse

Like work, marriage also engenders social bonds and can serve as a turning point
in the lifecourse. Sampson and Laub (1993) found that for married members of the
sample, attachment to a spouse assumed greater relative importance than job stability (see
also Laub et al., 1998). Marital attachment was negatively and significantly related to all
measures of deviance. Homey, Osgood, and Marshall (1995:665) also found that
cohabitation with one’s wife doubles the odds of desistance from criminality; whereas,
leaving the marital home increases the likelihood of offending. The relationship between
marriage and desistance is not instantaneous. Instead, the influence of marriage is
gradual over time (Laub et al., 1998). The timing of marriage is also important.
Individuals who married young were more likely to refrain fiom criminality (Laub et al.,
1998). Ouimet and Le Blanc (1996) argued that the relationship between marriage and
desistance varies by age. Cohabitation at an early age was positively associated with

crime; however, marriage after age 21 acted as a force toward desistance.

34

The transition from incarceration to marriage has also not been widely addressed
in the literature. The majority of work conducted in this area has been qualitative and has
examined the detrimental effect that incarceration has on current marital bonds (Goeke,
1980; Schafer, 1994). The literature in this area has also been based on research that used
small sample sizes and only examined the relative effect of prison stays on marital bonds.
To date, research has not been conducted on the relationship between the timing, nature,
or length of incarceration on marital development.

Although marriage has traditionally not been examined as a measure of resiliency
following incarceration (with exception Horney et al., 1995) including measures of
marital bonds in this study provides a comparison point for both the effect of the
incarceration event itself and as an additional measure of resilience following
incarceration. As Sampson and Laub (1993) found, marriage can mitigate the
relationship between work and criminality; hence, it is also important to examine the

relationship between marital bonds, employment, and incarceration.

Summary

The literature reviewed in this chapter highlights the importance of adult social
bonds in mitigating criminal careers. Researchers have found that individuals who
develop strong bonds to work, family, and other social institutions are significantly less
likely to be involved in criminality (Sampson & Laub, 1990, 1992). In the same light,
attainment of social bonds in adulthood can interrupt criminal careers (see 1.an et al.,

1998).

35

The study of desistance and social bond development has important implications
for the study of incarceration. The research on social bond development that has been
reviewed in this chapter points to the difficulties in attaining positive social bonds
following incarceration. In general, the research suggests that incarcerated persons are
less likely to develop bonds to marriage; however, the relationship between incarceration
and employment is more complex. Recent research has discovered that offenders are
generally able to attain employment following incarceration, but the quality of the work
attained and the wage levels garnered are reduced as a result of incarceration (e.g.
Western & Beckett, 1999). This is an important finding because the development of
social bonds has been consistently linked to desistance from crime.

Further research is needed on the relationship between social bond development
and incarceration. The existing literature has not considered how the timing and nature
of incarceration is related to the development of social bonds to marriage and work.
Moreover, the cumulative effect of multiple terms of incarceration also has been ignored.
This omission is striking in that results from the recidivism literature indicate that the
majority of incarcerated individuals experience more than one stay of incarceration
during their lifetime. The present study expands on the current incarceration and social
bond literature by exploring social bond development following incarceration.
Specifically, this research has been developed to aid in the understanding of the relative

effect of incarceration on social bond development to work and marriage.

36

Chapter 3: Methodology

Current Research

Few researchers have examined incarceration as a lifecourse event. Studies of
incarceration in general and as a lifecourse phenomenon have been limited by three
primary factors. First, most of the existing research conducted to date has
operationalized incarceration as a dichotomy. For example, incarceration is modeled at
one point in time and reflects the absence or presence of the event during a certain period
(e.g. Freeman, 1991). Second, much of the research has been cross-sectional and
therefore has not examined the possible cumulative effect of multiple stays of
incarceration. Third, the scope of the previous work has been limited by only focusing
on negative transitions following incarceration (i.e. recidivism). The current study has
been designed to overcome many of the limitations associated with previous correctional
research. Unlike other studies performed, this current research examines the relationship
between the frequency and timing of incarceration and resiliency as measured by positive
social bonds to marriage and employment.

The following research questions guide the analyses presented below.

1) How does incarceration affect life trajectories?
a. Specifically, are members of the sample able to attain bonds to marriage and
work following incarceration?
b. How does the development of bonds among the incarcerated group vary from

that of a Similarly situated group that did not experience incarceration?

37

2) Does incarceration remain a significant predictor of social attainment, even after
controlling both time-varying covariates and static factors (e. g., race)?

a. What is the relative effect of incarceration on social bond development in
contrast to other lifecourse events (e.g., military participation)?

3) Does the timing of the incarceration event differentially affect the likelihood of
developing bonds to work and marriage, and is the nature of the bond diminished
over time as a result of the event?

a. Does incarceration at an earlier age have a differential effect on social bond
development when compared to incarceration at early or middle adulthood?
b. How do multiple stays of incarceration affect the likelihood of and the nature

of the bond over time?

In many ways, this research is guided by the theoretical work proposed by
Sampson and Laub (1993); although the current study also serves to expand their
research in a number of ways. First, a large, nationally representative sample was used to
test the hypotheses. The nature of the data allowed the researcher to make
generalizations to a more diverse population, including females, and the large sample size
facilitated the development of both a matched and control sample that was not possible
with the Glueck data utilized by Sampson and Laub (see Glueck & Glueck, 19503). The
development of two comparison samples allowed the researcher to examine the relative
effect of incarceration on outcomes for an incarcerated group, a matched group that did

not experience incarceration, and a comparison group.

38

The current study also utilized a relatively contemporary data set. A guiding
assumption of lifecourse theories is that age-graded transitions are embedded in social
institutions and are subject to historical change (Elder, 1975). It is important to consider
the differential effect that lifecourse events may have on individual trajectories within
different historical epochs. Researchers have amassed considerable evidence on the
effect that economic and political changes can have on correctional policy and
incarceration rates (Elder, 1975; Jacobs & Helms, 1996; Lessan, 1991; Michalowski &
Carlson, 1999; Rusche & Kirchheimer, 1939). It follows that the influence of
incarceration as a lifecourse event may also vary over time. The dramatic change in
penal policy of the United States over the last two decades has changed the manner in
which individuals experience incarceration (Irwin & Austin, 1994). Rehabilitation
services have been replaced in prison with mechanisms of control. Consequently, the
relative effect of incarceration on social bond development may be substantially different
for an individual who was incarcerated in 1980 when compared with someone
incarcerated in the 1950’s. It is important to continue to examine the impact that
incarceration has on lifecourse trajectories during different historical, economic, and
political periods.

More importantly, this analysis expands the current body of research on the
relationship between incarceration and social bond development. Although Sampson and
Laub (1993) included measures of incarceration in their research, the current study
extends their theoretical and empirical model. The addition of multiple measures of
incarceration and the use of advanced statistical techniques allowed social bond

development to be examined in relationship to the timing and nature of incarceration.

39

Sampson and Laub present a superb research study; however, their research was not
designed to examine the intricacies of imprisonment or to end all research on lifecourse
phenomenon. The goal of this research iS to further validate the work of Sampson and
Laub while contributing to the theoretical development of lifecourse perspectives on

incarceration.

Overview of the N LSY

Data from the National Longitudinal Survey of Youth 1979 (NLSY) were used as
the primary dataset in this research. The NLSY is a nationally representative sample of
12,686 young men and women who were 14-22 years old when they were first surveyed
in 1979 (Center for Human Resource Research, 2001). Respondents have been
interviewed for over two decades and were between the ages of 36 and 44 during the last
interview period in 2001. Although the survey was originally designed to collect data on
employment for a varied workforce, the survey currently includes extensive data on
education, substance abuse, fertility, and family structure. Data for this study were
collected yearly from 1979 to 1994, and biannually from 1996 to the present. In total,
data have been collected on nineteen separate occasions.

The NLSY is comprised of three sub-samples. The main sample includes 6,111
respondents and was designed to be representative of the non-institutionalized civilian
segment of young people living in the United States in 1979 and born between January 1,
1957, and December 31, 1964. All members of the first sample remained eligible to be

interviewed during all data collection years.

40

The NLSY sampling design also facilitates research on groups such as Hispanics,
Blacks, and the economically disadvantaged. A second sample includes 5,295
respondents and was designed to oversample civilian Hispanic, Black, and economically
disadvantaged non-Black/non-Hispanic youth living in the United States during 1979.
Following the 1990 interview, none of the 1,643 members of the economically
disadvantaged, non-Black/non-Hispanic subsample were selected for interviews. The
decision to eliminate both this subsample and part of the military sample was made for
economic reasons and does not jeopardize the representativeness of the total sample.

Thirdly, a sample of 1,280 respondents was also included to represent the enlisted
population. This sample includes individuals born between January 1, 1957, and
December 31, 1961, who were enlisted in one of the four branches of the military (Army,
Navy, Air Force, Marines) as of September 30, 1978. Following the 1984 interview,
1,079 members of the military subsample were no longer eligible for interview; 201
respondents randomly selected from the entire military subsample remained in the

survey.

Description of Data Collection

A multi-stage stratified area probability sample of dwelling units in the United
States was used to select eligible respondents for the research project. Researchers from
NORC used two sampling procedures to select research participants. To select the
representative sample, interviewers from NORC identified 1,818 sample segments within
202 primary sampling units. From the sample segments, over 75,000 dwellings were

chosen for preliminary interviews. A random sample of these homes was then selected as

41

the sampling flame. NORC also extracted a random sample of Department of Defense
records to ensure the survey was representative of members of the military. A basic
screening questionnaire was administered to members of the households that were
identified in this phase of the sampling design.

Information collected from the screening questionnaire was used to aid in the
development of the final sample. Data on name, age, sex, race, and address were
collected on more than 155,000 people. The demographic information was then used to
identify all individuals aged 14 to 21 as of December 31, 1978. Each appropriately aged
individual was then assigned to one of the sample groups. NORC interviewers invited all
persons on this list to participate in the first NLSY79 interview, and each respondent who
completed the first round interview was considered a member of the NLSY79 cohort. In
total, 12,686 (87%) of the individuals who were identified as eligible for participation in
the program completed the first round interview.

Data for the study were collected primarily through in person interviews;
however, telephone contact was made with respondents under certain circumstances
(e.g.,, where the respondent resides in a remote area or was incarcerated). During recent
years respondents have expressed a preference for phone interviews, so more and more
interviews are now conducted by phone. For example, in 1980 less than 5% of the
sample was interviewed by telephone, but in 2000 over a third were contacted by phone.
The interview takes approximately an hour to complete and each respondent is given a
small monetary incentive to participate.

A high level of sample retention was maintained throughout the study. A

retention rate of nearly 90% was sustained for the first 16 waves of the survey. The

42

retention rate dropped to 86% in 1996 and 80% in 2000. Excluding those individuals
who have been dropped from the sample completely, respondents have completed, on
average, 17.4 of the 19 interviews. In 2000, 64% of the sample had fully completed a
survey in each of the data collection years.

Overall, the NLSY is a well-designed, comprehensive dataset that is appropriate
to examine the research questions proposed by this study. The longitudinal nature of the
study is especially well suited for the research at hand. Longitudinal data are essential for
the study of continuity and change over the lifecourse. The nature of the data allow
researchers to examine the natural history of development over the lifecourse, provides
information on time ordering of events, and enhances a researchers ability to infer

causation (Blumstein et al., 1988b).

Sample

This study utilized a sample of the data obtained through the National
Longitudinal Survey of Youth 1979. The sample population was selected in two phases.
The first phase of the process included the removal of participants from the dataset who
became ineligible for participation due to project modifications, interview non-response,
extended stays of incarceration, or death. During the course of the data collection period,
two large sub-groups of the sample became ineligible for interviews because of changes
to the study protocol. As of 1984, 1,079 of the 1,280 members of the military sample
were no longer interviewed. In 1990, 1,643 individuals from the non-Black, non-
Hispanic economically disadvantaged group were excluded from the sample. In addition,

335 people fi'om the total sample had died during the course of data collection and were

43

removed from the sample. Seven sample members, incarcerated eleven or more time
periods, were also excluded from the final sample. Finally, individuals who failed to
participate in data collection six or more times were also omitted from the analysis group.
A total of 8,872 individuals remained after the first phase of sample selection of which
493 (5%) had been incarcerated at some point from 1983-2001.

The second phase of sample selection involved the selection of three distinct
subject groups. The first group, the incarcerated group, included all individuals who had
been incarcerated during one or more interview periods from 1983-2000. The second
group was constructed as a direct match to the incarcerated group. This group serves to
partial out sample heterogeneity, and allows the researcher to consider social bond
attainment for a similarly disadvantaged group that had not experienced adult
imprisonment. The third group represented the median or ‘average’ person in the sample.
Inclusion of this group in the study, aids the researcher in understanding the effect of
lifecourse events on social bond deve10pment for an average person. Together, the
matched and median groups were selected to provide greater context to the results
obtained from the incarcerated group. For example, is the effect of military participation
on marriage the same for the incarcerated group as the median and matched group, or
does the incarceration event mitigate all lifecourse experiences?

After the matching had been conducted, three sample groups remained. In total,
the sample included 1,504 individuals. The incarcerated and matched groups included
493 individuals, and the comparison group consisted of 518 individuals. A detailed
description of the statistical technique used for the selection of both the matched and

comparison group is presented below.

Matching

Incarceration, social capital, and social bonds are not assigned to members of the
sample at random. Instead, social events and relationships develop through a process
involving many unobservable or unknown variables. The central problem in
observational studies is that treatment and control units may not be comparable prior to
an event or intervention; hence, differences in outcomes between groups may or may not
indicate the effect of a treatment (Rosenbaum, 1989). The concern with sample
heterogeneity is further confounded with the process by which the intervention or
treatment affect the outcome (Berk & Newton, 1985). For example, if all incarcerated
sample members came from similar socioeconomic backgrounds, it would be difficult to
differentiate the effect of incarceration from that of socioeconomic status. Matching
serves as a tool to partial out the unobserved heterogeneity in the model. Thus, matching
provides a statistical control by balancing the distributions of all relevant pretreatment
variables (Rosenbaum & Rubin, 1983).

Traditionally, researchers have used individual matching techniques to select a
comparison group. With this technique, researchers most often identified a small set of
variables and then individual-by-individual selected a matching case based on identical
matches of all identified matching criteria. For example, Sampson and Laub (1993) used
gender, race, age, and neighborhood characteristics to develop their matched sample.
Individual matching techniques are valuable in that they control for unobserved
heterogeneity and can enhance the validity of the study. Traditional matching techniques
also have many limitations. Individual matching techniques regard all covariates as

equally important. Each individual is matched based on a set of characteristics and

45

primacy is usually not given to one variable. The algorithm used in this type of matching
requires matches to be selected based on a series of separate decisions. If a match cannot
be found, most researchers using this technique will discard the individual from the
sample (Rosenbaum, 1989). Removing subjects from a dataset not only reduces
statistical power, but can also bias the final analysis. Traditional matching techniques
are also very labor intensive. With this technique researchers must search the database
for exact matches based on a number of individual characteristics.

More recently, researchers have developed applied statistical techniques that
allow a larger number of variables to be used in the estimation process while also
facilitating the ease of the matching process. The matching technique used in this
research involved the use of propensity scores. A propensity score is defined as the
predictive probability of a single term of incarceration given a vector of observed
covariates. In short, a model was developed to estimate the likelihood of incarceration
based on a set of identified exogenous predictors. From the estimated likelihood,
individuals can then be separated into groups and a matched and median group can be
selected.

A two-step modeling technique is commonly used in developing propensity
scores. First, the assignment process must be modeled using a logit formulation (Berk &
Newton, 1985). The goal of this phase of the matching process is to identify a group of
variables theoretically and empirically associated with the outcome and then to estimate a
logistic regression based on those predictors. In this analysis, estimated propensity scores
were developed from a vector of covariates that have been associated with incarceration

risk. A relative likelihood of incarceration was produced from the logistic regression for

46

each member of the sample. The likelihood estimate can also be understood as the
predicted probability of an individual experiencing a stay of incarceration based on the
variables included in the model. The likelihood estimate was then used as the propensity
score in all subsequent analyses.

The propensity score can then be used in two different ways to minimize the error
fi'om sample heterogeneity. First, the propensity score can be included in statistical
models to condition any analyses of treatment effects. In this technique, the propensity
score becomes another factor, or the sole covariate, in a statistical model and serves as a
general control (see Berk & Newton, 1985). This technique is most appropriate for
samples with little variation in the propensity score or if the treatment selection is
partially deterministic (e.g., quasi-experimental designs) (Morgan, 20012351). In these
cases the homogeneity risk is less severe than that of the current research question; hence,
it is more appropriate to use the propensity score as a control in the model.

Propensity scores were used in the current research to aid in the construction of a
matched and comparison sample. The samples were examined using a series of sub-
group analyses. To develop a matched sample, the propensity score for each individual
in an identified group is matched to an individual with the same (or similar score).
Because incarceration is both a rare and seminal event in the lifecourse, subgroup
analyses were selected in lieu of conditioning the results with propensity scores.
Including the propensity score as a control in the model would mitigate error associated
with homogeneity of variance, but with this technique, analyses of within group variance
are more difficult. The goal of this research was to examine the relative effect of

incarceration on social bond development among similarly situated groups; hence, it is

47

more important to use propensity scores to develop comparison groups than to condition
the entire sample.

Using propensity scores as a method for developing a matched sample offers
many advantages over traditional techniques. One of the primary benefits of the
propensity score model is the simplicity of developing a matched sample. Constructing
individual matched samples can be labor intensive in that all elements must be matched
pairwise; however, matching on propensity scores allows the researchers to match groups
by the score itself or by mean scores of individuals within groups. The inclusion of a
propensity score in a model also enhances the parsimony of a model. Individually
controlling for all of the appropriate covariates in a model can be cumbersome, especially
in analyses of groups with small sample sizes. Traditional matching techniques also
regard all interactions among covariates as equally important; however, researchers (see
Rosenbaum & Rubin, 1985b) have demonstrated that this methodology iS not as efficient
or valid as calculating a scalar of X as is conducted when computing propensity scores
(see also Rubin & Thomas, 1996).

The propensity score matching algorithm is especially appropriate for the current
study. Incarceration as an event of ‘treatrnent’ is very non-deterministic. Incarceration
is not a forgone conclusion of the commission of a crime or even a criminal conviction;
hence, it is important to control for a number of predictors that may be generally related
to the outcome of incarceration. Propensity score techniques facilitate the inclusion of a
large number of covariates; thus, use of the scores improves the validity of the
comparison group. Moreover, the continuous nature of the score allows the researcher to

understand the relative risk or inclination toward criminality of an individual in

48

comparison to all remaining sample members. The propensity scores were used in this
model to develop a valid and efficient matched sample. In addition, the hierarchical
nature of the scores also facilitated the selection of a comparison group comprised of
individuals with a median propensity score. The development of such a group would not

be possible with traditional matching techniques.

Propensity Score Model Specification

Variables were selected for this model based on their theoretical association with
lifetime risk of incarceration. Overall, two categories of exogenous measures, including
criminal history status and individual characteristics, were used in the model. In addition,
individual measures of regional unemployment rate and family economic status were also
included. All variables included in the model have been used as covariates in recidivism
and decisionmaking studies because of their strong correlation with incarceration and
criminal involvement (e.g. Gendreau et al., 1996; Sampson & Laub, 1993; Darrell
Steffensmeier et al., 1998). Although researchers (Rosenbaum & Rubin, 1983, 1985b)
have argued that the statistical Significance of a predictor is not important when
constructing a propensity score, it is important to consider parsimony when making the
final model. Large models with extraneous variables may limit the ability of the
researcher to develop an optimal matched sample. Multicollineary is also a substantial
concern when estimating models; hence, a correlation matrix of the all variables was
constructed and variance inflation scores were estimated for each of the variables in the

model.

49

In total, twelve variables were selected for use in the development of the final
propensity scorez. All exogenous predictors were measured prior to 1983 and were
designed to represent individual demographic and contextual characteristics. A general
description of the model variables and the theoretical significance for their inclusion in
the model is presented below. In addition, a description of the variables included in the
propensity score model can be found in Appendix A and descriptive statistics for the total

sample are included in Table 1.

Criminal Histog

Three measures of early involvement in delinquency and experience with the
criminal justice system were included in the model. Researchers continue to stress the
link between early criminality and subsequent adult criminal careers (Blumstein et al.,
1986; Tracy & Kemph-Leonard, 1996; Wolfgang et al., 1987). It follows that those
individuals who were involved in delinquency or with the criminal justice system at an
early age would be of greater risk of firture incarceration.

In 1980, a special self-administered crime addendum was included with the
questionnaire to measure involvement in criminal behavior and contact with the criminal
justice system among individuals in late adolescence and early adulthood. Data from this
module were then used to develop the indicators of delinquency and involvement with
the criminal justice system. The first measure, self reported delinquency, was designed to

capture involvement in major forms of delinquency using self-reported measures. Each

 

2 Five additional variables were also included in preliminary models. Two demographic and contextual
variables including region of the country, and urban classification were removed fiom the model because
they were not significantly associated with the outcome. In addition parental education and family size
were removed because of lack of statistical significance. Finally, parental employment

was omitted because ofmulticollinearity.

50

respondent was asked if during the past year he or she had (1) assaulted someone (2)
shoplifted (3) committed petty theft (4) sold drugs (5) committed auto theft (6)
committed fi'audulent behavior and (7) committed grand theft. Only five percent of the
total sample (n=366) reported involvement in any form of delinquency and only 18% of
individuals reporting delinquency indicated that they had participated in more than one
delinquent activity. Because of the limited variation in the predictor, self-reported
delinquency was operationalized as a dummy variable with those individuals who
reported involvement in criminality as the reference category.

In addition to self-reported delinquency, two dichotomous measures of official
involvement with the criminal justice system were also incorporated into the model.
Individuals were asked if they had (1) ever had official contact with the police and (2) if
they had been incarcerated at any point prior to their participation in the research study.
Although the measurement of both variables was not optimal, the indicators did serve as
proxies for involvement in the criminal justice system during late adolescence and early
adulthood. Consistent with the state dependence framework, one would expect that prior
experience with the criminal justice system would increase the likelihood of subsequent
incarceration. As such, the previous incarceration and contact with the police variables
were also included in the model as dichotomous controls.

Individuals were also queried as to their involvement with drugs as part of the
1980 addendum. Drug use is an important element in this model in that it has been
highly correlated with criminal behavior (Menard, Mihalic, & Huizinga, 2001; H. White,
Brick, & Hansell, 1993; H. White & Gorman, 2000; H. R. White, 1990). Drug use in this

research was dichotomized into those individuals whohad reported use of any illegal

51

substance during the last 30 days and those that did not report drug use of any kind. Only
twenty percent of the sample reported drug use; hence, a more complex measure was not
possible with the existing data.

Demographic Influences

Education represents an important turning point in the lifecourse of individuals in
late adolescence and early adulthood. Research has reinforced the consequence of
education in relationship to both desistance and abstinence from crime. Individuals who
complete high school and reported high to moderate attachment to school are
significantly less likely to be involved in criminality or to be incarcerated in the firture
(Arum & Beattie, 1999; Sampson & Laub, 1993). Although some of the research on the
relationship between cognitive ability, education, and criminality has been controversial
(see M. R. Gottfredson & Hirschi, 1990; Hirschi & Hindelang, 1977; Wilson &
Hermstein, 1985), education has proved to be a salient element in recidivism risk
assessment instruments (e.g. Gendreau et al., 1996). AS such, education and cognitive
ability serve as central elements in the model.

Measures of both educational attainment and attachment to school were included
in the model. The first measure is a general education measure that examines the
attainment of a high school degree. High school graduate is a dummy measure with
those reporting successful completion of high school equal to one. In addition to the
global education predictor, a measure of school suspension is also included. School
suspension was included in the model to serve as a proxy for attachment to school. The
suspension indicator is dichotomized into self-reported suspension from school or

otherwise (reference category).

52

Scores on the Armed Forces Qualifications Test (AFQT) were used to measure
cognitive ability. Each respondent in the sample was administered the Armed Services
Vocational Aptitude Battery (ASVAB) and results from sections of this battery were used
to construct an Armed Forces Qualifications Test (AFQT) score for each youth. The
ASVAB includes ten sections on general science, arithmetic reasoning, word knowledge,
paragraph comprehension, numerical operations, coding speed, auto and Shop
information, mathematics knowledge, mechanical comprehension, and electronics
information. The AF QT measure is first constructed as an additive score based on the
arithmetic reasoning, word knowledge, paragraph comprehension, and one half of the
score from the numerical operations sections of the ASVAB. The individual is then
assigned a ranked percentile score based on his or her performance on the instrument.
This measure was developed as an indicator of trainability for the armed forces and is
used as the primary criterion for enlistment eligibility in the United States armed forces
(Center for Human Resource Research, 2001).

Gender and race were also included in the model as demographic influences. The
importance of extra-legal factors in the incarceration decision has been highlighted in
current incarceration and decisionmaking literature. Minority males, after controls are
included to reflect the seriousness of the offense and criminal history, are significantly
more likely to be incarcerated than any other group (Spohn & Holleran, 2000; Darrell
Steffensmeier et al., 1993; Darrell Steffensmeier et al., 1998; Zatz, 1987). This research
is further reinforced by statistical estimates of lifetime likelihood of incarceration.

Researchers have found that Blacks are Six times more likely than Whites to be

53

incarcerated at some point during their life and men are eight times more likely to
experience the incarceration event (Bonczar & Beck, 1997).

A series of dummy variables were constructed for the race and gender variables in
the model. Gender was dichotomized into male and female (reference category). Two
dichotomous measures Of race, including African American and Hispanic3, were also
included in the model. White and Other races served as the reference categories for the

race variables.

Unemployment Rate of the Region

Neighborhood social context has emerged as an important predictor in the
examination of criminality. A central assumption of social disorganization theory is that
neighborhoods characterized by breakdown in the social institutions of the community
will have higher rates of crime and disorder (Bursik & Grasmick, 1993; Shaw & McKay,
1942). With the breakdown in social institutions, individuals are less likely to experience
informal forms of social control and are more likely to engage in criminality. The macro-
individual link is yet to be fully understood; however, neighborhood context remains an
important theoretical construct that should be considered when examining risk of
incarceration.

A measure of the regional unemployment rate was included in this model as an
indicator of neighborhood context. For each member of the sample, a measure of the
employment rate of the local labor market was included. The variable consists of six
categories with (1) <3.0%, (2) 30-59%, (3) 60-89%, (4) 9.0-11.9%, (5) 12.0-14.9%, (6)

>15.0%. The unemployment rate measure incorporated in this model is dichotomous.

 

3 The Afi'ican American and Hispanic variables are mutually exclusive.

54

Individuals living in regions with an unemployment rate greater than nine percent were

coded as one with unemployment rates below nine percent as the reference category.

F_amily Economic Status

Family capital has emerged in the research as an important correlate of
criminality, especially among juveniles. Researchers have tied low capital in families to
a number of outcomes including deviance, behavioral problems, homelessness, and
involvement with delinquent peers (see Coleman, 1990; Hagan & McCarthy, 1997;
Parcel & Menaghan, 1993, 1994; Wright et al., 2001).

Family poverty status was included in this model as a measure of family human
capital. It is important to include family poverty status in the model in that parental
earnings provide the foundation for family social structures on which other forms of
capital develop. Economic deprivation has also been shown to be negatively associated
with favorable child development (Duncan & Brooks-Gunn, 1997; Duncan, Brooks-
Gunn, & Klebanov, 1994).

The family poverty status measure was derived from two sources including (1) the
respondent’s family income for the past calendar year and is derived from either the total
income information provided by the parent or guardian in the home in which the
respondent was living or (2) the total income reported by the respondent if living apart
from the family. The final measure of family poverty status was dichotomous with those

children living in families above the poverty level as the reference category.

55

Table 1. Descriptive Statistics for the Variables Used in the Development of the
Propensity Scores (N=8,872)

 

 

Variable Mean SD. Minimum Maximum
Male 0.50 0.50 0.00 1.00
Hispanic 0.19 0.40 0.00 1.00
Black 0.29 0.45 0.00 1.00
Cognitive Ability 38.83 27.88 1.00 99.00
High School Graduate 0.77 0.42 0.00 1.00
Suspension 0.25 0.43 0.00 1.00
Poverty 0.25 0.43 0.00 1.00
Prior Incarceration 0.02 0.14 0.00 1.00
Contact with Police 0.18 0.38 0.00 1.00
Self-reported Delinquency 0.04 0.20 0.00 1.00
Drug Use 0.21 0.41 0.00 1.00
Unemployment Rate 0.15 0.36 0.00 1.00

 

Propensity Score Model Statistical Estimation

After the final variables for the propensity score model had been selected, they
were regressed against the dichotomous incarceration measurement using a traditional
logit model. AS stated above, the goal of the propensity score technique is to derive the
estimated likelihood of incarceration, or any event or decision, based on a series of
exogenous predictors.

The results from the logit model are consistent with previous incarceration
research (see Table 2). Minority males had the highest risk (or likelihood) of

incarceration at some point during the study period. Prior incarceration and drug use

56

were also strong predictors. The unemployment rate of the region played little role in the
model. Overall, the sum of the risk variables was effective in classifying individuals into

groups with 94.4% of the sample correctly classified".

Table 2. Logit Model for the Development of Propensity Scores (N=8,872)

 

 

B S.E. Wald df Sig. Exp(B)
Constant -4.646 0.218 452.902 1 0.000 0.010
Demographic Influences
Black 0.855 0.141 36.658 1 0.000 2.352
Hispanic 0.490 0.155 9.976 1 0.002 1.633
Male 2.084 0.165 159.542 1 0.000 8.037
Cognitive Ability -0.023 0.003 49.850 1 0.000 0.978
High School Graduate -0.606 0.1 18 26.393 1 0.000 0.546
Criminal History
Contact with the Police 0.334 0.117 8.154 1 0.004 1.396
Suspension 0.768 0.1 13 46.562 1 0.000 2.155
Delinquency 0.492 0. 188 6.824 1 0.009 1.636
Drug Use 0.710 0.111 41.319 1 0.000 2.035
Previous Incarceration 1.400 0.218 41.281 1 0.000 4.055
Context Influences
Family Poverty 0.298 0.113 6.953 1 0.008 1.348
Unemployment Rate of -0.118 0.157 0.563 1 0.453 0.889
Region

 

 

Model Chi-square = 1153.436" Percentage Correctly Classified = 94.4

 

4 The large size of the sample may increase the likelihood of significant relationships to be found between
variables in the model. As such, a series of logit models were conducted using sub-samples of the data NO
substantial differences in the strength or direction of the relationship were found in the analyses of the sub.
samples.

57

AS part of the logit model estimation, a propensity or likelihood score was
generated. This score was calculated for each sample member and was used in later
analyses to select a comparison and matched group. The score represents the theoretical
likelihood of incarceration during the study period based on the vector of covariates
selected for the model. The score itself cannot be interpreted in an exact sense; however,
it can be used to examine the relative distance between individuals and to understand
clustering of individuals within the sample. A series of tables was constructed for both
the entire sample and a number of sub-groups to help illustrate the general distribution
and variation of the propensity scores. Not only did the data provide a general
description of the total sample, the scores also point to prominent sub-groups within the
sample.

As evidenced in Figure 2 and Table 3, the distribution of propensity scores for the
total sample was positively skewed. This is not surprising in that incarceration is a very
rare event. Few sample members could be characterized by the totality of variables that
have been associated with incarceration. What is critical is that substantial variation in
scores exists for both the incarcerated and non-incarcerated groups (see Table 4).
Although the propensity scores for the incarcerated sample were positively skewed, there
were members from both the incarcerated and not-incarcerated group with very low and
very high scores. This finding highlights the non-deterministic quality of the
incarceration event. Although there are a number of predictors that can aid in the

estimation of incarceration risk, the sample of incarcerated offenders was heterogeneous.

58

In the same light, there are a number of people who appear to have all the characteristics

consistent with the incarceration group, but were not imprisoned during the study period.

Predicted probability

 

6000

Std. Dev = .11
Mean = .06
N = 8872.00

 

 

Frequency

0.00 .10 .20 .30 .40 .50 .60 .70 .80 .90
.05 .15 .25 .35 .45 .55 .65 .75 .85

Figure 2. Distribution of propensity scores for the total sample

The most significant differences in propensity scores came from the gender
comparisons (see Table 5). There was substantial variation in the scores for both men
and women; however, the propensity score distribution for the female respondents was
truncated in comparison to that for the male sample. Both the minimum and median
scores for the male group were nearly ten times larger for that of the female group. The

maximum score was also over three times greater for the male group.

59

 

 

 

 

 

 

3000.0 00000.0 22% _ .0 ~m~000 0~0m0.0 0020.0 00m _ 0.0 NBC-00.0 0800 $30 :02
E 92.0 000000 N 210.0 3000.0 3300.0 0Nm00.0 0m _ 00.0 3000.0 0300.0 33.6 5an
«dgecxmm $033.54 $228.54 emcee-Eel 3:23.54 Shemcxmm
5:53.33 500 50% £00 :50»: E0» 50m 1 E353: 2
33550 50:00 I "5:38:35 80 meeoom baseman 05 mo 20:2:me .v 2an
I i 0o>oEom
0:20 0328085
x0x~x0 mmx _ 0.0 m. _ 000.0 0_ 050 SN 0 0.0 0300.0 m0m000 00000.0 00000.0 0xm.x ES» 295%
mcofiom
mm00x0 mxm0x0 ”023.0 0033.0 33:0 0.02.0 «0000.0 0500.0 0200.0 M00 0880885
30000 2000.0 3000.0 0mm~00 0mm _ 0.0 00000.0 mmm000 00000.0 04000.0 000.0 295% 020.0
Stratum 3.220246% mESeSQ 3.2:?me ENEROSQ SEARESQ
E25333 S00 E90 E00 2:03: E0» s0m i 53:05: Z

 

 

:EEBOBOE 50 mDSOm @6535 05 00 8:30:35 .m 2an

60

 

 

 

 

 

 

61

 

00000.0 00000.0 00000.0 0002.0 00002.0 00000.0 0003.0 0005.0 00000.0 000.~ 0:852
00000.0 00000.0 00000.0 00000.0 0200.0 0020.0 00000.0 00000.0 0000.0 000.0 00E?
:03
02000.0 00000.0 00000.0 02020.0 00020.0 00000.0 00000.0 0300.0 0000.0 02.0 0:00:02
3 000.0 00000.0 00000.0 00000.0 0 _ 000.0 00 2 00.0 00200.0 00000.0 0300.0 000.0 0:55
22:9:
00:20.0me 03200000 0320390 0002.00.50 00020300 0320.0.me
535%: “=3 Em 0% SS: .02. E00 00 2:535 z
0000 0:0 00:00 I 0008000005 00.0 00500 00301080 00 00 20030505 .0 0300
0008000005
00000.0 2000.0 0000 2.0 0300.0 00 30.0 00000.0 0020.0 00000.0 0000.0 000.0 82
00000.0 2000.0 00000.0 0300.0 00000.0 0002.0 2000.0 0020.0 00000.0 000 0008000005
:0:
000800.505
0 2 0000 00000.0 2 00 2 0.0 0000000 _00000 000000 0200.0 00000.0 0300.0 000.0 82
3000.0 2000.0 0200.0 0003.0 00000.0 0020.0 00000.0 0200.0 0200.0 00 0008000005
«$53:
SNCSPuLNAN wwatmutswm mxmutmbkml 3&thka m>2§wbkmm m~§2m9kmnx
535%: “0% “0% RE :00»: “a: :00 r: SEES z

 

 

500000.805 0:0 00:00 I 000800.805 000 00800 0080080 05 00 00035me .0 030.0

When the propensity scores were fiirther sub—divided by the incarceration event,
the disparity in gender groups was even more apparent (See Table 5). For the non-
incarcerated samples, the difference in minimum and medium scores was over seven
times greater. The maximum male score was over three times higher than that of the
female non-incarcerated group. Although women as a whole have a lower likelihood of
incarceration, the divergence within groups for the incarcerated samples was not
dramatically different from that of the male group. The median propensity scores for the
female incarcerated group were six times higher than that of the non-incarcerated group.
The disparity between groups of males was five-fold.

The range of scores did not vary substantially within gender groups when scores
are further subdivided by gender and race (See Table 6), although the distribution of
scores did vary by race. For women, there was little difference in the minimum and
maximum scores by race, but for the twentieth through eightieth percentiles, the
propensity scores for minority women were at least four times higher for that of the non-
minority women. A similar pattern emerged for men. The minimum and maximum
scores were slightly higher for minority men; however, the true difference in scores came
in the median region of the distribution. For the twentieth through eightieth percentiles,
the propensity scores were at least four times higher for minority men when compared to
non-minority males.

The examination of the propensity scores in such a manner is helpful for
understanding risk of incarceration in general and provides firrther context for the final
statistical analyses. One of the most salient differences found in the propensity score

analysis was that of gender. Overall, the scores across gender groups varied

62

substantially. More importantly, the median scores for the incarcerated male sample
were nearly seven times that of the female sample, and the maximum scores were over
three times larger. The disparity in scores speaks to the relative differences of women in
the sample. Women, in general, were less likely to be considered at risk of incarceration
based on this model. The dramatic differences in scores can also be seen as a signal of
poor model fit for the women in the sample. A true, valid model of incarceration risk
would be able to equally predict incarceration for women as well as men.

The difference in gender contrasts is not surprising in that most research on
corrections has been based on male experiences (see Farr, 2000; Owen, 1998 for
exceptions). Models of incarceration risk for women need to be considered more
thoroughly. In order to further examine the validity of the propensity scores for the
women in the current sample, a series of sub-group analyses were conducted. The
statistical models are identical to those for the final propensity score model except gender
has been removed from the models.

As shown in Table 7, there are substantial differences in the logit models for
males and females. The coefficients for the male subgroup analysis were very similar to
those of the original propensity score model. For women, less than half of the exogenous
predictors were found to be significant. Only suspension from school, self reported
delinquency, drug use, and cognitive ability were found to be significant in the female
model. Nearly all (99.9%) of the women were correctly classified using the model;
whereas, 90.3 percent of the male sample was correctly classified. The results of the
model fit statistics are not surprising in that the female sample was substantially more

homogenous than that of the male sample. In addition, only 45 (1%) of the female

63

sample were incarcerated at some point during the study period; hence, there was also
little variation in the dependent variable in which to model. Although the models are
limited by the small sample of incarcerated women, this analysis highlights the difference
in propensity for men and women. It appears that risk models for women may be
different for that of men. Further research is needed on this subject; however, for this
research it is important to note the gender differences when interpreting the final results

of the statistical analyses.

Table 7. Logit Model for the Development of Propensity Scores by Gender

 

 

 

Female Male Total Sample
(N=4,484) (N=4,388) (N=8,872)
b S.E. b SE. B S.E.

Constant -4.75** 0.52 -2.51** 0.18 -3.25** 0.17
Demographic Influences

Black 0.21 0.44 0.95" 0.15 0.86" 0.14

Hispanic 0.46 0.44 0.47" 0.17 0.49" 0.15

High School Graduate 052 0.35 -0.63** 0.13 -0.62** 0.12

Cognitive Ability -0.03 * 0.01 -0.02** 0.00 -0.23** 0.00
Criminal History

Contact with the Police 0.76 0.39 029* 0.12 0.70“ 0.11

Suspension 1.50M 0.35 0.67" 0.12 0.86M 0.11

Delinquency 1.4 l * 0.56 043* 0.20 0.71 * * 0.18

Drug Use 0.97" 0.33 0.66M 0.12 0.82M 0.11

Previous Incarceration 1.23 0.73 1.46** 0.23 1.46* * 0.21
Context Influences

Family Poverty -0.15 0.35 0.33“ 0.12 0.16 0.11

Unemployment Rate -0.05 0.43 -0. l 2 0.17 -0.17 0.15

 

 

* p<.05 Mp <.01 (two-tailed tests)
Female Sample -
-2 log likelihood = 400.92
99.9% correctly classified
Male Sample -
-2 log likelihood = 2,231.35
90.3% correctly classified
T otal Sample -
-2 log likelihood = 2,888.07
94.7% correctly classified

65

Matched and Comparison Group Estimation
After the propensity scores were constructed, they were used to aid in the

development of a matched and a comparison sample. As discussed above, in this study,
each incarcerated person was matched to a control individual who had not experienced
incarceration using the nearest available matching (Rubin, 1973) on the estimated
propensity score. The optimal matching algorithm would involve exact matches for all
members of the group, but this was not possible with the existing data set. Only 178
individuals in the control sample (36%) were selected based on exact matches and the
nearest available match was utilized for the remaining 315 individuals (see Table 8). The
nearest available match was selected by identifying the individual with the nearest
propensity score to that of the incarcerated group. The selected score can be higher or
lower than that of the treatment individual; the goal is to find the absolute closest match
in regard to the propensity score. Although it is best to select exact matches for all
members of the incarcerated sample, the technique of nearest available matching has
been demonstrated as robust in previous research of this type (see Rosenbaum & Rubin,
1985a). As mentioned earlier, it is also advantageous to select a person with a close

match than to omit the dyad from the sample completely.

Table 8. Number of Exact Propensity Score Matches

 

 

Frequency Valid Percent
Proximate Match 315 63.9
Exact Match 178 36. 1
Total 493 1 00.0

 

 

For a small portion of the sample, multiple nearest available or exact matches

were found. In this special case, the propensity score served as the primary filter for the

66

development of the control sample. When multiple suitable matches were found,
matching was conducted on the basis of gender, age, and race. Gender was given
primacy in matching so that the gender balance of the sample could be maintained for
statistical analysis. If the sample needed to be narrowed fiirther, the match was then
selected based on age. Finally, matches were selected according to race, if necessary.
Overall, the propensity score technique was successful in selecting an appropriate
matched sample (see Table 9). The mean difference in propensity scores between groups
is 0.0199. Only the measure of juvenile incarceration was significantly different for the
matched sample when compared to the incarcerated sample. This is not surprising in
that those individuals who have served one stay of incarceration are significantly more
likely to go on to be incarcerated again in the fiJture (Langan & Levin, 2002). Overall,
the matching technique was appropriate for this model in that a number of close matches
were obtained, and the incarcerated group is not statistically different from that of the

matched group.

67

Table 9. Descriptive Statistics for the Variables Used in the Development of Propensity Scores

 

 

 

by Group
Incarcerated Matched Median Group
Variable (N=493) (N=493) (N=518)
Mean SD. Mean S. D. Mean SD.
Male 0.91 0.29 0.92 0.28 0.47 0.50
Hispanic 0.23 0.42 0.23 0.42 0.24 0.43
Black 0.53 0.50 0.53 0.50 0.27 0.45
Cognitive Ability 18.12 17.27 17.98 17.12 36.59 26.84
High School Graduate 0.44 0.50 0.43 0.50 0.77 0.42
Suspension 0.63 0.48 0.65 0.47 0.22 0.41
Poverty 0.41 0.49 0.38 0.49 0.24 0.43
Prior Incarceration" 0.18 0.38 0.1 l 0.32 0.00 0.00
Contact with Police 0.43 0.49 0.41 0.48 O. 12 0.32

Self—reported Delinquency 0.20 0.40 0.15 0.36 0.02 0.14

Drug Use 0.43 0.50 0.42 0.49 0.19 0.39
Unemployment Rate 0.13 0.33 0.12 0.33 0.17 0.37
Propensity Score 0.26 0.22 0.24 0.19 0.01 0.00

 

*The incarcerated and matched group are statistically different at p<0.05

As part of the matching phase of the analysis, a comparison group was also
selected (N = 518). In order to develop the comparison group, all individuals with
propensity scores equal to the median score of the total sample were chosen for inclusion
in this group. In addition, individuals whose scores were within 0.0025 from the score

were also incorporated. This technique facilitated the development of a median group

 

5 This number was chosen to aid in the selection of a comparison group that was similar in number to the
incarcerated and matched group.

68

that was large enough for comparison with the incarcerated and matched groups.
Descriptive statistics for the median group are presented in Table 9. This group was
developed to represent the experiences of an average person in the sample. Because of
the skewed nature of the data, the median group was selected in lieu of a comparison
sample that was selected based on mean scores. This group provides further comparative

information on the relative effect of lifecourse events on social bond development.

Sample Characteristics

As presented in Table 9, the characteristics of the incarcerated and matched
sample are very similar. Members of the matched and comparison groups are primarily
minority males with little education, living under the poverty level. Less than ten percent
of the matched and incarcerated groups are female, one half are Black, and one quarter
Hispanic. Two-fifths had incomes or lived in a family with economic means below the
poverty level. Less than half of the sample members attained a high school degree, and
three-fifths reported being suspended fiom school at some point.

Not surprisingly, members of both the matched and incarcerated groups reported
substantial involvement in criminality prior to the sample period. Two-fifths of both
groups reported prior drug use and contact with the police. Of the incarcerated group,
20% reported involvement with delinquency as a young adult or late adolescent, and 15%
of the matched group reported similar behaviors. As discussed previously, the two
groups are significantly different in their experiences with incarceration as a young adult

and juvenile. Nearly one-fifth (18%) of the incarcerated group reported being imprisoned

69

prior to the study period; whereas 11% of the matched sample indicated previous
incarceration.

The median group was much more homogeneous when compared to the
incarcerated and matched groups. Nearly half of the group members are White males.
The remaining sample members are split between Hispanic and African American racial
groups. Three quarters of sample members graduated from high school and 25% reported
suspension from school. In addition, a quarter of the sample reported living under the
poverty level.

Members of the median group also reported substantially less involvement in
criminality. None of the members of the median group experienced incarceration prior to
the study period. Moreover, one-fifth reported previous drug use, 12% had prior contact
with the police, and only 2% of the sample indicated involvement in delinquency.

The descriptive statistics reinforce the validity of the sample selected for the
current study. The characteristics of the matched and incarcerated groups are very
similar. Members of both groups reported considerable involvement with delinquency
and the criminal justice system. They were also more likely to live in poverty and have
less attachment to school. The nature of the incarcerated and matched groups facilitates
the comparison of bond development between groups. In contrast, the median group was
characterized by modest participation in delinquency and antisocial behavior. Members
of the median group were also more likely to have completed high school and less likely
to live in poverty when compared to the incarcerated and matched samples. The addition
of the median group to the study firrthers the consideration of the relative effect of

incarceration, in comparison to other social events, on social bond development.

70

Analysis Plan
The larger goal of this research is to consider the effect of incarceration on the

development and nature of social bonds to marriage. The research was designed to
answer two primary research questions. First, are sample members able to attain bonds to
employment and marriage following incarceration? Second, how is the quality of social
bonds affected by the incarceration event? The following section details the hypotheses
considered by the research. In addition, a discussion of the methodologies used to test
the hypotheses is presented. A number of exogenous variables were included in each of
the models to condition the estimation of the incarceration event on social bond
development. A detailed description of each of the variables used in the models is

provided in the following section.

Attainment of Social Bonds

The first part of the analysis plan addresses the attainment of social bonds to work
and marriage following incarceration. As discussed above, incarceration can shut off
legitimate opportunities, thus reducing the likelihood of developing social bonds
(Coleman, 1988; Laub & Sampson, 19883, 1993; Sampson & Laub, 1992, 1993). The
following hypotheses have been suggested by the research conducted to date and serve to
guide the analyses.

0 H1: Incarcerated persons will be less likely to attain employment of any kind

following incarceration6.

 

6For each of the hypotheses, it is expected that the effect of incarceration on social bond attainment be
examined in light of both time invariant controls and other lifecourse events. Including the static factors in
the model allows the researcher to consider the relative effects of both continuity and change in the model.

71

0 H2: Incarcerated persons will be less likely to develop bonds to marriage

following incarceration.

A series of logit models were estimated to address hypotheses H1 and H2.
Because of the dichotomous nature of the dependent variables, the logit model is the most
appropriate statistical model to address the research question (Long, 1997). This
technique allows the researcher to understand the probability of marriage or work at one
point in time after controlling for lifecourse, demographic, criminal history, and
contextual variables. In the current research, the effect of incarceration on the likelihood
of marriage and procurement of fulltime work in 2000, net of control variables, was
considered.

A specialty form of hierarchical linear modeling that incorporates the logit
function was also used to examine the relationship between incarceration, lifecourse
events, and social bond attainment (Raudenbush & Bryk, 2002). This modeling
technique allows conclusions to be made on the change in the probability of bond
attainment over time in reference to both static (e.g., gender, race) and dynamic (e. g.,

incarceration by year) predictors.

Nature of the Social Bond
Not only are individuals less likely to attain bonds to marriage and work
following incarceration, current research has suggested that the quality of the bond

attained may also be attenuated by the incarceration event (Freeman, 1991; Waldfogel,

 

The addition of lifecourse events in the models facilitates the comparison of the predictive power of the
state dependence and population heterogeneity perspectives.

72

1994; Western & Beckett, 1999; Western et al., 2001). Based on this literature, two

hypotheses are presented.

0 H3: The slope of the quality of employment measure will grow more slowly for

the incarcerated group when compared with the matched and comparison groups.

The change in the quality of bonds to employment following incarceration was
tested with this model. Growth curve modeling, a specialty form of HLM, was used to
address this research question. The analysis procedure allows the researcher to examine
the nature and rate of growth in social bonds for sample members. The current research
includes incarceration as an intervention point and explores the character of change

following the event.

0 H4: Individuals who have been incarcerated will report a lower quality marital

bond when compared with the non-incarcerated group.

Finally, the nature of the marriage bond developed following incarceration was
studied. For individuals who had attained marriage following a stay of incarceration, a
factor score was constructed based on the number of marriages the individual entered into
after incarceration, the length of those marriage(s), and the age at first marriage. The
operationalization of this aspect of the social bond is not optimal; however, attitudinal
measures as to the quality of the marriage (e.g., how often the couple reports fighting) are

not available with the current data set. The quality of marriage measure does expound on

73

the existing theoretical work conducted in this area and provides a proxy for a quality of
marriage variable.

Ordinary Least Squares regression was used to test this hypothesis. In this model,
demographic, criminal history, lifecourse events, and contextual indicators were

regressed on a continuous factor score computed to represent the nature of marriage.

Measurement of Variables

The following section outlines the independent and dependent variables used in
the current study. One of the secondary goals of this analysis was to understand how the
predictive models of social bond quality and attainment vary for both the marriage and
work outcomes. As such, the same statistical model was estimated for each of the
dependent variables. A detailed description of the variables used in the current study is

presented in Appendix A.

Measurement of Dependent Variables
M

Measures of both the attainment of work and the indicators of the nature of
employment attained were included in the model. Employment has been found to be
significantly related to desistance from crime (Homey et al., 1995; Sampson & Laub,
1993; Uggen, 2000). Because of the relationship between work and desistance, it
becomes important to examine what factors best predict employment following
incarceration. Specifically, does the timing and nature of incarceration have differential

effects on both the likelihood and nature of employment? Most of the current literature

74

indicates that incarceration has a small and ofien insignificant effect on the likelihood of
employment in adulthood, but a negative relationship, albeit weak, has been confirmed
between imprisonment and quality of employment (e.g. Waldfogel, 1994; Western,
2002). Current research on the timing of incarceration has also suggested that
incarceration at an early age is more detrimental to employment outcomes when
compared to later incarceration (Freeman, 1991; Western & Beckett, 1999). It appears
that incarceration can negatively affect employment outcomes, but the true nature of the
relationship should be explored further.

Three measures of employment were included in this research. First, a
dichotomous measure of employment was developed. The employment measure was
dichotomized into those individuals who reported working fulltime (more than 35 hours a
week) during the study year and those that did not work or worked part time (reference
category). Two additional constructs were developed to capture the nature and stability
of employment. First, a measure of tenure with employer was developed which indicates
the length of employment, in weeks, with the respondent’s primary employer. An
additional indicator was included which captured the number of jobs worked during the
study year. The number of j obs is an additive measure. For example, the number of jobs
worked in 1985 construct reflects the number of unique employment experiences
reported by the respondent in 1983, 1984, and 1985.

The examination of the nature of employment in the current research is an
important addition to the existing literature on social bond development. To date, most
researchers have only examined the attainment of employment and wages garnered from

work (Freeman, 1991; Waldfogel, 1994; Western, 2002; Western et al., 2001; Western &

75

Pettit, 2000). As suggested by social capital and control theorists, employment in and of
itself is not as important to consider as is the attachment to work (Coleman, 1988, 1990;
Hirschi, 1972; Laub et al., 1998; Sampson & Laub, 1993). It is the strong bond to work

that insulates against further criminality.

Larital Attachment

Social bonding in the form of marriage has been linked with desistance (Homey
et al., 1995; Laub et al., 1998). As such, the attainment of the social bond to marriage
was used as an outcome measure in this research. The marriage dependent variable is
dichotomous allowed the researcher to consider attainment of the bond of marriage at
each of the fifteen interview points.

A measure of the nature of the marital bond was also included. Similar to the
employment model, it was important to include an indicator of the nature of the marital
bond in the model because researchers have associated stronger bonds with higher
instances of desistance. The nature of the marital bond variable was operationalized
using a three item factor score that included 1) the length of the respondent’s first
marriage in years and 2) the number of marriages entered into during the study period
and 3) if the individual had been divorced at any point from 1983-2000 (eigenvalue 2.10,
factor loadings > 0.757). To ease the interpretation of the dependent measure, the factor
scores were multiplied by —1. As a result of the transformation, high scores on the

dependent measure corresponded with strong bonds to marriage.

76

Measurement of Independent Variables

Two sets of independent variables were incorporated in this research. The first
group included dynamic variables for which data had been obtained during each year of
the study. Measures of incarceration, work, marriage, age and military involvement were
added to the model to represent salient lifecourse events. The second group of
independent variables comprised static measures and captured data on each individual
prior to the study period. The static measures served largely as controls; although, these
constructs also serve as proxies for the criminal propensity as suggested within the
population heterogeneity framework (e.g. Gottfredson & Hirschi, 1990; Wilson &
Hermstein, 1985). The static predictors were selected to represent pivotal adolescent,
individual characteristics that have been shown to have both direct and indirect effects on
subsequent adult criminality. Descriptions of all variables included in the final statistical

models are presented in Appendix A.

Dynamic Influences
Incarcegtion

Incarceration was included as the primary independent variable in the models.
The NLSY does not include a specific measure of incarceration in the interview
instrument, but imprisonment can be measured in relationship to the current residence of
the individual at the time of interview. Respondents who were interviewed at a
correctional facility, either by phone or in person, were considered as incarcerated during
that particular year. Unfortunately, the NLSY does not capture information on

individuals who may have been incarcerated for part of the year, but were not imprisoned

77

at the time of the interview. In light of the data collection protocol, the incarceration
measures are more likely to represent those individuals who had been incarcerated for
longer periods of time (e. g., prison) than an individual who experienced a short stay (e.g.,
jail).

Two separate measures of incarceration were used in this analysis. The first
incarceration construct is a time-varying dummy variable. For each time wave, the
incarceration variable was coded one if the individual was incarcerated at that time point
and zero if the individual was not currently incarcerated. Using this type of construct as
an independent variable allows incarceration to be viewed as a form of intervention in
that it is possible to examine change in lifecourse trajectories following the event. This
construct also facilitates the study of the timing of incarceration in relationship to social
bond outcomes.

The second incarceration construct was developed to represent the cumulative
effect of incarceration over time7. In addition to the dichotomous measures, an additional
variable was included in the model to examine the cumulative nature of incarceration.
The incarceration count measure was calculated for each subject at all fifteen time
periods, and represents the number of time periods the individual was incarcerated prior
to the current time period.

Research on incarceration as both an outcome and as a predictor has traditionally

treated incarceration as a binary outcome, but over half of the current sample was

 

7 In the original research proposal, a number of categorical variables had been considered as representations
of the nature of incarceration. Because of the lack of information on the length and character of
confinement (e.g., jail v. prison and the number of days incarcerated), the original categorical variables
were removed from the final analyses. Instead, the cumulative incarceration measure was constructed.

This measure speaks to the additive effect of incarceration while not overstepping the bounds of the data
available for the current study.

78

incarcerated multiple times. Research has indicated that the majority of offenders
released to the community will be returned to prison within three years (Langan & Levin,
2002). It stands to reason that an extended stay of incarceration or multiple terms of
incarceration would have a differential effect on work and family outcomes when
compared with one incarceration event. The addition of a time varying covariate for the
discrete and cumulative measure of incarceration extends the current knowledge of the

relative effect of imprisonment on lifecourse trajectories.

Milim Participation
Consistent with the work of Sampson and Laub (1996), participation in the
military was used as a dynamic control in the model. Like prison, the military represents
an important social institution. Involvement in this institution can affect lifecourse
trajectories. Military service also removes an individual fi'om the conventional workforce
and can limit avenues for traditional employment and interaction with the opposite sex.
In addition to acting as a control in the models, military service also served as a

comparison for the institutional effect of incarceration.

Work_and Mame
Employment and marital status functioned as both statistical controls and as
exogenous variables in the model. Researchers have argued that the presence of one type
of social bond increases the likelihood of developing subsequent bonds (see Coleman,
1988; Coleman, 1990). One would expect that individuals whom are gainfully employed

would also be more likely to attract a marriage partner; therefore, it is important to

79

control for marital status when modeling bonds to employment and vice versa. In order
to control for social bond development, a series of dummy variables were constructed, by

year, for both marriage and employment.

A89

A measure of age at each interview point was also used as a dynamic predictor in
the model. The age variable in the model was centered using an arbitrary point of
reference. For each individual, the age measure was calculated by subtracting 21 years
from the age variable. The centering point of 21 was selected because it represents the
midpoint of the age range for the respondents at the first phase of data collection for the
current study. This centering technique is common in the literature and serves to reduce
multicollinearity when multiple model fit estimators are included (J ang, 1999;
Raudenbush, 2001). In addition, a polynomial of the centered age predictor was also
incorporated into select models to control for non-linear growth (Murphy & Welch,
1990). The most appropriate functional form of these variables are further examined in

Chapter 4 of this research.

Static Influences

A series of demographic measures were also included in the model as controls.
The static influences chosen for this model serve as the basis of the population
heterogeneity model and allow researchers to fiirther explore the effect of life events net
of immutable individual differences. Two categories of individual difference predictors

were included. The first group incorporated cognitive ability, education, race, and gender

80

and was used as the primary measure of individual difference. A set of criminal history
predictors was also included. With the exception of antisocial behavior measure, the
operationalization of the static influences did not vary fi'om those included in the

propensity score models. A general discussion of the variables is provided below.

Demographic Influences

Similar to the propensity score model, education and cognitive ability were
included as predictors. Graduation from high school represents a seminal event in the
lifecourse (Arum & Beattie, 1999). Cognitive ability, although controversial, has been
included as a central exogenous predictor in the state dependence models (see
Gottfredson & Hirschi, 1990; Hirschi & Hindelang, 1977; Wilson & Hermstein, 1985).
As such, high school graduate and cognitive ability were included in the social bond
models.

Gender and race were also incorporated in the model as demographic influences.
The incarceration decisionmaking literature has highlighted the importance of extra-legal
factors in the decision to incarcerate (Spohn & Holleran, 2000; Steffensmeier et al., 1998;
Zatz, 1987). A series of dummy variables were constructed for race and gender
variables in the model. Gender was dichotomized into male and female (reference
category). Two dichotomous measures of race, including African American and
Hispanic, were also included in the model. White and Other races served as the reference

categories for the race variables.

81

Criminal Historvfiand Involvement in Delinquency

 

Two measures of involvement in delinquency and criminality, including prior
incarceration and antisocial behavior, were included in the models. Contact with the
criminal justice system and participation in delinquency and drug use at a young age can
mortgage an individual’s tie to society, reducing the likelihood of development of
positive social bonds in the future (Coleman, 1988; Sampson & Laub, 1993). In addition,
antisocial behavior, even in adolescence, has been linked with work and family problems
(Moffitt, Caspi, Harrington, & Milne, 2002). The criminal history measures were used in
the analyses as controls and allow the relationship between incarceration and bonds to be
examined, net of deviance and delinquency as youth and young adults.

Both criminal history measures were included in the model as dummy variables.
Antisocial behavior was dichotomized into those individuals who had reported
involvement in criminality of any kind, including drug use, before 1980 and those that
did not indicate delinquency involvement. The previous incarceration measure queried
individuals if they had been incarcerated at any point before their involvement in the

survey.

Analytic Technique
Three multivariate analytic techniques were used in this research including
Ordinary Least Squares regression, Logistic regression, and Hierarchical Linear

Modeling. A description of each technique is provided below.

82

Ordinary Least Squares Regression

Ordinary Least Squares (OLS) regression was used in this analysis to understand
the relationship between incarceration and the nature of the marital bond. OLS is an
appropriate statistical technique for models that involve a continuous dependent variable
and multiple independent variables. OLS regression is generally used to provide an
estimate of the dependent variable as a linear function of a number of dependent variables
plus error (Bachman & Paternoster, 1997).

The OLS model rests on a number of assumptions. In OLS models, all variables
must be normally distributed and observations should be randomly selected. The error
term should be randomly distributed, constant across all levels of x, independent of all
exogenous predictors, and have an expected value of zero. Exogenous variables must also
not be correlated with each other (Bachman & Paternoster, 1997).

OLS regression was the most appropriate technique to examine the nature of the
marital bond because the research question involved examining the matrimonial
relationship at one point in time; hence, a longitudinal form of analysis was not
appropriate. The functional form of the dependent variable was also well suited for OLS
regression. Because the nature of marriage indicator was derived from factor analyses,
the dependent measure had a mean of zero and a standard deviation of one. The factor
score was also normally distributed. In addition, regression diagnostics did not uncover
possible problems due to multicollinearity (see Appendix A).

The OLS model was only used in the estimation of the nature of marriage in 2000.

The remainder of the models involved dependent measures of a dichotomous nature or

83

the analysis questions necessitated the use of longitudinal methodologies; hence, OLS

regression was not appropriate for these models.

Logistic Regression

Logistic regression is a specialized form of OLS and is most appropriate for
testing models that involve a dichotomous dependent variable. This technique was used
in this analysis to determine the likelihood of marriage and employment in 2000 in
relationship to lifecourse events, demographic predictors, criminal history indicators, and
contextual influences. In many ways, the OLS and logistic models are similar. Both
require multiple independent variables to be included in the model, observations to be
independent, and data to be randomly selected (Bachman & Paternoster, 1997).

The distribution of the dependent variable separates logistic regression from that
of OLS. In logistic regression the functional form of the dependent measure is
dichotomous. In the current study, respondents were either classified as married or not
and employed or unemployed. The coefficients derived from the model allow the
researcher to ascertain the probability of an event (e. g., marriage) occurring during a
specified time period. The method required to estimate the coefficients is also different
for logistic regression. In logit models, coefficients are estimated using the maximum-
likelihood estimation method (MLE) (Long, 1997).

Several logistic regression models were estimated. The first set involved the
examination of the attainment of the bond to marriage and work in 2000. These analyses
comprised the entire sample population. A series of sub-group analyses by sample

groups were also conducted to examine the differences in the likelihood of attainment of

84

social bonds for the incarcerated, matched, and comparison groups. In addition, a
specialty form of logistic regression was used in conjunction with Hierarchical Linear

Modeling. Specifics on this technique are presented below.

Hierarchical Linear Modeling

A specialty form of hierarchical linear modeling (HLM), deemed growth curve
analysis, was used as the primary analytic technique for this research (Raudenbush &
Bryk, 2002). This technique was selected to facilitate the examination of change in social
bonds over time. This statistical technique is designed to model longitudinal data and can
be considered an advanced method for understanding repeated measures data. Growth
curve analysis models are based on the assumption that the underlying path of change is
similar for all individuals in the study; however, this technique also allows researchers to
model the effect that dynamic factors (e.g., incarceration) have on trajectories of change
net of time-invariant individual differences (e.g., cognitive ability). This technique is
appropriate for the current study in that it allows the researcher to consider the
relationship between incarceration and the likelihood of social bond attainment over time,
while controlling for the effect of individual demographic characteristics (e.g., race).

HLM models are traditionally estimated on two levels, making it especially well
suited to evaluate theories that incorporate both continuity and change (e. g., lifecourse
perspectives). The first level of the HLM model is designed to assess individual
trajectories of growth as characterized by an intercept and a rate of change. The second
level of the model is a between-persons model with the slope and intercept of the Level I

variables as outcome variables and static factors as predictors (Johnson, Hoffmann, Su, &

85

Gerstein, 1997). The objective of the Level 11 model is to identify factors that are related
to the extent of individual differences in growth curve parameters. Level 11 describes
how these relationships vary over a population of persons. The factors included in the
second level of the model are time-invariant.

HLM is an appropriate statistical technique to address the proposed research
questions. The multi-level nature of the technique allows social bond development to be
modeled simultaneously in relationship to static time-invariant variables (e. g., age or
gender) and time-variant covariates (e.g., incarceration). HLM was used in the current
analyses to test two hypotheses. First, the logistic form of HLM was employed to
examine the change in the likelihood of marriage and firlltime employment. The second
set of models was designed to consider the variation in the nature of employment over

time as measured by number of j obs employed and tenure with current employer.

Research Implications

The research plan has been designed to facilitate the discussion of a number of
research propositions. The fixed-effect logit models allow the researcher to understand
social bond development among sample members. In this model, the probability of
fulltime employment and marriage in 2000 were modeled in relationship to lifecourse
events, demographic influences, criminal history, and context influences. In addition, a
series of sub-group analyses were also estimated to explore the variation in social bond
attainment across sample groups.

This analysis was also designed to examine the social bond development in a new

light. By using longitudinal data in conjunction with growth curve modeling techniques,

86

the researcher was able study change in the likelihood bond development and the nature
of bonds attained over the lifecourse, while controlling for time invariant factors (e.g.,
race). The use of longitudinal data is especially important in light of the current
desistance literature. Desistance and bond development do not occur spontaneously (see
Laub & Sampson, 1988b). Instead, bonds develop over time.

Finally, using longitudinal data that include yearly measures also allows the
researcher to consider the effect of timing and nature of the incarceration event on social
bond development. Specifically, the cumulative effect of incarceration was examined in
this research. Traditionally, researchers have compared juvenile to adult incarceration
(e. g. Western & Beckett, 1999); however, the nature of this research design will facilitate
conclusions on the specific timing of incarceration and social bond development. A

pictorial of the general analytical fi'amework is presented in Appendix B.

87

Chapter 4: Results of Analyses

As indicated in Chapter 1, the goal of this research was to examine the attainment
of social bonds to marriage and work following incarceration. A secondary focus was to
explore how the nature of these bonds was affected by the incarceration event. As such, a
series of statistical equations were estimated that model both the attainment and the
nature of the bond in relationship to incarceration and other salient life events, net of
controls. The hypotheses were tested using data from the National Longitudinal Survey
of Youth. The research design included three samples. The incarcerated sample consists
of all individuals incarcerated at some point from 1983-2000 (n=483), the matched
sample represents similarly situated individuals to the incarcerated group that have not
experienced incarceration (n=483), and a comparison group (n=512) includes individuals
with a median propensity of incarceration. In total, the sample comprises 1,478
individuals.

The analyses were conducted in two phases. The first phase included a series of
logit models designed to examine social bond attainment. The second aspect of the
research design was centered on the estimation of the nature of the social bond.
Hierarchical Linear Modeling (HLM) and ordinary least squares regression were used for
this phase of the research. Theoretically, it was expected that those individuals who
experience a stay of incarceration would be less likely to develop positive social bonds to
marriage and work. In the same light, of people who attain social bonds, it was

hypothesized that the nature of the bond would be negatively affected by incarceration.

88

Fixed-Effect Models of Social Bond Development
Marital Bonds

The goal of the first phase of the analysis was to examine the attainment of the
bond to marriage. It was hypothesized that individuals who had experienced a stay of
incarceration would be less likely to attain social bonds. The incarceration event, like
other salient life experiences (e.g., substance abuse), can sever ties to conventional social
relationships reducing the likelihood of attainment of social bonds (Sampson & Laub,
1993). As such, it has been theorized that those individuals who experienced one or more
incarceration events during the study period would be less likely to be married, even after
controlling for individual demographic characteristics, criminal history, and contextual
influences.

In order to understand attainment of the marital bond, marriage was
operationalized using a binary measure. The marriage dependent variable was coded as
“1” if the respondent was married at some point from 1983-2000. As shown in Table 10,
incarcerated persons were less likely to become married during the sample period with
38% of the sample reporting at least one marriage during the study period. Nearly half
(47.3%) of the match group reported being married, and 62% of the median group
attained marriage during the study period. Based on the results of the chi-square test, the
differences between groups are statistically significant (78:57.59, d.f.=2, p<0.001). In
short, individuals who have had at least one stay of incarceration are less likely to achieve

the bond of marriage.

89

Table 10: Attainment of Marital Bond (N=1,478)

 

 

IMarried during Study Period
Yes Percent
Matched group 228 47.3%
Incarcerated sample 182 37.8%
Median group 317 61.6%

 

 

78:57.59, d.f.=2, p<0.001
Aginment of Marriage

As evidenced by the chi-square analysis, the incarceration event is an important
factor in understanding the attainment of marital bonds. Since the bivariate relationship
has been established, it is important to fiirther specify the relationship between
incarceration and marriage. The following research questions were addressed in the
subsequent analyses. Does the relationship between incarceration and marriage remain
significant after controlling for lifecourse events, demographic influences, criminal
history and context influences? Moreover, does the association between exogenous
predictors and marital attainment vary across sample groups and with the timing and
nature of the incarceration?

A series of logistic regression models were estimated to address each of the
research questions. In order to understand the causal order of lifecourse events, marriage
was operationalized in this model as a binary measure and dichotomized into individuals
who were married in 2000 and otherwise. The absence from work and military status
variables were also dichotomized with individuals reporting an absence from work or
participation in the military from 1983 to 1998 as the reference category. In addition,
employment was included in the model as a continuous count measure representing the

number of study periods in which the subject reported being employed on a fulltime

90

basis. A subject was considered employed fulltime if over 35 hours of work per week
were reported.

Before the models of social bond attainment were estimated, general diagnostic
procedures were used to assess the general statistical validity of the predictive models. In
order to examine the statistical appropriateness of the exogenous predictors, a correlation
matrix for each of the social bond models was constructed and VIP values estimated.
Results from these analyses for the marriage model are presented in Appendix C and D.
Although some of the tolerance and VIF statistics are higher than would be optimal, the
coefficients do not suggest severe problems with multicollinearity that would bias the
statistical estimates 8.

Judging from the results from the logistic regression model, incarceration remains
a salient predictor in the model after controlling for lifecourse events, demographic and
context influences, and criminal history (see Table 1 1). Incarceration during the research
period was significantly and negatively associated with marriage, but incarceration that
occurred before the sample period had no effect. The effect of incarceration was large.
Individuals who had been imprisoned at one point during the study period were 62% less

likely to be married in 2000.

 

8 In addition to examining the VIF and Tolerance scores for the model, polychoric correlations were also
estimated. A number of borderline multicollinearity problems were highlighted by this analysis. In
specific, it appears that previous incarceration is highly correlated with antisocial behavior. Although these
variables will not be removed from the current analyses, the problem of multicollinearity with the
predictors will be addressed in future research.

91

Table 11. Marital Attainment in 2000 for Total Sample (N=1,478)

 

 

 

B s.e. Odds
Constant -0.24 0.55 0.79
Lifecourse Events
Incarceration -0.98*** 0.16 0.38
Work 013*" 0.02 1.14
Military 0.68“ 0.24 1.97
Absence from Workforce 0.06 0.15 1.06
Demographic Influences
Age 0.00 0.03 1.00
Hispanic 0.02 0.17 1.02
African American -0.94*** 0.17 0.39
Male -0.26 0.16 0.78
Cognitive Ability 0.00 0.00 1.00
High School Graduate 001 0.15 0.99
Criminal History
Antisocial Behavior -0.32** 0.13 0.73
Previous Incarceration -0.27 0.25 0.77
Context Influences
Family Poverty -0.34 0.14 0.71

 

*** p<.001 ** p<.01 (two tailed tests)
-2 log likelihood = 1655.35
Nagelkerke RZL = 0.26

Cox &Snell = 0.18

 

92

In contrast to incarceration, both employment and military participation increased
the likelihood of marriage. The effect of military participation was particularly strong.
Individuals who had participated in the military at some point during the study period had
nearly two times the log odds of being married. Although there is little theoretical
precedence to this finding, it appears that individuals who maintain steady work or
associations with stable social institutions like the military are more likely to develop
bonds to marriage. It may be that marriage and employment are interrelated. The
coalescence of employment, marriage, and participation in social institutions like the
military aid the individual in presenting a ‘respectability package’ (Giordano, Rossol, &
Cernkovich, 2003) that increases the likelihood for development of subsequent social
bonds.

Both demographic influences and criminal history predictors also conditioned the
marital attainment model. African American sample members were significantly less
likely to become married. Based on the odds ratio statistic, the explanatory power for
race was similar to that of incarceration. African American persons who had been
incarcerated at some point during the sample period were dramatically less likely to
become married. Furthermore, individuals who demonstrated antisocial tendencies as

young adults through drug use and criminality were also less likely to be married.

Marital Attainment by Group
In order to further understand the deve10pment of the bond to marriage across
groups, sub-group analyses were conducted. Consistent with the models presented

above, logistic regression was used as the primary analytic technique. In addition, Z

93

scores were calculated for each of the pairs of exogenous predictors in the models
according to the formula presented by Brame, Paternoster, Mazerolle, and Piquero (1998;
1998). The results provided a richer understanding of the possible differential effects of
lifecourse events, demographic characteristics, criminal history, and context influences
on marriage for the incarcerated, matched, and comparison samples.

The most striking finding from the analyses is that work is significantly and
positively associated with marriage across groups (see Table 12). Individuals who were
employed fulltime were more likely to attain a marital bond. Based on the standardized
coefficients, the strength of the relationship is small, but the consistency in the
relationship across sub groups is important. This finding speaks to the importance of
examining existing social bonds when studying the development of social relationships.

Military participation and absence from work were also important predictors in
the models. Enlistment in the military was significantly associated with marriage for the
matched sample but not for the incarcerated and median groups. Absence from the
workforce was negatively associated with marriage for the incarceration model, but did
not achieve a level of significance in the median and matched groups. Interestingly, for
the absence from work variable, the Z coefficient for the median, incarcerated group
contrast was statistically significant (see Table 13). This outcome suggests that absence
from work had a disparate, negative impact on marital bond attainment for the
incarcerated group when compared to the median group. It may be that the absence from
participation in the traditional workforce confounds the effect of incarceration making
attainment of marriage less likely. The multiplicity of absence from the workforce and

incarceration reduces the time spent by the individuals in conventional society; thus, the

94

time allowed for courtship would be reduced and the social presentation of respectability
compromised.

Race was the only salient demographic predictor of marital bond attainment in the
model. Across groups, Afiican American respondents were significantly less likely to be
married during the study period. The results from the odds ratio calculations suggest that
race may have a stronger impact on the probability of marriage for the incarcerated
sample; although, additional analyses would need to be conducted to further validate this

claim.

95

Table 12. Marital Attainment in 2000 by Group (n=1,478)

 

 

 

Incarceration Match Median
(n=484) (n=484) (n=5 18)

b s.e. odds b s.e. odds b s.e. Odds
Constant -l.83 1.29 0.16 -0.23 0.97 0.80 0.11 0.88 1.11
Lifecourse
Events
Work 013* 0.05 1.13 0.15M 0.03 1.16 0.13" 0.03 1.14
Military 0.16 0.58 1.18 0.80* 0.36 2.23 0.77 0.43 2.15
Workforce -0.80* 0.42 0.45 -0.07 0.23 0.93 0.42 0.25 1.52
Demographic
Influences
Age 0.01 0.06 1.00 -0.00 0.05 0.99 0.01 0.05 1.00
Hispanic -0.09 0.35 0.91 0.23 0.30 1.25 -0.27 0.30 0.76
African -1.17*** 0.35 0.31 -0.64* 0.28 0.53 -l.07** 0.39 0.01
American
Male 0.74 0.59 2.10 -0.50 0.39 0.61 -0.20 0.58 0.82
Cognitive 0.00 0.01 1.00 0.00 0.01 1.00 -0.00 0.01 0.99
Ability
High School 0.78 0.30 0.01 -0.13 0.23 0.88 -0.61 0.32 0.54
Criminal
History
Antisocial -0.46 0.30 0.63 -0.20 0.21 0.82 -0.36 0.31 0.70
Behavior
Previous 0.19 0.39 1.21 -0.49 0.36 0.62 **** **** ****
Incarceration
Context
Influences
Family -0.21 0.29 0.81 -0.27 0.22 0.76 -0.64** 0.27 0.53
Poverty

-2 log likelihood = -2 log likelihood = -2 log likelihood =

370.243 602.09 646.90

 

Nagelkerke RZLZO. l 8

Cox &Snell = 0.1 1

 

 

Nagelkerke R2L=0.16

Cox &Snell = 0.10

 

Nagelkerke RZLZO. l 7

Cox &Snell = 0.11

 

*p<0.05 "p<0.01 ***p<0.001 (two-tailed test)

*"* None of the members of the median group reported incarceration prior to 1980; hence,
prior incarceration represents a constant and was removed from the analysis.

96

Across models, criminal history variables were not significantly associated with

marriage, but the contextual measure of family poverty was negatively related with

attainment of the marital bond for the median group. Individuals from the median group

that grew up in impoverished circumstances nearly half as likely to be married. The Z

scores from the poverty model were not significantly different. Despite the difference in

the significance levels of the coefficients, the effect of adolescent or early adulthood

economic status and marriage did not vary significantly between groups.

Table 13. Z Score Coefficients for Sub Groups Comparisons — Marital Attainment Model

 

 

Incarcerated and Matched Groups

Incarcerated and Median Groups

 

 

 

 

Marriage 0341 0.00
Military 094 -0.84
Absence from -1 .52 -2.50*
Workforce

Age 0.13 0.00
Hispanic 069 0.39
Black -1.18 -0.19
Male 1.75 1.14
Cognitive Ability 0.00 0.00
High School 2.41* 3.17*
Antisocial Behavior -0.71 -0.23
Previous Incarceration 1.28 ****
Poverty 0.16 1.09
* p<.05

**** This contrast was not estimated because no members of the median group reported a
stay of incarceration prior to the study period.

97

Iimfig and Nature of Incarcenation

Based on the results of the analyses presented to this point, it is evident that
incarceration can affect life trajectories. Individuals who experienced incarceration at
some point during the study period were significantly less likely to become married, net
of traditional control measures. What is not known from this research is the possible
differential relationship between the nature and timing of the incarceration event and the
attainment of social bonds to marriage. The theoretical foundation of the work is drawn
fi'om social capital literature as posited by Sampson and Laub (1993) and Coleman
(1990; 1988). Based on this framework, one would expect that individuals who were
incarcerated at an early age would be less likely to attain marriage because avenues to
traditional interaction (e.g., employment, school) would be limited. For example,
individuals incarcerated at an early age are less likely to participate or complete
secondary or post-secondary education (Arum & Beattie, 1999). Degree completion not
only increases one’s human capital, it also affords opportunities to build relationships that
may increase social capital (e.g., marriage). In the same light, multiple periods of
incarceration would limit participation in society; thus, reducing the likelihood of
developing positive social bonds to marriage.

In order to test the relationship between the timing of the incarceration event, the
number of incarceration periods, and marriage a series of logit models were estimated.
The models are identical to those estimated previously except for the inclusion of a
multiple incarceration and an age at incarceration variable. The multiple incarceration

measure is dichotomous with those individuals incarcerated at more than one point from

98

1983-2000 as the reference category. The age at incarceration represents the age of the
subject at their first adult incarceration experience.

The age at incarceration and multiple incarceration measures had disparate effects
on the attainment of marriage. Age at incarceration was not significantly related to
marriage. Including the age at incarceration actually reduced the explanatory power of
the model by nearly half. In addition, when age at incarceration was added to the model,
the work variable became insignificant and the high school graduate measure became
insignificant.

As expected, those individuals who were incarcerated at more than one time
period during the study period were significantly less likely to be married in 2000 (see
Table 14). Within the incarcerated group, individuals who were incarcerated multiple
times were 44% less likely to become married. While the inclusion of this particular
variable did not change the model fit dramatically, the results do illustrate the importance
of examining the nature of incarceration. What is more interesting is that addition of this
variable did not change the relationship between work and marriage. Individuals with
steady employment histories were significantly more likely to be married, net of the

incarceration experience.

99

Table 14. Nature of Incarceration and Marriage — Incarcerated Sample Only (n=418)

 

 

 

Base Model Age at Incarceration Multiple Incarceration
b s.e. odds b s.e. odds b s.e. odds
Constant -1.83 1.29 0.16 -0.01 0.93 -1.64* 0.74 0.19
Lifecourse
Events
Work 013* 0.05 1.13 0.10 0.06 1.11 0.10* 0.05 1.11
Military 0.16 0.58 1.18 0.47 0.61 1.61 0.04 0.58 1.05
Workforce -0.80* 0.42 0.45 -1.01* 0.47 0.36 -0.63 0.42 0.53
Multiple —- -- -- -- -- - -0.58* 0.30 0.56
Age at -- -- -- -0.05 0.03 0.95 -- -- --
Incarceration
Demographic
Influences
Age 0.01 006 100 **** ##1## **** ##3## **** ****
Hispanic -0.09 0.35 0.91 -0.09 0.38 0.91 -0. 12 0.35 0.89
African -1.17*** 0.35 0.31 -1.06** 0.38 0.35 -1.15** 0.35 0.32
American
Male 0.74 0.59 2.10 0.84 0.68 2.31 0.86 0.58 2.36
Cognitive 0.00 0.01 1.00 0.00 0.01 1.00 0.01 0.01 1.01
Ability
High School 0.78 0.30 0.01 0.83“ 0.32 2.30 -0.12 0.35 0.89
Criminal
History
Antisocial -0.46 0.30 0.63 -0.46 0.32 0.63 -0.48 0.30 0.62
Behavior
Previous 0.19 0.39 1.21 0.20 0.44 1.22 0.29 0.39 1.34
Incarceration
Context
Influences
Family 021 0.29 0.81 -0.39 0.31 0.68 -0.15 0.29 0.86
Poverty
-2 log likelihood = -2 log likelihood = -2 log likelihood =
370.243 319.71 366.42

 

Nagelkerke R2L=0.18
Cox &Snell = 0.10

 

Nagelkerke R2L=0.10
Cox &Snell = 0.06

 

Nagelkerke R2L=0.19
Cox &Snell = 0.10

 

*p<0.05 **p<0.01 ***p<0.001 (two-tailed tests)
I"""""‘Age was excluded from Model 11 and 111 because of multicollinearity with the added

variables.

100

Nature of the Marifital Bond

The results presented above suggest that incarceration can attenuate the
development of social bonds. In the same light, the possession of existing social bonds
increases the likelihood of developing subsequent social relationships. The research
presented to this point has only focused on the attainment of bonds; however, researchers
have suggested that it is not the bond per se that mitigates involvement in negative
behaviors (e.g., delinquency, drug use). Rather, it is the capital and the informal social
control that is derived from a strong and involved bond that increases the likelihood of
both the initiation of a criminal career and desistance from criminality and other problem
behaviors (Coleman, 1988, 1990; Sampson & Laub, 1993).

The goal of the following analysis is to understand the relationship between
incarceration and the nature of the marital bond. Based on the research conducted to
date, it is hypothesized that for those individuals who do become married, that the nature
of the marriage will be diminished as a result of incarceration.

In the current study, the nature of the marital bond was operationalized using a
factor score. The three item factor score included 1) number of marriages reported 2)
length of first marriage and 3) if the respondent had been divorced at any time in the
study period (eigenvalue 1.94, factor loadings > 0.74). The variable was then
transformed by multiplying the factor score by -1. As a result of the transformation,
individuals that reported a stronger bond to marriage had a higher score for the dependent
measure. The independent predictors included in the model are identical to those

included in the previous marital bond models.

101

Similar to the results fi'om the logit models, individuals who had been
incarcerated at some point during the study period were less likely to attain strong marital
bonds; whereas employment enhanced the bond to marriage (see Table 15). Contrary to
the dichotomous model, military participation was associated with a diminished social
bond. Individuals who entered the military were more likely to become married but the
nature of the relationship was diminished as a result of the event. Judging from the
standardized Beta coefficients, employment had the strongest relationship to quality of
marriage of all the lifecourse predictors.

Individuals of Hispanic ethnicity or African American race also reported a strong
bond to marriage. These findings contradict the results from the dichotomous marriage
model, where African Americans were significantly less likely to attain the bond to
marriage. It appears that once an Afiican American individual attains marriage, that the
bond to marriage is likely to be stronger and to be sustained for a longer period of time.
In addition, the relationship between race, ethnicity and marriage was strong. The Beta
coefficient for Afiican American race and Hispanic ethnicity was identical to that of the
employment measure and greater than that of incarceration.

Antisocial behavior was the only other significant predictor in the model with
individuals reporting involvement in criminality or drug use prior to the study period less
likely to maintain quality marital relationships. Together, the variables explained little

variation in the dependent measure as the R2 for the model was small at 0.06.

102

Table 15. Nature of Marital Bond in 2000 for Married Sample (N=810)

 

 

 

 

b s.e. Exp(B)
Constant -0.67 0.38
Lifecourse Events
Incarceration -0.25* 0.12 -0. 10
Work 004’” 0.14 0.14
Military 031* 0.15 -0.09
Absence from Workforce 0.09 0.10 0.05
Demographic Influences
Age 0.02 0.02 0.05
Hispanic 0.31" 0.1 l 0.14
Afiican American 0.30“ 0.11 0.14
Male 0.15 0.1 l 0.07
Cognitive Ability 0.00 0.00 0.04
High School Graduate —0. 17 0.10 008
Criminal History
Antisocial Behavior -0.22* 0.09 -0.10
Previous Incarceration -0. l 8 0.16 005
Context Influences
-0.10 0.10 005
Family Poverty
* 0.05 **p<0.01 (two tailed tests)
R = 0.062

103

Bonds to Employment

Similar to the bond to marriage, an individual’s ability to secure employment may
also be tempered by the incarceration event (e.g. Freeman, 1991; Kling, 1999; Waldfogel,
1994). The goal of this phase of the research was to examine the procurement of
employment following incarceration. Based on the extant research, it was expected that
individuals who experienced one or more stays of incarceration would be significantly
less likely attain a job and the stability and nature of the employment would be
diminished in comparison to those individuals who did not experience incarceration.

The overwhelming majority of individuals were employed for at least one full
year over the study period (Table 16). Less than five percent of the matched (4%) and
median groups (8%) did not garner employment during the study period; while, twenty-
three percent of the incarcerated group remained unemployed from 1983-2000. Despite
the similarity in employment for matched and median groups, the differences between
groups in reference to employment over the study period are statistically significant
06:94.44, d.f.=2, p<0.001).

Table 16. Attainment of Emplognent by Group (N=1 ,47 8)

 

 

 

mployed during Study Period

es Percent
Incarcerated Group 374 77.3%
Matched Group 464 95.9%
Median Group 478 92.3%

 

 

£293.44, d.f.=2, p<0.001

104

To further explore the relationship between employment and incarceration, a
series of logit models were estimated. The first model includes the entire sample and was
designed to test the relative effect of incarceration on employment in relation to other
lifecourse events (e. g., marriage). For example, does the incarceration event outweigh
that of obtaining a high school diploma or serving in the armed forces? Second, separate
predictive models were estimated for individuals in the median, incarcerated, and
matched groups. As part of this analysis, the equivalency of coefficients across groups
was also tested. Finally, the relationship between the timing and nature of incarceration
and employment was further examined using the sample of incarcerated persons.

Consistent with the operationalization of measures for the marriage dependent
variable, the employment measure used in the following models is dichotomous and
represents fulltime employment (employed for more than 35 hours per week) in 2000.
Each of the lifecourse event variables is dichotomous and represents the presence of the
event during the study period from 1983 to 1998. In addition, a series of control
variables were included in the models. The control variables were measured prior to the
study period and were designed to further condition the social bond analyses.

The bivariate analysis of attainment of employment in 2000 further confirms the
negative relationship between incarceration and work (see Table 17). One third of the
incarcerated group reported fulltime participation in the workforce. The employment
rates for the median and matched groups were nearly identical with nearly 70% of the
sample members in each group demonstrating firlltime employment in 2000. The
difference between sample groups was statistically significant (x2=180.01, d.f.=2,

p<0.001).

105

Table 17. Fulltime Employment in 2000 (N=1,478)

 

 

 

Yes Percent
Incarcerated Group 156 32.2%
Matched Group 329 68.0%
Median Group 362 69.9%

 

x2=180.01, d.f.=2, p<0.001

Attainment of Emploment

As evidenced by the chi-square analysis, the incarceration event is an important
factor in understanding the attainment of employment. The following logit models
further specify the relationship between incarceration and employment. General
diagnostics were conducted prior to the estimation of the final models. The correlation
matrix and VIF and tolerance estimates are presented in Appendix E and F. As with the
marriage model, a number of variables had moderate tolerance estimates, but none of the
coefficients were such to suggest a serious threat to the validity of the model as a result of
multicollinearity.

The contribution of lifecourse events in understanding the attainment of
employment was further reinforced in the logit model for the total sample (see Table 18).
Each lifecourse event including marriage, military participation, incarceration, and
absence from workforce was significantly associated with the dichotomous measure of
employment. Individuals who had spent time in the military, been absent from the
workforce, or had been incarcerated were substantially less likely to be employed in

2000. The strength of the relationships between the variables and employment was also

106

very strong. Incarceration reduced the likelihood of employment by 69% and military
participation and absence from the workforce reduced the odds by nearly half. Military
service, incarceration, and absence fi'om workforce all represent events that separate the
individual from the conventional employment market. It appears that removal from the
workforce at any period can mediate meaningfirl employment.

Conversely, marriage was found to be both significantly and positively related to
employment; although, the relationship was not strong. This finding demonstrates the
reciprocal nature of social bond development. Individuals who have attained social
bonds were more likely to impart a feeling of social respectability. The results also
substantiate the relationships observed between employment and marriage in previous
models. Even after controlling for negative lifecourse events (e.g., incarceration) married
individuals were more likely to attain fulltime employment.

Demographic indicators also conditioned the employment model. Individuals
who tested high on cognitive ability tests and had a high school diploma were more likely
to be employed. This is not surprising in that both education and cognitive abilities are
representations of human capital. It would follow that individuals with more human
capital would also be more likely to attain employment. In addition, individuals of
Hispanic ethnicity were less likely to be employed; whereas, gender was positively
associated with employment. Males were also one and a half times more likely to be
employed.

Interestingly, the criminal history predictors and context influences were not
significantly associated with employment. It appears that sample members were able to

overcome the experience of incarceration and involvement in adolescent antisocial

107

behaviors when those events occurred in early adulthood and late adolescence; however,
incarceration in adulthood significantly reduced the likelihood of obtaining employment.

Table 18. Employment in 2000 for Total Sample (N=1,478)

 

 

 

b s.e. EprB)
Constant 1.14 0.54
Lifecourse Events
Incarceration -1 .17M 0.14 0.31
Marriage 0.06" 0.01 1.06
Military -0.62* 0.24 0.54
Absence fiom Workforce -0.86** 0.15 0.42
Demographic Influences
Age 003 0.03 0.97
Hispanic -0.33* 0.17 0.72
Afiican American -0.03 0.17 0.97
Male 040* 0.15 1.49
Cognitive Ability 0.01* 0.00 1.01
High School Graduate 0.37" 0.14 1.45
Criminal History
Antisocial Behavior -0.03 0. l 3 0.97
Previous Incarceration -0.10 0.22 0.90
Context Influences
Family Poverty -0.19 0.13 0.83

 

*p<.05 "p<.01 (two-tailed tests)
-2 log likelihood = 1715.55
Nagelkerke R2L=0.26

Cox &Snell '2 0.19

108

Employment by Group

In the sub-group analyses, lifecourse events remained significant predictors of
marriage. Across all models, married individuals were significantly more likely to attain
fulltime employment (see Table 19). The results from the military participation and
workforce absence models varied across groups. Individuals in the median group that
had participated in the military at some point in the study period were significantly less
likely to be employed fulltime. Military participation was not significant for the
incarcerated or matched groups. Absence from the workforce was also negatively
associated with employment for the incarcerated and matched groups. This finding is not
surprising in that separation from the workforce limits the human and social bonds that
can be built through steady employment, thereby, reducing the likelihood of employment.

The demographic influences associated with employment vary widely across
sample groups. For the median group, males were significantly more likely to have
attained employment. Cognitive ability was significantly and positively associated with
employment for the matched sample; while, high school education was the sole
significant, demographic predictor for the incarcerated group.

Criminal history variables played a very small role in the models. Antisocial
behavior was the only significant predictor of employment and only for the median
model. Individuals in the median group that reported drug use or involvement in criminal
behavior as youth were significantly less likely to be employed in 2000. In addition, the
Z score was significantly different for contrast of the incarceration and matched groups

(see Table 20).

109

Table 19. Employment in 2000 by Sample Group (N=1,478)

 

 

Incarceration Match Median
(n=484) (n=484) (n=518)
b s.e. odds B s.e. odds b s.e. Odds

Constant -0.43 0.99 0.65 2.05* 1.03 7.73 0.53 0.94 1.69
Lifecourse
Events
Marriage 0.07* 0.03 1.07 0.07" 0.02 1.07 0.05* 0.02 1.05
Military -0.53 0.49 0.59 -0.20 0.38 0.82 -1.29** 0.43 0.28
Absence -0.47 0.36 0.62 —0.89** 0.22 0.41 -1.09** 0.27 0.34
from
Workforce
Demographic
Influences
Age -0.03 0.05 0.97 -0.07 0.05 0.94 0.01 0.05 1.00
Hispanic -0.49 0.30 0.62 -0.30 0.31 0.75 -0.02 0.33 0.98
African -0.29 0.27 0.75 0.01 0.31 1.00 0.67 0.42 1.98
American
Male 0.76 0.42 2.14 -0.08 0.39 0.92 1.49* 0.64 4.46

Cognitive -0.00 0.01 0.99 002* 0.01 1.02 -0.00 0.01 0.99
Ability
High School 0.54* 0.22 1.72 0.18 0.24 1.20 -0.12 0.34 0.98
Graduate

 

Criminal
History
Antisocial -0.12 0.22 0.89 -0.23 0.22 0.80 0.75* 0.36 2.11
Behavior
Previous -0.08 0.29 0.93 -0.01 0.35 0.99 **** **** ****
Incarceration
Context
Influences
Family -0.14 0.21 0.87 0.02 0.22 1.02 -0.40 0.27 0.67
Poverty
-2 log likelihood = -2 log likelihood = -2 log likelihood =
582.60 564.16 544.78
Nagelkerke R2L=0.07 Nagelkerke R2L=0.12 Nagelkerke R2L=0.22
Cox & Snell=0.08 Cox & Snell =0.04 Cox & Snell=0.l6

 

 

 

 

*p<0.05 **p<0.01 p<0.001 (Two-tailed tests)

**** None of the members of the median group reported incarceration prior to 1980; hence,
prior incarceration represents a constant and was removed from the analysis.

110

It is also important to consider the relative fit for the models. Based on the
Nagelkerke RZL, the totality of variables in the median model explain more variance in
employment than either the matched or incarcerated groups. In fact, the model for the
median group explained three times the variance when compared to the incarcerated
group and twice that of the matched sample. This finding reinforces the need for
additional research on models of employment for individuals who tread non-traditional
employment paths (e.g., seasonal employment, part-time work). Clearly, models that are
adapted from research predicting participation in the fulltime, mainstream employment

market are not as efficient in predicting employment for the incarcerated and matched

groups.

111

Table 20. Z Score Coefficients for Sub Groups Comparisons — Employment Attainment Model

 

 

 

 

 

 

Incarceration and Matched Incarceration and Median
Marriage 0.00 0.55
Military 053 1.17
Absence from 1.00 1.38
Workforce
Age 0.57 057
Hispanic 044 -1.05
Black 073 -1.92
Male 1.47 -0.95
Cognitive Ability -1.41 0.00
High School 1.11 1.63
Antisocial 0.35 -2.06*
Behavior
Previous 015 ****
Incarceration
Poverty -0.53 0.76
* p<.05

**** This contrast was not constructed because zero members of the median group reported a
stay of incarceration prior to the study period.

Timing and Nature of Incarceration
Consistent with the modeling techniques used for the marriage dependent
variable, it is also important to examine the relationship between the timing and nature of
incarceration and employment. While research conducted in this area is still in its
infancy, current thinking suggests that early incarceration may be more detrimental than

that of adult incarceration (Kling, 1999; Waldfogel, 1994; Western & Beckett, 1999).

112

 

Based on the social capital framework, one would expect that individuals who were
incarcerated multiple times would be less likely to be able to attain the human capital to
obtain a job. In the same light, incarceration at an early age may hinder an individual’s
ability to successfully secure fixture employment by limiting opportunities to gain an
education and to build social relationships.

A series of logit models were estimated to examine the relationship between
employment and the nature and timing of incarceration. The models are identical to
those estimated previously except for the inclusion of a multiple incarceration and age at
incarceration variable. The multiple incarceration indicator is dichotomous with those
individuals incarcerated at more than one point from 1983-2000 as the reference
category. The age at incarceration represents the age of the subject at their first adult
incarceration experience.

Both the number of stays of incarceration and the timing of those events were not
significantly associated with employment for the incarcerated the sample. What is more
interesting is that inclusion of the multiple incarceration variables into the model changes
the relationship between marriage and work. It appears fiom this model that multiple
terms of incarceration do counteract in some manner the positive effects of marriage.
This is an important finding in that the majority of sample members experienced multiple
stays of incarceration. The model fit and coefficients did not change dramatically by

adding the age at incarceration measure.

113

Table 21. Nature of Incarceration and Employment ~— Incarcerated Sample Only (n=418)

 

Constant

Lifecourse

Events
Marriage
Military
Absence
from

Workforce
Multiple

Age at
Incarceration
Demographic
Influences
Age
Hispanic
Afiican
American
Male
Cognitive
Ability
High School
Graduate
Criminal
History
Antisocial
Behavior

Previous
Incarceration

Context
Influences
Family

Poverty

Base Model

Age at Incarceration

Multiple Incarceration

 

b s.e. odds

-0.43 0.99 0.65

0.07 0.03 * 1.07
-0.53 0.49 0.59
-O.47 0.36 0.62

-0.03 0.05 0.97
-0.49 0.30 0.62
-0.29 0.27 0.75

0.76 0.42 2.14
-0.00 0.01 0.99

0.54 0.22“ 1.72

-0.12 0.22 0.89

-0.08 0.29 0.93

-0.14 0.21 0.87

B s.e. odds

-0.10 0.94 0.99

0.53 023* 1.70
-0.38 0.51 0.56
-0.63 0.40 0.53-

-0.04 0.02 0.96

**** ***** ****

-0.44 0.32 0.64
-0.11 0.29 0.90

0.73 0.45 2.01
-0.00 0.01 0.99

0.54 0.24* 1.71

-0.16 0.23 0.85

-0.26 0.32 0.77

-0.12 0.22 0.89

 

 

-2 log likelihood =
582.60

Nagelkerke R2L=0.07
Cox & Snell 20.04

 

-2 log likelihood :
5 19.78

Nagelkerke R2L=0.08
Cox & Snell =0.04

 

b s.e. odds
-0.90 0.59 0.41
0.39 0.22 1.47
-0.59 0.49 0.55
-0.52 0.36 0.59
-0. 12 0.22 0.89
***# **** ****
-0.48 0.29 0.62
-0.28 0.27 0.76
0.79 0.42 2.19
-0.00 0.01 0.99
0.50 0.22* 1.65
-0. 13 0.22 0.88
-0.06 0.29 0.95
-0. 13 0.21 0.88
-2 log likelihood =
544.12

Nagelkerke R2L=0.07
Cox & Snell 20.04

 

*p<0.05 "p<0.01 ***p<0.001

**** Age was excluded from Model 11 and 111 because of multicollinearity with the age at
incarceration and multiple variables.

114

Summary

From the series of logit models constructed, it is apparent that incarceration
during the lifespan has great potential to mitigate the development of bonds to marriage
and work. Across all models estimated, incarceration remained both a strong and a
negative predictor of social bond development. In the marriage model, multiple terms of
incarceration also reduced the likelihood of bond attainment; however, the inclusion of
the additive incarceration measure did not improve the model fit or explanatory power.

Surprisingly, previous incarceration was not found to be significant in any of the
models estimated. Incarceration as an adolescent or young adult had little effect on the
development of bond to marriage or work. It appears that individuals were able to
overcome the handicap of incarceration in adolescent and early adulthood; however,
incarceration in early to middle adulthood mortgaged the respondents’ ability to develop
positive social bonds. This finding is inconsistent with other research that has identified
a strong, negative relationship between youth incarceration and employment, but have
not found similar negative effects for adult incarceration (Freeman, 1991; Western &
Beckett, 1999).

The importance of lifecourse events in the development of social bonds was also
reinforced by the analyses. The reciprocal nature of the social bond development to work
and marriage was the most surprising finding of the analyses. Individuals who had been
married were significantly more likely to attain fulltime employment. In the same light,
employment was significantly associated with marriage.

Military participation was also significant across models; however, the

relationship between military participation and social bond development varied by model.

115

Military participation was positively associated with marriage, but individuals who had
participated in the military were less likely to report fulltime employment in 2000. In the
marriage model, military participation appears to represent stability and social attachment
to a potential mate; thereby, increasing the likelihood of marriage. In relation to
employment, military participation removes the individual from the traditional workforce
limiting avenues for participation in the mainstream workforce.

Overall, lifecourse events are important in understanding social bond
development. None of the demographic, criminal history, or poverty measures had a
consistent effect on social bond attainment to marriage or work. Only African American
race was a significant predictor of marriage across models. The results of the analyses
provide support for the lifecourse theoretical perspective. Although individual,
demographic influences were significant predictors in a few of the models, most of the
model variation was due to the inclusion of the lifecourse predictors, specifically

incarceration, employment, and marriage.

Models of Individuals Change

As stated above, lifecourse events had a consistent and salient effect on social
bond attainment to work and marriage. The following analyses firrther explore the
relationship between lifecourse events, demographic influences, and social bonds. First,
a series of models were estimated to understand the nature of change in the probability of
attainment of bonds to marriage and work over time. These models served to validate
the findings of the logit models and to further examine the relationship between

incarceration and the attainment of social bonds. The second group of models was

116

designed to examine the change in the nature and stability of employment over time as
measured by tenure with primary employer and number of j obs worked. Together the
analyses aid in the understanding of both individual differences at the initiation of the
research and the character of change in social bonds over time. The structure of the
analyses also allows the researcher to draw conclusions across models as to similarities in
salient predictors of social bond attainment. For example, does incarceration affect the
likelihood of employment over time, net of controls? In addition, does the predictive
model for the probability of employment model differ from that estimated for the tenure
with employer dependent measure?

The theoretical questions posed by this research were tested using Hierarchical
Linear Modeling (HLM) (Raudenbush & Bryk, 2002). The first part of the following
section outlines the statistical technique. The subsequent sections present the results
from both the probability of social bond attainment and the nature of employment

models.

The HLM Model

Hierarchical Linear Modeling (HLM) was selected as the primary analytic
technique for this research. HLM is the most appropriate technique because it allows
data from individual periods in the lifecourse to be nested within persons while modeling
variation in the outcomes of marriage and work. Traditionally, researchers were forced
to either examine the nature of individual change over time without controlling for a
number of time-invariant demographic variables (e. g., repeated measures design) or to

aggregate individuals into groups to model change. Both methodologies involve

117

eliminating data points; hence, reducing the power of the statistical model and the
validity of the results. HLM is a superior modeling technique in that it allows
researchers to study the effect of time-varying covariates on dependent measures while
controlling for time-invariant measures. In this model, HLM was used to estimate the
relationship between lifecourse events, specifically incarceration, and social bond
attainment, net of traditional controls. HLM is also more flexible than traditional
longitudinal modeling techniques. For example, the nested structure of the HLM model
facilitates the valid estimation of models when both the spacing and the number of
observations vary by individual. This fimctionality of the HLM technique is important
for this research in that 65% (968) of respondents could not be interviewed at one or
more data collection points during the study period.

The modeling functions for growth curve analysis used in this research were
conducted in a manner that is consistent with the techniques proposed by Raudenbush
and Bryk (2002:160-204). The general methodology employed in Hierarchical Linear

Modeling is presented below.

Estimation of the HLM Model
Level 1 Model
Growth curve models were estimated on two levels). The first level of the model
is a within-individual model and includes repeated measures of individuals. In the
current study, individuals were observed yearly and bi-annually and each measurement

point represents a data point in the model. Respondents were interviewed a total of

 

9 It is possible to include additional levels in a HLM model. The research questions presented do not
warrant the estimation of a third level; hence, the discussion will be limited to the traditional two-level
model.

118

fifteen times over the study period. It is important to note that the individual respondent
is not the subject in the Level I model. Instead, the Level I model serves to examine the
individual at multiple points in time; hence, the time or interview event is the focus of the
Level I model (Cherlin, Chase-Lansdale, & McRae, 1998). The primary goal of the
Level 1 model is to assess the individual trajectory of growth characterized by an
intercept and a rate of change (Raudenbush & Xiao-Feng, 2001). A general
representation of the Level 1 model estimated in the following analyses is presented

below.

Tenure with employer“ 2 no; + It“ agen+ 1112i Incarceration“ + e“ ( 1)

For each individual in the model, Y“ represents the likelihood of employment,
likelihood of marriage, tenure with employer, or the number of j obs worked. Each
dependent measure was estimated for each individual i at time t and was modeled in
relationship to both growth curve and random error. Models were constructed for 1': 1, .
. . , 1,46610 individuals. The constant for the model is represented by run and the slope as
1c“, The error term e“ is assumed to have a simple error structure. Error terms estimated
for each individual are assumed to be independently and normally distributed with a
mean of zero with a variance (3'2 constant across individuals.

The nature of the grth trajectory was modeled using age of the participant
(ageti). The age variable in the model was centered using an arbitrary point of reference.
For each individual the age measure was calculated using age-21. The centering point of

21 was selected because it represents the midpoint of the age range for the respondents at

 

'0 Twenty individuals were excluded from the final analyses because they did not report employment at any
period in the study; hence, there was zero variation in which to model in the tenure and jobs worked
models.

119

the first phase of data collection for the current study. The age term was centered to aid
in the interpretation of the coefficients. In addition, centering the age measure and its
squared term reduces the colinearity between the two constructs (Cohen & Cohen, 1983).
This modeling technique is common in criminology and the social sciences (e.g. Jang,
1999; Raudenbush & Xiao-Feng, 2001).

In addition to the general model discussed above, a series of time-varying
covariates were also included in the Level I model. The variables are dichotomous and
represent the presence of a lifecourse event for person i at time j. Five time-varying
covariates were included at Level I in the final models presented below. Consistent with
the logit models presented above, dichotomous measures of incarceration, employment,
marriage, and absence from the workforce were estimated for subject i at time t. In
addition to the dichotomous measures, an additional variable was included in the model
to examine the cumulative nature of incarceration. The incarceration count measure was
also calculated for each subject i at time j, and represents the number of time periods the
individual was incarcerated prior to the current time period.

The inclusion of time-varying covariates is not commonly used in HLM models.
In traditional time-series models, only the polynomial of age or time is included in the
Level 1 model as an indicator of change with the remainder of covariates included in the
model at Level II. Integrating covariates at Level 1 allows researchers to model the time
varying effect of a predictor. For example, the addition of incarceration as a time-varying
covariate improves on the current incarceration research in that the nature of the

imprisonment experience is examined in light of both the presence and the timing of the

120

event. This modeling technique improves the statistical power of the model, in addition

to, the validity of the conclusions that can be drawn from the results.

Level H Model

The Level 11 model is designed to ascertain the nature of the change in trajectories
across the population (Raudenbush, 2001). In the Level II model, the slope and intercept
of the Level I variables are included as outcome variables and static factors as the
predictors (Johnson et al., 1997). The objective of the Level H model is to identify
predictors that are related to the extent of individual differences in both the coefficients
for the base rate and the grth curves. The factors included in the second level of the
model are time-invariant. The final Level 11 model is presented belowl '.

troi = Boo + 1301 (Hispanic) + [302 (African American) + [303 (male)

+ [304 (cognitive ability) + 1305 (high school graduate)

+ [306 (antisocial behavior) + [307 (previous incarceration)

+ [303 (family poverty) +rm (2)
1:“ = 1310 + B 11 (Hispanic) + 1312 (African American) + [313 (male)

+ [314 (cognitive ability) + 1315 (high school graduate)

+ [316 (antisocial behavior) + [3.7 (previous incarceration)

+ [313 (family poverty) +rn

 

” Research on propensity scores has suggested that it is most appropriate to include the propensity score at
Level II as the sole predictor of the outcome (e.g., Morgan, 2002). In the current analyses, a set of
individual-level variables was included it Level II in lieu of the propensity scores. This decision was made
to facilitate the discussion of the specific effects of individual variables on the marriage and work
outcomes. For example, this modeling decision allowed the author to describe the attainment of work in
light of antisocial behavior instead of a representation of propensity. To better examine the validity of the
current modeling technique, the final models were replicated using propensity scores. Results are
presented in Appendix G and H.

121

The inclusion of predictors at Level 11 allows the researcher to estimate the effect
of a variable on both the intercept of the line in addition to the slope. In equation 2, a
group of predictors were included to represent the effect of demographic influences,
human capital, criminal history, and family poverty on an outcome at the beginning of the
study and the change the dependent measure as a function of age.

In summary, HLM involves the estimation of models on two levels. Level 1
represents change in which all covariates are allowed to vary for each time or interview
period. The Level 11 model represents the ‘person’ model; hence, all predictors included
are time-invariant (e. g., race) and are incorporated to provide greater context to period-
specific change. Because of the multi-level nature of the estimation technique, a number
of preliminary models must be constructed before the final model can be estimated.
Consistent with traditional HLM methodology (Raudenbush & Bryk, 2002), three
preliminary models were estimated for each dependent measure. The following section

outlines the general modeling technique used for the preliminary and final models.

Model Estimation

In total, four separate models were estimated for each dependent variable
including three preliminary models and the final estimation. The first model constructed
was an unconditional base model. The unconditional model includes only the variables
included as estimates of individual change. An essential part of the estimation of the base
model involves the proper identification of the appropriate coefficients of change. The
nature of the growth curve can be determined using theoretical expectations for change in

conjunction with graphical analyses of the data. Two measures of growth were included

122

in the models presented below. In each model, an, represents linear growth over time. In
select models, an additional variable was included to capture the acceleration or the non-
linear change in the growth trajectory. For example, in the likelihood of marriage model,
the growth trajectory for the total sample increases dramatically in the first half of the
study period and then levels off and even declines. Because of the curvilinear nature of
the trajectory, the quadratic function of age (1211 was included as an estimate of the grth
model. The quadratic function of age was also included in the tenure model. The use of
the quadratic function to model trajectories of this nature is common in the
criminological literature (see Jang, 1999; Raudenbush & Chan, 1992).

The unconditional model also provides estimates of model fit that serve as the
baseline for future analyses. First, a reliability coefficient is estimated for the growth
parameter(s). The coefficient provides the researcher with an estimation of the ratio of
true parameter variability to model error. In addition, a test for homogeneity of the slope
and intercept is also conducted. The test involves the use of the x2 statistic and identifies
significant individual variation in the growth parameters. Together, the model fit
estimates can be used to judge the validity of the use of HLM for a specific statistical
model. Models with significant variation in the growth parameters and moderate to high
estimates of reliability are appropriate to examine using HLM.

The second model estimated is the random-coefficient model, and includes the
examination of the nature of the individual variation in the predictors over time. This
model tests if the relationship between the time-variant variables and the dependent
measures vary across the population of subjects. For example, is the relationship

between incarceration and tenure different for individual 1 at time 3 when compared to

123

individual 66 at time 4? If variance in the slopes is significantly different than zero, it is
assumed that the variation in the time-level slopes is a result of individual differences
(Kreft and De Leeuw, 1998, p. 43).

The third preliminary model estimated was the fixed-effect model. All
independent, period specific, covariates were considered “fixed” in this model that also
serves as the final Level 1 model. The individual-level variables were used as controls in
the final models, so there was little utility in modeling the variation in these variables
between individuals. In addition, meaningful variation was not found in the covariates;
thus, constraining the variation for these variables improved the efficiency of the model.
The age variables were not considered fixed in the preliminary or the final model. In
each of the model estimates, there was significant variation in the model fit coefficients;
hence, it was important to allow them to vary in all subsequent models.

The final model estimated merges the final Level 1 model with the demographic,
human capital, criminal history, and family poverty indicators as predictors of the slope
and intercept. From this model it was possible to ascertain the relative impact of
individual-level predictors on social bond attainment, net of the time-variant covariates.
Because the Level 11 model contained a number of variables consistent with the
population heterogeneity framework (e.g, cognitive ability), the estimation of the final
model allowed the researcher to understand the relative predictive power of variables
associated with the population heterogeneity framework, net of predictors associated with
the state dependence theory. For example, does cognitive ability have a significant effect
on the change in both the base rate and the likelihood of employment over time, even

after controlling for salient life events?

124

Li_kelihood of Attainment of Social Bonds to Marriage and Work

 

As found in the logit models, incarceration had a significant, negative effect on
social bond attainment. The following models expanded on the logit models by
examining the change in likelihood of marriage and fulltime employment over the
lifecourse. The importance of examining social bond attainment in a longitudinal manner
has been reinforced in the literature. Researchers have found that incarceration can have
a dramatic initial effect on employment, but that the impact of the event subsides over
time (Western, 2002). The opposite appears to be the case for marriage. Laub, Sampson,
and Nagin (1998) found that the deterrent effect of marriage was initially weak, but

gradually grew in strength over time.

Model Estimation

Consistent with traditional HLM methodology, three general statistical models
were estimated for each of the dependent variables. It is important to note, that the
following estimation techniques diverged from that of traditional HLM methodology. A
linear model was not appropriate for the test of this hypothesis due to the binary nature of
the dependent variable. Instead, the Bernoulli probability model was used to estimate the
likelihood of marriage and employment.

The first equation estimated was the unconditional model; however, before the
final base model can be estimated it was important to examine the nature of the grth

parameters. In the employment model, the pattern of growth approximated a linear

125

pattern; therefore, the centered age measure was included in the model as an estimate of

change'2 (see Figure 3).

 

.60

.10

.00

 

 

 

 

 

 

 

 

 

 

I

I

I

r j I I f I fl j 7

1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1996 1998 2000

 

Figure 3. Likelihood of Employment 1983-2000 (N=1,466)

The growth trajectory for the marriage dependent measure was not linear (see

Figure 4). Instead, the likelihood of marriage increased rapidly for the first six years of

the study period and then increased slowly. This finding is not surprising in that most

people enter into marriage in their 20’s and there are fewer people that enter into marital

relationships later in life. Because of the acceleration in the dependent measure, the

centered age variable was included as a measure of linear growth and the quadratic

firnction of this variable was used to estimate acceleration.

 

'2 In all four models, individual grth trajectories were also graphed to ascertain the true nature of change
over time. Each of the individual figures further validated the nature of change as suggested in the

aggregate graphs.

126

 

 

 

.45

 

 

 

 

 

 

 

1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1996 1998 2000

 

 

 

Figure 4. Likelihood of Marriage 1983-2000 (N=1,466)

Now that the general form of the model has been designated, it is important to
consider the reliability and variability of the coefficients. As shown in Table 23, the
reliability estimate of the sample mean for marriage and employment was very high. The
sample mean for marriage was >.79 and >82 for employment. Both dependent measures
are appropriate indicators of the sample mean. Finally, the variability of the coefficients
was estimated using the homogeneity of variance test. For both models, significant
differences were found in the average likelihood of marriage and employment. Because
of the va1iability in the initial statuses and growth rate of participants, the use of HLM

was an appropriate technique for addressing the hypothesis proposed.

127

Table 22. Variance Components for Random Effects - Likelihood of Marriage and

 

 

 

Work Models
Unconditional Model Conditional Model

Reliability Variance x2 Reliability Variance x2
Marriage
Intercept 0.79 6.80 7,924.13*** 0.75 5.80 7,103.84***
Age 0.33 0.33 2,130.87*** 0.31 0.31 l,997.26***
(Age)2 0.23 0.00 1627.14" 0.22 0.00 1,587.30*
Work
Intercept 0.82 2.61 7,498.33*** 0.66 1.19 4,633.63***
Age 0.43 0.02 2,826.07*** 0.25 0.01 1,937.27***

 

 

*p<0.05 **p<0.01 ***p<0.001 (two-tailed tests)

Next, two random-coefficient models were estimated (results not shown).
Significant variation was not found in any of the time-varying covariates included in both
models estimated; therefore, they were considered fixed in the final models. In addition,
a separate random coefficient model for the dichotomous incarceration measure was
estimated. Even in the reduced model, the effects of incarceration on marriage and work
did not vary significantly by individual. The incarceration event had the same effect for
an individual imprisoned at age 25 as it did for someone aged 35 and incarcerated in
2000. This is an important finding in that the researches within the lifecourse perspective
have argued that lifecourse events are embedded in social institutions and are subject to
historical change (e. g., Elder, 1985). Based on this assumption, one would expect that
the effect of incarceration, or a similar event like marriage, would change over time. This

was not the case with the current analyses and warrants further examination.

128

The third model estimated was the fixed-effects model. This model is similar to
the random-coefficient model, although in the fixed-effects models, occasion of
measurement-level slopes were constrained to be constant across individuals. In this
model, the lifecourse event variables were modeled on the intercept and slope across
individuals. The model fit predictors were allowed to vary in this and the final model.

As shown in Table 23, incarceration was a significant predictor of both the
likelihood of employment and marriage. Individuals who were incarcerated at least one
time during the study period were significantly less likely to attain a bond to marriage or
work. In fact, individuals who had been incarcerated were approximately 30% less likely
to be married and 34% less likely to be employed fillltime. The cumulative number of
incarceration events was also significantly associated with a reduced likelihood for
employment but had no effect on marriage.

Marriage was also positively related employment; likewise, work was a
significant predictor of marriage. Judging from the standardized coefficients, it appears
that effect of work on marriage is larger than that of marriage on work. Being married
increased the log odds of employment by 1.63, but employment only increased the log
odds of being married by 1.16.

The remainder of the lifecourse event predictors had disparate effects on the
attainment of bonds to work and marriage. In both models, military participation was
strongly related with the likelihood of social bond attainment; however, the effect was
positive for the marriage model and negative for the employment model. Individuals
who had served in the military were significantly more likely to be married but less likely

to be employed. In the work model, the effect of military participation far outweighed

129

the influence of incarceration on social bond development. Military participation reduced
the likelihood of employment by 78%; incarceration was associated with a subsequent
reduction of 34%. Absence from the workforce also had a strong negative effect on
employment. Individuals who had been absent from the workforce at any point were
approximately 80% less likely to become employed.

Consistent with the hypothesis presented, incarceration reduced the likelihood of
attaining bonds to marriage and work. In reference to marriage, lifecourse events that tie
individuals to conventional social institutions, like work and the military, increase the
likelihood of marriage. For the employment model, positive social bonds to marriage can
also improve the likelihood of social bond attainment (e.g., marriage); yet, events that
remove the individual from the traditional employment pool reduced the probability of
employment quite dramatically.

In addition, it appears that the Level I predictors, taken together, are better able to
explain variation in the likelihood of marriage than work. The Nagelkerke R21, for the
work model was 0.34; whereas, lifecourse events only explained 12% of the variation in
the likelihood to develop bonds to marriage. This finding is surprising in light of the
results fi'om the dichotomous logit models. The pseudo- R2 measures for the two social
bond development models were nearly identical at 0.26. Not only did the HLM
employment model explain more individual variation than the marriage model, it also
accounted for more variation than either of the logit models, using less than half the
number of variables. By using HLM to estimate the variability in lifecourse events over
time, the power of the model to explain variance has been enhanced, especially for the

dichotomous employment model.

130

Table 23. Fixed Effects of Occasion of Measurement Variables on Likelihood of Attainment of
Social Bonds (N =19,823)

 

 

 

 

 

 

Marriage Work
Independent Variable [3 SE Exp(B) [3 SE Exp(B)
Intercept -0.70*** 0.07 -0.15** 0.04
Lifecourse Events
Incarceration -0.36*** 0.08 0.70 041*" 0.09 0.66
Incarceration -0.07 0.04 0.93 -0.22*** 0.03 0.80
Count
Marriage -- -- 049* * * 0.05 1.63
Work 015*" 0.04 1.16 -- --
Military 037* 0.18 1.45 -2.09*** 0.19 0.12
Absence from -0.06 0.05 0.94 -1.60*** 0.05 0.20
Workforce
Model Fit
Age 012*" 0.02 1.23 0.10*** 0.00 0.90
Age Squared -0.00*** 0.00 1.00 -- --
Variance Explained 14%‘3 34%
at Level I
*p<0.05 **p<0.01 ***p<0.001 (two-tailed tests)

Results from the final model are presented in Table 24. The purpose of this
model was to test if the relationship between lifecourse events, marriage, and work were
maintained after individual-level controls had been added. Does incarceration remain a

significant predictor of employment and marriage even after controlling for criminal

 

'3 The proportion of variance explained for both the marriage and work dichotomous models was
developed using the Nagelkerke R2,”, The Cox & Snell coefficients for the work and marriage are 0.25 and
0.10 respectively. Because of the experimental nature of the pseudo-R2 coefficients, it is important to
interpret the variance estimates with a great deal of caution. The coeflicients should be used to provide a
general estimate of the variance explained and should not be used to make general statements regarding the
generalizability of the model.

131

history and demographic factors? A set of eight fixed-effect predictors at Level II was
included in the model. It was hypothesized that the predictors were related to both the
initial status and the growth rate.

There were substantial differences in the estimated models for the work and
employment dependent variables. In the work model, human capital, criminal history
and context influences had the largest effect on the initial likelihood of employment.
Male, high school graduates with high AFQT test scores had a significantly higher
probability of employment at the outset of the study; whereas, previous incarceration,
antisocial behavior, and family poverty reduced the base rate of employment. Gender
and education had particularly strong effects on the initial likelihood of employment.
Males with high school diplomas were nearly 1.5 times more likely to have been
employed at the onset of this study when compared with females and individuals who did
not complete high school.

Collectively, criminal history predictors reduced both the probability of
employment at the onset of the study and negatively affected the growth in the likelihood
of work over time. Previous incarceration had the largest effect on the initial status of
participants. Individuals who had experienced a stay of incarceration prior to the study
period were 40% less likely to be employed in 1983. Previous incarceration also had a
significant effect on the rate of growth over time, but the strength of the relationship was
very small. Participation in antisocial behavior also reduced the initial likelihood of
employment by 34%; but a similar relationship was not found between the variable and
the rate of change over time. The coefficient for the previous incarceration was opposite

of what would be expected. Individuals who were incarcerated prior to the study were

132

actually more likely to attain employment over time. Even though the effect was very
small, the finding is surprising in light of the current literature on this topic. Further
examination of this phenomenon is needed.

African American race and antisocial behavior were the only significant
predictors in the marriage model. African individuals had a 43% lower chance of being
married at the onset of the study; whereas, participation in antisocial behavior reduced the
likelihood of employment by 34%. In fact, antisocial behavior had a much stronger
effect on the initial likelihood of marriage than for work. None of the demographic,
human capital, criminal history or context influences were significant predictors of the
nature of change over time for the marriage model.

As a whole, the demographic, criminal history, and contextual factors explained a
moderate amount of variation in parameter variance at the initial status. In the marriage
model, a total of 17% of the parameter variance in the intercept was explained by the
variables, and 20% of the variance in initial status was explained in the work model.
Very little variance was explained by the variables for the estimates of the slope. The
variables, as a whole, explained about five percent of the variation in the slope for the
marriage indicator and less than one percent for the work model. It appears that the
static-individual level predictors did have some effect on the initial work and marriage
statuses of the sample, but the variables taken together explained little variation in the

rate of change.

133

Table 24. The Effects of Individual-Level Variables on Likelihood of Marriage and Employment
(N = 1,466)

 

 

 

 

 

Marriage Work
Independent Variable L SE Exp(B) B SE Exp(B)
Intercept -0.13 0.21 -0.80*** 0.12
Hispanic 006 0.18 0.94 -0.05 0.10 0.95
Afiican American -0.56** 0.17 0.57 0.08 0.09 1.08
Male 025 0.15 0.78 0.47*** 0.08 1.60
Cognitive Ability 0.00 0.00 1.00 0.01*** 0.00 1.01
High School Graduate 0.02 0.14 1.02 0.35*** 0.08 1.42
Antisocial Behavior -0.41** 0.13 0.66 -0.17* 0.07 0.84
Previous Incarceration 0.09 0.22 l .09 -0.52*** 0.12 0.59
Family Poverty -0.23 0.13 0.79 -0.25** 0.07 0.78
Slope 0.09" 0.03 0.12*** 0.01
Hispanic 0.00 0.02 1.00 -0.01 0.01 0.99
Afiican American -0.01 0.02 0.99 -0.01 0.01 0.99
Male 0.02 0.02 1 .02 -0.01 0.01 0.99
Cognitive Ability 0.00 0.00 1.00 0.00 0.00 1.00
High School Graduate 0.02 0.02 1-02 -0.01 0.01 0.99
Antisocial Behavior -0.01 0.02 0.99 -0.00 0.01 1.00
Previous Incarceration -0.02 0.02 0.98 003* 0.02 1.03
Family Poverty -0.02 0.02 0.98 0.01 0.01 1.01
Variance Explained at
Level 11 - Initial Status”
17% 20%
Variance Explained at
Level II - Growth Rate
5% <1 %

 

 

*p<0.05 **p<0.01 ***p<0.001 (two-tailed tests)

 

'4 The proportion of variance explained is the difference between the total parameter variance obtained
from the unconditional model and the residual parameter variance from the fitted model divided by the total
parameter variance (Raundenbush & Bryk, 2002:168).

134

Finally, it is important to consider the results from the Level 11 models in light of
the Level I predictors that are also included in the model. The coefficients for the Level 1
model (see Table 25) remained relatively unchanged as a result of the inclusion of the
Level II predictors. Even after including a host of individual factors in the model, the
time-varying covariate predictors remained significant predictors of the likelihood of
employment and marriage. In the same light, the relationship between early antisocial
behavior, incarceration, and the social bonds were maintained even after controlling for
lifecourse events at the event level. It is also important to note that, in the employment
model, the Level II predictors accounted for substantially less variation when compared
with that of Level I factors. In the marriage model, Level II predictors explained nearly
the same amount of variation as the Level I variables.

The results from this analysis confirm the theoretical propositions presented by
lifecourse researchers. Negative life events in early adolescence, like drug use and crime,
can initially close doors to firture opportunities like employment and marriage. However,
judging from the explained variance for the estimates of the slope, individual traits have
little effect on the subsequent change in the likelihood of bond development over time.
In contrast, the lifecourse event variables, especially in the employment model, explained
a considerable proportion of the total model variation. Static predictors aided in
determining the initial status of an individual, but life events were more efficacious in

explaining variation over time.

135

Table 25. Fixed Effects of Occasion of Measurement Variables on Likelihood of Attainment of
Social Bonds — Level II (N =19,823)

 

 

 

 

Marriage Work
Independent Variable B SE Exp(B) B SE Exp(B)
Lifecourse Events
Incarceration -0.34** * 0.08 0.71 -0.43* ** 0.09 0.65
Incarceration -0.04 0.04 0.96 -0.22*** 0.03 0.80
Count
Marriage -- -- 0.48** * 0.05 1.62
Work 015*" 0.04 1.16 -- ~-
Military 041* 0.18 1.51 -2.28*** 0.18 0.10
Absence from -0.07 0.05 0.93 -1.66*** 0.06 0.19
Workforce
Model Fit
Age 009" 0.03 1.09 0.12*** 0.01 l. 13
Ag; Squared -0.00*** 0.00 1.00 -- --
*p<0.05 **p<0.01 ***p<0.001 (two-tailed tests)

Nature of Employment
The purpose of the following analyses was to examine the relative effect of
incarceration on the nature and stability of employment. Consistent with social bond
theory, it is important to understand the strength and nature of social bonds. The research
presented to this point has only focused on the attainment of bonds; however, researchers
have suggested that it is not the bond per se that mitigates involvement in negative
behaviors (e.g., delinquency, drug use). It is the capital and the informal social control

that is derived from a strong and involved bond that increases the likelihood of both the

136

initiation of a criminal career and desistance from criminality and other problem
behaviors (Coleman, 1988, 1990; Sampson & Laub, 1993).

The statistical models estimated for the following analyses varied from those used
in the binomial change models. Both dependent variables, tenure with primary employer,
and number of jobs worked represent counts of an event; hence, the binomial and
traditional OLS algorithms used in HLM are inappropriate. In conventional HLM
methodology Ordinary Least Squares regression is used; however, OLS assumes that
random errors are independent, normally distributed, and have constant variance (Bryk &
Raudenbush, 1992:15). In both sets of models, the assumption was violated. The
distribution for both tenure with employer and number of j obs worked were highly
skewed. Significant overdispersion of the dependent variables violated the OLS
assumptions. An overdispersed Poission model was used instead of the default OLS
model in this analysis to compensate for the distribution of the dependent variables.
Overdispersed Poisson models calculate a factor to correct the inferential statistics;
therefore, statistical reliability is not compromised (Gardner, Mulvey, & Shaw, 1995).

The first task in the estimation of an HLM equation is the delineation of the most
appropriate growth model. The aggregate grth models for the tenure and jobs worked
model are presented in Figures 5 and 6. As shown, the growth rate for the tenure model
was not linear (see Figure 5). The growth rate increased quickly for the first four to five
time periods, then lags, and increases again during the last fourth of the time period. The
non-linear grth curve is firrther confirmed when individual grth trajectories are
examined (not shown). To account for the non-linear grth in the dependent measure,

two variables were included in the model as estimates of change. The centered age

137

variable was included to capture linear growth; whereas, the quadratic form of age was

used to model non-linear acceleration.

 

 

25000

/'

2001K) l/lr”"‘/’

 

 

3
9
8

 

 

Tenure in Weeks
'5
F’
8

 

501K110—

 

 

 

-00 I T 1 r I I I I I I T I I I

1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 I994 1996 1998 2000

 

 

 

Figure 5. Tenure with Primary Employer 1983-2000 (N=1,466)

As presented in Figure 6, the rate of growth for number of jobs worked was
linear. Because of the nature of the trajectory, only the age variable was included in the

model as an estimate of growth.

138

 

 

 

 

Number of Cummulative Jobs

 

 

 

1983 1984 1985 1986 1987 1988 1989 1990 I991 1992 1993 I994 I996 1998 2000

 

 

 

Figure 6. Number of Jobs Worked 1983-2000 (N=1,466)

Now that appropriate growth models have been determined, it is important to examine the
reliability and variation in the coefficients for the base, unconditional model. As
evidenced in Table 26, there was significant and reliable mean variation in both the
estimates of the slope and intercept. The reliability estimates of the intercept and slope
predictors were high with 0.99 for the intercept, 0.85 for the age predictor, and 0.79 for
the quadratic measure. For the number of j obs worked, the reliability estimate for the
intercept was 0.96 and 0.41 for age. The homogeneity test was also significant for both
the slope and the growth rate. Based on these statistics, it is possible to conclude that
individuals vary significantly in their tenure with employer and the number of j obs
worked at the beginning of this study. In addition, there was also sufficient variation in
the mean grth rates among sample members. Judging from the estimates presented in

Table 25, it was valid to proceed with the analyses using HLM.

139

Table 26. Variance Components for Random Effects for Nature of Employment Models

 

 

 

Unconditional Model Conditional Model
Reliability Variance x2 Reliability Variance x7-
Tenure
Intercept 0.99 1.67 1,119,617*** 0.99 1.49 958,861“**
Age 0.85 0.03 20,227.53*** 0.85 0.03 20,165.1***
(Age)2 0.79 0.00 12,423.45*** 0.79 0.00 12,566.9***
Jobs
Intercept 0.96 0.37 54,542.46*** 0.96 0.33 48,269.6***
Age 0.41 0.00 2,681.31*** 0.40 0.00 2,572.39***

 

 

*p<0.05 **p<0.01 ***p<0.001 (two-tailed tests)

Consistent with findings fi'om the dichotomous growth curve models, all of the
time-varying covariates were considered fixed in each of the following tenure and jobs
worked models. In the random-coefficient analyses (not shown), significant variation in
the time-varying covariates was not found. The effect of incarceration did not vary by
individual. The incarceration event had the same effect on the nature of employment for
individual incarcerated in 1983 or in 2000. The model fit estimates were allowed to vary
in all of the models.

The results of the fixed effect models are presented in Table 27. Although the
tenure and jobs worked variables were designed to reflect the stability of employment by
the respondents, the predictive models vary substantially. The findings obtained from the
tenure model are very similar to that of the logit and HLM employment models.
Confirming the hypothesis presented, both the dichotomous and additive incarceration
measures of incarceration were significant, negative predictors of tenure. The strength of

the relationship between incarceration and tenure was moderately low. In fact, the

140

dichotomous incarceration measure played a larger role in the likelihood of work model
than in the tenure model. The standardized coefficient for incarceration was 0.65 in the
dichotomous employment model and 0.90 in the tenure model.

Marriage and participation in the military were both significant predictors for the
tenure model. Consistent with the results from the likelihood of employment models,
married respondents were more likely to maintain long-term relationships with
employers, and enlistment in the military further reduced the length of tenure with
employer. The relative size of the standard coefficients for the marriage and military
predictors was relatively small for the tenure model, in comparison to the likelihood of
employment model. Contrary to the findings fi'om the dichotomous employment model,
absence from the workforce was not significantly associated with employee tenure.

Participation in the military and absence from the workforce were the only
significant predictors of the number of jobs worked. Individuals who had been absent
from the workforce or participated in the military reported a smaller number of unique
employment experiences. Although the explanatory power of both variables was small,
military participation was the strongest predictor in the jobs worked model. Participation
in the military reduced the number of jobs worked by l 1%.

Interestingly, both the dichotomous and additive incarceration predictors were not
significantly associated with the number of jobs worked. It appears that while
incarceration reduced the likelihood of employment and the opportunity to build long-
term bonds with an employer, incarceration did not impact the number of unique job

experiences.

141

Table 27. Fixed Effects of Occasion of Measurement Variables on Nature of Work Dependent

Variables (N =1 ,466)

 

 

Tenure with Employer Number of Jobs Worked
Independent Variable B SE Exp(B) B SE Exp(B )
Intercept 0.06*** 0.03 2.07 *** 0.02
Lifecourse Events
Incarceration -0.10** * 0.01 0.90 0.02 0.01 1.02
Incarceration -0.16*** 0.00 0.85 -0.00 0.00 1.00
Count
Marriage 0.03*** 0.00 1.03 0.01 0.01 1.01
Military 01 l*** 0.02 0.90 -0.23*** 0.03 0.79
Absence from -0.01 0.00 0.99 -0.07*** 0.01 0.93
Workforce
Model Fit
Age 031*" 0.00 1.36 0.05*** 0.00 1.05
Age Squared -0.01*** 0.00 0.99 -- --

 

 

***p<0.001 (two-tailed tests)

Results fiom the two full hierarchical models are presented in Table 28. The goal

of these analyses was to test the effects of demographic, human capital, criminal history,

and contextual influences on both the base rate of tenure with employer and the number

of jobs worked during the study period and the change in the nature of the bonds over

time.

The Level H models estimated for the jobs worked and tenure with employer

dependent variables were very different. Human capital and criminal history predictors

were the most significant predictors of both the intercept and the rate of growth for the

tenure model. Males and high school graduates and individuals with a high AFQT score

142

had a higher initial tenure measure; whereas, family poverty and antisocial behavior
negatively affected the initial starting point for the tenure measure. Although all but two
of the variables were significant predictors of the intercept, the predictors accounted for
very little variance as a whole. Together, the variables explained only 11% of the
variation in the slope.

In addition, females and Afi'ican American individuals and respondents with a
high AFQT score were most likely to report a positive growth in tenure over time;
however, the predictive power of the variables was very low. The standardized
coefficients for these variables range from 0.99 to 1.03. Together, the variables
explained less than one percent of the variation in the slope.

Previous incarceration was also found to be a significant predictor of both the
intercept and the slope in the tenure model. Individuals who had experienced
incarceration prior to the study period reported less established relationships with their
employer at the onset of this study. In fact, previous incarceration reduces initial tenure
with employer by 33%. Surprisingly, incarceration at an early age increases the
likelihood of positive growth in tenure with employer. Although the relationship is quite
weak, this finding in contrary to the findings from established research studies and the
current work. Future examination of this relationship is needed.

The original hypotheses were not confirmed for the number of jobs worked
model. It was expected that individuals who had experienced a stay of incarceration
would be more likely to have transient work experiences and move from job to job. This
was not the case. Neither the additive of the dichotomous measures of incarceration were

significantly associated with number of jobs worked. Instead, demographic influences

143

played the largest role in this model. Minority females reported the lowest number of
jobs worked at the beginning of the study. In contrast, respondents who indicated
participation in antisocial behaviors reported a high number of unique employment
experiences at the onset of the study. The predictive power of these variables was very
low, only nine percent of the variation in the slope coefficients could be accounted for
with these variables.

In reference to the growth in employment experiences over time, Caucasian
respondents with a high school diploma and elevated AFQT scores were significantly less
likely to experience a substantial increase in the number. Similar to the tenure model,
inclusion of the predictors of the slope added little to the study. The individual predictors

explained only six percent of the variation in the slope for the jobs worked model.

144

Table 28. The Effects of Individual-Level Variables on Nature of Work 01 = 1,466)

 

 

 

Tenure with Employer Number of Jobs Worked
Independent Variable B SE Exp(B) B SE Exp(B)
Intercept 5.88*** 0.1 l 1.89*** 0.05
Hispanic 0.07 0.09 l .07 -0.09* 0.04 0.91
Afiican American 0.06 0.09 1.06 -0.15** 0.04 0.86
Male 031*“ 0.08 1.36 0.23*** 0.04 1.26
Cognitive Ability 0.01** 0.00 1.02 -0.00 0.00 1.00
High School 029*" 0.07 1.34 -0.00 0.04 1.00
Graduate
Antisocial Behavior -0. 16* 0.07 0.85 020*" 0.03 1 .22
Previous 040“ 0. 12 0.67 0.07 0.05 1.07
Incarceration
Family Poverty -0.32*** 0.07 0.73 0.01 0.03 1.02
Slope 0.30*** 0.01 0.06*** 0.00
Hispanic 0.01 0.01 1.02 0.00 0.00 1.00
Afiican American 0.02** 0.01 1.02 0.01** 0.00 1.02
Male 001* 0.01 0.99 -0.00 0.00 1.00
Cognitive Ability 0.00** 0.00 1.00 -0.00* 0.00 1.00
High School 0.01 0.01 1.02 -0.00** 0.00 1.00
Graduate
Antisocial Behavior -0.00 0.00 1 .00 -0.00 0.00 1.00
Previous 003*" 0.01 1 .03 -0.00 0.00 1.00
Incarceration
Family Poverty -0.01 0.01 0.99 0.00 0.00 1.00
Variance Explained at
Level II — Initial Status 1 1% 9%
Variance Explained at
Level II — Growth Rate <1 % 6%

 

*p<0.05 **p<0.01

***p<0_00] (two-tailed tests)

145

Like the previous growth curve models presented, inclusion of the Level II

predictors in the model did little to change the Level I coefficients (see Table 29). Even

after controlling for demographic influences, absence from the workforce and military

participation remained significant predictors of number of jobs worked. Incarceration,

marriage, and military participation endured as significant predictors of tenure with

employer.

Table 29. Fixed Effects of Occasion of Measurement Variables on Nature of Work Dependent
Variables — Level II (N =1,4@

 

 

 

Tenure with Employer Number of Jobs Worked
Independent Variable B SE Exp(B) B SE Exp(B )
Lifecourse Events
Incarceration -0.10*** 0.01 0.90 0.01 0.01 1.01
Incarceration -0.16*** 0.01 0.85 -0.01 0.00 0.99
Count
Marriage 0.03*** 0.00 1.03 0.02* 0.01 1.02
Military 011*” 0.01 0.90 -0.25*** 0.03 0.78
Absence from 0.01 0.00 1.01 -0.07*** 0.01 0.93
Workforce
Model Fit
Age 030*“ 0.01 1.35 0.06*** 0.00 1.06
Age Squared -0.01*** 0.00 0.99 -- --

 

“*p<0.001 (two-tailed teStS)

146

Summary

In Chapter 3, three specific research questions were outlined. The relevant results
for each question posed are presented below. First, the relationship between
incarceration and social bond development was considered. Specifically, were
incarcerated sample members able to develop social bonds to work and marriage and how
did the development of bonds among the incarcerated group vary from that of a similarly
situated group that did not experience incarceration? Apart from the number of jobs
worked model, incarceration was a significant, negative predictor of both the likelihood
of social bond attainment and the nature of the bond developed. The strength of the
relationship between incarceration and social bond outcomes was found to vary by
model. Judging from the standardized coefficients, it appears that incarceration had a
stronger effect on the likelihood of work and marriage than in the nature of the bond
models. Incarceration reduced the likelihood of employment and marriage by nearly one
third in both the logit and dichotomous grth curve models, but the relationship was
very weak for the nature of marriage and tenure with employer models. The current
research highlights the negative effect that incarceration can have on opportunities to
develop bonds. That said, once an individual has attained the bond, the stability or nature
of the bond is not reduced substantially as a result of the incarceration event.

The relationship between previous incarceration and social bond attainment was
not as strong when compared to the effect of adult imprisonment. In fact, the relationship
between incarceration that occurred before the study period and marriage was not
significantly significant, in either the dichotomous or nature of bond models. Previous

incarceration did have a moderate effect on employment outcomes, especially in the

147

growth curve models. Individuals who had been incarcerated prior to the study period
were nearly 40% less likely to report bonds to work at the onset of the research and
tenure with employer was reduced by nearly a third at the initiation of the research as a
result of this event. Prior incarceration was also significantly related to the growth in
both the likelihood of attainment and the stability of employment overtime; however, the
relationship between the variables was contrary to what was hypothesized. Individuals
who had been incarcerated prior to the study period had significantly more growth in the
likelihood and stability of employment over time, although the relationship was quite
weak. It may be that grth in the parameters over time serves to compensate for the
disadvantage posed by previous incarceration on the initial employment status of the
individual, but this theory cannot be validated with the available data.

The findings from the present analyses augment the current incarceration
literature. Consistent with prior research, incarceration at an early age does appear to
close doors to adult employment opportunities, at least initially (Fagan & Freeman, 1999;
Grogger, 1995). However, the effect of previous incarceration does not have a strong
effect on change in the nature and likelihood of bonds over time. The effect of previous
incarceration on the slope for the tenure and dichotomous work models was very small
and in the opposite direction than was hypothesized. Incarceration during the study
period did have strong enduring effects on the change in the likelihood and the nature of
bonds to work over time. Individuals who were incarcerated during the study period
were less likely to attain bonds to marriage and work and the quality of the social bonds

developed were also diminished.

148

The second goal of the research was to understand the relative effect of
incarceration on social bond development in relationship to both lifecourse event
variables and demographic controls. For example, can incarceration outweigh the effect
of marriage or educational attainment? In the logit and growth curve models,
incarceration remained a significant, negative predictor of social bonds to marriage and
work even after controlling for a host of individual, static predictors (e.g., race).
Antisocial behavior was the only individual demographic predictor that was consistently
related to both the attainment and quality of social bonds. Individuals who had reported
delinquency or drug use prior to the study were significantly less likely to attain social
bonds to work or marriage and the quality of the social bonds developed were reduced as
result of the behavior. Afiican American race was the only predictor consistently related
to marital outcomes. Afiican American respondents were significantly less likely to be
married, but once the bond had been achieved, they were more likely to report a positive
growth in the bond over time. Not surprisingly, deficits in human capital (e.g., cognitive
ability, high school education) identified prior to the study were also associated with a
lower likelihood and stability of work. The findings from the analyses highlight the
importance of certain individual traits, like cognitive ability, in understanding the
attainment of social bonds. Although, the significant, negative effect of static factors on
social bond outcomes were not permanent. Individual, time invariant factors were only
found to be significant predictors of the initial status and were not associated with change
over time.

Together, the lifecourse event variables were consistently related to both social

bond development and stability. Except for the number of j obs worked model,

149

employment and marriage, as exogenous predictors, were significant, positive predictors
of both the attainment and nature of marriage and work. Individuals who were employed
were more likely to develop bonds to marriage; conversely, respondents who reported
being married had an increased likelihood of subsequent employment. Even though the
relationships were relatively weak, the findings highlight the importance of considering
multiple lifecourse event indicators when studying social bond outcomes. Even in the
analyses that involved only the incarcerated group, employment and marriage remained
significant, further highlighting the impact that life events can have on social bond
development.

Military participation also played a substantial role in the models, although the
effect of the predictor varied by dependent measure. Military participation increased the
likelihood of obtaining and achieving a stable marriage, but the opposite relationship was
found for the employment models. Absence fiom the workforce was also a significant,
negative predictor in all of the employment models. The military represents an important
social institution; hence, it would be expected that participation would increase social
capital and the likelihood of obtaining social bonds. The hypothesized relationship holds
true for marriage, but it appears that any event or institution (e. g. absence from work) that
removes the individual from the traditional workforce will affect employment. Taken as
a whole, the findings speak to the importance of social events in understanding both the
likelihood of obtaining bonds and the nature of change in social relationships over time.

Finally, the research was designed to consider if the timing of the incarceration
event differentially affected the likelihood of developing bonds to work and marriage,

and if the nature of the bonds were diminished over time as a result of the event.

150

Surprisingly, the timing of the incarceration event was not significantly related to the
nature or development of social bonds in any of the models estimated. In the
dichotomous logit models, a predictor of age at first incarceration was included in a sub-
group analysis of the incarcerated sample, but this predictor was not significantly related
to attainment of marriage or fulltime employment in 2000. In the HLM models, a
preliminary, random-coefficient model was estimated. This model allowed the researcher
to test if the effect of incarceration on marriage and work varied by person. In the
dichotomous and nature of work models, the effect on incarceration did not vary by
person. For example, the effect of incarceration on the likelihood of marriage for a
subject who was incarcerated in 1983 is not different from someone who was
incarcerated in 1998. Because the timing of the event is an important topic for
consideration under the lifecourse framework, filrther research on this topic is warranted.
The research was also developed to test if the likelihood of attainment and nature
social bonds was further reduced for individuals who had experienced multiple stays of
incarceration. Multiple terms of incarceration did reduce the likelihood of marriage in
2000, but a similar relationship was not found in the employment model. The effect of
multiple terms of incarceration further reduced the likelihood of obtaining fulltime
employment, but had no effect on the probability of marriage. It does appear that
multiple terms of incarceration can reduce the likelihood and nature of social bonds to

marriage in some manner, but the relative effect varies across models.

151

Chapter 5: Conclusions

The difficulty in integrating individuals back into society following a stay of
incarceration has challenged the criminal justice system since the inception of penal
confinement. The issue of prison reentry has once again entered the political arena due to
changes in the nature of incarceration. Never before have more people been incarcerated
(Beck et al., 2002). On average, terms of incarceration are also increasing. In 1990
inmates served on average 28 months before being released; whereas, inmates released
from prison in 1999 served a total of 34 months prior to release (Hughes, Wilson, &
Beck, 2001). More importantly, the current criminal justice system has few resources to
address the needs of this population. Most offenders are not avowed of services while
imprisoned (Lynch & Sabol, 2001). Once released, offenders often get little supervision
or guidance. Parole caseloads have almost doubled in the last three decades. An average
caseload in 1970 included 45 parolees; today, most caseloads have risen to about 70
(Petersilia, 2000). Large caseloads do not facilitate effective supervision of parolees; nor,
is the parole officer able to effectively link the offender with needed services. Under the
current conditions, prison remains a revolving door.

The costs of imprisonment and subsequent reentry are very high. State prison
expenditures have risen dramatically. Spending on corrections increased nearly five fold
in 15 years. In 1982, total prison expenditures totaled $9 billion in 1982 and rose to $44
billion in 1997 (81 S, 2001). Economic costs are not the only consequence of rising
incarceration rates; imprisonment also exacts costs on the community. For example,

reentry has collateral consequences on the family. More than half of all men incarcerated

152

have at least one child; approximately eighty percent of incarcerated women are mothers
of young children with the majority being under the age of 10 (Snell, 1994).

In light of the dramatic rise in incarceration over the past twenty years and the
collateral consequences that this event can have on the community, there has been a
heightened demand for programming that promotes successful reintegration (Petersilia,
2001). Reducing the number of individuals returned back to prison each year could
alleviate some of the economic burden placed on state and local governments to house
such a burgeoning population. In addition, developing programming that is designed to
facilitate long-term reintegration could have far-reaching benefits for prisoners, their
families and communities (Travis, Solomon, & Waul, 2001).

Despite the possible social and economic advantages of promoting successful
reentry, most of the research conducted to date on prisoner reentry has centered on the
study of recidivism (e. g. Gendreau et al., 1996). Although important, this research does
not consider factors associated with resiliency following incarceration and eventual
desistance from crime. The current study was designed to address some of the limitations
of current research. Using a large, longitudinal dataset allowed the researcher to examine
how incarceration alters life trajectories. Specifically, the research was focused on the
effect that incarceration has on the attainment and the nature of social bonds to work and
marriage, in light of other lifecourse events (e.g. military participation) and static,

demographic influences.

153

Results

As discussed in Chapter 2, three central theoretical perspectives have been
developed to explain changes in involvement in criminality and other behaviors over the
lifecourse. The population heterogeneity framework rests on the assumption that latent
traits, developed in early childhood, determine involvement in criminality over the
lifecourse (e.g. M. R. Gottfiedson & Hirchi, 1986; Wilson & Hermstein, 1985).
According to research conducted within this perspective, individual characteristics (e. g.,
antisocial personality, I.Q.), solidified by age four or five, determine the life chances of
an individual. Because individual traits become immutable after early childhood, there is
very little that can be done to divert lifecourse trajectories, especially in adulthood.
Conversely, state dependence researchers assert that behaviors arise out of direct
interaction with the environment and are not determined in any way by personal
disposition (e.g. Blumstein et al., 1986; Hirschi, 1972). Individual traits do not play a
role in the state dependence framework. Instead, life events (e. g., marriage,
imprisonment) determine social outcomes.

The lifecourse perspective blends the state dependence and population
heterogeneity perspectives and posits that criminality arises from both continuity and
change (Sampson & Laub, 1993). Individual, time-invariant traits (e.g., cognitive ability)
can limit avenues for future social bond attainment, but lifecourse events in adulthood
serve to determine an individual’s ultimate trajectory. For example, individuals who
exhibit antisocial behaviors in early childhood may be more likely to be removed from
traditional educational settings, reducing the likelihood that the individual will develop

positive bonds to school. Reduced social bonds to school have been associated with

154

subsequent involvement in criminality and incarceration in adulthood (Arum & Beattie,
1999; Sampson & Laub, 1993). According to researchers in this perspective, antisocial
behaviors, and related behavioral traits, do not determine life chances. Instead, individual
traits can close doors to positive bond development, increasing the likelihood of
involvement in deviance or criminality.

The lifecourse perspective was used in the current study to explore the effect of
incarceration on social bond development. Specifically, it was hypothesized that
individuals who experienced one or more stays of incarceration would be less likely to
attain bonds to marriage and work and that the nature of these bonds would be affected as
a result of the imprisonment experience. Using longitudinal data and advanced statistical
techniques allowed the researcher to compare the effect of incarceration on social bonds,
relative to other lifecourse events and net of time-invariant individual traits.

The results from this research confirm the theoretical propositions presented by
lifecourse theorists, and consequently also substantiate the hypotheses outlined at the
initiation of the research study. In each of the dichotomous social bond models
estimated, individuals who had experienced incarceration during the study period had a
lower likelihood of marriage and work. This finding confirms the original hypothesis
that stated that individuals who were incarcerated during the study period would be less
likely to attain social bonds to work and marriage. In two of the three models estimated,
incarceration was also negatively associated with the nature of the bond attained.
Individuals who had been incarcerated during the study period were significantly more
likely to report improved bonds to marriage and lengthy tenure with employer. A

significant relationship was not found between number of jobs worked and incarceration.

155

A secondary goal of this study was to test the relative effect of incarceration in
reference to other lifecourse events. Each of the lifecourse predictors was significantly
related to social bond outcomes in at least one model estimated. One of the most striking
findings from this study was the significant, strong effect that participation in the military
had on marriage and work outcomes. Military participation was positively associated
with both the likelihood of development and the nature of marital bonds, but was coupled
with a decrease in the probability of employment and reduced tenure with employer. The
negative relationship between military participation is especially striking in that previous
researchers have observed that participation in the military is one of the few social
institutions that can spark social change (Caspi & Moffltt, 1993; Sampson & Laub,
1996)

Work and marriage were also significantly and positively related to social bond
development in all models, except the jobs worked analysis. Individuals who were
employed were more likely to develop bonds to marriage; conversely, respondents who
reported being married had an increased likelihood of subsequent employment. Finally,
individuals who reported being absent from work at one time were less likely to have
positive employment outcomes. The findings from the current analysis highlight the
importance that development of multiple social relationships can have on subsequent
social bond achievement. Although interaction analyses were not conducted as part of
the current study, it does appear that social bonds can work in concert to increase the
likelihood of positive social outcomes. It may be that after individuals have been able to
develop one positive social bond, even after controlling for negative life events, would be

better able to present a public image of respectability (Giordano et al., 2003); thereby

156

increasing the probability of developing positive social relationships. Negative lifecourse
events are important, but the relative effect of the incarceration event may not outweigh
that of positive social bonds.

In addition to the examination of the relative effects of lifecourse events on social
bond attainment, a number of individual-level variables were included in the models as
controls. Consistent with the lifecourse framework, individual, demographic
characteristics were related to the life circumstances of the participants at the beginning
of the study, but the same variables added little to the explanation of change over time.
For example, individuals who were incarcerated prior to the study period were less likely
to be employed at the onset of the study, but previous incarceration had little effect on the
change in the likelihood of work over time. There was also little consistency in the
predictors significantly associated with social bond development. Antisocial behavior
was the only significant predictor of social bond development across models. The
variance estimates obtained for growth curve models highlight the diminished power that
static, individual factors can have in explaining model variation, especially that for
growth over time. Together, the static individual predictors account for less than one-
fifth of all variance in the initial status of respondents and less than six percent of the
total variation in estimates of the slope. Individual, static influences can affect social
bond outcomes, but the impact is early and does not endure over time.

Overall, the results fiom this analysis confirm the theoretical propositions
presented by lifecourse researchers. Negative life events in early adolescence, like drug
use and crime, can initially close doors to future opportunities. However, judging from

the explained variance for the estimates of the slope, individual traits have little effect on

157

the subsequent change in the likelihood of bond development over time. In contrast, the
lifecourse event variables, especially in the employment model, explained a considerable
proportion of the total model variation. Static predictors aid in determining the initial
status of an individual, but life events are more efficacious in explaining variation over
time.

In summary, incarceration in adulthood can have significant, negative effects on
the chances of realizing employment and marriage, as well as, the nature of the bond
attained. The relative effect of predictors associated with the population heterogeneity
model (e.g., cognitive ability) on social bonds is less clear. The results from the current
research suggest that individual traits can shut doors to social bond attainment in early
adolescence, but do little to explain meaningful individual change over time. Although
the character of the relationship needs to be considered further, the results of the analyses
also point to the importance of positive life events as predictors of social bond
development. It appears from the current research that involvement in social institutions
(e.g., military participation, marriage or work) can ameliorate some of the negative

effects of incarceration on subsequent bond development.

Limitations

Before research implications can be drawn fiom the current study, it is important
to note any limitations of the work. The most substantial limitations of the current study
result because of measurement bias. The NLSY was designed to track changes in labor
patterns and was intended largely for econometric purposes; therefore, detailed data on

incarceration, social bond development, and child psychopathology were not collected.

158

Additional limitations of the study also arise due to the scope of the research and the
nature of the sample. The following section outlines the limitations of the current study
and provides insight into possible improvements that could be addressed by future

research.

Measurement of Dependent Variables

The operationalization of the nature of social bonds constructs was not optimal.
For example, the marriage factor only included indicators of the timing of the marriage
and the number and length of first marriage. Unlike the construct developed by Sampson
and Laub, (1993) data were not available on the attitudinal assessment of the relationship.
For example, subjects were not queried as to the frequency of fighting, or the quality of
communication, or the characteristics of the conjugal relationship. The nature of
employment measures also fail to capture the qualitative aspect of the social bond.
Tenure with employer and number of j obs worked speak to the stability of employment,
but the attachment to the job or employer was not addressed with this construct. In light
of social bond theory, it is important for future researchers to capture data that provide
adequate representations of the attachment, involvement, and belief that individuals have
for social institutions. This problem can be resolved by including both quantitative and
qualitative elements in future studies, similar to the methodologies employed by Sampson
and Laub (1992).

The scope of the social bond model is also limited. First, only two indicators of
social bonds were included. Although marriage and work have been the most studied of

all social bonds indicators, researchers have also linked participation in religious

159

institutions (Coleman, 1988, 1990) and strong peer networks (Warr, 1998) to desistance.
Including multiple measures of social bonds will refine social bond theory and can
facilitate broader generalizations fi'om the research. As such, future researchers should

consider incarceration in reference to a number of different social bonds.

Measurement of Independent Variables

The precision of the incarceration measure was also limited. In the current
research, incarceration was considered in relationship to the respondent’s housing
situation at the time of the interview. If an individual, when contacted for the interview,
was living in a correctional facility, he or she was coded as imprisoned for that year or
interview period. Persons that may have been incarcerated at another point during the
year were not identified as incarcerated. Conversely, subjects that were released flour a
correctional facility shortly following the interview were described as imprisoned.
Because of the nature of the data collection protocol, the imprisonment measure largely
reflects individuals who had been incarcerated for a longer period of time; the
experiences of subjects who had experienced a short stay of incarceration (e. g., jail) may
not have been fully captured by this data.

From this research it is possible to assert that longer stays of incarceration have a
strong negative effect on social bond attainment; however, caution should be exercised
when making generalizations from this study to shorter stays of imprisonment. Future
data collection efforts should be designed to capture data on both the nature of the
incarceration event (e.g., jail V prison) and the length of the stay (e.g., number of days

served). Without this information, it is impossible to understand the nature or the dosage

160

of the incarceration event. Inclusion of these variables in future research would enhance
the precision of the studies.

Second, this study lacked adequate statistical controls for involvement in
delinquency and with the criminal justice system. In specific, the previous incarceration
measure was not optimal. This construct was developed as a proxy for juvenile
incarceration; however, because of the cohort nature of the data collection design, the
meaning of the measure varies by cohort. For example, the previous incarceration
measure represents the presence of a stay of incarceration prior to age 15 for the youngest
cohort but indicates imprisonment prior to age 23 for the oldest cohort. The lack of
consistency and precision in measurement of this variable severely limits the
generalizations that can be made on the impact of previous incarceration on social bond
development. In addition, information was not available in the dataset on the type of
conviction that lead to the incarceration event. The goal of this research was to examine
the effect of incarceration as a lifecourse event on social bond development; hence, the
nature of the conviction has little theoretical significance to the study at hand. That said,
researchers have found that incarceration for certain classes of crimes (e. g., white collar
crimes) had disparate effects on employment opportunities (Kling, 1999); therefore,
researchers should strive to collect detailed criminal history information in future
research studies.

In the same light, the validity of the latent trait predictors was compromised
because of the character of the data. Individuals in the sample were 15 to 23 at the outset
of data collection; therefore, the latent trait predictors represented adolescent conditions

for some of the sample but not all. This research was also lacking a pre-adolescent

161

baseline of information on family structure, antisocial behaviors, and demographic
influences. Although this is a common limitation in research of this type (Cernkovich &
Giordano, 2001; Nagin & Paternoster, 1991), it is important to consider the results of the

current study in light of this omission.

Nature of the Conceptual Model

A limitation of this specific study, and most of lifecourse research in general, is
the simplicity of the statistical tests used in proportion to the complexity of the theoretical
models. In the current study, marital and work outcomes were modeled in relationship to
a series of time-variant and invariant covariates. The assumption of this model is that
entrance into a social state (e.g., incarceration), net of time-invariant demographic
characteristics, will consequently alter an individual’s chances of attainment of marriage
and work over time. The analytic model, although easily tested with modern statistical
methods, fails to capture the dynamic and emotional nature of change. This criticism is
especially poignant when considering desistance fiom crime. Individuals do not develop
social bonds and immediately desist from crime. Change takes time and is most often
accompanied by psychological and emotional modifications (see Maruna, 2001). Most
lifecourse theorists have failed to consider the slow, socio-emotive changes that
accompany desistance.

The limited breadth of the current predictive model was evidenced in the model
fit statistics. The R2 and pseudo R2 statistics presented for the estimated equations are
moderately low; hence, the statistical models as presented fail to explain a substantial

variation in the dependent variables. Considering the emotional changes that accompany

162

a lifecourse event may improve model fit (see Maruna, 2001). Nonetheless, researchers
should strive to develop both theoretical and statistical models that capture the dynamic
nature of change. The challenge of this research is to involve a large sample to facilitate
generalizations to a broad subject population while maintaining the integrity and
complexity of individual-level data collection that would be needed to firrther consider

lifecourse change.

Nature of the Sample

Finally, it is also important to consider the results fiom the current study in light
of the sample that was used for the analyses. As discussed in Chapter 3, three sample
groups were selected for the present study. The incarcerated group represents all
individuals in the study that had been incarcerated one or more times during the data
collection period. Using a propensity score matching technique, an individually matched
sample was also selected. The final sample group included individuals with median
propensity scores and served as a comparison group in the study. The matched group
was selected for this study to partial out sample heterogeneity. In addition, the
comparison group was included as a representation of the average sample member.
Taken together, the data represent only half of the distribution of the total sample.
Individuals with a low propensity for incarceration (e.g., White women) are not fully
represented in the dataset. The nature of the sample does not pose any potential problems
for the current study; however, it is important to consider the nature of the study sample

when making generalizations to other populations.

163

Implications
Theoretical Implications

This research has important implications for the study of criminology in general
and the lifecourse perspective in specific. The findings from this research reinforce the
lifecourse theoretical perspective. Specifically, the results as a whole suggest that
lifecourse events in adulthood can have dramatic consequences for likelihood of
attainment of social bonds to marriage and work. Individual predictors, or latent traits,
most often associated with the population heterogeneity perspective (e. g. cognitive
ability) had little effect in the models. In the growth curve models, the predictors were
able to explain a small amount of variation in the initial statuses of the sample members,
but together the variables were not able to explain a substantial amount of variation over
time. Consistent with the lifecourse framework, individual, static predictors are
important as controls, but adult lifecourse events are more valuable for understanding
change in lifecourse trajectories.

The importance of social bond attainment in adulthood is further reinforced when
considering the relative effect of incarceration. Across all of the models estimated, the
effect of incarceration during the study period far outweighed that of previous
incarceration. Research of late has downplayed the relationship between adult
incarceration experiences and employment outcomes (e.g., Grogger, 1995; Western &
Beckett, 1999); however, the results from this study confirm that incarceration in
adulthood can interrupt life trajectories, especially in relationship to employment

outcomes.

164

The use of multiple predictors of lifecourse events is also reinforced by the results
of the current study. The results of this research point to the importance of considering
the effect of incarceration experiences in light of other life events. Current lifecourse
research has also reinforced this finding. For example, Giordano, Rossol, and
Cernkovich (2003) argue that resiliency is best understood in relationship to the total
presentation of self. Employment and marriage work together to develop a respectability
package that can sustain a move away from criminality. Again, incarceration is an
important lifecourse event; however, research that fails to incorporate measures of
positive social bonds in statistical models may overestimate the impact of the
imprisonment event.

The findings from the research also have important methodological implications.
Because both cross-sectional and longitudinal analyses were conducted as part of the
research study, it was possible to compare the efficacy of the models in explaining
differences in social bond outcomes. From the results of the analyses, it appears that the
lifecourse models are better able to explain variation in the likelihood of marriage and
work. Although the pseudo- choefficents for the models estimated are not optimal, they
do suggest that the growth series models are able to explain more variation in the
outcomes when compared to the cross-sectional models. By using HLM to estimate the
variability in lifecourse events over time, the power of the model to explain variance has
been enhanced. In addition, using longitudinal methodologies allows the researcher to
separate the relative effect of the predictors on the initial status to that of the slope. This
technique is especially well suited for lifecourse studies and should be utilized by future

researchers who examine phenomenon of this type.

165

Policy Implications

A number of policy implications also follow fiom this study. First, the
importance of employment related services and programming was confirmed by this
study. The incarceration event is terribly stigrnatizing. Researchers have discovered that
employers indicate that they are significantly less likely to hire recently paroled
individuals even when compared with individuals on welfare and with less relevant
experience (Holzer, 1996). As shown in the current study, sample members that had
experienced a stay of incarceration were significantly less likely to obtain work and the
quality of employment was diminished as a result of the event. In addition, incarceration
also limits avenues for developing positive social bonds to marriage, further restricting an
individual’s ability to gain employment and weakening already fi'agile ties to society.

Research on work and crime has confirmed the importance of employment in both
initiating and maintaining desistance (see Pagan & Freeman, 1999 for a review). Even in
controlled experimental studies, participation in employment programming reduced the
likelihood of subsequent criminality (Berk, Lenihan, & Rossi, 1980; Uggen, 2000).
Employment can play a powerful role in assisting individuals in overcoming negative life
events and can fiu'ther promote the development of social capital.

Despite the link between employment and desistance, prisons are currently not
equipped to provide work-related programming for inmates. Lynch and Sabol (2000)
found a decrease in services provided to state inmates from 1991 to 1997, despite the rise
in imprisonment. Around 27% of the 1997 cohort participated in vocational

programming and 12% of the population were offered prerelease services. In being

166

removed from the community, the social resources and capital of the inmates are reduced,
and the lack of reentry and social services further handicaps the released offender.

This research also highlights the importance of multi-faceted rehabilitative
programming. As demonstrated by the results of the current study, lifecourse events do
not work in isolation. Attaining one social bond increases the likelihood of developing
subsequent bonds. For example, addressing the employment needs of a released
individual will not be as effective as a program that incorporates employment training,
with family programming or substance abuse services. Incarceration can have collateral
consequences on the offender’s family, social environment, and community. Addressing
multiple correlates of positive social bonds will increase the likelihood of resilience and
eventual desistance.

The research findings also have important implications for the study of parole.
With the increase in use of determinate sentencing in many states, prisoners are being
released after they have completed their sentence and are not being afforded services of
any kind. Nearly twenty percent of parolees that left prison in 1998 did so without any
post custody supervision (Beck, 2000). Of those that are placed on parole, only 6% were
assigned to an intensive parole (Beck, 2000). The remainder were given regular parole
which rarely affords the offender more than two short contacts with the parole agent a
month (Petersilia, 1999). This is not enough time to address the problems associated with
reentry and to provide needed services to ensure success.

The point of discharge from prison represents an opportune time to intervene in
the life of offenders. In many ways release fiom prison is as much of a turning point in

the lifecourse as is incarceration. Despite the importance of the topic, researchers have

167

not considered this subject with the same detail that has been afforded to the study of
incarceration and even probation. Research that incorporates the study of parole will not
only aid in the understanding of why people do well following incarceration, it may also
put pressure on policymakers to invest in the post release supervision of inmates
(Petersilia, 2001).

Finally, the results from the current study also point to the importance of
considering the unique imprisonment experiences of women and minorities. The rise in
incarceration and subsequent reentry has had a disproportionate effect on minorities. The
statistics on minority imprisonment are striking. Each day, one out of eight Black males
between the ages of twenty-five to thirty-four is in prison (Mauer & Chesney-Lind,
2002). Compared with White men, Afiican American males are six times more likely to
be admitted to prison during their lifetime (Bonczar & Beck, 1997). Western and Pettit
(2000) found that on an average day in 1996 more African American high school
dropouts aged 20 to 35 were in some form of custody than employed.

Despite the dramatic grth in the incarceration of minorities, few researchers
have examined the differential effect of incarceration on social outcomes for this group.
Western (2002) found that the wages for Afiican Americans when compared to Whites
were initially reduced as a result of incarceration; however, the discrepancy between
groups closed over time. This study provides a solid framework for future analyses, but
additional research is needed in this domain.

Women have also been differentially affected by incarceration. Women are
statistically less likely to be incarcerated; however, the rate of growth in imprisonment

for women has been nearly twice that of men (Mauer & Chesney-Lind, 2002). Afiican

168

American women have also been disproportionately affected by the rise in incarceration.
In addition to the rise in incarceration, there is initial evidence that women may
experience incarceration different from that of men. Women’s facilities are typically
older and offer less services than that of men’s (Chesney-Lind, 2002). The combination
of isolation from family and lack of adequate services for women further increases the
pains of imprisonment for women. It is important for future research on correctional
outcomes and penal research in general, to consider the disparate effect that
imprisonment can have on women and subsequent reentry outcomes.

Research conducted to date on the relationship between incarceration and work
experiences have omitted women from the sample (e. g., Western, 2002). Women are
already marginalized when they enter the employment market. It would follow that
women may be disproportionally affected by a stay of incarceration because of their
minority status. The current research is an improvement over the existing research
because women are included in the sample; however, further sub-group analyses are
needed to fully explore the relationship between incarceration and social bond

development.

169

APPENDICES

170

Appendix A: Description of Variables

 

Variable

Outcome Variables

Employment
Tenure with
Employer

Number of Jobs
Worked

Marriage

Quality of Marriage

Dynamic Predictors

Incarceration

Marriage

Employment

Military

Out of Labor Force

Age

Age

 

Description

A dummy variable with fulltime employment (>35
hours) during the current year = 1.

Length of tenure in weeks with current employer.

A cumulative measure of number of jobs worked
since the inception of the study.

A dummy variable with marriage during the current
year = 1.

Three item factor score 1) length of the
respondent’s first marriage in years 2) number of
marriages the subject entered into during the study
period 3) a dichotomous measure of divorce during
the study period (eigenvalue 2.10, factor loadings >
0.757)

A dummy variable with incarceration at any point
during the past year =1.

A dummy variable with marriage at any point
during the past year =1.

A dummy variable with fulltime employment (>35
hours) during the current year = 1.

A dummy variable with participation in any branch
of the military during the current year =1.

A dummy variable with respondent unable to
participate in the workforce due to disability,

attendance in school, or family responsibilities = 1.

A centered variable with age measured in years —
22.

The squared function of the centered age variable

171

Years
Collected

1983-2000

1983-2000

1983-2000

1983-2000

1983-2000

1983-2000

1983-2000

1983-2000

1983-2000

1983-2000

1983-2000

 

Appendix A: Description of Variables (cont’d)

 

Static Predictors

Demographic
Influences

Gender
African American
Hispanic

Cognitive Ability

High School

School Behavior

Criminal History
Prior Incarceration
Contact with the
Police

Self-reported
Delinquency

Drug Use

Antisocial Behavior

Contextual Predictors
Unemployment Rate

Poverty

 

A dummy variable with male = 1.
A dummy variable with African American = 1.
A dummy variable with Hispanic = 1.

Percentile score on the Armed Forces Qualification
Test.

A dummy variable with graduation from high
school = 1.

A dummy variable with suspension from school at
any time = 1.

A dummy variable with incarceration prior to 1980
= l.

A dummy variable with prior history of contact with
the police = 1.

A dummy variable with self-reported delinquency =
1.

A dummy variable with self-reported drug use = 1.
A dummy variable with self-reported delinquency

ordrug use= l.

A dummy variable with individuals living in
regions with an unemployment rate greater than
nine percent = l.

A dummy variable with family income under the
poverty level = 1.

172

1983

1983

1983

1981

1983

1980

1980

1980

1980

1984

1980

1980

 

 

@0005 0050500

 

 

 

 

00000505 050200509

 

 

 

 

0000002 000
0:03 00 000002 .0600

 

 

 

acaaeeeaa 08:2

 

 

 

 

 

 

0000002 000
0003 00 00:0m 05000

 

 

 

 

000800.805

 

 

.682 850$ Mm 588%

173

 

 

000. 2 2020002290
000. 000.
2 00. 000.- 2 0220.200
000. 000. 000.

002.- 022.- 002.- 2 0:
002. 000. 000. 000.
3.0. 08.- $0 2.8..- 2 82:82:
000. 000. 000. 000. 000.

000.- 0 2 0. 000. 000. 000.- 2 V20302
000. 000. 000. 000. 000. 000.
00 2. 002. 000.- 000. 002.- 000.- 2 00:00
000. 000. 000. 000. 000. 000. 000.
002. 002. 000. 202.- 020.- 002. 000. 2 0:2
000. 00 0 . 000. 000. 0 2 0. 000. 000. 000.

002.- 0 20.- 02 0.- 000. 000.- 000.- 200. 000.- 2 0000
000. 00 2. 000. 000. 000. 000. 000. 000. 000.
02 2. 000. 000.- 20 2. 000.- 000. 000.- 000.- 00 2. 2 00<
000. 2 00. 000. 000. 0 2 0. 000. 000. 000. 000. 000.
000. 000. 002. 002.- 000. 000. 002 .- 000. 20.- 002.- 2 SO
00 2. 000. 0 00. 000. 0 2 0. 000. 000. 000. 000. 000. 000.

000.- 202.- 200.- 002. 000.- 020. 202. 000.- 002. 000.- 002.- 2 0.2822222
000. 000. 000. 000. 2 00. 000. 000. 000. 000. 000. 000. 00 0.

202.- 000.- 2 2 0.- 200. 000.- 000.- 00 2. 000.- 000. 02 2. 000.- 000.L 02.203

051000 202000000. E0260 0: 0200002: 0200202 000000 05 0.0000 00< SO 00022222

20002 0002.502 02 000m 202000 000 02.002 0000200000 .0 020000000

174

Appendix D. Collinearity Diagnostics for the Bond to Marriage Model

 

 

Tolerance VIF

Work .540 1.853
Military .881 1.136
Absence from .656 1.525
Workforce

Age .900 1.1 l l
Antisocial Behavior .526 1.901
Gender .725 1.380
Black .517 1.934
Hispanic .646 l .549
High School .694 1.440
Family Poverty .854 1.171
Antisocial .852 1.174
Previous .812 1.232
Incarceration

Incarceration .670 1 .492

 

175

xvc.

 

 

Sc.- 009000-232
000. 000.
002.- 200. 0200002:
000. 000. 000.
2.200.- 000.- 000.- v22.202
000. 000. 000. 000.
02.0. 02 0.- 0 2 0.- 00 2. 222
000. 000. 002. 000. 000. -
00 2. 000.- 000 020.- 00 2. 222 20
000. 000. 000. 02.0. 000. 000.
002. 0:.- 000.- 020. 002. 000. 22802290
000. 000. 000. 000. 000. 200. 000.
000.- 002.- 020. 000. 00.0. 200. 000.- 0220.50
000. 200. 000. 000. 000. 000. 000. 000.
000. 000. 0 2 0.- 3.0. 20 2 .- 002.- 02 2.- 2.02 .- 0:
000. 000. 020. 000. 000. 000. 2.00. 000. 000.
200. 2- 2 0. 000.- 000.- 000.- 002.- 0 2 0.- 02 0.- 20020. 00%.
000. 000. 000. 000. 000. 000. 002. 000. 000. 000.
000.- 00 2. 000.- 000. 000.- 02 2. 000. 000.- 20 2. 002. 0000
000. 000. 020. 000. 000. 002. 000. 000. 000. 000. 000.
2:. 000. 000.- 020. 000.- 000.- 202.- 200.- 002. 002. 000.- 0222.2
000. 000. 020. 000. 000. 000. 200. 000. 000. 000. 000. 000.
002.- 000.- 2-00. 000. 000. 000. 02.0. 002. 002.- 020.- 002.- 002.- so
5080 0:02 U200.202: v20020 0222 22212.2 20280222 0:260 022 a? 00.0 02.00202

20002 Eofixoafim 2 000m _Soom 00.0 x2002 2820022000 .00 0020:2500

176

Appendix F. Collinearity Diagnostics for the Bond to Employment Model
Tolerance VIF

 

 

Absence from Workforce .776 1.289
Military .928 1.078
Age .888 1.126
Cognitive Ability .535 1.868
High School Graduate .710 1.408
Poverty .861 1.162
Antisocial .845 1 . 1 84
Previous Incarceration .819 1.221
Incarceration .704 1 .420
Hispanic .646 1 .548
Black .504 1.983
Marriage .804 1 .244
Gender .766 1 .306

 

177

Appendix G. Final Statistical Models for the Likelihood of Work and Marriage
Dependent Variables — Propensity Score Only Analyses (N =1 ,466)

Independent
Variable

Intercept
Lifecourse Events
Incarceration

Incarceration
Count

Marriage

Work

Military

Absence from

Workforce
Model Fit

Age

Age Squared

Level II
Propensity
Score
(Intercept)

Propensity
Score
(Slope)

 

 

Employment Marriage

[3 SE [3 SE
-0.05 0.06 -0.43 0.09***
-0.21 0.08* -O.28 008*"
-0.18 0.03*** -0.04 0.04
0.31 005*“ -- __

-- -- 0.11 0.04“
-1.45 0.18*** 0.25 0.18
-0.97 0.05*** -0.05 0.05
0.09 0.01*** 0.14 0.02***

-- -- -0.00 0.00***
-1.50 0.23*** -1.39 0.32***
0.01 0.02 -0.09 004*

 

***p<0.001 **p<0.01 *p<0.05 (two-tailed tests)

178

Appendix H. Final Statistical Models for the Nature of Work Dependent Variables —

Propensity Score Onl Analyses (N =1 ,466)

 

Independent
Variable

Intercept
Lifecourse Events
Incarceration

Incarceration
Count

Marriage
Military

Absence from
Workforce

Model Fit
Age
Age Squared
Level II
Propensity
Score

(Intercept)

Propensity
Score (Slope)

 

 

Tenure with Employer Number of Jobs Worked
13 SE [3 SE
0.46 0.04*** 1.97 0.02***
-0.10 001*“ 0.01 0.01
-O.16 0.01*** -0.00 0.00
0.03 000*“ 0.02 0.01*
-0.11 002*" -0.21 003*"
0.01 0.00 -0.06 001*"
0.31 001*" 0.05 0.00***
-0.01 0.00*** -- --
-1.01 0.17*** 0.51 0.08***
-0.01 0.01 0.01 0.00

 

***p<0.001 Mp<0.01 *p<0.05 (two-tailed tests)

179

BIBLIOGRAPHY

Allan, E., & Steffensmeier, D. (1989). Youth, Underemployment, and Property Crime:
Differential Effects of Job Availability and Job Quality on Juvenile and Young
Adult Arrest Rates. American Sociological Review, 54, 107-123.

Andrews, D. A., & Bonta, J. (1994). The Psychology of Criminal Conduct. Cincinnati,
OH: Anderson.

Andrews, D. A., Bonta, J ., & Hoge, R. D. (1990). Classification for Effective
Rehabilitation: Rediscovering Psychology. Criminal Justice and Behavior, 1 7,
19-52.

Arum, R., & Beattie, I. R. (1999). High School Experience and the Risk of Adult
Incarceration. Criminology, 37(3), 515-537.

Bachman, R., & Paternoster, R. (1997). Statistical Methods for Criminology and
Criminal Justice. New York: McGraw-Hill.

Baron, S. W., & Hartnagel, T. F. (1997). Attributions, Affect, and Crime: Street Youths'
Reactions to Unemployment. Criminology, 30, 47 -87.

Beck, A. J. (2000). State and Federal Prisoners Returning to the Community: Findings
fiom the Bureau of Justice Statistics. Washington, DC: Bureau of Justice
Statistics.

Beck, A. J ., Karberg, J. C., & Harrison, P. M. (2002). Prison and Jail Inmates at Midyear
2001. Washington, DC: Bureau of Justice Statistics.

Becker, G. S. (1964). Human Capital: A Theoretical and Empirical Analysis, with
Special Reference to Education. New York: National Bureau of Economic
Research.

Berk, R. A., Lenihan, K. J ., & Rossi, P. (1980). Crime and Poverty: Some Experimental
Evidence from Ex-Offenders. American Sociological Review, 45(5), 766-786.

180

Berk, R. A., & Newton, P. J. (1985). Does Arrest Really Deter Wife Battery? An Effort
to Replicate the Findings of the Minneapolis Spouse Abuse Experiment.
American Sociological Review, 50(2), 253-262.

B] S. (2001). Direct expenditure on criminal justice, by level of government, 1 982-9 7,
from http://www.ojp.usdoj .gov/bj s/glance/expgov.txt

Blumstein, A., Cohen, J ., & Farrington, D. P. (1988a). Criminal Career Research: Its
Value for Criminology. Criminology, 26(1), 1-35.

Blumstein, A., Cohen, J ., & Farrington, D. P. (1988b). Longitudinal and Criminal Career
Research: Further Clairifications. Criminology, 26(1), 57-74.

Blumstein, A., Cohen, J ., Roth, J. A., & Visher, C. A. (1986). Criminal Careers and
"Career Criminals ". National Academy Press: Washington DC.

Bonczar, T. P., & Beck, A. J. (1997). Lifetime Likelihood of Going to State or Federal
Prison. Washington, DC: Bureau of Justice Statistics.

Bourdieu, P. (1986). The Forms of Capital. In J. G. Richardson (Ed.), Handbook of
T heory and Research for the Sociology of Education (pp. 241-256). New York:
Greenwood Press.

Brame, R., Bushway, S., & Paternoster, R. (1999). On the Use of Panel Research Design
and Radom Effects Models to Investigate Static and Dynamic Theories and
Criminal Offending. Criminology, 37(3), 599-641.

Brame, R., Paternoster, R., Mazerolle, P., & Piquero, A. (1998). Testing for the Equality
of Maximum-Likelihood Regression Coefficents between Two Independent
Equations. Journal of Quanitative Criminology, 14(3), 245-261.

Bursik, R. J. J ., & Grasmick, H., G. (1993). Neighborhoods and Crime: The Dimensions
of Effective Community Control. New York: Lexington.

Bushway, S., Brame, R., & Paternoster, R. (1999). Assessing Stability and Change in
Criminal Offending: A Comparison of Random Effects, Semiparametric, and
Fixed Effects Modeling Strategies. Journal of Quanitative Criminology, 15(1),
23-61.

181

Caspi, A., & Bern, D. (1990). Personality Continuity and Change across the Life Course.
In L. A. Pervin (Ed.), Handbook of Personality: Theory and Research (pp. 549-
575). New York: Guilford.

Caspi, A., & Moffitt, T. (1993). When do Individual Differences Matter? A Paradoxical
Theory of Personality Coherence. Psychological Inquiry, 4, 247-271.

Center for Human Resource Research. (2001). National Longitudinal Survey of Youth
Data. Columbus, OH: The Ohio State University.

Cernkovich, S. A., & Giordano, P. C. (2001). Stability and Change in Antisocial
Behavior: The Transition fi'om Adolescence to Early Adulthood. Criminology,
39(2), 371-410.

Cherlin, A. J ., Chase-Lansdale, P. L., & McRae, C. (1998). Effects of Parental Divorce
on Mental Health Throughout the Life Course. American Sociological Review, 63,
239-249.

Chesney-Lind, M. (2002). Imprisoning Women: The Unintented Victims of Mass
Imprisonment. In M. Chesney-Lind (Ed.), Invisible Punishment: The Collateral
Consequences of Mass Imprisonment (pp. 79-94). New York: The New Press.

Chiricos, T. (1987). Rates of Crime and Unemployment: An Analysis of Aggregate
Research. Social Problems, 334, 187-212.

Cohen, J ., & Cohen, P. (1983). Applied Multiple Regression/Correlation Analysis for the
Behavioral Sciences (2nd ed.). Hillsdale, N.J.: Lawrence Erlbaum.

Coleman, J. S. (1988). Social Capital in the Creation of Human Capital. American
Journal Of Sociology, 94, 895-8120.

Coleman, J. S. (1990). Foundations of Social Theory. Boston, Massachusets: Harvard
University Press.

DeJong, C. (1997). Survival Analysis and Specific Deterrence: Integrating Theoretical
and Empirical Models of Recidivism. Criminology, 35, 561-575.

182

Duncan, G. J ., & Brooks-Gunn, J. (Eds.). (1997). Consequences of Growing Up Poor:
Russell Sage Foundation.

Duncan, G. J ., Brooks-Gunn, J ., & Klebanov, P. K (1994). Economic Deprevation and
Early Childhood Development. Child Development, 65, 296-318.

Duster, T. (1987). Crime, Youth Unemployment, and the Black Urban Underclass. Crime
& Delinquency, 33, 300-316.

Elder, G. H. (1975). Age Differentiation and the Life Course. Annual Review of
Sociology, I, 165-190.

Elder, G. H. (1985). Life Course Dynamics. Ithaca, NY: Cornell University Press.

Elliott, D. S., Ageton, S., & Canter, R. J. (1979). An Integrated Theoretical Perspective
on Delinquent Behavior. Journal of Research in Crime and Delinquency, 16, 3-
27.

F.B.I. (1990). Age-Specific Arrest Rates and Race-Specific Arrest Rates for Selected
Oflenses. Washington, DC: United States Department of Justice.

Fagan, J ., & Freeman, R. B. (1999). Crime and Work. In M. Tonry (Ed.), Crime and
Justice: A Review of Research (Vol. 25, pp. 225-290). Chicago, Illinois: The
University of Chicago Press.

Farr, K A. (2000). Classification for female inmates: Moving forward. Crime &
Delinquency, 46, 3-17.

Farrington, D. P. (1986). Age and Crime. In Crime and Justice: An Annual Review of
Research (Vol. 7). Chicago: University of Chicago Press.

Fishbein, D. H. (1990). Biological Perspectives in Criminology. Criminology, 28, 27-72.

Freeman, R. B. (1991). Crime and the Employment of Disadvantaged Youths. Cambridge,
MA: National Burea of Economic Research.

183

Gainey, R. R., Payne, B. K, & O'Toole, M. (2000). The Relationships between Time in
Jail, Time on Electronic Monitoring, and Recidivism: An Event History Analysis
of a Jail Based Program. Justice Quarterly, 1 7(4), 732-752.

Gardner, W., Mulvey, E., & Shaw, E. (1995). Regression Analysis of Counts and Rates:
Poisson, Overdispersed Poisson, and Negative Binomial Models. Psychological
Bulletin, 118, 392-404.

Geis, G. (2000). On the absence of self-control as the basis for a general theory of crime.
Theoretical Criminology, 4(1), 35-53.

Gendreau, P., Little, T., & Goggin, C. (1996). A Meta-Analysis of the Predictors of Adult
Offender Recidivism: What Works! Criminology, 34(4), 575-607.

Giordano, P. C., Rossol, J ., & Cernkovich, S. A. (2003). Emotion, Cognition, and
Desistance From Crime. Unpublished Manuscript.

Glaze, L. E. (2002). Probation and Parole in the United States, 2001. Washington, DC:
Bureau of Justice Statistics.

Glueck, S.,, & Glueck, E. (1950a). Unraveling Juvenile Delinquency. New York: The
Commonwealth Fund.

Glueck, S., & Glueck, E. (1950b). Unravelling Juvenile Delinquency. Cambridge,
Massachusets: Cambridge University Press.

Goeke, J. C. (1980). Factors Related to the Alienation of the Male Married Oflender from
His Family. Saratoga, California: Century Twenty One Publishing.

Good, D. F., Pirog-Good, M. A., & Sickles, R. C. (1986). An Analysis of Youth Crime
and Employment Patterns. Journal of Quanitative Criminology, 2, 219-236.

Gottfiedson, D. M., Gottfredson, M. R., & Garafolo, J. (1977). Time Served in Prison
and Parole Outcomes for amoung Parolee Risk Categories. Journal of Criminal
Justice, 5, 1-12.

184

 

Gottfredson, M. R., & Hirchi, T. ( 1986). The True Value of Lambda Would Appear to be
Zero: An Essay on Career Criminals, Criminal Careers, Selective Incapacitation,
Cohort Studies and Related Topics. Criminology, 24(2), 213-234.

Gottfredson, M. R., & Hirschi, T. (1990). A General Theory of Crime. Stanford,
California: Stanford University Press.

Greenberg, D. F. (1985). Age, Crime, and Social Explaination. American Journal of
Sociology, 91(1), 1-21.

Grogger, J. (1995). The Effect of Arrests on the Employment and Earnings of Young
Men. The Quarterly Journal of Economics, 1 1 0(1), 5 1-71.

Hagan, J ., & McCarthy, B. R. (1997). Mean Streets: Youth Crime and Homelessness.
New York, New York: Cambridge University Press.

Hagan, J ., & Palloni, A. (1988). Crimes as Social Events in the Life Course:
Reconceiving a Criminal Contreversy. Criminology, 26, 87-100.

Hirschi, T. (1972). Causes of Delinquency. Berkeley: University of California Press.

Hirschi, T., & Gottfredson, D. M. (1983). Age and the Explaination of Crime. American
Journal of Sociology, 89(3), 552-584.

Hirschi, T., & Gottfredson, D. M. (1985). Age and Crime, Logic and Scholarship:
Comment on Greenberg. American Journal Of Sociology, 91(1), 22-27.

Hirschi, T., & Gottfredson, M. R. (2000). In Defense of Self Control. Theoretical
Criminology, 4(1), 55-69.

Hirschi, T., & Hindelang, M. J. (1977). Intelligence and Delinquency: A Revisionist
Review. American Sociological Review, 42, 571-586.

Hogan, D. P. (1978). The Variable Order of Events in the Life Course. American
Sociological Review, 43, 573-586.

Hogan, D. P. (1980). The Transition to Adulthood as a Career Contingency. American
Sociological Review, 45, 261-276.

185

Holzer, H. J. (1996). What Employers Want: Job Prospects for Less- Educated Workers.
New York: Russell Sage.

Homey, J ., Osgood, D. W., & Marshall, I. H. (1995). Criminal Careers in the Short-
Term: Intra-Individual Variability in Crime and Its Relation to Local Life
Circumstances. American Sociological Review, 60(5), 655-673.

Hughes, T. A., Wilson, D. J ., & Beck, A. J. (2001). Trends in State Parole, [990-2000.
Washington DC: The Bureau of Justice Statistics.

Irwin, J ., & Austin, J. (1994). It's About Time: America's Imprisonment Binge. Belmont,
California: Wadsworth.

Jacobs, D., & Helms, R. E. (1996). Toward a Political Model of Incarceration: A Time-
Series Examination of Multiple Examinations for Prison Admission Rates.
American Journal Of Sociology, 102(2), 323-357.

J ang, S. J. (1999). Age-Varying Effects of Family , School, and Peers on Delinquency: A
Multilevel Moeling Test of Interactional Theory. Criminology, 37(3), 643-685.

Johnson, R. A., Hoffmann, J. P., Su, S. S., & Gerstein, D. R. (1997). Grth Curves of
Deviant Behavior in Early Adolescence: A Multilevel Analysis. Journal of
Quanitative Criminology, 13(4), 429-467.

Jones, M., & Sims, B. (1997). Recidivism of Offenders Released from Prison in North
Carolina: A Gender Comparison. Prison Journal, 77, 33 5-348.

Kling, J. R. (1999). The Effect of Prison Sentence Length on the Subsequent Employment
and Earnings of Criminal Defendants. Princeton, New Jersey: Woodrow Wilson
School, Princeton University.

Langan, P. A., & Levin, D. J. (2002). Recidivism of Prisoners Released in 1994.
Washington, DC: Bureau of Justice Statistics.

Laub, J ., Nagin, D. S., & Sampson, R. (1998). Trajectories of Change in Criminal
Offending: Good Marriages and the Desistance Process. American Sociological
Review, 63, 225-238.

186

Laub, J ., & Sampson, R (1988a). Unraveling Families and Delinquency: A Reanalysis of
the Gleuck's Data. Criminology, 26(3), 355-3 80.

Laub, J ., & Sampson, R. (1988b). Unraveling Families and Delinquency: A Reanalysis of
the Glueck's Data. Criminology, 26(3), 355-3 80.

Laub, J ., & Sampson, R. (1993). Turning Points in the Life Course: Why Change Matters
to the Study of Crime. Criminology, 31(3), 301-325.

Laub, J ., & Sampson, R. (2001). Understanding Desistence from Crime. In M. Tonry &
N. Morris (Eds), Crime and Justice: A Review of Research (pp. 1-69).

LeBlanc, M., & Loeber, R. (1993). Developmental Criminology Updated. In M. Tonry
(Ed.), Crime and Justice: A Review of Research (V 01. 23). Chicago: University of
Chicago Press.

Lessan, G. T. (1991). Macro-Imprisonment Determinants of Penal Policy: Estimating the
Unemployment and Inflation Influences on Imprisonment Rate Changes in the
US Crime Law and Social Change, 16(2), 177-198.

Loeber, R. (1982). The Stability of Antisocial Child Behavior: A Review. Child
Development, 53, 143 1 -1446.

Loeber, R., & Le Blanc, M. (1990). Toward a Developmental Criminology. In M. Tonry
& N. Morris (Eds), Crime and Justice: A Review of Research (Vol. 12, pp. 375-
473). Chicago: University of Chicago Press.

Long, J. S. (1997). Regression Models for Categorical and Limited Dependent Variables.
Thousand Oaks, CA: Sage.

Lynch, J. P., & Sabol, W. J. (2000). Prisoner reentry and the Consequences of
Sentencing Reform, fi'om www.urban.org/news/tuesdays/ 12-00/lynch.html

Lynch, J. P., & Sabol, W. J. (2001). Prisoner Reentry in Perspective. Washington, DC:
Urban Institute Press.

Manning, P. (1994). The Police: Symbolic Capital, Class, and Control. In G. S. Bridges
& M. A. Myers (Eds), Inequality, Crime, and Social Control. Oxford: Westview.

187

Maruna, S. (2001). Making Good: How Ex-Convicts Reform and Rebuild their Lives.
Washington, DC: American Psychological Association.

Marx, K. (1887). Capital: A Critical Analysis of Capitalist Production. London: Lowrey.

Mauer, M., & Chesney-Lind, M. (2002). Invisible Punishment. New York: The New
Press.

Menard, S., Mihalic, S., & Huizinga, D. (2001). Drugs and Crime Revisited. Justice
Quarterly, 18, 269-299.

Michalowski, R. J ., & Carlson, S. M. (1999). Unemployment, Imprisonment, and Social
Structures of Accumulation: Historical Contingency in the Rusche-Kirchheimer
Hypothesis. Criminology, 3 7(2), 217-249.

Moffitt, T ., Caspi, A., Harrington, H., & Milne, B. J. (2002). Males on the Life-Course-
Persistent and Adolescence-Limited Antisocial Pathways: Follow-up at Age 26
Years. Development and Psychopathology, 14, 179-207.

Moffitt, T. E. (1993). Adolescence-limited and Life-course-persistent antisocial behavior:
A Developmental Taxonomy. Psychological Review, 100, 674-701.

Moffitt, T. E., Caspi, A., Dickson, N., Silva, P. A., & Stanton, W. R. (1996). Childhood-
Onset Versus Adolescent-Onset Antisocial Conduct Problems in Males: Natural
History from Ages 3 to 18 Years. Development and Psychopathology, 8, 399-424.

Morgan, S. L. (2001). Counterfactuals, Causal Effect Heterogeneity, and the Catholic
School Effect on Learning. Sociology of Education, 74, 341-3 74.

Murphy, K M., & Welch, F. (1990). Empirical Age-Earnings Profiles. Journal of Labor
Economics, 8, 202-209.

Nagin, D. S., & Farrington, D. P. (1992a). The Onset and Persistence of Offending.
Criminology, 30, 501-523.

Nagin, D. S., & Farrington, D. P. (1992b). The Stability of Criminal Potential from
Childhood to Adulthood. Criminology, 30, 235-260.

188

Nagin, D. S., & Paternoster, R. (1991). On the Relationship of Past to Future
Participation in Delinquency. Criminology, 29(2), 163-189.

Needels, K E. (1996). Go Directly to Jail and Do Not Collect? A Long Term Study of
Recidivism, Employment, and Earnings Patterns amoung Prison Releases.
Journal of Research in Crime and Delinquency, 3 3, 471-496.

Olweus, D. (1979). Stability of Agressive Reaction Patterns in Males: A Review.
Psychological Bulletin, 86, 852-875.

Orsagh, T., & Chen, J. (1988). The Effect of Time Served on Recidivism: An
Interdisciplinary Theory. Journal of Quanitative Criminology, 4, 155-171.

Ouimet, M., & Le Blanc, M. (1996). The Role of Life Experiences in the Continuation of
the Adult Criminal Career. Criminal Behaviour and Mental Health, 6(1), 73-97.

Owen, B. (1998). In the Mix: Struggle and Survival in a Women's Prison. Albany: State
University of New York Press.

Parcel, T. L., & Menaghan, E. G. (1993). Family Social Capital and Children's Behavior
Problems. Social Psychology Quarterly, 56(2), 120-135.

Parcel, T. L., & Menaghan, E. G. (1994). Early Parental Work, Family Social Capital,
and Early Childhood Outcomes. American Journal Of Sociology, 99(4), 972-
1009.

Paternoster, R., Brame, R., Mazerolle, P., & Piquero, A. (1998). Using the Correct

Statistical Test for the Equality of Regression Coefficents. Criminology, 36(4),
859-866.

Paternoster, R., & Bushway, S. (2001). Theoretical and Empirical Work on the
Relationship Between Unemployment and Crime. Journal of Quanitative
Criminology, 1 7(4), 391-407.

Paternoster, R., Dean, C. W., Piquero, A., Mazerolle, P., & Brame, R. (1997). Generality,
Continuity, and Change in Offending. Journal of Quanitative Criminology, 13(3),
231-266.

189

Patterson, G., & Yoerger, K. (1993). Developmental Models for Delinquent Behavior. In
S. Hodgins (Ed.), Mental Disorder and Crime. Newbury Park, California: Sage.

Paxton, P. (1999). Is Social Capital Declining in the United States? A Multiple Indicator
Assesment. American Journal of Sociology, 105(1), 88-127.

Petersilia, J. (1999). Parole and Prisoner Reentry in the United States. In M. Tonry & J.
Petersilia (Eds), Prisons (pp. 479-529). Chicago: University of Chicago Press.

Petersilia, J. (2000). Prisoners Returning to Communities:
Political, Economic, and Social Consequences. Washington, DC: Urban Institute.

Petersilia, J. (2001). Prisoner Reentry: Public Safety and Reintegration Challenges. The
Prison Journal, 81(3), 360-375.

Raudenbush, S. (2001). Comparing Personal Trajectories and Drawing Causal Inferences
from Longitudinal Data. Annual Review of Psychology, 52, 501-525.

Raudenbush, S., & Bryk, A. S. (2002). Hierarchical Linear Models: Applications and
Data Analysis Methods. Thousand Oaks: Sage.

Raudenbush, S., & Chan, W.-S. (1992). Growth Curve Analysis in Accelerated
Longitudinal Design. Journal of Research in Crime and Delinquency, 29(4), 387-
411.

Raudenbush, S., & Xiao-Feng, L. (2001). Effects of Study Duration, Frequency of
Observation, and Sample Size on Power in Studies of Group Differences in
Polynomial Change. Psychological Methods, 6(4), 387-401.

Reisig, M. D., Holtfreter, K L., & Morash, M. (2002). Social Capital among Women
Offenders: Examining the Distribution of Social Networks and Relationships.
Journal of Contemporary Criminal Justice, 18(2), 167-187.

Rindfuss, R., Swicegood, C. G., & Rosenfeld, R. (1987). Disorder in the Lifecourse: How
Common and Does it Matter? American Sociological Review, 52, 785-801.

Rose, D., & Clear, T. R. (1998). Incarceration, Social Capital, and Crime: Implications
for Social Disorganization Theory. Criminology, 36(3), 441-479.

190

Rosenbaum, P. R. (1989). Optimal Matching for Observational Studies. Journal of the
American Statistical Association, 84, 408.

Rosenbaum, P. R., & Rubin, D. B. (1983). The Central Role of the Propensity Score in
Observational Studies for Causal Effects. Biometrika, 70, 41-55.

Rosenbaum, P. R., & Rubin, D. B. ( 1985a). The Bias Due to Complete Matching.
Biometrics, 41(1), 103-116.

Rosenbaum, P. R., & Rubin, D. B. (1985b). Constructing a Control Group Using
Multivariate Matched Sampling Methods that Incorporate the Propensity Score.
The American Statistician, 39, 33-38.

Rosenfeld, R., Messner, S. F., & Baumer, E. P. (2001). Social Capital and Homicide.
Social Forces, 80(1), 283-309.

Rowe, A., & Tittle, C. R. (1977). Life Cycle Changes and Criminal Propensity.
Sociological Quarterly, 18, 223-236.

Rubin, D. B. (1973). Matching to Remove Bias in Observational Studies. Biometrics, 36,
293-298.

Rubin, D. B., & Thomas, N. (1996). Matching Estimated Propensity Scores: Relating
Theory and Practice. Biometrika, 52(1), 249-264.

Rusche, G., & Kirchheimer, O. (1939). Punishment and Social Structure. New York:
Columbia University Press.

Sampson, R., & Laub, J. (1990). Crime and Deviance over the Life Course: The Salience
of Adult Social Bonds. American Sociological Review, 55, 609-627.

Sampson, R., & Laub, J. (1992). Crime and Deviance in the Life Course. Annual Review
of Sociology, 18, 63-84.

Sampson, R., & Laub, J. (1993). Crime in the Making: Pathways and Turning Points
through Life. Cambridge, MA: Harvard University Press.

191

Sampson, R., & Laub, J. (1996). Socioeconomic Achievement in the Life Course of
Disadvanted Men: Military Service as a Turning Point, Circa 1940-1965.
American Sociological Review, 61, 347-367.

Schafer, N. (1994). Exploring the Link between Visits and Parole Success: A Survey of
Prison Visitors. International Journal of Offender Therapy and Comparative
Criminology, 3 8( 1), 17-32.

Schmidt, P., & Witte, A. D. (1988). Predicting Recidivism Using Survival Models. New
York: Springer-Verlag.

Schultz, T. W. (1961). Investment in Human Capital. American Economic Review, 51, 1-
I7.

Shaw, C. R., & McKay, H. D. (1942). Juvenile Delinquency and Urban Areas: A Study of
Delinquents in Relation to Differential Characterisitics of Local Communities in
American Cities. Chicago: University of Chicago Press.

Snell, T. L. (1994). Women in Prison. Washington, DC: Bureau of Justice Statistics.

Spohn, C., & Holleran, D. (2000). The Imprisonment Penalty Paid by Young,
Unemployed Black and Hispanic Male Offenders. Criminology, 38(1), 281-306.

Steffensmeier, D., Kramer, J ., & Streifel, C. (1993). Gender and Imprisonment Decisions.
Criminology, 31(3), 41 1-446.

Steffensmeier, D, Uhner, J., & Kramer, J. (1998). The Interaction of Race, Gender and
Age in Criminal Sentencing: The Punishment Cost of Being Young, Black and
Male. Criminology, 36(4), 763 7-97.

Thomberry, T. P., & Christenson, R. L. (1984). Unemployment and Criminal
Involvement: An Investigation of Reciporical Causal Structures. American
Sociological Review, 56, 609-627.

Tracy, P. E., & Kemph-Leonard, K (1996). Continuity and Discontinuity in Criminal
Careers. New York: Pleunm.

192

Travis, J ., Solomon, A. M., & Waul, M. (2001). From Prison to Home: The Dimensions
and Consequences of Prisoner Reentry. Washington, DC: The Urban Institute.

Uggen, C. (2000). Work as a Turning Point in the Life Course of Criminals: A Duration
Model of Age, Employment, and Recidivism. American Sociological Review, 67,
529-546.

Uggen, C., & Piliavin, I. (1998). Asymmetrical Causation and Criminal Desistance. The
Journal of Criminal Law and Criminology, 88(4), 1399-1422.

Waldfogel, J. (1994). The Effect of Criminal Conviction on Income and the Trust
'Reposed in the Workmen'. Journal of Human Resources, 29(1), 62-81.

Warr, M. (1998). Life-Course Transitions and Desistance from Crime. Criminology, 36,
183-216.

West, D. J ., & Farrington, D. P. (1973). Who Becomes Delinquent? London: Heinemann.

Western, B. (2002). The Impact of Incaceration on Wage Mobility and Inequality.
American Sociological Review, 6 7, 526-546.

Western, B., & Beckett, K (1999). How Unregulated is the US. Labor Market? The
Penal System as a Labor Market Institution. American Journal Of Sociology,
104(4), 1030-1060.

Western, B., Kling, J. R., & Weiman, D. F. (2001). The Labor Market Consequences of
Incarceration. Crime & Delinquency, 47(3), 410-427.

Western, 8., & Pettit, B. (2000). Incaceration and Racial Inequality in Men's
Employment. Industrial and Labor Relations Review, 54, 3-16.

White, H., Brick, J ., & Hansell, S. (1993). A Longitudinal Investigation of Alcohol Use
and Agression in Adolescence. Journal of Studies on Alcohol, 11, 62-77.

White, H., & Gorman, D. M. (2000). Dynamics of the Drug Crime Relationship: The
Nature of Crime Continuity and Change. Washington, DC: US Department of
Justice.

193

White, H. R. (1990). The Drug-Use Delinquency Connection in Adolescence. In R. A.
Weisheit (Ed.), Drugs, Crime, and Criminal Justice. Cincinatti: Anderson
Publishing Company.

Wilson, J. Q., & Hermstein, R. (1985). Crime and Human Nature. New York: Simon and
Schuster.

Wolfgang, M., Figlio, R., & Sellin, T. (1972). Delinquency in a Birth Cohort. In.
Chicago: University of Chicago Press.

Wolfgang, M., Thomberry, T. P., & Figlio, R. (1987). From Boy to Man: From
Delinquency to Crime. Chicago: University of Chicago Press.

Wright, J. P., Cullen, F. T., & Miller, J. T. (2001). Family Social Capital and Delinquent
Involvement. Journal of Criminal Justice, 29, 1-9.

Zatz, M. (1987). The Changing Forms of Racial/Ethnic Biases in Sentencing. Journal of
Research in Crime and Delinquency, 24(1), 69-92.

194

 

1293 02461