Você está na página 1de 16

Race, Family Structure, and Delinquency: A Test of Differential Association and Social

Control Theories
Author(s): Ross L. Matsueda and Karen Heimer
Source: American Sociological Review, Vol. 52, No. 6 (Dec., 1987), pp. 826-840
Published by: American Sociological Association
Stable URL: http://www.jstor.org/stable/2095837
Accessed: 19-01-2018 21:46 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at

American Sociological Association is collaborating with JSTOR to digitize, preserve and

extend access to American Sociological Review

This content downloaded from on Fri, 19 Jan 2018 21:46:36 UTC
All use subject to http://about.jstor.org/terms


University of Wisconsin-Madison

Studies of the relationship between race and delinquency have typically found that
broken homes lead to greater delinquency among blacks than whites, but have not
demonstrated empirically why this is so. This paper derives theoretical mechanisms
from differential association theory and social control theory, specifying how broken
homes may influence delinquency among both blacks and nonblacks. The analysis
specifies a structural equation model of delinquency (Matsueda 1982), derives
competing hypotheses from the two theories, and estimates a cross-population model
for blacks and nonblacks using data from the Richmond Youth Project. Consistent with
previous research, we find that broken homes have a larger impact on delinquency
among blacks than nonblacks, but, unlike previous studies, our model explains this
effect completely. In both populations, the effects of broken homes and attachment to
parents and peers are mediated by the learning of definitions of delinquency, a finding
that supports differential association over social control theory.

Although race is a critical variable in many sons with delinquency data (Hirschi 1969). This
theories of crime, little empirical research has implies that any cross-race comparison must
examined competing explanations of the race- consider differential errors of measurement
delinquency relationship. There are perhaps (Bielby, Hauser, and Featherman 1977).
three reasons for this. First, given the history of Most previous research on black-white differ-
racial discrimination in the United States, any ences in delinquency has focused on the
examination of black-white differences in unlaw- structure of the family. Stimulated by the
ful behavior is likely to be politically sensitive Moynihan Report (1965), which hypothesized
and controversial (Wilson and Herrnstein 1985; that black youths commit more delinquent acts
Wilson 1985). Second, differences in criminal in part because of a tangle of pathology
and delinquent behavior, as measured by official originating in female-headed households, unem-
statistics, have been attributed to racial bias in ployment, illegitimacy, and differential social-
the criminal justice system. Third, racial ization, such research has examined the joint
disparities in delinquency have been difficult to relationships among race, broken homes, and
measure reliably. Indeed, researchers disagree delinquency. The conclusions have been mixed:
over the extent to which rates of unlawful most researchers find that broken homes have a
behavior vary by race: official statistics and larger effect on delinquency among blacks
victimization surveys show wide disparities, (Monahan 1957; Moynihan 1965; Rosen, Lalli,
while self-report surveys show few differences and Savitz 1975); some find a greater effect
(Hindelang 1978; Hindelang, Hirschi, and Weis among whites (Toby 1957; Chilton and Markle
1979, 1981). Moreover, because the responses 1972; Austin 1978); still others find little
of blacks to survey questions contain more difference by race (Tennyson 1967; Berger and
random variability than those of whites, some Simon 1974). This literature has been preoccu-
have cautioned against making racial compari- pied with the demographic question of whether
the effect of broken homes on delinquency
varies by race. From a theoretical standpoint, a
* Direct all correspondence to Ross L. Matsueda,
more significant question concerns the causal
Department of Sociology, University of Wisconsin,
Madison, WI 53706.
mechanisms intervening between broken homes
This is a revised version of a paper prepared for and delinquency for both races. What is needed,
presentation at the 1987 Annual Meetings of the then, is a theoretical model that can explain
American Sociological Association, Chicago. The re- these relationships.
search was partially supported by the Graduate School of This paper examines delinquent behavior
the University of Wisconsin-Madison. For their insight-
among blacks and nonblacks using a causal
ful comments on an earlier draft, we thank William T.
model derived from two dominant sociological
Bielby, Rosemary Gartner, Charles N. Halaby, and two
anonymous ASR reviewers. We also thank Travis
theories of delinquency: differential association
Hirschi, Joseph G. Weis, and Carol A. Zeiss for making theory and social control theory. The model
the data available. builds on a statistical model previously esti-

826 American Sociological Review, 1987, Vol. 52 (December:826-840)

This content downloaded from on Fri, 19 Jan 2018 21:46:36 UTC
All use subject to http://about.jstor.org/terms

mated to test differential association against organization consists of neighborhood organiza-

control theory (Matsueda 1982). We use the tion, family processes, peer relationships, and
model to examine differences in parameters the distribution of age, race, and class.
across populations of black and nonblack youth, At the individual level, Sutherland maintained
to focus on the relationship between family that normative conflict is translated into indi-
structure and delinquency, and to test the vidual acts of delinquency through differential
efficacy of differential association versus social association. Definitions favorable and unfavor-
control theory across race. The first section able to delinquent behavior are learned through
discusses the implications of differential associ- communication, primarily in intimate groups.
ation and social control theories for explaining Whether delinquency occurs depends on the
the relationships among race, broken homes, ratio of learned definitions favorable and
and delinquency. Here we derive several unfavorable to that act. Moreover, each defini-
testable hypotheses from the competing theo- tion is weighed by four modalities: frequency,
ries. The second section presents a structural duration, priority, and intensity. Definitions
equation model of these relationships, estimates presented more frequently, for a longer time,
the model's parameters, and tests key hypothe- earlier in life, and from a more prestigious
ses. The third section discusses the implications source receive more weight.
of the results for theorizing about race, social Taken together, the individual and group
structure, and delinquency. components of differential association explain
the organizational and learning mechanisms by
which race and family status influence delin-
quent behavior. The learning mechanism (differ-
ential association process) should be invariant
across race, although the context or source of
Our task is to develop a social-psychological that learning, such as parents, peers, or
explanation of the joint relationships among neighborhoods (differential social organization),
race, broken homes, and delinquency. Two may vary by race. For example, if a broken
distinct mechanisms can explain such relation- home impedes parental supervision and attach-
ships. First, race and broken homes could ment, it could indirectly increase a child's
interact in their effects on delinquency: the contact with prodelinquent definitions from
effect of broken homes and other determinants delinquent boys and other influences outside the
of delinquency could be greater among blacks. home (Sutherland and Cressey 1978, p. 219-24;
Second, race could influence delinquency indi- Shaw and McKay 1931). Furthermore, broken
rectly through its effects on broken homes. The homes may hamper the formation of attach-
latter assumes that the effect of broken homes ments to parents (prestige) and the transmission
on delinquency does not vary by race; conse- of antidelinquent definitions from parent to
quently, testing the interaction effect is logically child; thus, the prodelinquent organization of
prior. For this reason, and because prior studies the community or neighborhood would not be
suggest that both measurement and substantive offset by antidelinquent influences within the
processes vary by race, we will examine a home. Since racial segregation often limits
cross-race model of delinquency. Previous blacks to inner-city neighborhoods with low
research suggests that the effects of race and socioeconomic status and abundant definitions
broken homes must be disentangled from the favorable to street crimes (Sutherland and
influences of socioeconomic status and neighbor- Cressey 1978, p. 220), the influence of broken
hood processes (Shaw and McKay 1969; homes on delinquency may be particularly acute
Monahan 1957; Moynihan 1965; Berger and for blacks. The important point is that for both
Simon 1974). Therefore, we need to locate blacks and nonblacks, structural variables such
those intervening social-psychological processes as broken homes and neighborhood organization
explaining such relationships. affect delinquency by influencing the dynamic
According to Sutherland's (1947) theory of process of learning definitions favorable and
differential association, delinquency is rooted in unfavorable to crime.
normative conflict. Modem industrial societies In contrast to differential association, Hir-
contain conflicting structures of norms, behavior schi's (1969) social control theory denies the
patterns, and definitions of appropriate behavior existence of normative conflict and ignores the
that give rise to high rates of crime. At the importance of motives for delinquency, such as
group level of explanation, Sutherland posited prodelinquent definitions. Control theory posits
that normative conflict is translated into group a single conventional moral order in society and
rates of delinquency through differential social assumes that the motivation for delinquency is
organization: the extent to which a group is invariant across persons. The question is not,
organized for or against delinquency determines "Why do some people violate the law?" since
its rate of law violation. This differential we are all equally motivated to do so, but rather,

This content downloaded from on Fri, 19 Jan 2018 21:46:36 UTC
All use subject to http://about.jstor.org/terms

"Why do most people refrain from law tural-level counterpart of bonding as social dis-
violation?" Hirschi's answer is that they are organization-a community's inability to control
dissuaded by strong bonds to conventional the behavior of juveniles because of weak and
society: attachment, commitment, involvement, unlinked institutions-we can hypothesize about
and belief. racial differences in bonding and its determinants
Attachment to others dissuades persons from (Kornhauser 1978; Shaw and McKay 1969). Bro-
delinquency through a moral process: those with ken homes, lower socioeconomic classes, and
warm relationships with their parents or friends high-crime neighborhoods (disorganization)
are likely to consider their reactions to the should influence delinquency by impeding the
unlawful act. Because only a single moral order formation of strong attachments, commitments,
exists, that reaction will always be negative. involvements, and beliefs. Because nonintact
Commitment to conventional lines of action homes undermine parent-child relations, attach-
reflects an investment of time and energy in ment to parents-perhaps the most important el-
procuring an education, developing a business, ement of the bond-should be the principal in-
or building a virtuous reputation. The greater the tervening variable between broken homes and
investment, the less likely the person will delinquency (Hirschi 1969, 1983).1 In turn, at-
jeopardize it by violating the law. Involvement tachments to parents should generalize, allowing
in conventional activities simply limits one's attachments to form among peers and reinforcing
time to contemplate and execute illegal acts. strong moral beliefs. If Moynihan and others are
Finally, belief in the moral order directly taps an correct that blacks are ensnarled in a tangle of
individual's internalization of conventional mo- pathology, then social control theory would claim
rality. Here, Hirschi reconceptualizes Suth- that this pathology is a reflection of disorganiza-
erland's definitions of delinquency to conform tion and that broken homes, social class, and
to the assumptions of control theory: since there neighborhood delinquency will produce more de-
is only one moral order, beliefs concerning linquency among blacks by inhibiting the forma-
delinquency are all conventional, and the greater tion of strong attachments and beliefs.
the belief the less likely the deviation. In sum, control theory and differential
Each of these components of the bond, while association make different predictions of the
intercorrelated, are said to affect delinquency causes of delinquency among black and nonblack
independently and additively (Hirschi 1969, pp. males. Social control theory predicts that, for
27-30). While differential association theory both blacks and nonblacks, delinquency is
implies that attachments, involvements, and determined by the independent effects of the
commitments will affect delinquency only elements of the social bond. Family structure
indirectly through their effects on definitions may affect the elements of the social bond
(belief), control theory maintains that each differently across race, but each element of the
element of the bond itself affects delinquency bond should exert a unique effect on delin-
directly (Jensen 1972; Kornhauser 1978; Mat- quency for both races. The relative importance
sueda 1982). of these bonds, however, may vary across race,
Control theory implies that the causes of due to a different emphasis on socialization
delinquency (social bonding) are the same for all practices, which in turn stems from social
racial groups (Hirschi 1969, p. 80). The theory disorganization. Differential association, how-
would receive strong support if the absolute ever, predicts that, for both blacks and nonblacks,
effect on delinquency of each element of the delinquency is. determined by learning defini-
bond were identical for all races. This would tions of the legal code (beliefs), which mediate
imply that the theory describes a deep invariant the influence of attachments, commitments, and
structure that persists in the face of racial involvements. The sources of that learning,
segregation and discrimination. But confirming however, are determined by individuals' group
control theory may not require such invariance, location in the social structure, which organizes
instead requiring only that the elements of the their patterns of interactions, and which may
bond explain the probability of delinquent differ by race. This implies that the determinants
behavior. Thus, we might expect socialization of a person's learned definitions, such as being
practices or belief systems to vary across racial from a broken home, a trouble-ridden neighbor-
groups, causing attachment, commitment, in- hood, a close family unit, or a delinquent peer
volvement, and belief to affect delinquency
differently by race.
l Hirschi (1969, 1983) has argued that single-parent
Furthermore, the relative strength of structural
families should have similar rates of delinquency as intact
determinants of social bonding may also vary by
families, since, all things being equal, one parent should
race. Here, we are on less-solid ground, since be as effective as two in socializing children. Neverthe-
Hirschi (1969, p. 113) had little to say about less, all things are never equal, and logically, for social
factors affecting the strength of elements of the control theory, if broken homes influence delinquency,
bond. Nevertheless, if we conceptualize the struc- they do so by attenuating the elements of the social bond.

This content downloaded from on Fri, 19 Jan 2018 21:46:36 UTC
All use subject to http://about.jstor.org/terms





Differential Association and//I \\\
Social Control Theories1 / /I \\ \
- - - Social Control Theory Only FRPI KUP
---Neither Social Control nor

Fig. 1. Path Diagram of the Full Structural Equation Model of Delinquency

group, may vary across race, but the determi- 1,588 nonblack males and 1,001 black males.2
nant of delinquency-an excess of definitions Self-report measures, described in Appendix A,
favorable to delinquency-will not. were obtained through questionnaires adminis-
tered in schools.3 (For further details of the data
collection procedures and characteristics of the
A CROSS POPULATION MODEL OF RACE sample, see Hirschi 1969.)
AND DELINQUENCY Our causal model of delinquency, depicted in
Figure 1, consists of a substantive model of the
Our investigation analyzes Matsueda's (1982) mechanisms generating delinquent behavior and
causal model of differential association, control a measurement model of the process by which
theory, and delinquency by replicating the model underlying substantive concepts generate observ-
on the black population of the Richmond Youth able measures. The measurement model, indi-
Project. We first examine whether the model as cated by the paths connecting latent variables to
a whole varies across race, then test key hypoth-
eses about substantive parameters both within
and across groups. We examine two substantive 2 The response rate for nonblacks was 75 percent, for
issues: (1) the model's ability to explain the in- blacks, 68 percent. Hirschi (1969) examined potential
bias due to nonresponse, finding that nonresponse was
fluence of family structure on delinquency; and
evenly distributed among permission denied by parent,
(2) the relative efficacy of differential associa-
no response by parent, transfers and dropouts, and
tion versus social control theory.
absentees. Furthermore, while respondents were less
The data were collected in 1965 as part of the likely than nonrespondents to have a police record, this
Richmond Youth Project, which sampled a large effect did not vary much by race. Therefore, nonresponse
number of students in 11 junior and senior high should not bias our cross-population results appreciably.
schools of Contra Costa County in California (Upon request, covariance matrices of observable vari-
(Wilson 1965). These data are particularly well ables are available from the authors.)

suited to the issues at hand: 1965 marked the 3 In using a sample stratified by race, estimating
separate models for nonblacks and blacks, and fixing the
publication of Moynihan's report; the population
validity coefficient of self-reported delinquency to be
is a large heterogeneous metropolitan area
larger for nonblacks than blacks, we are following the
containing substantial numbers of lower- recommendations of Hindelang et al. (1981), who argue
income, inner-city blacks; and the random that, after taking these steps, self-reports of minor forms
sample was stratified by race, as well as school, of delinquency are reasonably reliable and valid for
sex, and grade. Our analyses will focus on the testing theories.

This content downloaded from on Fri, 19 Jan 2018 21:46:36 UTC
All use subject to http://about.jstor.org/terms

observable indicators, allows us to estimate and broken homes and other background variables
control statistically for the biasing effects of mea- on delinquency may differ by race. For
surement error in substantive constructs.4 Such a example, as some previous research has found,
model can be crucial for cross-population anal- broken homes may exert a larger effect on
yses because it can reveal differential measure- delinquency for blacks than for nonblacks.
ment processes across populations, which, if not Whatever the magnitude, however, social con-
dealt with, can obscure cross-population com- trol and differential association theories specify
parisons. Therefore, before we proceed to our intervening mechanisms to account for the total
hypotheses derived from differential association effects. The most significant hypotheses for
and social control theory, we will examine our social control theory are that attachment to
measurement models for the two populations. parents, attachment to peers, and belief in
The substantive component of our model morality each have a direct effect on delin-
consists of three blocks of variables: four quency and together should mediate the influ-
exogenous background variables describing de- ence on delinquency of background characteris-
mographic characteristics of individuals, four tics such as broken homes, age, SES, and
intervening variables representing the social neighborhood trouble. These hypotheses should
control and differential association processes, hold equally for blacks and nonblacks. Further-
and an outcome variable of self-reported delin- more, control theory allows the relative effects
quency.5 We specify the intervening variables- of these variables on delinquency to differ by
parental supervision (attachment to parents), race, reflecting, for example, differential social-
delinquent friends, attachment to peers, and ization practices across racial groups. The
definitions of delinquency-as linear functions foregoing can be expressed as two hypotheses:
of our background variables: age, socioeco-
HYPOTHESIS 3. The effects on delinquency of
nomic status, broken homes, and neighborhood
broken homes and the other background
trouble (see Figure 1). In both social control and
variables are mediated by variables repre-
differential association theories, these effects,
senting social bonding.
representing the influence of social structure on
HYPOTHESIS 4. Attachment to parents, attach-
socialization processes, may vary by race.
ment to peers, and belief all have signifi-
HYPOTHESIS 1. The effects of background cant effects on delinquency.
variables, including family structure, on
In contrast, the crucial proposition of differ-
social bonding (attachment and belief) vary
ential association theory is that the effects of
by race.
definitions of delinquency on delinquent behav-
In addition, differential group organization ior should be racially invariant and, for both
predicts that attachment to parents and peers and blacks and nonblacks, should mediate the
delinquent friends may influence definitions effects on delinquency of all other variables (see
differently by race: Figure 1). The antecedent variables, including
background characteristics and other elements of
HYPOTHESIS 2. The effects of background
the social bond, reflect elements of social
variables and parental and peer processes
organization that structure the differential learn-
on definitions of delinquency vary by race.
ing of behavior patterns. Consequently, if age,
According to both theories, the total impact of broken homes, or parental supervision have
large total effects on delinquency, it is because
they represent an important source of learning
4 The measurement model of definitions of delin- definitions of delinquency. These propositions
quency conceptualizes Sutherland's concept of a ratio of translate into two testable hypotheses:
definitions favorable and unfavorable to delinquency as a
unidimensional construct, which generates fallible indica- HYPOTHESIS 5. For both blacks and nonblacks,
tors. Each indicator, measured on a single continuum a person's learned ratio of definitions
from highly antidelinquent to highly prodelinquent, is mediates the effects of other antecedent
assumed to capture one domain of the ratio of definitions.
variables in the model, including the effect
After controlling statistically for response errors, the
of broken homes.
common variation across our measures should adequately
HYPOTHESIS 6. The effect of definitions of the
tap such a construct (see Matsueda 1982 for details).
5 The causal ordering among our variables within a law on delinquency is racially invariant.
cross-sectional design follows previous research using
these data (Hirschi 1969; Jensen 1972; Matsueda 1982).
This ordering is consistent with both differential RESULTS
association theory and social control theory. Some recent
We estimated the parameters of our measure-
evidence on this issue within a longitudinal framework
confirms the causal priority of attachment to parents on
ment and substantive models jointly as a single
delinquency (Liska and Reed 1985; Agnew 1985) and system using the maximum likelihood estimator
definitions of delinquency (belief) (Agnew 1985; Elliott, of Joreskog's LISREL V program (Joreskog and
Huizinga, and Ageton 1985), but see also Minor (1984). Sorbom 1984). Our analysis of the measurement

This content downloaded from on Fri, 19 Jan 2018 21:46:36 UTC
All use subject to http://about.jstor.org/terms

models, discussed in Appendix B, reveals larger total effects of age, broken homes, and
measurement errors for blacks than nonblacks. neighborhood trouble are mediated by the joint
Thus, the failure to correct for attenuation due to effects of attachment to parents, delinquent
unreliability could lead to greater downward friends, attachment to peers, and moral beliefs
biases in regression coefficients among blacks (line 15). Thus, being older, from a broken
than nonblacks. Overall, the model fits better home, and from a troubled neighborhood
for nonblacks (L2 = 129.04 d.f. =71) than increases the likelihood of delinquency by
blacks (L2 = 216.22; d.f. = 71).6 Both findings attenuating attachments to parents and peers,
are consistent with other similar response increasing the number of delinquent friends, and
models (Bielby et al. 1977). reducing the strength of conventional beliefs.
Nevertheless, our fourth finding, which ad-
dresses our crucial test of differential association
Estimation of the Model for Nonblacks
versus social control theory (Hypothesis 4),
Our discussion of the substantive model will fo- supports differential association theory. Both
cus on the above six hypotheses. We first high- attachment to parents and peers have substantial
light the results for nonblacks, then present the and statistically significant indirect effects on
findings for blacks in more detail, emphasizing delinquency through definitions. Moreover, the
differences across race. Table I presents the un- remaining unmediated direct effects of the
standardized parameter estimates of our baseline attachment variables are not only nonsignificant
model for nonblacks in their reduced, semi- and small in magnitude, but, from the stand-
reduced, and structural forms; their standardized point of social control theory, implausibly
counterparts appear in Table 2. These estimates positive in sign (line 15 of Table 1). Thus, as
reveal four principal findings. First, the model differential association predicts, youths who are
explains substantial variation in definitions of de-closely supervised and develop warm friend-
linquency (R2 = .66). Friends picked up by the ships commit fewer delinquent acts because they
police, attachment to peers, and parental super- are exposed to fewer prodelinquent definitions.
vision exert substantial direct effects on the In addition, none of the background variables
learned ratio of definitions (line 10 of Table 2) has a significant direct effect on delinquency in
and also mediate the effects of certain back- the structural form (Hypothesis 5). The number
ground variables. More precisely, the total effect of delinquent friends, however, does have a
of neighborhood trouble is mediated by supervi- substantial and statistically significant influence
sion and delinquent friends (compare line 7 with on delinquency.7 This direct effect is smaller
lines 8 and 9). Thus, living in a troubled neigh- than the effect of definitions, and about as large
borhood exposes nonblacks to more delinquent as the indirect effect of delinquent friends
definitions by attenuating parental supervision and through definitions of delinquency; neverthe-
increasing the number of delinquent compan- less, the result provides some negative evidence
ions. Also, the total effect of broken homes on for differential association.8 Although irrelevant
definitions, modest in size but statistically sig- to the debate between Hirschi and Sutherland,
nificant, is mediated by parental supervision. the finding supports a group process explanation
Second, the model also does well in explain-
ing variation in delinquent behavior (R2 = .56):
every variable in the model except socioeco- 7 This is the only finding inconsistent with Matsueda
nomic status has a significant total effect on de- (1982), who found that the influence of delinquent
linquency. The largest total effect is exerted by friends on delinquent behavior was mediated by defini-
delinquent friends, followed by definitions of de- tions. The discrepancy between our model for nonblacks
linquency, supervision, and neighborhood trou- and Matsueda's (1982) is due to a different method of
handling missing values. Here, to insure comparability
ble. Broken homes exert a small but statistically
with the sample of blacks, we used pairwise deletion for
significant total effect.
nonblacks, while Matsueda (1982) used listwise deletion.
Consistent with social control theory (Hypoth- We also estimated cross-population models using listwise
esis 3), our third finding is that the significant deletion for both groups, and, while the sample size was
reduced by 40 percent, the results were identical for
blacks. Thus, missing values do not substantially
6 This holds even though we have a larger sample ofinfluence the overall pattern of results.
nonblacks, and, thus, greater statistical power to detect 8 This direct effect of delinquent friends on delinquent
departure from the hypothesized model. For comparabil- behavior also results in three variables having indirect
ity, the model for blacks includes the same measurement effects on delinquency through delinquent friends, not
error correlations specified by Matsueda (1982) for the definitions of delinquency. The effects are modest in
nonblacks-some of which were nonsignificant. A size, however, particularly in comparison to similar
sensitivity analysis, however, revealed that a better- indirect effects through definitions. The relative indirect
fitting model did not alter the substantive picture in anyeffects through delinquent friends and definitions,
meaningful way. Thus, it appears that the overall respectively, are: .07 and .08 for age, .09 and .19 for
goodness-of-fit statistic is sensitive to trivial departures neighborhood trouble, and -.07 and -.20 for supervi-
from uninteresting restrictions. sion.

This content downloaded from on Fri, 19 Jan 2018 21:46:36 UTC
All use subject to http://about.jstor.org/terms

- r- W t r t r - c o c c o
N 1 >>^^

CO 0 0 t- 00 ,At 0) 00 WI ~ - '
C- i c 00 ci C- ci1 ON 0 c

o .Y
~O 0ON 0
Wot 0
0 c'N
O N00 00
00 o0
cN Q
0 ^' C'-
N r-
?-t 00
0 C'-

ON 00 ci ON - 00 C" Ot 0 b 00 bC N0 C
- -m 0 0 - c e'N N0 XC - - ?t ?

9 ci 00
C--4 00 r~l 0'IC
cl~ M0\O c 00 --00
00 C- '-~ c

ce, O O M ,At~~~~~~~~~~~~~~0 O

o oo N o oo > t o i'oON (10

00 o N 8 ooN oo 00
rq O

ON~~~O ONONlON ON O n -'(C r- \O

0.0. 0. 0. )0. t0. t0.

2 ~ C> - -o eoo co io o o cO ci 0-mF
~~~~~~ ~~~~~~~~~C/) I ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ t e l)Cf )W

u~~~~ ~~~~ rl kn Nt 0t 00 W)\ t ent r- C\ kn C' N


O ~ C"-0 ~0ciCOC'\.00ON0C cqiCO Dci-ci-VC"-ON00 00 - 00 0 cO

c: s_ o N Nq \C t \C r- ( ) (I M (1 oo cq C1 0 oo A - cq \C It N It M I ^t W) t
-~~~~~~~ ~~~~~~0 III ~ ~ ~ ~ ~~~q0

E) 4 O I I I I

a)c oN 0. omo o ootooo^Nc o^o

4-4 "orzot o0 ( \. o. WI o. o0 on o on o o oN oN o. ON o ON oq o, o M o0

C) Z

oo- 0- C
CO 0~ N - 00000- ooC\e
^^oo l\c qc 00
oto coci-o-
Eboo 00


C - Ct ; 0ON C C - '(oN CON (o0 ci0 c -O

v ~ ~ ~ ~~~~~- 0

- C C) CL

C)COC/)< <: <: a

CO C) C ci C^ .t '( ?O C" 00 ON 0 i N ^ t m
F-' W

This content downloaded from on Fri, 19 Jan 2018 21:46:36 UTC
All use subject to http://about.jstor.org/terms

Table 2. Standardized Parameter Estimates of the Substantive Model: Nonblack Males (N = 1,558)

Predetermined Variables

Dependent BROK- YOUNG-


1. SUPER -.123 .017 -.140 -.210

2. FRPICKUP .208 -.033 .010 .278
3. FRPICKUP .178 -.029 -.025 .225 -.250
4. ATTACHPE .106 .092 .037 -.193
5. ATTACHPE .132 .088 .066 -.150 .206
6. ATTACHPE .180 .080 .059 -.089 .138 -.271
7. DEF .160 -.063 .109 .384
8. DEF .109 -.056 .051 .296 -.419
9. DEF .011 -.040 .064 .173 -.282 .548
10. DEF .054 -.022 .078 .152 -.249 .484 -.235
11. DEL .108 .000 .090 .311
12. DEL .075 .005 .052 .255 -.267
13. DEL -.030 .022 .067 .123 -.120 .589
14. DEL -.018 .027 .071 .117 -.111 .572 -.063
15. DEL -.044 .037 .033 .043 .011 .336 .051 .487

of delinquency (Short and Strodtbeck 1965; our substantive variables and then the causal
Briar and Piliavin 1965). structure explaining those total effects. Note that
delinquent friends have a slightly larger total
effect in our model for nonblacks than for
Estimation of the Model for Blacks
blacks. Perhaps the most striking racial differ-
Parameter estimates of our substantive model ence, however, is in the reduced-form effects of
for blacks appear in Table 3 in unstandardized broken homes and neighborhood trouble: the
form, and Table 4 in standardized form. Our former is three times larger among blacks, while
discussion will focus on our cross-population the latter is five times smaller. Thus, consistent
hypotheses. To test these hypotheses, we use with much previous research, broken homes are
likelihood-ratio statistics, which are distributed more influential in producing delinquency among
chi-square in large samples and are obtained by blacks than nonblacks.
subtracting the pooled likelihood-ratio statistic Paralleling our findings for nonblacks, we
of our baseline model (L2 = 345.26; d.f. = find that, along with delinquent peers, the
142) from that of the model with cross-group elements of the social bond-attachments to
constraints. Using the overall test of invariance, parents and peers and belief in morality-
we reject the hypothesis that all substantive collectively mediate the influence of our back-
parameters are the same for blacks and nonblacks ground variables on delinquency (line 15). The
(L2 = 427.79; d.f. = 82; p < .001) and then indirect effects of age, broken homes, and
proceed to more specific cross-group compari- neighborhood trouble are substantial, while the
sons. Our first comparison hypothesizes that the remaining unmediated effects are either trivial in
determinants of the processes of social bonding size (broken homes) or opposite in sign than
and differential association vary by race. For the anticipated (age and neighborhood trouble).
social control process, the effects of background Again, this is consistent with social control
variables on elements of the social bond theory (Hypothesis 3).
(Hypothesis 1) appear invariant across groups We can assess Hypotheses 4 and 5, which test
(L2 = 20.13; d.f. = 12; p > .05). For the control theory against differential association,
differential association process, however, we by comparing lines 11-15 in Tables 3 and 4.
find that the determinants of definitions of the Line 14 reveals that before adding definitions of
legal code vary by race (L2 = 18.37; d.f. = 7; the legal code into the equation, our model
p < .01). This finding (Hypothesis 2) is due accounts for a substantial amount of variation in
primarily to the larger effects of broken homes, delinquency (R2 = .31). Thus, our test of
parental supervision, and neighborhood trouble differential association-the extent to which
on the process of learning definitions among definitions mediate the effects of other variables
blacks (compare line 10 in Tables 3 and 4). on delinquency-is a strong one, since substan-
Thus, from the standpoint of differential associ- tial total effects must be mediated. As noted
ation, the neighborhood and family organization above, in the black sample, the reduced-form
of blacks is most telling in the process effect of broken homes on delinquency is
producing definitions of delinquency. substantial (line 11 of Table 4), as is the
Turning to the equations predicting delinquent semi-reduced form (line 14). Before adding
behavior, we first examine the total effects of definitions into the equation, then, broken

This content downloaded from on Fri, 19 Jan 2018 21:46:36 UTC
All use subject to http://about.jstor.org/terms

-z ~J n if) 0) 0 0) CA CA CA CA In) In) In) I) I

00 O I Ct CA C1 C C CC-
o CC-C- CC CC- In m 0 m 0 0 0 0 00

OC if m OC m O N kr 00 V m m

t *'e oo x t ? o ~~~~N CP ff r-- C\ m?

7e M C C) m r-- Mo ONmo

> Cl ?t 0 0 - CC CC t Nt Co CC mf ot In

o e N x m ~~~~~~~~~~~~~~~t 1r) 1 xot

} -J 00 In CC- 0 c CC- ON In 00 In N C-- In
vO >In
) ; o
n m m O IN 0 ( omn
t00 \C 0 00
In r-
Cl 1s
I vt


r \ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ooN m osmCC-
~~ 00 00 ~~~t ~ \C 0 \C 0 CA
0C nC m
C- Oa\v)
CAl0 v)
)\ In cn m
n In\

Z ON O CC 00 C- InOO

O - Cl 00 I NM t 00 In C-- CC 00 ON \ t- \ c \C\C- \C -

> I
t ~~~~~~0O 000 ON- -O 00 o)tc -C S- ^
n ; o O > ? ~~~00C InC ooI~ 00I InC CC-C--to ?m
11~~~~~0 -t - Cl O CO O- o)c)_c

c) HClO OO?N O O ONO CCC oI- o IN t ON In C- 0 >O- - CC?

CO l I0 00 I O- If In C -I CI- C- -~

CC CO C In CCM~ C)l 0 O1 tN CCN O)N C- ON M -MN 0 In

70 :.)) I C C C-U

o W 0 I nl O mN CC-ON Inl ~tl Cm ot~ 0I oot C- t C-C- CI)C- O- -C--o000


U t ? m o oU U 7

0J 0 C) 0:3 06 0~~~~~~~~~~~~~C-6 CC- C~ 0Cl

O o 000 C- COC -n Cl0 C-0ON oo~ > oa ntl ~t t C-t >0C 0I OC- oNmo n
- -I 2t~ COl m~ CN 00C OC O 00 ONCO OC- C CC- O m m

m O C CCC -C 00 00 00 In / t C- / f CO In In In--Cl

CO CCI Cl NIn Cl Cl Cl C C-CC- t Cl Qt Cl In0 Ct ?C t C-t 00 t Cl n

s C)

01 I I /I yI I /.

N c

r -C)CO 0. X H H H ;' ;' ;' ;I C- I- C- C- 1- ..

s Q-O0 cT ; < < ? 0 0 0 01 0 0 0 0 0
- Col CC- ~ t In) C C- 00 2 ON 0 - Cl CC-

This content downloaded from on Fri, 19 Jan 2018 21:46:36 UTC
All use subject to http://about.jstor.org/terms

Table 4 Standardized Parameter Estimates of the Substantive Model: Black Males (N = 948)

Predetermined Variables

Dependent BROK- YOUNG-


1. SUPER -.070 .039 -.180 -.135

2. FRPICKUP .162 -.039 .106 .136
3. FRPICKUP .141 -.027 .051 .094 -.306
4. ATTACHPE .006 .121 .043 -.125
5. ATTACHPE .019 .114 .076 -.100 .183
6. ATTACHPE .062 .106 .092 -.071 .089 -.306
7. DEF .228 .097 .348 .372
8. DEF .197 .115 .268 .312 -.444
9. DEF .151 .123 .252 .281 -.345 .323
10. DEF .165 .146 .272 .266 -.325 .257 -.216
11. DEL .006 .101 .291 .074
12. DEL -.014 .112 .241 .036 -.280
13. DEL -.068 .122 .221 .000 -.162 .385
14. DEL -.059 .137 .234 -.010 -.150 .342 -.-139
15. DEL -.164 .044 .062 -.178 .056 .180 -.003 .632

homes have a large and significant effect on the total effect of attachment to peers is small
delinquency. After adding definitions, however, and statistically nonsignificant, the indirect
that effect becomes trivial in size and statisti- effect through definitions is significant, render-
cally indistinguishable from zero (line 15 of ing the direct effect on delinquency virtually
Table 3). As differential association predicts, nonexistent. Differential association is again
broken homes influence delinquency by imped- supported over control theory.
ing the transmission of antidelinquent defini- Finally, we tested Hypothesis 6, derived from
tions and increasing the transmission of prodelin- differential association theory, which postulates
quent patterns. Similarly, in accord with control that the effect of definitions on delinquency is
theory, attachment to parents (supervision) has a invariant across race. That test confirmed the
large total effect on delinquency that works hypothesis: the point estimates are indistinguish-
partly indirectly through delinquent friends and able from one another at conventional levels of
partly directly before adding our definitions significance (L2 = .17; d.f. = 1; p > .50).
variable. But the structural form (line 15) Thus, differential association theory again
reveals that, after adding definitions into the receives strong support.
equation, the effect of supervision becomes
nonsignificant, and, from the standpoint of
control theory, implausibly positive. Again, this
is consistent with differential association theory: For both black and nonblack samples, our
supervision influences delinquency by influenc- models support differential association theory
ing the ratio of learned definitions of delin- over social control theory. Contrary to Hirschi's
quency. Furthermore, delinquent friends exert a (1969) postulate that each element of the social
large and significant effect on delinquency bond shows a unique and substantial effect on
before adding definitions, but a comparatively delinquency, we find that the effects of
small and nonsignificant effect in the presence attachment to parents and peers operate indi-
of our definitions construct (compare lines 14 rectly through the process of learning an excess
and 15). Thus, in contrast to our findings for of definitions favorable to delinquency. This
nonblacks, delinquent friends influence delin- finding is consistent with differential association
quency by presenting definitions of the legal theory, as are the findings that across racial
code; this finding supports differential associa- groups, the effect of definitions on delinquency
tion theory over group process theories.9 While is invariant, and within groups, definitions
mediate the influence on delinquency of our
other explanatory variables.
I The nonsignificance of this parameter estimate for
Of more interest are the differences between
blacks could be due to type II error, given the smaller our models for blacks and nonblacks. The most
size of the black sample. To investigate this, we
conducted a power analysis, following the recommenda-
tion of Matsueda and Bielby (1986). We found that the (power = .50), we can detect a standardized coefficient
model for blacks had ample statistical power (.95) to of .23 (power = .65). Thus, we have sufficient
detect a metric coefficient the size of the estimate for protection against type II error, assuming a nontrivial
nonblacks. But, although we cannot detect, with (larger than .20) effect of delinquent friends on
reasonable power, a standardized coefficient of .20 delinquency in our black population.

This content downloaded from on Fri, 19 Jan 2018 21:46:36 UTC
All use subject to http://about.jstor.org/terms

striking difference is that the total effect of each category of family status led to multicol-
broken homes on delinquency is much larger for linearity and unstable estimates. Other research
blacks than nonblacks. Yet in both racial groups suggests that the etiology of the break, such as
nonintact homes influence delinquency through death, divorce, or desertion, can influence
a similar process -by attenuating parental super- subsequent behavior (McLanahan 1985). Further-
vision, which in turn increases delinquent more, many argue that the pertinent variable is
companions, prodelinquent definitions, and, marital and familial discord, which could have
ultimately, delinquent behavior. But to a much an adverse effect on intimacy, supervision, and
greater extent, broken homes directly foster an the transmission of antidelinquent behavior
excess of definitions favorable to delinquency, patterns, and which could also cause a marital
which then increases delinquent behavior. This breakup. Since marital severance is also likely
effect, being much larger among blacks, ac- to cause discord, cross-sectional research de-
counts for the greater total effect of broken signs are likely to confound the causes and
homes on delinquency among blacks. consequences of family disruption. Longitudinal
A second racial difference is the total effect of designs are needed to disentangle the reciprocal
neighborhood trouble on delinquency, which is effects of family process, family structure, and
much larger in the model for nonblacks. Among delinquent behavior. Based on the results of our
nonblacks, that effect works partly through models, we expect that the key intervening
delinquent friends, but largely through defini- mechanism explaining the effects on delin-
tions of delinquency; among blacks, a large quency of such family processes is the learning
indirect effect operates solely through defini- of delinquent and antidelinquent definitions.
tions. We also examined an interaction hypoth- But the link between definitions of delin-
esis between neighborhood trouble and broken quency and social structure may be more
homes: Do broken homes influence delinquency complex than implied thus far and may suggest
only in the context of a trouble-ridden, high another empirically testable divergence between
delinquency neighborhood? Entering a product the theories of differential association and social
variable representing the interaction effect, we control. More precisely, social control theory,
found evidence of a conditional effect among based on a consensus model of social order,
blacks but not nonblacks. Blacks from broken denies the efficacy of competing subcultural
homes who also live in troubled neighborhoods norms and assumes that only conventional
are more likely than those residing in trouble- norms and definitions of morality influence
free neighborhoods to associate with delin- behavior. In contrast, differential association
quents, learn an excess of definitions favorable theory, based on a group conflict model of
to delinquency, and, consequently, violate the social organization, specifies that subcultural
law. We were unable to locate such an groups may differ on two dimensions of
interaction in the nonblack model, perhaps due definitions of delinquency-the dimension of
to multicollinearity among main and interaction the weight of the definition, and, more impor-
effects. 10 tantly, the dimension of the meaning or content
We should note that, following a long history of the definition (Matsueda 1982). The latter
of research on family structure and delinquency, implies that groups located at different junctures
we have used a single dichotomous variable to in the social structure may communicate and
distinguish intact from nonintact homes. Re- behave according to very different definitions of
cently, some have argued that the impact of unlawful behavior. In particular, the content of
family structure on delinquency may vary definitions of delinquency may vary by race,
depending on the nature of that structure, such neighborhood, and social class. 11 To explore
as whether a step-parent is present or whether this issue, researchers must first use in-depth
the mother or father is absent (Rankin 1983; interviews to induce the content of such
Johnson 1986; Wells and Rankin 1986). We definitions for distinct communication groups
were unable to examine the joint relationships and then develop empirical measures to tap such
among different forms of family structure, our definitions. Structural equation methods within
intervening variables, race, and delinquency
because the small number of cases falling into
" We attempted to explore this inductively using the
Richmond data. That is, with our confirmatory factor
10 Large bivariate correlations between product vari-
models of definitions, we examined the possibility that
ables and their constituents suggest the problem of some indicators that are valid for nonblacks are invalid
multicollinearity in disentangling interaction effects from for blacks, and vice versa. By and large, we did not find
main effects in both samples. Thus, we treat these results such differences in validity across race; what was a strong
with caution. We also failed to unearth interaction effects indicator in one population was generally strong for the
among SES, broken homes, and neighborhood trouble, other. This is not surprising, of course, since the
which was expected, given the null effects of SES on the measures are global, and designed to apply across general
endogenous variables. populations.

This content downloaded from on Fri, 19 Jan 2018 21:46:36 UTC
All use subject to http://about.jstor.org/terms

the LISREL framework exist for making however, such theorizing is speculative. Never-
cross-population comparisons when indicators theless, such speculation is consistent with our
for concepts differ across groups (Allison 1985). principal findings that the influence of broken
Viewed in broader perspective, our results homes on delinquency is greater among blacks;
raise larger questions concerning the role of that this influence is explained by the process of
social structure on race, cultural norms, and learning definitions of delinquency; and that, for
delinquency. That is, given that delinquency is both blacks and nonblacks, differential associa-
largely determined by the learning of definitions tion theory is supported over social control
of the legal code, what are the wider structural theory.
determinants of that learning process? Our
ability to explain remarkably large amounts of
variation in definitions with a small number of
variables suggests that such a learning process is Key to Variable Labels

tightly structured. When examining a single AGE: Age of respondent. 0 = 12 years or younger, 1
cross-section of individuals, we find that the = 13 years, 2 = 14 years, 3 = 15 years, 4 = 16 years,
learning process is structured by elements of 5 = 17 years, 6 = 18 years, 7 = 19 years, 8 = 20 years
social organization such as age, neighborhoods, or older.

families, and peers. Moreover, the differential BELIKFR: "Would you like to be the kind of person
your best friends are?" 0 = not at all, 1 = in a few
impact of these structures accounts, in large
ways, 2 = in most ways.
part, for racial disparities in delinquent behav-
BROKHOME: A dummy variable coded as one if
either the mother or the father did not live with the
From both a scientific and policy standpoint, respondent.
a more significant issue may be the historical DEL: An index of delinquency committed during the
emergence of social and economic structures last year containing the following six items:
that give rise to distinct racial patterns of social
organization. Thus, the racial cleavages in BATTERY: "Not counting fights you may have had
with a brother or sister, have you ever beaten up on
normative definitions of delinquent behaviors
anyone or hurt anyone on purpose?"
may derive from a history of restricted opportu-
CARTHEFT: "Have you ever taken a car for a ride
nities, a sense of resignation, and, ultimately,
without the owner's permission?"
new ways of adapting to a bleak situation THEFT2: "Have you ever taken little things (worth
(Cloward and Ohlin 1960). For example, less than $2) that did not belong to you?"
William Julius Wilson (1985) argues that THEFT250: "Have you ever taken things of some
increasing social dislocations among the urban value (between $2 and $50) that did not belong to
underclass were a culmination of a number of you?"

demographic, economic, and cultural changes. THEFT50: "Have you ever taken things of larger
value (over $50) that did not belong to you?"
Specifically, the increasing disparity in crime
VANDALSM: "Have you ever banged up something
across race is a result of historic not contempo-
that did not belong to you on purpose?"
rary discrimination, the unabated migration of
Southern blacks to the centers of Northern
DELHURT: "Most things that people call 'delin-
cities, the drop in age structure among inner-city quency' don't really hurt anyone." Strongly disagree,
blacks, and a general economic shift from a disagree, undecided, agree, strongly agree.
manufacturing to a service economy. In turn, EVNBREAK: "Policemen try to give all kids an even
these broad historical trends have led to different break." Strongly agree, agree, undecided, disagree,
patterns of social organization among the urban strongly disagree.
underclass, which influence rates of delin- FRPICKUP: "Have any of your close friends ever

quency. For example, we have shown that been picked up by the police?" 0 = no or don't know, 1
= one friend has, 2 = two friends have, 3 = three
delinquency is in part spawned by broken
friends have, 4 = four or more friends have.
homes, unsupervised family life, ineffective
GETAHEAD: "To get ahead, you have to do some
neighborhood organization, and, ultimately, things which are not right." Strongly disagree, disagree,
differential association. If this historical expla- undecided, agree, strongly agree.
nation is correct, and the critical learning OKLAW: "It is alright to get around the law if you can
process is indeed interwoven in the fabric of get away with it." Strongly disagree, disagree, unde-
such historical trends, it should be no surprise cided, agree, strongly agree.
that simplistic policies of rehabilitation and PARWITH: A composite asked regarding each parent:
"Do your parents know who you are with when you are
deterrence have failed to stem the tide of rising
away from home?" 0 = never-never, 0.5 = sometimes-
rates of delinquency. Sweeping social and
never, 1.0 = sometimes-sometimes, 1.5 = usually-
economic reforms may be necessary to reverse
sometimes, 2.0 = usually-usually.
the strong currents of law violation (Wilson PARWHERE: Same as above but with the question:
1985). "Do you parents know where you are when you are away
In the absence of a substantial body of from home?"
empirical research verifying these propositions, RSPECTFR: "Do you respect your best friend's

This content downloaded from on Fri, 19 Jan 2018 21:46:36 UTC
All use subject to http://about.jstor.org/terms

opinions about the important things in life?" 0 = not at reflect a validity of .95, and fixed the error variances of
all, 1 = a little, 2 = pretty much, 3 = completely. SES, BROKHOME, YOUNGTRO, FRPICKUP, and
RSPECTPO: "I have a lot of respect for the Richmond DEL to reflect a validity of .80 for nonblacks and .70 for
police." Strongly agree, agree, undecided, disagree, blacks. These values follow previous research, which
strongly disagree. finds larger measurement error variances among blacks
SES: Father's occupation measured on the Duncan than nonblacks for measures of socioeconomic status
Scale; if there is no father living in the home, mother's (Bielby et al. 1977) and delinquency (Hindelang et al.
occupation is used. For the few cases in which father's 1981). A sensitivity analysis varying the validity
occupation had a missing value, and father's education coefficients from .95 to .60 for nonblacks, and from .85
was reported, values of father's occupation were to .60 for blacks, did not appreciably change our
predicted by regressing occupation on education. substantive parameter estimates.
SUCKERS: "Suckers deserve to be taken advantage Column 5 presents the metric (unstandardized) slopes
of." Strongly disagree, disagree, undecided, agree, of our measures. We found that the metric slopes for
strongly agree. indicators of parental supervision and attachment to peers
TROUBLE: "I can't seem to stay out of trouble no are statistically indistinguishable across groups (lambda
matter how hard I try." Strongly disagree, disagree, invariant) (13 = 5.06; d.f. = 2; p > .05), a finding that
undecided, agree, strongly agree. allows us to make straightforward cross-population
YOUNGTRO: [In my neighborhood] "Young people comparisons of metric coefficients involving supervision
are always getting into trouble." Strongly disagree, and attachment. Because we have to normalize our
disagree, undecided, agree, strongly agree. indicators (here, by fixing the metric slope of one
indicator of each construct to unity), only the ratios of
APPENDIX B metric slopes are identified. Consequently, lambda
invariance literally means that the ratios of metric slopes
Analysis of the Measurement Models
of a given construct are invariant across groups (Bielby
Two issues concerning our measurement models require 1986). We also found the two metric slopes of
attention: (1) whether, as previous studies suggest, blacks supervision statistically indistinguishable within race
respond with greater random variation; and (2) whether (tau-equivalent) (13 = 5.43; d.f. = 2; p > .05), but
the metrics of latent variables appear equivalent across rejected tau-equivalence for indicators of attachment to
groups, allowing straightforward cross-population com- peers (13 = 7.38; d.f. = 2; p < .05).
parisons. Parameter estimates of our measurement For indicators of DEF, however, metric slopes appear
models appear in Table Al. Column 2 reveals that, with dissimilar across groups (13 = 19.42; d.f. = 6; p <
the exception of age, the means of our observable .005) and across indicators (L2 = 81.96; d.f. = 12; p <
variables all differ significantly across race. On average, .001). Relative to the reference indicator (TROUBLE),
nonblacks have higher socioeconomic status, fewer most of the other indicators in the black population have
broken homes, less troubled neighborhoods, more flatter slopes. Thus, relative to TROUBLE, blacks
parental supervision, fewer delinquent friends, more scoring high on the other indicators tend to understate
attachment to peers, a lower ratio of definitions favorable their true definitions of delinquency, and those scoring
and unfavorable to delinquency, and fewer self-reported low tend to overstate. By and large, the opposite holds
delinquent acts. Our analyses leave the observable means for nonblacks: relative to TROUBLE, most of the other
unconstrained. indicators have steeper slopes. This suggests that blacks
Column 3 reveals that the observed indicators show use slightly different metrics than nonblacks in interpret-
more variation among blacks than nonblacks. This is dueing the Likert-scale indicators of definitions. In turn, this
to uniformly larger random response errors among implies that cross-population comparisons of our unstan-
blacks, as indicated in column 4, a finding that is dardized regression coefficients among substantive con-
consistent with previous research. A formal test of structs could vary depending on which indicator we
invariant error variances revealed that invariance is choose to normalize on (Bielby 1986; Williams and
rejected for measures of the two attachment constructs Thomson 1986). Therefore, we performed a sensitivity
(L = 41.98; d.f. = 4; p < .001), and also rejected for analysis, varying the reference indicator for the latent
our measures of definitions (L2 = 77.10; d.f. = 7; p < construct underlying our measures of definitions of
.001). The validity coefficients indicate a similar delinquency. That analysis revealed no substantial
ordering of accurate indicators for both races: RSPE- differences in black-nonblack comparisons of regression
CTFR is a better indicator of attachment to peers than coefficients, suggesting that our results are not sensitive
BELIKFR; and OKLAW, RSPECTPO, and TROUBLE to the choice of reference indicator. Given this, it seems
are more accurate measures of definitions. Following reasonable to assume that we can make meaningful
Matsueda (1982), we fixed the error variance of AGE to cross-population comparisons of metric coefficients.

This content downloaded from on Fri, 19 Jan 2018 21:46:36 UTC
All use subject to http://about.jstor.org/terms

0 t; o Wo t o 8 0t ?? e) - r-0
ON S 0 00 000 0 -ooW ) n t W ) c cn 00

:r s z

0? t O O O 00 0~
mN N N - 00 N t x _ N tt N 0t N


Q~~~~ or-
0 0 0 W - e C _

00 00 ? F cw> N ^ W ? t m t oN o
o o ON N en l $ o ) c) a cn ^ a cO
O N o. o __ ooN ) W) o \oo en o) oo o ON

oo c>m 0 o - O l _- _o

m eM o o p
ON N0. _n ___W) o ^

ca4 ca a C'T NN a -

12 a O N W) N _N _n
wD ti^ z -cm
00 U

0% -W O I W X Xm_ N Rt ON a N
r 0% O Ch t 0 % C l00 N On I) m0 0o 4-

i X~~~~~0 _ON ^ ^ t ? t m t m t Wo r )r )r
oO .D,


Q _00o m? _F t__ P I mtE

2 Q
2 8 ss_^x~~~~~~~~~~~~otooon-^ N eo Os~~~~~~~~~~~~~0

m rY N ^o N NN - N N N NN N No O

Q~~~~~~~V VI VIt

r~~~~~~~~l el; 4 tr 6X:0 6 6 C 4 4

This content downloaded from on Fri, 19 Jan 2018 21:46:36 UTC
All use subject to http://about.jstor.org/terms

REFERENCES Conventional Institutions and Delinquency: Estimating

Reciprocal Effects." American Sociological Review
Agnew, Robert. 1985. "Social Control Theory and 50:547-60.
Delinquency: A Longitudinal Test." Criminology Matsueda, Ross L. 1982. "Testing Control Theory and
23:47-62. Differential Association: A Causal Modeling Ap-
Allison, Paul D. 1985. "Estimation of Linear Models proach." American Sociological Review 47:489-504.
with Incomplete Data." Revised version of a paper Matsueda, Ross L. and William T. Bielby. 1986.
presented at the 1981 Annual Meetings of the "Statistical Power in Covariance Structure Models."
American Sociological Association. Pp. 120-58 in Sociological Methodology 1986, edited
Austin, Roy L. 1978. "Race, Father-Absence, and by N.B. Tuma. San Francisco: Jossey-Bass.
Female Delinquency." Criminology 15:487-504. McLanahan, Sara. 1985. "Family Structure and the
Berger, Alan S. and William Simon. 1974. "Black Reproduction of Poverty." American Journal of
Families and the Moynihan Report: A Research Sociology 90:873-901.
Evaluation." Social Problems 22:145-61. Minor, W. William. 1984. "Neutralization as a Harden-
Bielby, William T. 1986. "Arbitrary Metrics in Multiple ing Process: Considerations in the Modeling of
Indicator Models of Latent Variables." Sociological Change." Social Forces 62:995-1019.
Methods and Research 15:3-23. Monahan, Thomas. 1957. "Family Status and the
Bielby, William T., Robert M. Hauser, and David L. Delinquent Child: A Reappraisal and Some New
Featherman. 1977. "Response Errors of Black and Findings." Social Forces 35:250-58.
Nonblack Males in Models of the Intergenerational Moynihan, Daniel P. 1965. The Negro Family: The Case
Transmission of Socioeconomic Status." American for National Action. Washington, DC: Office of Policy
Journal of Sociology 82:1242-88. Planning and Research, U.S. Department of Labor.
Briar, Scott and Irving Piliavin. 1965. "Delinquency, Rankin, Joseph H. 1983. "The Family Context of
Situational Inducements, and Commitment to Con- Delinquency." Social Problems 30:466-79.
formity." Social Problems 13:35-45. Rosen, Lawrence, M. Lalli, and Leonard Savitz. 1975.
Chilton, Roland and Gerald Markle. 1972. "Family "City Life and Delinquency: The Family and Delin-
Disruption, Delinquent Conduct, and the Effects of quency." Unpublished report submitted to the National
Sub-Classification." American Sociological Review Institute for Juvenile Justice and Delinquency Preven-
37:93-99. tion, Law Enforcement Assistance Administration.
Cloward, Richard A. and Lloyd E. Ohlin. 1960. Shaw, Clifford R. and Henry D. McKay. 1969. Juvenile
Delinquency and Opportunity. New York: Free Press. Delinquency and Urban Areas. Rev. ed. Chicago:
Elliott, Delbert S., David Huizinga, and Suzanne S. University of Chicago Press.
Ageton. 1985. Explaining Delinquency and Drug Use. Short, James F., Jr. and Fred L. Strodtbeck. 1965.
Beverly Hills, CA: Sage. Group Process and Gang Delinquency. Chicago:
Hindelang, Michael J. 1978. "Race and Involvement in University of Chicago Press.
Common Law Personal Crimes." American Sociologi- Sutherland, Edwin H. 1947. Principles of Criminology,
cal Review 43:93-109. 4th ed. Philadelphia: Lippincott.
Hindelang, Michael J., Travis Hirschi, and Joseph G. Sutherland, Edwin H. and Donald R. Cressey. 1978.
Weis. 1979. "Correlates of Delinquency: The Illusion Criminology, 10th ed. Philadelphia: Lippincott.
of a Discrepancy." American Sociological Review Tennyson, R.A. 1967. "Family Structure and Delinquent
44:994-1014. Behavior." Pp. 57-69 in Juvenile Gangs in Context,
. 1981. Measuring Delinquency. Beverly Hills, edited by Malcolm W. Klein. Englewood Cliffs, NJ:
CA: Sage. Prentice-Hall.
Hirschi, Travis. 1969. Causes of Delinquency. Berkeley: Toby, Jackson. 1957. "The Differential Impact of Family
Free Press. Disorganization." American Sociological Review
. 1983. "Crime and the Family." Pp. 53-68 in 22:505-12.
Crime and Public Policy, edited by J.Q. Wilson. San Wells, L. Edward and Joseph H. Rankin. 1986. "The
Francisco: ICS Press. Broken Homes Model of Delinquency: Analytic
Jensen, Gary F. 1972. "Parents, Peers, and Delinquent Issues." Journal of Research in Crime and Delin-
Action: A Test of the Differential Association quency 23:68-93.
Perspective." American Journal of Sociology Williams, Richard and Elizabeth Thomson. 1986.
78:562-75. "Normalization Issues in Latent Variable Modeling
Johnson, Richard E. 1986. "Family Structure and with Multiple Populations." Sociological Methods and
Delinquency: General Patterns and Gender Differ- Research 15:24-43.
ences." Criminology 24:65-80. Wilson, Alan B. 1965. Richmond Youth Project
Joreskog, Karl G. and Dag Sorbom. 1984. LISREL V: [MRDF]. Berkeley: Survey Research Center, Univer-
Analysis of Linear Structural Relationships by Maxi- sity of California, Berkeley [Producer].
mum Likelihood and Least Squares Methods. Chicago: Wilson, James Q. and Richard J. Herrnstein. 1985.
National Educational Resources. Inc. Crime and Human Nature. New York: Basic Books.
Kornhauser, Ruth R. 1978. Social Sources of Delin- Wilson, William Julius. 1985. "Cycles of Deprivation
quency. Chicago: University of Chicago Press. and the Underclass Debate." Social Service Review
Liska, Allen E. and Mark D. Reed. 1985. "Ties to 59:541-59.

This content downloaded from on Fri, 19 Jan 2018 21:46:36 UTC
All use subject to http://about.jstor.org/terms