Download Comparing Control Theory and Differential Association Theory: Evidence and Testing and more Study notes Sociology in PDF only on Docsity!
TESTING CONTROL THEORY
AND DIFFERENTIAL ASSOCIATION:
A CAUSAL MODELING APPROACH*
Ross L. MATSUEDA
Universityof California,Santa Barbara
A number of strong theoretical statements have been based on analyses of delinquency data from the Richmond Youth Project. Hirschi (1969) and Jensen (1972), in particular, found that Hirschi's control theory was empirically supported over Sutherland's theory of differential association. This paper reanalyzes these data and reassesses this negative evidence pertaining to differential association theory. It is shown that the ratio of learned behavior patterns favorable and (^) unfavorable to violation of legal (^) codes, the critical variable in Sutherland's (^) theory, can be operationalized by (^) explicitly modeling its measurement error structure. In turn, this allows the testing of specific hypotheses derived from the theory. The analysis based on this strategy finds differential association theory supported over control theory. Specifically, the unobservable construct representing the ratio of learned behavior patterns successfully mediates the effects on delinquency of the model's other variables.
A major contemporary controversy in the sociology of crime and delinquency concerns two dominant theories of criminal behavior: Sutherland'stheory of differentialassociation, and Hirschi'scontrol theory. The most signifi- cant researchaddressingthis issue is Hirschi's (1969) landmark study. Hirschi developed, operationalized,and empiricallyconfirmedhis control theory, and presented evidence that seriously questioned the empiricalefficacy of differentialassociation theory. Jensen (1972) reanalyzed these data from the Richmond Youth Project and, focusing on the relation- ships among parents, peers, and delinquency, also found Sutherland'stheory unsupported. The objective here is to develop a better strat- egy for testing differentialassociation theory, to investigate certain measurementproperties of the Richmond data, and to reassess this negative evidence pertainingto the theory.
DIFFERENTIAL ASSOCIATIONVERSUS CONTROLTHEORY Based on a conception of modern society as heterogeneous and segmented into conflicting
groups, the theory of differentialassociation asserts that crime is rooted in normativecon- flict (Sutherland, 1947:19). In industrialized societies, at least, definitions of legal codes that favor law violation exist alongside defi- nitions unfavorableto law violation. Sutherlandgave the name "differentialasso- ciation"to the process by which persons expe- rience these conflicting definitions about ap- propriatebehavior. Thus, definitionsfavorable and unfavorableto delinquentor criminalbe- havior are learned through interaction (com- munication)in intimatepersonalgroups.' This differential learning includes the specific di- rection of motives, drives, rationalizations, and attitudes-whether toward viewing legal codes as rules to be observed or broken. "A person becomes delinquentbecause of an ex- cess of definitionsfavorableto violationof law over definitions unfavorable to violation of law" (Sutherlandand Cressey, 1978:81). Both favorable and unfavorabledefinitions (behaviorpatterns)are weightedby frequency, duration,priority,and intensity. Thus, behav- ior patternspresented with greaterfrequency, presentedfor a longertime, presentedearlierin life, and presented from a more prestigious source will have more weight in the process producing delinquent or nondelinquent be- havior (differentialassociation). In developing differential association, Sutherlandattempted to account for both the distributionof crime rates and for individual cases of criminalbehavior (Sutherland,[1942] 1973:18-20; Cressey, 1960). Because true crime rates are summarystatementsabout the
- Direct all correspondence to Ross L. Matsueda, Department of Sociology, University of California, Santa Barbara, CA 93106. An earlier version of this paper was presented at the Annual Meetings of the American Sociological Association, Toronto, 1981. I thank William T. Bielby, Donald R. Cressey, and Thomas P. Wilson for helpful suggestions and invaluable comments on earlier drafts. I also thank James N. Baron, Gary F. Jensen, Bruce C. Straits, and anonymous ASR re- viewers for their comments; Nancy C. Jurik for her encouragement and thoughtful advice; and Travis Hirschi, Joseph G. Weis, and Carol A. Zeiss for their assistance in obtaining the data.
I (^) Our references to crime also pertain to juvenile delinquency, and vice versa. American Sociological Review (^) 1982, Vol. 47 (August:489-504) 489
490 AMERICAN SOCIOLOGICALREVIEW
frequencyof individualcriminalacts, they are determinedby the proportionsof persons re- ceiving an excess of criminalbehaviorpatterns throughthe differentialassociationprocess. In other words, the extent to which a group or society is organized in favor of crime, as against the extent to which it is organized against crime, determines its crime rate. Sutherlandgave the name "differentialsocial organization"to this process whereby certain structures translate normative conflict into various rates of crime. Moreover,he proposed that structuralconditions such as class, age, sex, ethnicity, and family status affect individ- ual criminality (and, thus, aggregate crime rates) only by affecting the probability of learningbehavior patternsfavorableand unfa- vorable to law violation (Sutherland, [1944] 1973:31).Thus, any effects that these factors have on either criminalityor crime rates are mediatedby the process of learningdefinitions favorable and unfavorableto delinquency. In contrast, Hirschi's control theory insists thatdefinitionsof the legalcode do not mediate all such factors. Instead of asking why some persons engage in crime-as do most theories of deviance-control theory asks why most persons refrain from criminalbehavior. Delin- quency is taken for granted;conventionalbe- havior is problematic (Hirschi, 1969:10).Put positively, control theory maintainsthat per- sons conform to legal codes because they are bonded to society. Accordingly, when a per- son's bond to society is broken or weakened, he or she is free to engage in delinquency-but is not requiredto do so. For Hirschi, then, the motivationto commit delinquentand criminal behavioris constant across persons, and thus, is not an explanatory variable (Hirschi, 1969:10-11, 24-25, 32). The bond to society is a strongcord consist- ing of four interwoven strands: attachment, belief, commitment,and involvement.Because the fibers or elements of each strandresemble those of the others, these strandsare positively intercorrelated.However, each affects delin- quency independently, so the four are ana- lytically separable (Hirschi, 1969:27-30). At- tachment, perhaps the most importantstrand in the bond, contains a moral dimension that dissuades persons from delinquency. For Hirschi there are no delinquent subcultures. Instead, there is variation in the extent to which people believe in society's norms, and the less their belief the more likely they are to engage in delinquency. Commitmentto con- ventional activity dissuades persons from de- linquency because, when considering delin- quent behavior, a person who has invested time and energy in a conventional activity- such as getting an education or buildingup a
business-calculates the risk of losing the in- vestment. Finally, involvement in (^) con- ventional activity reduces delinquency by limitingone's time to contemplateand commit delinquentacts. In summary, accordingto differentialasso- ciation theory, definitions of the legal code mediate the effects of structural factors on crime. According to control theory, in con- trast, definitions of the legal codes, which re- flect the degree of belief in the moralorder, do not mediate attachment, commitment, or in- volvement. This differenceprovidesa basis for empirically testing the comparative explana- tory efficacy of the two theories (Hirschi, 1969:98-100; Jensen, 1972:564; Hepburn, 1976:450-51;Kornhauser, 1978:238).
HIRSCHI AND JENSEN'S TESTS OF
DIFFERENTIALASSOCIATION
Hirschi's analysis of the self-report delin- quency data collected from the Richmond Youth Projectprovidesempiricalevidence that purportedlycontradictsdifferentialassociation theory and supports his control theory. Spe- cifically, two of his findingsregardingparents, peers, and delinquency directly question the explanatorypower of differentialassociation. First, Hirschifinds that the more intense the friendships-as measured by attachment- among boys with one or more delinquent friends, the less likely they will engage in de- linquency.4 Here, following Short (1957), Hirschi operationalizesdifferentialassociation theory by using associations with delinquents ratherthan associations with behaviorpatterns favorable to delinquency.5But because delin-
(^2) Sutherland maintained that neutral (con- ventional)behavioraffects criminalityin two princi- pal ways. First, neutral behavior occupies one's time, which prevents one from associating with either criminal or anticriminalbehavior patterns (Sutherland,1947:7).Second, in a given situation,a particularnoncriminalact can providean alternative to criminalbehavior and thereby prevent that be- havior from transpiring (Sutherland, (^) [1944] 1973:35-36). This constitutes part of the objective opportunityto engage in criminalbehavior. (^3) We are hereconcernedwithonly the (^) two specific findingspertainingto parentsand peers. We do not address other hypotheses, used by (^) Hirschi (1969:140)as indirecttests of differentialassociation, becausethey are quiteperipheral,if not irrelevant,to the theory. (^4) In a partialreplicationof Hirschi's study, Hin- delang (1973) found that, contrary to predictions from control theory, attachmentto peers increased the likelihoodof delinquency. (^5) Among the other studies operationalizingdif- ferential association using associations with delin- quent friends, rather than using behavior patterns
492 AMERICAN SOCIOLOGICAL REVIEW
Model Derived From DifferentialAssociation Theory
Background (^) Definitions (^) Delinquency Variables (^) n_ fn n D
Parental Peer Supervision - Relationships
Model Derived From ControlTheory
Background Definitions -) Delinquency Variables
Parenta l Peer Supervision -> (^) Relationships (Attachment (Attachment to Parents) to Peers)
Model Derived From MultipleFactor Theories Definitions
BackgroundBackground ---'A~~~^ ~~~~~~~~~~~~Delinq
Variables
Parental Peer Supervision -^ Relationships Figure 1. Causal Models of Delinquency
differentialassociation, it is assumed that the measure of definitions favorable and unfavor- able to delinquency is causally prior to delin- quent behavior and causally subsequent to other variablesin the model. This assumption cannot be tested with cross-sectional data; in- stead, a longitudinaldesign is required. (For evidence on this issue, see Cressey, [1953] 1973;Minor, 1981.) The reanalysisattempts,in three steps, to go beyond Hirschi and Jensen's tests of differen- tial associationtheory. First, certaincausal re- lationsimpliedby the theory are translatedinto a structuralequation model. Second, a mea- surement model, which treats definitions fa- vorable and unfavorableto delinquency as an
unobservablevariablewith multipleindicators, is incorporatedinto the system of equations. Third,specific hypotheses^ and^ the overall^ fit of the model are evaluated using both tests of point estimates and a more global goodness- of-fit test.
The Substantive Model To simplify matters, the causal relations specified in the substantive model can be dis- cussed in terms of the five variables dia- grammed in Figure 1. Primary focus will be given to the differential association model, using the alternativemodels derivedfrom^ con- trol theory and^ multiplefactortheory as a point of contrast. The four backgroundvariables, the age of the respondent (AGE), his parents' socioeco- nomic status (SES), whetheror not his home is intact (BROKHOME),and his perceptions of trouble in his neighborhood (YOUNGTRO), are commonly cited as importantdeterminants of crime and delinquency. However, as Sutherland and Cressey (1978) show, dif-
viduals who failed to complete the questionnaire were more likely to be delinquent. The critical ques- tion, however, is whether the relationships among relevant variables differ for those who failed to re- spond. Hirschi (1969:46) compared the relationships between certain school-related variables and delin- quency and found no significant differences between respondents and nonrespondents. For a detailed de- scription of the Richmond data, see Hirschi (1969).
TESTING CONTROL THEORY AND DIFFERENTIAL ASSOCIATION 493
ferentialassociationtheoryexplicitly stipulates the causal mechanismthat makes these back- groundvariables "work" to produce crime or delinquency. (See Appendix A for brief de- scriptions of the variables.) For example, after summarizingthe empiri- cal literatureon age and crime, Sutherlandand Cressey (1978:129-30) conclude that age has importantdirect or indirect effects on crimi- nality. The objective here is to determine whether age affects delinquency directly, or indirectlythroughits effects on parentalsuper- vision, peer relationships, and definitions fa- vorable and unfavorableto delinquency. Sutherland and Cressey (1978:220)further hypothesize that low socioeconomic status may affect delinquencyin two ways. First, be- cause poverty areas are often areas of high delinquency, a child froma low SES home has a greaterprobabilityof encounteringmany de- linquent behavior patterns. Second, being a member of a lower class may affect a child's denial or acceptance of conventional values. Similarly, they conclude that various home conditions, including broken homes, promote delinquency by increasing children's associ- ations with delinquent definitions, and de- creasing their associations with antidelinquent definitions (Sutherland and Cressey, 1978:219-24). Perceptions of trouble in the neighborhood (YOUNGTRO)is used as an indicatorof delin- quency area. Both Short (1957) and Jensen (1972) advocate using neighborhood delin- quencyor troubleasperceived by respondents. For Sutherlandand Cressey, what is important is actual trouble in the neighborhood,which indicates the number of delinquent behavior patterns in the area. But perhapsone's effec- tive neighborhood,as experiencedthroughas- sociation with persons and values of the area, is more significant for learning delinquency than is one's objective neighborhood.This is because some persons do not perceive trouble in their neighborhoodsbecause they are iso- lated from it. In the substantive model (Figure 1), each backgroundvariable is assumed to affect pa- rental supervision. From the-^ perspective of differentialassociation theory, supervisionre- duces delinquency by increasing exposure to antidelinquentdefinitions and decreasing ex- posure to delinquent definitions (Sutherland and Cressey, 1978:222). In contrast, both Hirschi and Jensen treat parentalsupervision as an indicatorof attachmentto parents, and find that, as control theory predicts, supervi- sion affects delinquent behavior directly (see Figure 1). The substantive model also allows parental supervision, as well as the backgroundvari-
ables, to affect peer relationships.Presumably, the kinds of friendshipsa boy makes, and the closeness of those friendships,are affected by the extent that his parents supervise him. Dif- ferential association theory predicts that be- cause delinquentbehavior,like all behavior, is learned primarilyin intimate groups, peer re- lationships have an important impact on learningdefinitions of the legal code. The two aspects of peer relationshipscon- sidered here are numberof delinquentfriends and attachmentto friends. Fromthe standpoint of differential association theory, delinquent friends increase one's delinquency because they are morelikely thannondelinquentfriends to transmit definitions favorable to delin- quency. Jensen, however, finds that numberof delinquentfriends affects delinquencyregard- less of the numberof definitionsfavorable^ to delinquency. This finding is reassessed. Accordingto differentialassociationtheory, behaviorpatternslearned in relationshipsthat are intense, emotional, and prestigious have more significancefor subsequentconduct^ than behavior patterns without these charac- teristics. Therefore, degree of attachment^ to friends should have an important^ effect on learningdefinitionsof legal codes. But the di- rectionof that effect dependson the content of the behaviorpatternslearned-whether favor- able or unfavorableto delinquency.Again, the importantpoint is that, from^ the standpointof differentialassociation theory, attachmentto friends affects delinquency only insofar as it affects (^) the learningof definitionsfavorableand unfavorableto law violation. In contrast, con- trol theory predicts that attachmentto peers affects delinquency regardless of what is learned (see Figure 1). In short, differentialassociation theory pre- dicts that the ratio of definitionsfavorableand unfavorableto delinquencymediatesthe effect of each prior variable on delinquency. Thus, the effects of these variablesshouldbe zero, or at least trivialin size comparedto the effect of definitions of the law. In contrast, multiple factor theories impute direct causal power to each variable, while control theory predicts that parental supervision (attachmentto par- ents) and peer relationships (attachment to peers) directly affect delinquent behavior. In the followinganalysis, each of these alternative hypotheses is examined. The full structuralequationmodel, contain- ing these relationships,is presentedin the path diagramof Figure2. The substantivemodel is a recursive system of five equations. Each dis- turbance is assumed to be independentof all other disturbancesin the model and indepen- dent of the predictorvariablesin its equation. The disturbances contain numerous omitted
TESTING CONTROLTHEORY AND DIFFERENTIAL ASSOCIATION 495
the underlyingconstruct, the inaccuracy can be taken into considerationby explicitly mod- eling the^ indicators'^ measurement^ error structure. However, this method of operationalizing the "ratio"concept does not directly compute a ratio. Sutherland([1942] 1973:22)specified "ratio"to indicate: that some persons who have many intimate contacts with criminals refrain from crime and that this is probably due to the coun- teracting influence of associations with anti-criminalbehavior. Actual participation in criminal behavior is a resultant of two kinds of associations, criminal and anti- criminal,or the associations directedtoward crime and the associations directed against crime.
Empirically,this implies that any operationali- zation must capture associations with both procriminaland anticriminalbehaviorpatterns. Because each definitionvaries by the weight^ it receives from frequency, duration, priority, and intensity, each can, at least in principle,be placed on a continuum. If each definition is placed on a single continuum, ranging^ from highly antidelinquent to^ highly delinquent, each can be measuredby the same scale. The theoretical construct, then, becomes a unidi- mensional variable measuring weighted defi- nitions favorable and unfavorable to delin- quency on^ a continuous scale.^ Although, strictlyspeaking,the theoreticalconstructhere is not a ratio, it is a monotonictransformation of a ratio, and, moreover, it measures both kindsof associations, which^ is what^ Sutherland intended to accomplish. In short, after eliminating measurement error, the common varianceamong^ these indi- cators should adequatelycapturethe variation among persons' definitions of the legal code. Consequently, the ratio of definitions of the legal code can be operationalized, and the theory of differentialassociation can be sub- jected to rigorous empiricalexamination. The measurementmodel is specified by the indicator-constructand^ indicator-errorpathsof Figure2. As in a factor^ analytic^ configuration, each indicatoris specified as a linearcombina- tion of the latent variable(DEF), plus a random measurement disturbance. The measurement errors are assumed to be uncorrelated^ with both structuralvariablesand structuraldistur- bances. The stochastic measurement error
componentconsists of two elements. One, the eight indicators are each assumed to be ran- domly generated from an infinite domain of definitions items, and the resulting sampling error is picked up by the disturbance. Two, imperfectionsin the measuringinstrument(the questionnaire)result in measurementerror(see Jdreskog, 1976). From a statistical standpoint, the two are treated the same way (Bielby and Hauser, 1977:144). There are two reasons to expect some mea- surement errors to be positively correlated. First, some of the items are substantively similar,so the components orthogonalto DEF tap some commonality.Second, the items were all measuredon a single occasion, measuredon identical scales, and (for the first four items) measuredconsecutively on the questionnaire. This procedurefor operationalizingDEF re- quires that the indicators show content validity-the items should tap the content of behaviorpatternsfavorableand unfavorableto delinquency as Sutherlandintended. In addi- tion, collectively, the sample of items should represent all strata of the domain of content (Bohrnstedt, 1970:92). The items were de- signed to measureattitudestowardthe law and Sykes and Matza's (1957) techniques of neu- tralization, both direct operationalizationsof Sutherland's delinquent definitions (Hirschi, 1969:198-212; see also Austin, 1977). Also specified as an unobservable variable with multiple indicators, the theoretical con- struct "parental supervision" (SUPER) is in- dicated by PARWITH and PARWHERE.^13
(^10) Jensen (1972), Hepburn (1976), Akers et al. (1979), and Johnson (1979) have tried to measure definitionsof the legal code on a continuous unidi- mensionalscale, but without modelingits measure- ment error structure.
(^11) Withdomainsamplingmodels the measuresare not literallysampledfroman infinitedomainof mea- sures. Rather, it is merely assumed that the sample of items is a randomrealizationof all possible mea- sures (see Nunnally, 1967;and Bohrnstedt,^ 1970). 12 Persons' responses along the continuous scale should also reflect aspects of their ratios of defi- nitions. For the sample of items used here, this as- sumptionappearsreasonable.For example,^ persons answering"strongly agree" to the statement, "It is all right to get around the law if you can get away with it," shouldbe those with highratiosof weighted definitionspertainingto this class of verbalizations. Conversely, those answering "strongly disagree" feel it is not "all^ rightto get aroundthe law^ if you can get away with it"-which reflects antidelinquentat- titudes. Thus, these persons should have low ratios of this class of weighted definitions. (^13) Parentalsupervisionwas originallyspecified as a perfect linear combinationof its two indicators, and similar substantive results were found. How- ever, if these indicatorscontain measurementerror, that specification could bias the model in favor of differentialassociation theory. Consequently,to in- crease the strengthof the test of differentialassocia- tion, the resultsof the presentspecification,allowing for errors of measurement,are reported.
496 AMERICAN SOCIOLOGICALREVIEW
Similarly, ATTACHPE is indicated by BE- LIKFR and RSPECTFR, and, (following Hirschi) labeled "attachmentto peers." In addition,the perhapsunrealisticassump- tion that AGE, SES, BROKHOME, YOUNGTRO, FRPICKUP, and DEL are perfectly (^) measuredvariables is relaxed. Since each of these constructs has but a single indi- cator, the measurementparametersof a model allowing for measurement error cannot be identified-a prerequisitefor estimation.Thus, these parameterscannot be estimatedfromthe data; (^) they can only be fixed to more plausible values. The validity coefficients of SES, BROKHOME,YOUNGTRO, and DEL were fixed to equal .80, and that of AGE to equal .95; the correspondingreliabilitieswere fixed to .64 and .90. These values appearlow enough to reduce adequatelythe chance that tests are biased in favor of differential association theory, and highenough to avoid obtainingim- plausible estimates of the remainingparame- ters of the model.'
ESTIMATIONOF THE MODEL
Following Hirschi and Jensen, the present study focuses on the 1588 nonblack males
sampled.After using listwise deletion, missing values reduced the sample size to 1140. The model's parameterswere estimated by using Joreskog and Sdrbom's (1978) LISREL IV program.Assuming that the joint distribu- tion of the 17 variables is approximatelymul- tivariate normal, the programcomputes con- sistent and asymptoticallyefficient maximum likelihood estimates of parameters for iden- tified models. The parametersof both the five equationsubstantivemodel and the seventeen equation measurementmodel were estimated jointly as a single system.' LISREL IV also allows one to test an over- identifiedmodel's ability to reproducethe ob- served variance-covariance matrix. Specifi- cally, the likelihood-ratio-teststatistictests the null hypothesis, that the model's overidentify- ing restrictionsare satisfied in the population, against the alternative, that the moments are actually unconstrained.In large samples, such as the one used here, this statisticis distributed approximately x2, with degrees of freedom equal to the number of moments minus the numberof parametersestimated. In addition, specific hypotheses, or overidentifyingrestric- tions, can be tested by nesting the hypoth- esized model within a less restrictive model. The difference in x2S provides a likelihood- ratio test of the restrictions, with degrees of freedom equal to the difference in degrees of freedom between the two models. Often the descriptive ratio of X2/dfis used to assess the relative fit of various models (cf. Wheaton et al., 1977;Isaac et al., 1980).This procedureis followed here. The programalso provides a matrixof first- order partial derivatives of the minimized function with respect to each fixed and free parameter,and a residualmatrixof discrepan- cies between observed and implied moments. These may be useful in respecifying a poor fitting model (see Sdrbom, 1975; Jdreskog, 1979). Table 1 lists the goodness-of-fittests for each model estimated. Model 1, derived from con- trol theory and multiplefactor theory, allows each variableantecedent to DEF to affect de- linquency directly and also restricts all mea- surementerrors^ to be uncorrelated.The x2 is over four times the degrees of freedom, indi- cating a relatively poor fit (p = .000). Model 2
(^14) The metric of an unobservableis arbitraryand must be normalizedfor identificationpurposes. For metric models, by constrainingone of the factor loadingsof each constructto equal unity, the metric of SUPER was arbitrarilyfixed to equal that of PARWITH,the metricof ATTACHPEwas fixed to equal that of BELIKFR,and the metricof DEF was fixed to equal that of TROUBLE. Thus the slope parametersof these indicatorscan be identifiedonly relative to one another(Bielby et al., 1977:1251).In addition,for standardizedmodels, the metricwas set by fixing the variance of all unobservablesto unity and freeingall factor (^) loadings(see Jdreskog, 1979). 's We also performed a sensitivity analysis on these error variances (see Duncan, 1975:110).With the exception of FRPICKUP, varying the validity coefficients from .95 to .60 does not alter the sub- stantive picture in any meaningfulsense. However, with a validityfixed at .70, FRPICKUPhas a direct effect on delinquency statistically distinguishable from zero, but still dwarfedby the effect of DEF; at .60, multicollinearitypreventsstable estimation.We believe that the reliability estimates we chose for these indicators are the most plausible values. Moreover, the values for SES, FRPICKUP, and DEL are similar to previously reported estimates (see Bielby et al., 1977;Hindelanget al., 1981). For metricmodels, reliabilitieswere fixed to an arbitrary value by fixing errorvariances:o-2^ = (1-Reliability) o-2,where o-2^ is the varianceof the observable. For standardizedmodels, reliabilitiesare fixed by fixing Pye =^ 1 -^ Reliability,where^ Pye is the squaredpath coefficient of the measurementdisturbance.
(^16) Estimationof a model based on a pairwisepre- sent covariance matrix yielded similar substantive results. 17 By using the full informationgiven by the model, this strategyprovidesefficient parameteres- timatesbut has (^) the drawbackthat misspecificationin one portion of the model can spill over and bias estimates of anotherportion.
498 AMERICAN SOCIOLOGICAL REVIEW
ot _0-^ oo oo 00
o o o > N
.o
CZ CZ:< en^ ON F o _ b U en o Io t V) o> c, b^ -^ en^ o^ _^ _^ en^ en^ en^ en^ VI)^ VI)
= (^) VI O-cr o en r- t- t- (^) en vo I", ao b O =^ oo^ eq^ oo oo t-? o 'I^ V)^ en 0 oo oo oo o Eob en^ _ o en^ en^
0
W) X o^ t-^ VI)m b~~~~~~~~~~~~~
S~~~~~~r 1-^0 r- - ON ---N
8~~~~~~~~~~~~~~~~~~~" 0 r
r0~~~~~~~~~~~~~~e (^) t- 00 00i S
S O < t o N m? N
CZ (^) Z.....NooN o oNo o o. No>o? ### D~~~~~~~~~ to^ C's ### z~~~~~~~~~~~~~~~aO7 ONtt>X -O^ m-N^ tUe<n 0 e t i- ### W~~~~~~0O0 o en goxoNo- - - - CA ## en > _nn888_~oooo__~ X~~~w' o o N.o ) eno (^) eno0 eno- o,- o- o- aso ooo oo o4 I",o o C E~~~~" (^) " "I^ - en t' t' " env" "^ en ^- ON ^ 00 o0 o^ONW) V^ ### rs o~~~~~~~~~~~~~ cd cl ts^ _^ CN-^ Pa^0 0 00 o^ _^ N^ e^ cd TESTING CONTROL THEORY AND^ DIFFERENTIAL^ ASSOCIATION (^499) Table 3. Standardized Parameter Estimates of the Substantive (^) Model 3: Nonblack Males (N = 1,140) Predetermined Variables Dependent BROK- Variable AGE SES HOME YOUNGTRO SUPER (^) FRPICKUP ATTACHPE DEF 1. SUPER -.128 -.018 -.126 -. 2. (^) FRPICKUP .243 -.043 -.032. 3. FRPICKUP .209 -.047 (^) -.066 .152 -. 4. (^) ATTACHPE .105 .095 .065 -. 5. ATTACHPE .137 .099 .096 -.121. 6. ATTACHPE (^) .183 .089 .082 -.088 .193 -. 7. DEF (^) .152 -.029 .064. 8. DEF (^) .090 -.036 .003 .260 -. 9. DEF (^) -.016 -.014 .037 .184 -.344. 10. DEF .035 (^) .011 .059 .159 -.291 .446 -. 11. DEL (^) .105 .008 .052. 12. DEL .067 .003 (^) .014 .244 -. 13. DEL -.034 .026 (^) .047 .170 -.163. 14. DEL -.017 (^) .034 .054 .162 -.144 .464 -. 15. DEL -.040 .027 (^) .014 .054 .053 .162 .094. tidelinquentdefinitions (compare lines 7 and 8). Also consistent with differentialassociation theory, number of friends picked up by the police has a large positive total impacton DEF (line 9 of Table 3). In fact, FRPICKUPhas the largestrelativedirecteffect (line 10of Table 3). In addition, FRPICKUP mediates nontrivial amounts of AGE, YOUNGTRO,and SUPER in theireffects on DEF. Therefore,beingolder, being in a neighborhoodperceived to be more trouble-ridden, and being supervised less causes boys to acquireslightlymore delinquent friendswhich, in turn,increases theirexposure to delinquentdefinitions. Adding FRPICKUP into the equationincreases the R2to over .60. The negative effect of ATTACHPEon DEF (line 10) indicates that, on the average, in closer friendships a higher proportionof an- tidelinquentdefinitionsis transmitted.As a re- sult of being more attached to their friends, boys who are from more trouble-free neigh- borhoods, who are more closely supervised, and who have fewer delinquentfriends tend to learn more antidelinquentdefinitions relative to delinquentdefinitions. With the addition of ATTACHPE (line 10), the structural form equation explains over 67 percent of the vari- ance in DEF. Thus, the model does well in accountingfor variabilityin the unobservable underlyingdefinitions of the legal code. Lines 11^ through 15 present estimates of the parameters predicting delinquency (DEL). Whilethe backgroundvariablesexplain 12 per- cent of the variance in delinquency (line 11), adding SUPER, FRPICKUP, and AT- TACHPEincreases this to 40 percent (line 14). Without DEF, then, the model still accounts for a substantial amount of variance in delin- quency. Also, the total effects of each variable except SES and BROKHOMEare nontrivial. These are importantfindings,for the empirical test of differential association-the ability of DEF to mediate the antecedent variables' ef- fects on delinquency-requires something nontrivialto be mediated. Adding DEF to the equation predictingde- linquency gives the structuralform (line 15). The impactof DEF on delinquencyis negative and, as revealed by the standardizedcoeffi- cients (line 15 of Table 3), comparativelylarge. Thus, as differential association theory specifies, increasingthe numberof definitions favorable to violation of law relative to unfa- vorable definitions increases delinquent be- havior. The model accounts for over half of the variance in delinquency. The hypothesis, derivedfrom differentialas- sociation theory, that DEF mediates the other variables' effects on delinquency, can be as- sessed by comparinglines 14 and 15. Line 14 reveals that before adding DEF to the equa- tion, AGE and SES already have trivial direct effects on delinquency. On the other hand, YOUNGTROhas a substantialand statistically significant direct effect. But when DEF is addedto the equation, the effect becomes both trivialin size and statisticallyindistinguishable from zero (line 15). Therefore, as differential association predicts, perceptions of neighbor- hood trouble increase delinquencyby increas- ing the probabilityof learning delinquent be- havior patterns. While modest in size and not quite significant, the direct effect of BROKHOMEon delinquencyis also mediated by DEF (compare lines 14 and 15). As both Hirschi and Jensen found, and as control theory predicts, attachmentto parents, indicatedby SUPER, has a negative effect on delinquency unmediated by variables repre- senting peer relationships(line 14). However, contraryto control theory, but consistent with TESTING CONTROL THEORY AND DIFFERENTIAL ASSOCIATION 501 they are intendedto tap, valid and reliablein- dicators of offending behavior. Consequently, generalizationof the findingsreportedhere to more serious forms of delinquencyis probably unwarrantedwithout furtherresearch. Fourth,this analysis is necessarily based on Hirschi and Jensen's assumptions about the content validity of the indicators of DEF. Ideally, the indicators should be consistent with Sutherland'sintended meaning of "defi- nitions favorable and unfavorableto violation of the legal code." Further,they should mea- sure, at the very least, the most dominantdefi- nitions of the legal code communicated,trans- mitted, and applied to delinquent behavior within the populationstudied. The content of such definitions is likely to vary across sub- culturalgroups, where communicationis dis- tant and impersonal,and yet remaininvariant within subcultural groups and geographic areas, wherecommunicationis relativelyopen, intimate, and personal. Although the measures used here appear reasonably consistent with Sutherland's spe- cification, it is not certainthat they are appro- priate for this particularpopulation, because the content of the items was determined de- ductively from a priori theoretical and con- ceptual considerations,ratherthan also induc- tively from a subset of the respondentsthem- selves. Finally, some of the evidence reportedhere raises questions about the validityand reliabil- ity of the indicatorsof DEF (see AppendixB). The relatively low reliability coefficients are similarto those reportedfor other social psy- chological data (cf. Miller et al., 1979; and Isaac et al., 1980),and in fact supportthe very reason for disentanglingunreliabilityfrom ac- tual variation in the theoretical construct (Kohn and Schooler, 1978).^ Nevertheless, the low reliabilitiescould also signal^ that the indi- cators have undesirable measurement prop- erties. Furthermore,although it is unreason- able to expect these measurementerrorsto be uncorrelated, the finding of many correlated errors among indicators within and between theoreticalconstructs could suggest problems of validity. Ultimately, however, questions about validity should be resolved on theoreti- cal ratherthan statistical grounds. In sum, this study has located possible sources of unreliability in the Richmond data and also has explicitly modeled that unreliabil- ity by relaxing a number of untenable restric- tions. This is important, for previous work that ignored these issues has produced misleading results. Thus, if one chooses to reject the Richmond data outright, this research locates possible grounds for doing so. If, on the other hand, one chooses to accept these data (which are still perhaps the best data addressing the theoretical issue at hand and easily the most widely cited) this research exploits these data more fully than previous work, takes various forms of unreliability into account, and thus provides an improved test of control theory versus differential association. APPENDIX A KEY TO VARIABLE LABELS AGE Age of respondent. 0 =^ 12 years or younger, 1 =^ 13 years, 2 =^14 years, 3 = 15 years, 4 = 16 years, 5 = 17 years, 6 = 18 years, 7 = 19 years, 8 = 20 years or older. SES Father's occupation measured on the DuncanScale; if there is no fa- ther living in the home, then moth- er's occupationis used. For the few cases in which father's occupation had a missing value, and father's education was reported, values of father's occupation were predicted by regressing occupation on educa- tion. BROKHOME A dummy variable coded as one if either the mother or father did not live with the respondent. FRPICKUP "Have any of your close friends ever been picked up by the police?" 0 =^ no or don't know, 1 =^ one friend has, 2 =^ two friends have, 3 = three friends have, 4 = four or more friends have. PARWITH A composite asked regarding each parent: "Do your parents know who you are with when you are away from home?" 0 =^ never- never, 0.5 =^ sometimes-never, 1.0 =^ sometimes-sometimes, 1.5 =^ usually-sometimes 2.0 = usually-usually. PARWHERE (^) Same as above but with the ques- tion: "Do your parents know where you are when you are away from home?" BELIKFR (^) "Would you like to be the kind of person your best friends are?" 0 = not at all, I = in a few ways, 2 = in most ways. RSPECTFR "Do you respect your best friend's opinions about the important things in life?" 0 = not at all, 1 = a little, 2 = pretty much, 3 = completely. (^19) In the strategy used here, specifying the items as fallible indicators of an unobservable variable amounts to correcting for attenuation due to unrelia- bility, a procedure requiring reasonably valid and reliable measures (Crouse et al., 1979:359-63). If in fact these indicators do not tap the theoretical do- main specified by Sutherland, any use of them for examining differential association theory is unwar- ranted. #### 502 AMERICAN SOCIOLOGICALREVIEW DEL An index of delinquencycommitted duringthe last year containingthe following six items: THEFT2 "Have you ever takenlittle things (worth less than $2) that did not belong to you?" THEFT250 "Have you ever taken things of some value (between $2 and $50) that did not belong to you?' THEFT50 "Have you ever taken things of large value (worth over $50) that did not belong to you?" CARTHEFT "Have you ever taken^ a car for a ride without the owner's permis- sion?" VANDALSM "Have you ever banged up somethingthat did not belong to you on purpose?" BATTERY "Not counting fights you may have had with a brotheror sister, have you ever beaten up on any- one or hurtanyone on purpose?" The following items are all measuredon the scale, "stronglyagree, agree, undecided,disagree,strongly disagree." YOUNGTRO [In my neighborhood] "Young people are always getting into trouble.' EVNBREAK "Policemen try to give all kids an even break." DELHURT "Most things that people call 'de- linquency' don't really hurt any- one." OKLAW "It is all rightto get aroundthe law if you can get away with it." RSPECTPO "I have a lot of respect for the Richmondpolice." GETAHEAD "To get ahead, you have to do some things which are not right." TROUBLE "I can't seem to stay out of trouble no matter how hard I try." SUCKERS "Suckers deserve to be taken ad- vantage of." APPENDIX B Analysis of the Measurement Model Parameterestimates of the measurementmodel ap- pearin Table4. All unrestrictedmetricslopes (listed incolumn3), except those of EVNBREAKandDEL- Table 4. Measurement Parameter Estimates of Model 3: Nonblack Males (N= 1,140) Variable (1)^ (2)^ (3)^ (4)^ (5) _____________________________ Observed Error Metric Validity Reliability Latent Observable Variance Variance Slope Coefficient Coefficient AGE AGE 3.012 .277f 1.000q .94Y .90() SES SES 5.663 2.037f 1.OW0 .80() .640f BROKHOME BROKHOME .204 .073f 1.0W .80() .640f YOUNGTRO YOUNGTRO .932 .336f 1.000f .800' .640f SUPER PARWITH .292 .125 1.00o0 .756. (.014) PARWHERE .279 .132 .938 .725. (.012) (.075) FRPICKUP FRPICKUP 2.235 .804f 1.00O .800 .640f ATTACHPE BELIKFR .374 .289 1.00O0 .477. (.019) RSPECTFR .499 .319 1.456 .600. (.033) (.244) DEF EVNBREAK 1.500 1.361 .904 .301. (.060) (.128) DELHURT 1.123 .989 .904 .348. (.044) (.113) OKLAW 1.079 .834 1.213 .476. (.042) (.130) RSPECTPO 1.189 .893 1.327 .496. (.044) (.136) GETAHEAD 1.222 .988 1.183 .436. (.047) (.130) TROUBLE .967 .801 1.00o .415. (.037) SUCKERS 1.219 1.010 1.132 .418. (.048) (.120) DEL DEL 1.200 .432f 1.000' .800 .640f Fixed coefficient. Note: Standard errors appear in parentheses. The following measurement errors were found to be significantly correlated: (^) p(ele8) = -.433 (^) p(elel0) = .467 (^) p(eje12) = .265 (^) p(ele15) = -.467 (^) p(elel6) = -. p(e2e 1) = (^) -.186p(e2el3) = (^) .137p(e,3e15)= .098p(e7e,0) = .166p(e10el3) = (^) .308p(el0el4) = (^) .077p(ellel6) =. p(ejje12) =^ .159 p(el2el3) =^ .054 p(el2el4) =^ .130 p(el2e16) = .104 p(el4e16) = .107 p(e,5e1,6)= .127. 504 AMERICAN SOCIOLOGICAL REVIEW Hirschi, Travis 1969 Causes of Delinquency. Berkeley: Univer- sity of CaliforniaPress. Isaac, Larry,ElizabethMutranand SheldonStryker 1980 "Politicalprotest orientationsamong black and white adults." American Sociological Review 45:191-213. Jensen, Gary F. 1972 "Parents, peers, and delinquent action: a test of the differentialassociation perspec- tive." American Journal of Sociology 78:562-75. Johnson, RichardE. 1979 JuvenileDelinquencyand its Origins.Cam- bridge:CambridgeUniversity Press. Jdreskog, Karl G. 1969 "A general approach to confirmatory maximum likelihood factor analysis." Psychometrika34:183-202. 1976 "Causal models in the social sciences: the need for methodologicalresearch."Univer- sity of Uppsala, Departmentof Statistics Research Report76-8, Uppsala, Sweden. 1979 "Analyzingpsychologicaldataby structural analysis of covariance matrices." Pp. 45- 100 in Karl G. Joreskog and Dag Sorbom (eds.), Advances in Factor Analysis and Structural Equation Models. Cambridge: Abt Books. J6reskog, Karl G. and Dag Sorbom 1978 LISREL IV: Analysis of Linear Structural Relationshipsby the Method of Maximum Likelihood. Chicago: National Education Resources, Inc. Kohn, Melvin L. and CarmiSchooler 1978 "The reciprocal effects of the substantive complexity of work and intellectual flexi- bility: a longitudinalassessment." Ameri- can Journalof Sociology 84:24-52. Kornhauser,Ruth R. 1978 Social Sources of Delinquency. Chicago: University of Chicago Press. Krohn, MarvinD. 1974 "An investigationof the effect of parental and peer associationson marijuanause: an empirical test of differential association theory." Pp. 75-89 in MarcReideland^ Ter- rence P. Thornberry(eds.), Crimeand De- linquency: Dimensions of Deviance. New York: Praeger. Miller,Joanne,CarmiSchooler, MelvinL. Kohn and Karen Miller 1979 "Woman and work: the psychological ef- fects of occupational conditions." Ameri- can Journalof Sociology 85:66-94. Minor, W. William 1981 "Techniques of neutralization: a recon- ceptualizationand empiricalexamination." Journal of Research in Crime and Delin- quency 18:295-318. Nunnally, Jum C. 1967 Psychometric Theory. New York: McGraw-Hill. Reiss, Albert J., Jr., and Lewis Rhodes 1964 "An empirical test of differentialassocia- tion theory." Journalof Researchin Crime and Delinquency 1:5-18. Short, James F., Jr. 1957 "Differentialassociationand delinquency." Social Problems4:233-39. 1958 "Differential association with delinquent friends and delinquent behavior." Pacific Sociological Review 1:20-25. 1960 "Differentialassociation as a hypothesis: problems of empirical testing." Social Problems8:14-25. Short, James F., Jr., and Fred L. Strodtbeck 1965 Group Process and Gang Delinquency. Chicago: University of Chicago Press. Sbrbom, Dag 1975 "Detection of correlated errors in lon- gitudinal data." British Journal of Mathe- matical and Statistical Psychology 28:138-51. Stanfield, Robert E. 1966 "The interaction of family variables and gang variables in the aetiology of delin- quency." Social Problems 13:411-17. Sutherland,Edwin H. [1942] "Development of the theory." Private 1973 paper published posthumously. Pp. 13- in Karl Schuessler (ed.), Edwin Sutherland on Analyzing Crime. Chicago: University of Chicago Press. [1944] "Critique of the theory." Private paper 1973 publishedposthumously.Pp. 30-41 in Karl Schuessler (ed.), Edwin Sutherland on Analyzing Crime. Chicago: University of Chicago Press. 1947 Principlesof Criminology.Fourth Edition. Philadelphia:Lippincott. Sutherland,Edwin H. and Donald R. Cressey 1978 Criminology.Tenth Edition. Philadelphia: Lippincott. Sykes, Greshamand David Matza 1957 "Techniquesof neutralization:a theory of delinquency." American Sociological Re- view 43:643-56. Voss, Harwin L. 1964 "Differential association and delinquent behavior." (^) Social Problems 12:78-85. Wheaton,Blair, Bengt Muthen,Duane F. Alwin and Gene F. Summers 1977 "Assessing reliabilityand stability in panel models." Pp. 84-136 in David Heise (ed.), Sociological Methodology 1977. San Fran- cisco: Jossey-Bass.
