Sie sind auf Seite 1von 8

Journal of Consulting and Clinical Psychology 1991, Vol. 59, No.

4, 499-506

Copyright 1991 by the American Psychological Association, Inc. 0022-006X/91/S3.00

Structure of Problem Behaviors in Preadolescence


Mary Rogers Gillmore, J. David Hawkins, Richard F. Catalano, Jr., L. Edward Day, and Melanie Moore
Social Development Research Group University of Washington, School of Social Work Robert Abbott Educational Psychology, University of Washington
This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Earlier research suggests that diverse adolescent problem behaviors, such as substance use, school problems, early sexual intercourse, and delinquency, reflect a single underlying dimension of behavior. Data from an ongoing longitudinal study were used to examine this issue in a previously unexamined sample (N = 426) of preadolescent sixth-grade youth. Ss included boys and girls from diverse socioeconomic and racial/ethnic backgrounds, whose average ages were 11 and 12. By using confirmatory factor analyses to test competing models, multiple factor structures were detected, suggesting that earlier findings supporting a single factor conceptualization may not be generalizable to this age group. Implications of the finding that problem behaviors may be more differentiated in late childhood than in adolescence are discussed.

Researchers have suggested that diverse adolescent problem behaviors, such as substance use, school problems, delinquency, and early sexuality, reflect a single underlying dimension of behavior (lessor & lessor, 1977), although empirical evidence evaluating this claim is limited. Whether different problem behaviors reflect a single underlying factor or whether they are better conceived as a multidimensional phenomenon has importance for both theory and practice. To the extent that problem behaviors represent a single factor, a general theory of problem behavior need not specify separate causal influences for different behaviors. On the other hand, to the extent that these behaviors are partially independent phenomena, individual theories, multiple factor approaches, or general theoretical frameworks that allow for specific variation become the proper direction for theory development (Osgood, Johnston, O'Malley, & Bachman, 1988) with parallel implications for intervention. The idea that different adolescent problem behaviors have the same underlying cause is evident in several theories of deviance (e.g., Elliott, Huizinga, & Ageton, 1985; Jessor & Jessor, 1977) and is consistent with a growing body of research that documents that positive relationships exist at the bivariate level between deviant behaviors. Positive relationships have been found between drug use and delinquency (Donovan & Jessor, 1985; Elliott, Huizinga, & Menard, 1989; Jessor & Jessor, 1977;

The research reported here and the preparation of this article were supported by Grant DA-03721 from the National Institute on Drug Abuse. We gratefully acknowledge the assistance of William Goldsmith and Marilyn Hoppe in data collection and management, and we appreciate the helpful comments of anonymous reviewers on drafts of this article. Correspondence concerning this article should be addressed to J. David Hawkins, Social Development Research Group, 146 North Canal Street, Suite 211, University of Washington, XD-50, Seattle, Washington 98103.
499

Johnston, O'Malley, & Eveland, 1978; Kaplan, 1985); early sexual intercourse and drug use among adolescents (Bentler & Newcomb, 1986; Donovan & Jessor, 1985; Elliott & Morse, 1987; Zabin, Hardy, Smith, & Hirsch, 1986); delinquency and sexual activity (Elliott and Morse, 1987); dangerous driving, crime, and drug use (Osgood, et al., 1988); and drug use and low educational performance (Bachman, O'Malley, & Johnson, 1978; Jessor, 1987; Smith & Fogg, 1978). Also, Loeber and Schmaling (1985) performed a meta-analysis on factor analyses of ratings of child psychopathology and concluded that various forms of antisocial behavior can be accounted for by a single dimension. Donovan and Jessor (1985) reported the results of a maximum likelihood test for one common factor for problem behaviors in four different samples. They first examined the factor structure of problem behaviors in two cohorts of youths: a high school sample and a college sample. Measures of problem behaviors included alcohol and marijuana use, precocious sexual intercourse, and general deviant behavior. Donovan and Jessor found that with the exception of 3rd-year college men, a onefactor solution provided an adequate fit to the data for both male and female subjects. To test the stability of this finding, they repeated the analysis using data obtained when the subjects were in their 20s. They again reported that one factor adequately accounted for the observed correlations. To examine the generalizability of this finding, they replicated the analysis using data from a national probability sample of 11th- and 12th-grade students. In this analysis they examined both problem behaviors (including cigarettes, alcohol, marijuana, and other illicit drugs and general deviant behavior) and conventional behavior (church attendance and school performance). Again, a single factor adequately accounted for the correlations among the variables for both boys and girls. More recently, Donovan, Jessor, and Costa (1988) replicated the analysis on a new sample of 11th- and 12th-grade students who were

500

GILLMORE ET AL. trict. In 1985, when subjects entered fifth grade, the panel was expanded to include all fifth-grade students in 18 elementary schools (N = 1,053). Of this eligible population of fifth-grade students, 919 (87%) completed the fifth-grade fall survey, and 608 (66%) of these students completed surveys in the spring of their sixth-grade year. Donovan et al. (1988) and Donovan and Jessor (1985) reported much lower initial response rates (typically approximately 53%) and comparable retention rates (65-73%) in their studies. Questionnaires were administered in classrooms by project personnel who read aloud each question and its associated response categories to students. Students had copies of the survey on which they indicated their responses to each question. At the conclusion of the sixthgrade school year, teachers were asked to fill out the teacher's form of the Achenbach and Edelbrock (1983) Child Behavior Checklist (CBCL) for each student in their classroom. Data from the 426 subjects for whom usable data were provided on both the sixth-grade student survey and the teacher CBCL were used in the analyses reported here. The sample of 426 subjects was 49.1% White, 21.3% Black, 19.7% Asian-American, and 9.9% other racial or ethnic groups. About half (52%) were male and half female (48%). According to official school district records, 33.3% qualified for the federally funded free or reduced fee lunch program. Most of the subjects (89%) were 11 or 12 years of age by the time of the survey. These demographic characteristics were virtually identical to those in the larger sixth-grade sample of 608 students. In addition, comparisons at fifth grade indicated that the sample used in the present analysis did not differ significantly from the fifth-grade panel on most sociodemographic and antisocial behavior variables including sex, racial composition, eligibility for free or reduced price lunches, substance use, delinquent behaviors, and school problems. Students in the present sample were slightly more likely to come from two-parent families and were rated as somewhat less aggressive by their teachers, however. Measures Our definition of problem behaviors is consistent with that of Donovan and Jessor (1985), Jessor and Jessor (1977), and Osgood et al. (1988), who viewed problem behaviors as behaviors that are socially defined as undesirable and typically evoke efforts at social control. We included multiple measures of three different kinds of behaviors that are proscribed in this age group: serious school misbehaviors, delinquent behaviors, and substance use. We included behavior problems at school because they have been shown to be related to both drug use and delinquency (e.g., Hawkins & Lishner, 1987; Hawkins, Lishner, Catalano, & Howard, 1986) and because early aggressiveness at school is predictive of later problem behaviors such as adolescent substance use and delinquency in adolescent males (e.g., Ensminger, Kellam, & Rubin, 1983). Certain problem behaviors included in previous studies, such as dangerous driving (Osgood et al, 1988) and sexual intercourse (Donovan & Jessor, 1985), were not included in the present analyses because they are extremely rare at ages 11-12. Although our measures did not exhaust the problem behaviors possible in this age group, they were chosen to represent a broad range of problem behaviors and are similar to those included in earlier studies. School Problems Three indicators of school problems were created including students' self-reported problems at school and two measures of teacherrated behavior problems at school. School trouble is an index composed of four student self-report items including (a) the extent to which the student got in trouble in school in the past year scored on a 4-point scale including NO.'(I), No (2), Yes (3), and YES!(4), indicating degree of endorsement with the statement; (b) the number of times in the past year the student was sent out of class for misbehavior (0-4); (c) the number of times the student was suspended or expelled in the past year (0-4); and (d) whether the student has ever hit a teacher (No/Yes).

surveyed in 1985. They again reported that a single factor accounted for the correlations among alcohol use, marijuana use, sexual intercourse, and general deviant behavior. Osgood et al. (1988) used LISREEs linear structural equation modeling approach to examine the fit of a factor model in which both shared, and unique components of deviant behaviors were examined. For this analysis, they used longitudinal data from a sample of White high school seniors for whom follow-up data were available. Measures of deviant behavior included illegal behavior directed at victims, heavy alcohol use, marijuana use, use of other illicit drugs, and dangerous driving. They found that when longitudinal relationships were included, a model in which both general and specific deviance factors were specified provided a better fit to the data than did a model in which only a general deviance factor was included. They concluded that each deviant behavior is in part not only a function of a general deviance factor, as suggested by Donovan and Jessor (1985), but also a function of factors unique to the particular behavior in question. Our research provides additional empirical evidence on the question of whether different problem behaviors reflect a single underlying factor or are better conceived as multidimensional phenomena. These analyses (a) expand on earlier work, (b) examine the structure of problem behavior in a multiethnic sample of preadolescents in sixth grade, and (c) use confirmatory factor analysis to explicitly test competing models of the factor structure underlying the measured variables. This approach is consistent with the suggestions of Bentler (1980) and Maruyama and McGarvey (1980) that absolute fitting of a model is less meaningful than comparing the relative fit of competing models. Four alternative factor models were tested (Figure 1). Model 1 represents a single-factor model in which three types of problem behaviors, school problems, delinquent acts, and substance use, were hypothesized to reflect a single underlying factor. This model is consistent with Jessor and colleagues' theory and empirical findings. Model 3 is a three-factor model in which these problem behaviors are specified as separate but correlated dimensions of deviance. Model 2a specifies delinquency and problem behavior at school as a single factor and substance use as a separate factor. This model is consistent with empirical findings that school problems (Jessor & Jessor, 1977; Kandel, Kessler, & Margulies, 1978) and delinquency often precede drug use (Elliott et al., 1989). Finally, because there is a strong positive association between delinquency and drug use in these data, as well as in data from older adolescent samples (e.g., Donovan & Jessor, 1985; Elliott et al., 1989; Johnston et al., 1978), we also estimated Model 2b, in which delinquency and drug use are viewed as one factor, and problem behavior at school represents a separate factor.

This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Method
Study Participants Data were collected as part of an ongoing longitudinal study seeking to identify childhood risk factors for adolescent drug use and delinquency and to test the effects of preventive interventions (Catalano & Hawkins, 1986). Data collection began in 1981 with a panel of 568 first-grade students in 8 elementary schools in the Seattle school dis-

PREADOLESCENT PROBLEM BEHAVIORS

501

Delinquency School Troubles F1 School Troubles Delinquency Substance Use F1

This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Model 1 One factor

Model 2a Two factors

Delinquency Substance Use F2

Model 2b Alternative Two factors

Model 3 Three factors

Figure 1. Four alternative models of the factor structure underlying problem behaviors. (F = factor; VI = school problems; V2 = aggression; V3 = acting out; V4 = delinquencybreaking and entering, picked fights or assaulted others; V5 = delinquencythrowing objects at people or cars and shoplifting; V6 = delinquencystealing and vandalism; V7 = tobacco use; V8 = alcohol use; V9 = illicit drugs.)

Higher scores on the items indicate more school problems. Eighty percent of the sample reported one or more of these school problems. Because the response formats differ, the items were first transformed to z scores, then averaged to form the index of school trouble. This transformation did not change the pattern of correlations among the variables. Factor analysis of the ratings on the teacher CBCL items found that 10 items loaded on the same factor and formed a conceptually coherent scale that could be labeled aggressiveness. Because multiple indicators have the dual advantage of providing more information and greater degrees of freedom, two indicators of teacher-rated aggressive behaviors were constructed from these 10 CBCL items. We would expect these indicators to be highly correlated, and indeed they are. However, this should not affect the comparisons of the models, and the advantage of having multiple indicators outweighs the disadvantages in view of the fact that the indicators are not perfectly correlated. Each item ranged from not true of the student (0) to very true of the

student (2). The first scale (Aggression) consisted of the average score on five items including gets in many fights, explosive or unpredictable, bullying or meanness, defiant or talks back, and swears or uses obscene language. The second scale (acting out) was the average score on five items including threatens others, argues a lot, doesn't get along with others, physically attacks others, and temper tantrums or hot temper. Twenty-five to 30% of the sample were rated by teachers as having engaged in at least one or more of the behaviors on each scale.

Delinquent Acts
Delinquent acts were measured by six items from the student survey asking how often in the past year the subject had stolen things, vandalized property, thrown objects at cars or people, shoplifted, broken and entered, or picked fights with others. These behaviors represent the more common forms of index offenses reported in this age group. Scores on the items range from never in the past year (1) to more than 4

502

GILLMORE ET AL.

times in the past year (4). To create multiple indicators for the delinquency factor, three 2-item indexes were formed from the six items by averaging responses on the relevant items. Because all items were positively intercorrelated, considerations of distribution largely dictated which two items to combine. The first indicator, labeled delinquency-1, consisted of breaking and entering and picked fights or assaulted others, and the second, labeled delinquency-2, included throwing objects at people or cars and shoplifting. Stealing and vandalism constituted the third indicator, labeled delinquency-3. Thirty-seven percent reported engaging in at least one of the behaviors constituting the first indicator, 33% reported engaging at least one of the behaviors in the second indicator, and 23% reported engaging in one or more behaviors in the third.
This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

For all models, paths that were not hypothesized were set to zero. It is necessary either to fix the regression weight of one indicator for each factor, or, alternatively, to fix the variance of each factor in order to identify each factor. Because, for our purposes, the estimates of factor loadings were of more substantive interest than were estimates of factor variances, we set the factor variances to 1.0 to define the scales. For all models that specified more than one latent factor (i.e., Models 2a, 2b, 3), the factors were allowed to correlate. Consistent with the work of Donovan and Jessor (1985), we analyzed the data separately for boys and girls. The raw data from each subsample were used as input for the analyses.

Substance Use
Three indicators of substance use were created: use of tobacco, use of alcohol, and use of other illicit drugs. Cigarette use ranging from never (1) to more than four times in the last month (4) and use of chewing tobacco (ever/never used) were combined into one measure of tobacco use. This variable was scored 0 if neither cigarettes nor chewing tobacco had ever been used, 1 if either chewing tobacco had been used or if cigarettes had been used but not in the past month, 2 if cigarettes had been used once or twice in the past month, 3 if used 3 or 4 times, and 4 if used more than 4 times. Alcohol use was also scored on a 4-point scale: 0 if alcohol had never been used, 1 if it had been used but not in the past month, 2 if it had been used 1 or 2 times in the past month, and 3 if it had been used more than 2 times in the past month. For the measure of illicit drug use, the drug with the highest frequency (using the scoring method described for alcohol) represented the measure. These drugs included marijuana, cocaine, psychedelics, sniffing glue or other inhalants, amphetamines, tranquilizers, and sedatives not prescribed by a physician. Although initiation of alcohol is not rare in this sample42% have used alcoholthe measure is somewhat skewed as would be expected in this age group. Tobacco use is more skewed, but even so, over a quarter (26%) of the sample had tried cigarettes or chewing tobacco by spring of sixth grade. As would be expected in this age group, use of illicit drugs is the most skewed; only 14% of the sample had ever tried any of these drugs. In the interest of space, we labeled the nine indicators of problem behavior VI-V9, respectively, in figures and tables.

Findings The zero-order correlations of the measured variables for boys are presented below the diagonal in Table 1. All correlations were positive, and most were significant. All subsequent analyses were based on the variance/covariance matrix. The standardized parameter estimates, corresponding z statistics, and overall model fit indices for the one-, two-, and three-factor models for boys are presented in Table 2. The decreases in chi-square for the two-factor models relative to the one-factor model were significant: x2(l, N= 222) = 16.20, p < .001, for Model 2a versus Model 1; x20, N= 222) = 283.68, p < .001, for Model 2b versus Model 1. Inspection of both the Bentler-Bonett normed and non-normed fit indices suggested that a two-factor model, which distinguishes school trouble from delinquency and drug use (Figure 2b), fitted the data better than the alternative two-factor model (Model 2a). There was slight improvement in fit by adding a third factor. The decrease in chi-square for Model 3 versus Model 2b, although significant, X2(2, N = 222) = 8.72, p < .05, was not very large, and the Bentler-Bonett normed fit index hardly changed at all (.86 to .87). In fact, the non-normed fit indices were virtually identical (.83, .83). As can be seen in Table 2, the correlations among the factors were positive and significant. Additionally, the normed fit indices for both Model 2b and Model 3 approached the .90 criterion proposed by Bentler and Bonett (1980) as representing a good fit in an absolute sense. Because our primary interest was in the relative fit of alternative models, we did not conduct specification searches for clues to improving absolute fit. A multifactor outcome also was found for girls, but the results are slightly different. The zero-order correlations of the measured variables for girls are presented above the diagonal in Table 1. As was true in the case of the boys, all relationships among measured variables were in the expected direction, and most were significantly different from zero. The standardized parameter estimates, corresponding z statistics, and overall model-fit measures for the one-, two-, and three-factor models for girls are presented in Table 3. Similar to the findings for boys, both two-factor models represented a significant improvement in fit over a single-factor model: x2(l, AT = 204) = 81.28, p < .001, for Model 2a versus Model 1; x20, N= 204) = 228.55, p < .001, for Model 2b versus Model 1. Judging by the Bentler-Bonett fit indices, Model 2b appears superior to Model 2a, just as for boys. Model 3 produced a significant improvement in fit relative to Model 2b as judged by

Results Model Estimation The models were estimated with the Bentler-Weeks model for confirmatory factor analysis (CFA), as implemented in the EQS computer program (Bentler, 1986). Maximum likelihood estimation was used although our indicators are somewhat skewed, because (a) this procedure has been shown to be robust with respect to violations of the normality assumption (see Huba & Harlow, 1986; Muthen & Kaplan, 1985; Tanaka & Bentler, 1985); (b) our sample size is reasonably large; and (c) for CFA models the distributions of the variables have little effect on the chi-square statistic or the standard errors of the factor loadings (Amemiya & Anderson, 1985; Anderson & Amemiya, 1985). Consistent with Osgood et al. (1988) and Wheaton (1988), we compared the fit of the alternative factor structure models using the Bentler-Bonett normed fit index (Bentler & Bonett, 1980), which is relatively independent of sample size and little affected by the distribution of variables, as well as using a chi-squared difference test.

PREADOLESCENT PROBLEM BEHAVIORS Table 1 Intercorrelations, Means, and Standard Deviations for Measured Indicators of Problem Behaviors Variable VI. School problems V2. Aggression V3. Acting out V4. Delinquency- 1 V5. Delinquency-2 V6. Delinquency-3 V7. Tobacco use V8. Alcohol use V9. Illicit drugs Boys M SD Girls M SD
VI V2 V3 V4 V5 V6 V7 V8

503

V9

_ .45* .49* .47* .43* .53* .31* .20* .12


0.15 0.75 -0.19 0.61

.44* .90* .31* .31* .39* .28* .11 .28* 0.19 0.39 0.07 0.23

.55* .84* .25* .21* .33* .18* .04 .15* 0.22 0.41 0.11 0.27

.52* .26* .33* .50* .54* .41* .34* .45* 1.41 0.60 1.22 0.44

.49* .20* .20* .51* .66* .38* .41* .36* 1.38 0.65 1.22 0.50

.39* .04 .08 .31* .51* .41* .36* .36* 1.31 0.68 1.14 0.39

.38* .20* .17* .29* .53* .44* .39* .37* 0.39 0.72 0.37 0.75

.28* .04 .07 .23* .41* .39* .61* .27* 0.73 0.94 0.56 0.82

This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

.37* .05 .04 .28* .44* .40* .60* .53* 0.22 0.56 0.21 0.57

Note. Coefficients above the diagonal are for girls (n = 204), below the diagonal are for boys (n = 222). *p<.001. the significant reduction in chi-square: x2(2, N = 204) = 38.8, p < .001, and the improvement in the Bentler-Bonett fit indices. Thus, school trouble, delinquency, and substance use were best represented as three separate but correlated factors for girls at this developmental stage. The correlations among the three factors (Table 3) were all positive and significant, suggesting that the factors share some, but not all, covariance in common. As was true of the best fitting model for boys, the Bentler-Bonett normed fit index for the best fitting three-factor model for the

girls (.86) approached the criterion recommended by Bentler and Bonett for an adequate absolute fit. With few exceptions, correlations among indicators of the same factor were higher than the correlations among indicators of different factors, suggesting that the measurement models have discriminant validity. Furthermore, the factor loadings for indicators of the best fitting models for both boys and girls were all significant, and all but one of the standardized factor loadings were greater than .3.

Table 2 Standardized Parameter Estimates and Measures of Fit for Four Models of Problem Behaviors (Boys) Model I Path
Fl, VI Fl, V2 Fl, V3 Fl, V4 Fl, V5 Fl, V6 Fl, V7 Fl, V8 Fl, V9

Model 2a Path
F1.V1 Fl, V2 Fl, V3 Fl, V4 Fl, V5 Fl, V6 F2, V7 F2, V8 F2, V9 Fl, F2

Model 2b Path
Fl, VI Fl, V2 Fl, V3 F2, V4 F2, V5 F2, V6 F2, V7 F2, V8 F2, V9 Fl, F2

Model3 Path
F1.V1 Fl, V2 Fl, V3 F2, V4 F2, V5 F2, V6 F3, V7 F3, V8 F3, V9 Fl, F2 Fl, F3 F2, F3

Coefficient
.64* .57* .49* .69* .73* .80* .55* .45* .48*

Coefficient
.66* .59* .52* .68* .72* .80* .68* .54* .55* .77*

Coefficient
.50* .97* .93* .69* .76* .80* .56* .50* .52* .44*

Coefficient
.50* .96* .93* .68* .77* .82* .64* .56* .56* .45* .34* .84* 128.12 24 222 5.34 0.87 0.83

x2

N *2/df NFI NNFI

df

420.52 27 222 15.57 0.56 0.43

404.32 26 222 15.55 0.58 0.44

136.84 26 222 5.26 0.86 0.83

Note. F = factor; VI = school problems; V2 = aggression; V3 = acting out; V4 = delinquency-1; V5 = delinquency-2; V6 = delinquency-3; V7 = tobacco use; V8 = alcohol use; V9 = illicit drugs. NFI = Bentler-Bonett (1980) normed fit index; NNFI = Bentler-Bonett non-normed fit index. *p<.001.

504

GILLMORE ET AL.

Table 3 Standardized Parameter Estimates and Measures of Fit for Four Models of Problem Behaviors (Girls)
Model I Path F1.V1 F1,V2 Fl, V3 F1,V4 F1.V5 Fl, V6 Fi,V7 F1.V8 F1.V9 Coefficient .66* .33* .36* .57* .74* .60* .73* .61* .64* Path F1.V1 Fl, V2 F1.V3 Fl, V4 F1.V5 Fl, V6 F2, V7 .F2, V8 F2, V9 F1.F2 Model 2a Coefficient .81* .52* .58* .64* .66* .50* .85* .71* .71* .61*
303.16 26 204 11.66 0.64 0.53

Model 2b Path F1.V1 Fl, V2 Fl, V3 F2, V4 F2, V5 F2, V6 F2, V7 F2, V8 F2, V9 F1.F2 Coefficient .56* .85* .99* .46* .70* .60* .80* .69* .70* .21**
155.89 26 204 6.00 0.82 0.78

Model3

Path
Fl, VI F1.V2 F1.V3 F2, V4 F2, V5 F2, V6 F3, V7 F3, V8 F3, V9 F1.F2 F1.F3 F2, F3

Coefficient .56* .85* .98* .58* .84* .62* .84* .72* .72* .29* .16*** .73* 117.09 24 204 4.88 0.86 0.83

This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

x2

df

x2/df
NFI NNFI

384.44 27 204 14.24 0.55 0.42

Note. F = factor; VI = school problems; V2 = aggression; V3 = acting out; V4 = delinquency-1; V5 = delinquency-2; V6 = delinquency-3; V7 = tobacco use; V8 = alcohol use; V9 = illicit drugs. NFI =

Bentler-Bonett (1980) normed fit index; NNFI = Bentler-Bonett non-normed fit index. *p<.001. **/><.01. ***p<.05.

These results contrast with those of Donovan and lessor (1985), who found that a single factor adequately accounted for the covariation among measures of deviance in older populations of teenagers and young adults. However, they used a different analytic techniqueexploratory factor analysis with maximum likelihood estimationto examine the factor structure of their data. It is possible that we might have obtained the same results with this technique. To examine this possibility, we reanalyzed the data using maximum likelihood estimation available in SPSSX. The results were consistent with the confirmatory factor analysis: A one-factor solution did not provide an adequate fit to the data for either the boys or the girls. As Donovan and lessor correctly noted, however, large samples may produce a significant chi-squared value with only trivial departures from the one-factor model. To guard against this possibility, we repeated our analyses with data from randomly selected samples of 100 girls and 100 boys. The results again indicated that a one-factor solution did not adequately account for the intercorrelations among the variables. We also reestimated the models using the generalized least squares solution available in EQS as a check on the possibility that the skewness in our measures may have influenced the results. Although there were some differences in the parameter estimates, the overall fit and patterns of residuals suggested that a one-factor solution did not provide an adequate fit to the data for either boys or girls. Thus, using three different methods of estimation, the same conclusion was reached. One caveat concerns the fact that two informants (teachers and students) provided data for the measures of school troubles, whereas measures of the other two factors were based solely on student re-

ports. Thus, the school factor may reflect both the informant and the behavior rated. This is not problematic so long as there is no interaction between informants and characteristics. Furthermore, the two-factor model that distinguished school troubles as a separate factor did not provide the best fit to the data. To further explore these findings, we examined the joint distributions of the variables to gain some insight into the extent to which these three different kinds of deviant behavior overlapped. For this purpose, we added the scores on all three indicators of a particular type of problem behavior to form a single index for school problems, delinquency, and substance use. For each of these three variables, we created a dichotomy of more serious deviance (defined as being 0.5 standard deviations above the mean on the variable) versus less serious deviance. The joint distributions of these three dichotomized measures of deviance supported the results of the factor analysis in that a higher proportion of youths engaging in serious deviant behaviors engaged in only one of the three types of behavior (56% of the boys and 60% of the girls) than were engaging in two (35% of the boys and 34% of the girls) or three (9% of the boys and 6% of the girls) of the behaviors. This was true when an even more stringent definition of deviance was used (those who are one standard deviation above the mean on the measure of deviance) or when a less stringent definition of deviance was used (ever engage in at least one of the behaviors making up the global measure).

Discussion
Confirmatory factor analysis was used to examine four hypothetical factor structures for the observed correlations among

PREADOLESCENT PROBLEM BEHAVIORS

505

different forms of problem behaviors in a sample of preadolescent sixth-grade students. Results for both the boys and girls suggested that a three-factor solution that differentiated school problem behaviors, delinquent behavior, and substance use represented the best fitting model, although for boys the improvement in fit over a two-factor model was minimal. Thus, while the three problem behaviors studied here are correlated as in previous studies, they do not appear to represent a single underlying dimension at this developmental period. Given the positive correlations among the factors, however, it may be that deviance is a unitary factor at a higher order of analysis. As Loeber (1985) noted, the developmental course of antisocial behavior is not well understood. Whether different antisocial behaviors have different etiologies or whether they are manifestations of a common etiology as suggested by lessor and colleagues is unclear. The empirical literature indicates that many, indeed most, youths outgrow the antisocial behaviors they engaged in during childhood and adolescence. Yet, for some, the pattern of antisocial behaviors persists into adulthood and is evident in career criminals. For the latter, Loeber postulates a dual track progression of antisocial behaviors. One group of youths engage in a variety of antisocial behaviors, typically at higher rates than other youths, whereas other youths tend to "specialize" in one type of antisocial behavior. These behaviors become more serious as the youth ages, but they remain similar in type. Our data indicate that of those who engaged in any antisocial behaviors, less than 10% engaged in all three types of problem behaviors. Whether these are the "versatiles" that Loeber described cannot yet be determined; however, we intend to observe this group as they progress through adolescence for evidence of stability of these antisocial behaviors. A somewhat larger group but still a minority of youths in our sample who engaged in any antisocial behavior engaged in two types of antisocial behavior (e.g., delinquency and drug use). If Loeber's hypothesis is valid, predicting which of these youths will progress toward increasing versatility in antisocial behavior and which will mature out of these behaviors becomes an important task for future research. Similarly, for those who engage in antisocial behavior, but only one typethe majority of our samplewe will want to understand why some become "specialists" and continue their antisocial behavior, while others outgrow these behaviors. Although there may be a common etiology that underlies the propensity to engage in antisocial behaviors as lessor and colleagues posit, to the extent that there are identifiable groups of youth who are "versatiles," others who are "specialists," and still others who "mature out" of their antisocial behaviors suggests that different etiological factors are involved in these different pathways. Clearly, more research is needed to better understand the developmental aspects of persistent antisocial behavior. In conclusion, the results from this study suggest that the "common factor" thesis in which different types of problem behaviors are hypothesized to reflect a single underlying proneness to deviance does not hold for preadolescent youths. Researchers studying younger populations should be cautious in combining measures of problem behaviors. In studying problem behaviors in younger populations, one would be prudent to obtain multiple measures of different behaviors and examine their structures before proceeding to analyses directed at substantive issues. Further study of the structure of problem behav-

iors among preadolescent youth is warranted. Should such studies demonstrate consistency with the findings obtained in our study, future research should examine the possibility of a higher order factor of deviance and a progression from a differentiated to a less differentiated structure of problem behaviors during adolescence. References
Achenbach, T., & Edelbrock, C. (1983). Manual for the child behavior checklist and revised child behavior profile. Burlington, VT: Department of Psychiatry, University of Vermont. Amemiya, Y., & Anderson, T. W (1985). Asymptotic chi-squaretests for a large class of factor analysis models (Technical Report No. 12). Stanford, CA: Stanford University. Anderson, T. W, & Amemiya, Y. (1985). The asymptotic normal distribution of estimators in factor analysis under general conditions (Technical Report No. 12). Stanford, CA: Stanford University. Bachman, J. G., O'Malley, P. M., & Johnson, J. (1978). Adolescence to adulthood: Change and stability in the lives of young men. Ann Arbor, MI: Institute for Social Research. Bentler, P. M. (1980). Multivariate analysis with latent variables: Causal modeling. Annual Review of Psychology, 31, 419-456. Bentler, P. M. (1986). Theory and implementation ofEQS: A structural equations program. Los Angeles: BMDP Statistical Software. Bentler, P. M., & Bonett, D. G. (1980). Significance tests and goodness of fit in the analysis of covariance structures. Psychological Bulletin, 88, 588-606. Bentler, P. M., & Newcomb, M. D. (1986). Personality, sexual behavior, and drug use revealed through latent variable methods. Clinical Psychology Review, 6, 363-385. Catalano, R., & Hawkins, J. D. (1986, November). The social development model: A theory of antisocial behavior. Paper presented at the Safeco lectureship on crime and delinquency, Seattle, WA. Donovan, J. E., & Jessor, R. (1985). Structure of problem behavior in adolescence and young adulthood. Journal of Consulting and Clinical Psychology, 53, 890-904. Donovan, J. E., Jessor, R., & Costa, F. M. (1988). Syndrome of problem behavior in preadolescence: A replication. Journal of Consulting and Clinical Psychology, 56, 762-765. Elliott, D, Huizinga, D, & Ageton, S. S. (1985). Explaining delinquency and drug use. Beverly Hills, CA: Sage. Elliott, D, Huizinga, D., & Menard, S. (1989). Multiple problem youth: Delinquency, substance use, and mental health problems. New \fork: Springer-Verlag. Elliott, D., & Morse, B. (1987). Drug use, delinquency, and sexual activity. In C. Jones (Ed.), Drug use and adolescent sexual activity, pregnancy, and parenthood (NIDA Research Monograph). Washington, DC: US. Government Printing Office. Ensminger, M. E., Kellam, S. G, & Rubin, B. R. (1983). School and family origins of delinquency. In K. T. Von Dusen & S. A. Mednick (Eds.), Prospective studies oj'crimeanddelinquency-(pp. 73-97). Boston: Kluwer-Nijhoff. Hawkins, J. D, & Lishner, D. (1987). Etiology and prevention of antisocial behavior in children and adolescents. In D. H. Crowell (Ed.), Childhood aggression and violence: Sources of influence, prevention and control (pp. 263-282). New York: Plenum Press. Hawkins, J. D, Lishner, D, Catalano, R. F., Jr., & Howard, M. (1986). Childhood predictors of adolescent substance abuse: Toward an empirically grounded theory. Journal of Children in Contemporary Society, 8, 11-48. Huba, G. J., & Harlow, L. L. (1986). Robust estimation for causal models: A comparison of methods in some developmental data sets. In P. B. Baltes, K. L. Featherman, & R. M. Lerner (Eds.), Life-span development and behavior (pp. 69-111). New York: Academic Press.

This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

506

GILLMORE ET AL. Muthen, B., & Kaplan, D. (1985). A comparison of methodologies for the factor analysis of non-normal Likert variables. British Journal of Mathematical and Statistical Psychology, 38,171-180. Osgood, D. W, Johnston, L. D., O'Malley, P. M., & Bachman, J. G. (1988). The generality of deviance in late adolescence and early adulthood. American Sociological Review, 53, 81-93. Smith, G. M., & Fogg, C. P. (1978). Psychological predictors of early use, late use and non-use of marijuana among teenage students. In D. B. Kandel (Ed.), Longitudinal research on drug use: Empirical findings and methodological issues (pp. 101-113). Washington, DC, Hemisphere. Tanaka, J. S., & Bentler, P. M. (1985). Quasi-likelihood estimation in asymptotically efficient covariance structure models. Proceedings of the 1984 meeting of the American Statistical Association (Social Statistics Section, pp. 659-662). Wheaton, B. (1988). Assessment of fit in overidentified models with latent variables. In J. S. Long (Ed.), Common problems/Proper solutions (pp. 193-225). Beverly Hills, CA: Sage. Zabin, L., Hardy, J. B., Smith, E., & Hirsch, M. B. (1986). Substance use and its relation to sexual activity among inner-city adolescents. Journal of Adolescent Health Care, 7, 320-331.

This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Jessor, R. (1987). Problem behavior theory, psychosocial development and adolescent problem drinking. British Journal of Addictions, 82, 331-342. Jessor, R., & Jessor, S. L. (1977). Problem behavior and psychosocial development: A longitudinal study of youth. New \brk: Academic Press. Johnston, L. D., O'Malley, P. M., & Eveland, L. K. (1978). Drugs and delinquency: A search for causal connections. In D. Kandel (Ed.), Longitudinal research on drug use: Empirical findings and methodological issues (pp. 137-156). Washington, DC: Hemisphere. Kandel, D, Kessler, R. C, & Margulies, R. Z. (1978). Antecedents of adolescent initiation into stages of drug use: A developmental analysis. In D. B. Kandel (Ed.), Longitudinal research on drug use: Empirical findings and methodological issues (pp. 73-99). Washington, DC: Hemisphere. Kaplan, H. B. (1985). Testing of a general theory of drug abuse and other deviant adaptations. Journal of Drug Issues, 15, 477-492. Loeber, R. (1985). Patterns and development of antisocial child behavior. Annals of Child Development, 2, 77-116. Loeber, R., & Schmaling, K. B. (1985). Empirical evidence for overt and covert patterns of antisocial conduct problems: A meta-analysis. Journal of Abnormal Child Psychology, 13, 337-352. Maruyama, G., & McGarvey, B. (1980). Evaluating causal models: An application of maximum-likelihood analysis of structural equations. Psychological Bulletin, 87, 502-512.

Received March 21,1990 Revision received September 4,1990 Accepted January 4,1991

Das könnte Ihnen auch gefallen