Journal o f Youth and Adolescence, Vol. 14, No. 6, 1985

Using Panel Data to Examine Sex Differences in Causal Relationships Among Adolescent Alcohol Use, Norms, and Peer Alcohol Use W i l l i a m R. D o w n s ~ Received February 19, 1985; accepted August 28, 1985

This paper uses longitudinal data and multiple regression of follow-up data on baseline data to identify direction o f causality among adolescent alcohol use, normative structure toward alcohol, and peer alcohol use. Baseline and follow-up data were collected on a random sample o f 100 adolescents (54 males). Separate regressions were performed on male and female respondents. A m o n g males, self-drinking and normative structure toward alcohol were f o u n d to have a reciprocal relationship over time. No significant relationship was f o u n d between self and peer alcohol use over time among males. A m o n g females, close-friend alcohol use was f o u n d to be causally prior to self drinking and other-friend drinking level. Normative structure toward alcohol was f o u n d unrelated to other variables over time among females.

INTRODUCTION Peer influence is widely assumed to affect the drinking behavior of adolescents. In addition, adolescents' normative structure concerning alcohol is believed to affect their use of alcohol. The alternative possibilities that adolescents influence the drinking levels of friends (either directly or by selecting friends based on self alcohol use) and that drinking behavior affects nor-

This research has been partially supported by grant #H84 AA 04026 from the National Institute on Alcohol Abuse and Alcoholism, Joan F. Robertson, principal investigator. ~Assistant Professor of Social Work, School of Social Work, State University of New York at Buffalo. He received his Ph.D. from the University of Wisconsin- Madison, School of Social Work. Research interests include effects of peer and family on adolescent deviant behavior, and effects of labeling on referral of adolescents to social service agencies. 469 0047-2891/85/[200-0469504.50/0 9

Plenum Publishing Corporation

470

Downs

mative structure are equally possible, but have received scant attention in the literature. A third possibility, that relationships among these three variables are reciprocal and complex, is also likely, but again, has received only scant attention in the literature. Numerous studies have used cross-sectional data to conclude that there is a relationship between peer alcohol use and adolescent drinking (Alexander, Jr., 1964; Biddle et al., 1980; Forslund and Gustafson, 1970; Globetti, 1973; Kandel et al., 1976; Kane and Patterson, 1972; Lassey and Carlson, 1980; Smart et a/., 1978; Sower, 1959; Wechsler and Thum, 1973). Other studies (Biddle et al., 1980; Maddox, 1964) have investigated adolescent norms as predictors of adolescent drinking behavior. Cross-sectional data, however, do not allow a clear ordering of the variables along a time dimension. Consequently, these studies could not establish the direction of causality among adolescent alcohol use, normative structure, and peer alcohol use. Despite a clear lack of empirical support for the hypotheses that attitudes toward alcohol and peer alcohol use are causally prior to self drinking among adolescents, even some studies with panel data have examined only the possibility of these variables affecting adolescent drinking. For example, Jessor and Jessor (1975) used longitudinal data to conclude that having friends who drink and having tolerant attitudes toward deviance predict onset of drinking among adolescents. The alternative hypotheses concerning the effect of alcohol use on attitudes and choice of friends were not tested. The alternative hypothesis that a reciprocal relationship exists among these variables also was not tested. The only panel studies testing causality in both directions have produced conflicting results. Igra and Moos (1979) used panel data to conclude that Time 1 dorm context (a measure of peer influence) predicted respondent Time 2 alcohol use, but that Time 1 alcohol use did not predict Time 2 dorm context. These results thus indicate a unidirectional effect o f peer influence on alcohol use. However, Britt and Campbell (1977) found that Time 1 respondent alcohol use had a much stronger effect on Time 2 peer influence than Time 1 peer influence had on Time 2 respondent alcohol use. These results thus support the opposite unidirectional effect of alcohol use on choice of peers. Consequently, direction of causality among peer alcohol use, adolescent alcohol use, and normative structure remains unclear. There are additional questions as well. Females are expected to be more attentive to interpersonal relationships than males (Bem, 1974) and thus may be more susceptible to peer influence than males. For example, Brown (1982) reported that adolescent females reported a greater degree of association between peer influence and drinking behavior than did males. Igra and Moos (1979) reported Time 1 peer drinking context to be a predictor of Time 2 college student female but not male drinking behavior. Urberg and Robbins

Adolescent Alcohol Use, Norms, and Peer Alcohol Use

471

(1982) reported that direction of peer influence on attitudes toward smoking differed across gender, with boys having more positive attitudes and girls more negative attitudes with increasing number of friends who smoked. Given these results and the differential sex roles in our society, direction of causality and degree o f reciprocity among peer alcohol use, adolescent use, and normative structure is likely to depend on the gender of the adolescent. Another question concerns level of association between the adolescent and the peer group. Braucht (1980) pointed out that the peer group as a whole may be a more important source o f influence than a single friend. Conversely, other studies have supported the importance o f closeness to friends in determining adolescent alcohol use (Margulies et al. 1977). These contradictory results imply the need to examine the differential effects of close-friend alcohol use and alcohol use by the remainder of the peer group on adolescent drinking. These issues have some important practical implications. Heavy drinking by adolescents may be related to delinquency (Schonfield, 1967), and may be a forerunner of drug use (Kandel et al., 1976). Prevention and intervention programs will differ depending on whether attitudes and peer alcohol use cause, result from, or are reciprocally related to, adolescent drinking; on whether close friends or the entire peer group has the greatest effect on alcohol use; and on whether causal direction and degree of reciprocity among these variables depends on the gender of the adolescent. The study uses panel data and multiple regression of follow-up data on baseline data to provide some answers to these questions.

METHOD Sample

The sample for this study was selected in two stages, in conjunction with an evaluation o f the Adolescent Drug and Alcohol Invervention and Referral (ADAIR) demonstration program. The ADAIR program has been described in more detail by Size et al. (1981). In the first stage, a telephone survey of a simple random sample (SRS) of 1014 adolescents aged 13-17 was completed using computer-assisted random-digit dialing. 2 By using randomdigit dialing, all households with telephones were included in the sample frame, thereby avoiding problems of adolescents who had been expelled or who otherwise had left school and o f households with unlisted phone n u m b e r s - b o t h a source of bias in studies of deviant behavior. :The 1980 sample and data were obtained by the Wisconsin Survey Research Laboratory through a subcontract with the grantee.

472

Downs

From this SRS sample, a comparison-group random sample for in-depth study was randomly selected from age-sex blocks to approximate the age-sex distribution of the A D A I R program. In this comparison-group random sample, which is the sample used in the present study, a total of 114 (64 male) adolescents (age 13-17) completed the baseline interview schedule and a total of 100 (54 males) completed the follow-up interview schedule for a response rate of 88%. Mean age at baseline among males was 15.8 years and among females it was 15.5 years. Respondents were administered a two-hour interview schedule. Readministration was scheduled, as closely as possible, for one-year time lag between baseline and follow-up.

Measure

of Variables

Three conceptually distinct sets of variables were used in this analysis: respondents' drinking behavior, respondents' attitudes toward alcohol, and drinking behavior of peers. There were two variables within each set, so that this study includes a total of six variables. Since this was a longitudinal study, each variable was measured both at baseline and follow-up. Table I contains the baseline and followup means and standard deviations of all six variables by sex of respondent.

Drinking Behavior of Respondents The Adolescent Alcohol Involvement Scale (AAIS) and Quantity Frequency (QF) Index were used. The AAIS (Mayer and Filstead, 1979) is a Table !. Means and Standard Deviations of Baseline and

Follow-Up Variables by Sex of Respondent Baseline Males Females AAIS 32.99 32.95 (9.52)" (9.18) QF index 0.34 0.27 (0.55) (0.32) Close-friend 2.19 2.34 drinking level (0.97) (0.94) Other-friend 2.52 2.54 drinking level (0.95) (0.99) Perceived harm 11.68 11.50 due to alcohol (1.75) (1.95) Attitudes toward 17.44 17.37 alcohol (4.74) (4.55) ~

Follow-up Males Females 34.54 33.02 (9.10) (8.40) 0.32 0.41 ( 0 . 4 0 ) (0.87) 2.55 2.59 (1.04) (0.92) 2.69 2.94 (0.86) (0.95) 10.91 11.57 (2.14) (1.78) 18.20 17.57 (4.99) (4.29)

in parentheses are standard deviations.

Adolescent Alcohol Use, Norms, and Peer Alcohol Use

473

14-item scale that ranges from 0 to 72, with high scores indicating high involvement with alcohol (as measured by items such as type of beverage, reason for drinking, age at first use, and negative consequences of drinking). Scores of 42 to 72 indicate alcohol problems. Coefficient alpha was 0.76 on the baseline and 0.79 on the follow-up sample. Moberg (1983) completed an extensive field test of the AAIS and found acceptable levels of internal consistency and test-retest reliability, as well as a high level of construct validity for the AAIS used as a research instrument. The QF Index provides an index of alcohol consumption for the past month by averaging the daily consumption alcohol in ounces (Jessor et al., 1975). The QF Index has a lower limit of zero (no alcohol consumed in past month), and high scores indicate a high amount of alcohol consumed in the past month. Test-retest reliability over an average six-week interval at followup was 0.85.

Attitudes To ward AIcohol The Attitudes Toward Alcohol scale was used (Goodstadt et al., 1978). This six-item scale measures level of support for alcohol use and ranges from 6 (rejection or negative attitudes) to 30 (support or positive attitudes). Coefficient alpha on the baseline sample was 0.69 and on the follow-up sample was 0.75. A second measure was the Perceived H a r m due to Alcohol scale (Johnston et al., 1977), which measures risk or harm of drinking alcohol. Risk or harm is considered in terms of physical, psychological, and social harm to self on 4-point items. The scale contains four items, and can range from 4 (no risk or harm perceived) to 16 (high risk or harm). A coefficient alpha of 0.71 was obtained on the baseline sample and an alpha of 0.59 was obtained on the follow-up sample.

Peer Drinking Behavior Peers were divided into close friends and other friends. Close friends were defined as those friends whom respondent trusted, confided in, and spent considerable leisure time with. A 5-point scale ranging from 1 (abstainer) to 5 (heavy drinker) was then used. Number of close friends at each point on the scale was obtained and multiplied by the number value of that scale point. Finally, the total value was divided by the total number of close friends. For example, a respondent with two close-friend abstainers (scale value = 1) and one close-friend occasional drinker (value = 2) would have an average

Downs

474

close-friend drinking level of 1.33. Test-retest reliability over an average sixweek interval at follow-up was 0.90. Other friends were defined as all other acquaintances with whom respondent spent leisure time and regarded as a friend, and basically included the remainder of the respondent's peer group. The range of number of other friends precluded a detailed measure of this variable (as obtained with close friends). Instead, respondents were asked to categorize "most o f their other friends" on the 5-point item ranging from 1 (abstainer) to 5 (heavy drinker). Test-retest reliability over an average six-week interval at follow-up was 0.87.

Data Analysis The purposes of this study include a longitudinal examination of the reciprocal effects among drinking behavior, normative structure, and peer alcohol use. Use of bivariate correlations in cross-lagged panel analysis has been criticized for its assumption of a system of variables that is at disequilibrium at Time 1 and equilibrium at Time 2 when other assumptions (e.g., equilibrium at Time 1 and disequilibrium at Time 2) are equally possible (Shingles, 1976). These alternative assumptions have strong adverse effects on the interpretation of the cross-lagged correlations (Shingles, 1976). In addition, use of bivariate correlations fails to control for rival hypotheses, such as attentuation and spurious effects due to external causes (Shingles, 1976), and ignores the causal influence of a variable on itself over time (Rogosa, 1979). Table II contains all synchronous and diachronous bivariate correlations by sex of respondent for those interested in these for such a crosslagged analysis. Multiple regression of follow-up variables on baseline variables is a more useful method to disentangle reciprocal effects. This method has the advantage o f controlling for spuriousness in nonrecursive systems that results from variables having opposite or incongruent effects on each other (Shingles, 1976). In addition, this method makes no assumptions as to equilibrium in the system of variables. However, comparison of cross-lagged regression coefficients still presents some difficulties. If the scales of the variables differ (as they do, see Table I), use of the unstandardized coefficients will tend to attribute greater causal effect to the variables with the larger units of measurement (such as the AAIS). This asymmetry cannot be ignored when assessing the relative influence o f two or more variables (Shingles, 1976). Similarly, the standardized coefficient is sensitive to differences in variances between variables. Therefore, comparison o f cross-lagged regression coefficients to make conclusions as to the relative strengths of the cross-lagged causal relationships

Adolescent Alcohol Use, Norms, and Peer Alcohol Use

475

I

66

o

6

I

I

I I

0

oo

o

c~

I

66

66 I

o

d

dd

I I

od

6

6

I

I

I

6 I

o

cs

I

I

I o

oo

o

o

,o

.

. oo

.

. o

6 I

66

o

6 ~

. o

6

~

6

~ cs

6

6 I

~

~

~

6

o

6 I

cs

E o

c5 "0

o

oo

o

o

6 I

oo

o

o

6 I

6

66

o

6

6 ]

6

66

6

o

6 I

6

e~

o

en

I I

I

I

I I

I I

I

I

o

I

0

c- o 0

Using panel data to examine sex differences in causal relationships among adolescent alcohol use, norms, and peer alcohol use.

This paper uses longitudinal data and multiple regression of follow-up data on baseline data to identify direction of causality among adolescent alcoh...
901KB Sizes 0 Downloads 0 Views