Addictive Behaviors, Vol. 16, pp. 21-28, Printed in the USA. All rights reserved.
1991 Copyright
0306-4603/91 $3.00 + .OO Q 1991 Pergamon Press plc
GRADE CHANGES
IN PEER INFLUENCE ON ADOLESCENT CIGARETTE SMOKING: A COMPARISON OF TWO MEASURES
KATHRYN
A. URBERG,
CHIEN-HUNG
CHENG,
and SHIANG-JEOU
SHYU
Wayne State University Abstract - The effects of peer influence on adolescent cigarette smoking were investigated in longitudinal study of 309 white, middle-class subjects in the 8th and 11th grades. Subjects provided data on their smoking behavior, the proportion of their friends who smoked, and the identity of their best friend. Data from the person named as best friend was used to measure peer smoking, rather than the adolescent’s perceptions of the friend’s smoking. Peer influence was defined as the difference between the subject’s smoking behavior and that of their best friend. This definition (PIl) minimized the confounding of peer influence with selective association. The effects of peer influence on change in smoking behavior were found to be stronger for 8th than 11th graders and for boys than girls. When the proportion of friends who smoke (PI2) was used as the measure of peer influence, the effects of peer influence were stronger for 1 lth than 8th graders. The relative merits of the two measures are discussed, and the argument is made that the difference between friend and adolescent smoking is a more appropriate measure of peer influence than the proportion of friends who smoke.
Since peer smoking is the single best predictor of adolescent smoking (Flay, d’Avemas, Best, Kersell, & Ryan, 1983), peer influence is generally considered to be the major cause finding, a causal of adolescent cigarette smoking. From this basically correlational relationship has been inferred. The assumption is usually made that the entire relationship between peer and adolescent smoking is due to peer influence, although little is known about how the influence may be actually transmitted. Modeling and peer pressure are the transmission pathways most often mentioned in the literature. However, some part of the relationship between peer and adolescent smoking may be due to selective association. That is, to the extent that selection of a friend is based on similar smoking status, the observed relationship between peer and adolescent smoking would be accounted for by selective association rather than by peer influence. Cohen (1983) points out that, depending on the degree of similarity, peer influence may act either to produce stability or change. If peers are similar on a dimension such as cigarette smoking, peer influence would result in stability or no change in the current behavior. However, if peers were discrepant with respect to smoking behavior, there would be pressure for one or both of the adolescents to change their behavior in the direction of the other. Selective association with peers who are perceived to be like the self is well documented in the sociology and social psychology literature (Cohen, 1977; Epstein, 1989; Kandel, 1978a, 1978b). Quite broadly based similarities between friends can occur because many traits and behaviors are not independent. For example, adolescents who smoke are also more likely than the average adolescent to drink, to be sexually active, and to have values that do not focus on school and the family (Jessor & Jessor, 1977). The more important the characteristic is to the adolescent, the more likely it is to serve as a basis for friendship
Research for this study was supported by NIH grand HD18425 awarded to Kathryn Urberg. We wish to thank Lori Violante for assistance in data gathering. We gratefully acknowledge helpful discussions with Thomas Bemdt, Laurie Chassin, Steven Sherman, and Clark Presson on these issues. Requests for reprints should be sent to Kathryn Urberg, Department of Psychology, Wayne State University, Detroit, MI 48202. 21
22
KATHRYN
A. URBERG,
CHIEN-HUNG
CHENG,
and SHIANG-JEOU
SHYU
selection. However, Cohen (1983) points out that some forms of similarity are too unimportant to result in friendship choice, so that on less important dimensions, a considerable degree of heterogeneity may be present among close friends. He also reports his earlier finding that peer conformity on a characteristic was negatively related to the importance of that characteristic for selection of friends. He concludes that it may be largely on the aspects of a relationship considered too unimportant for peer selection that the pressures for homogeneity are most effective in producing change. Two studies have examined the relative contributions of peer influence and selective association to similarity among adolescent friends (Cohen, 1977; Kandel, 1978b). Both concluded that selective association and influence each accounted for about 50% of the similarity, although Cohen found some characteristics affected by association but not by influence and vice versa. In most studies of adolescent smoking, the study design has not permitted the separation of the effects of peer influence from the effects of selective association. This has almost certainly resulted in overestimates of the amount of peer influence. Cohen’s (1983) discussion of peer influence suggests a way that a longitudinal study could be used to examine peer influence relatively uncontaminated with the effects of selective association. The initial difference between the adolescent and the peer on the variable of interest, according to Cohen’s view of influence, should serve as a measure of the strength of the pressure for one or both of the adolescents to change on that variable. Thus, since only diflerences between the adolescent and peer in smoking behavior are assumed to predict changes over a period of time in smoking behavior, any similarity as a result of selection on that variable does not contribute to predicting behavior change. This variable is only relatively unconfounded with selective association, since if selection for friends is made on some, as yet unidentified, variable that is causally related to cigarette smoking, peer influence would still be confounded to an extent with selective association. Another factor that has probably also contributed to the overestimation of the amount of peer influence in adolescent smoking is the common practice of using the subject’s estimates of the number of their friends who smoke as the criteria of peer smoking. Several studies (Evans. 1982; Sherman, Presson, Chassin, Corty, & Olshavsky, 1983) have documented the fact that adolescent smokers overestimate their friends’ smoking. To eliminate this source of bias in the results, data should be available from the person that the adolescent designates as a best friend so that the friends’ self-reports of how much they smoke can be used instead of, or in addition to, the report of the subject as to how much their friends smoke. This study is designed to examine the effects of peer influence on adolescent cigarette smoking when potentially confounding influences such as the influence of selective association and overestimation of friend smoking are minimized. For comparison purposes, data analysis will also be done with a frequently used measure of peer influence, the proportion of friends who smoke. Grade and sex differences in susceptibility to influence will also be examined. METHOD
Procedure
In Year 1 of the study, a questionnaire was administered to all 8th graders in a junior high school and all 1 lth graders in the high school fed by that junior high school. A passive consent form was used. About 1% of the parents refused permission for their child to participate, and another 1% of the students refused to participate on the day of the study or turned in unusable forms. The questionnaire was given in school during class time, at which time a saliva sample was also collected. The students were told that an analysis of the saliva
Peer influence
23
sample would reveal how much they had smoked in the past week. Subjects were also asked to write down the names of their two best friends in that grade in their school. The data for each student was matched to that of the first-listed best friend. If no match could be made to the friend listed first, an attempt was made to match to the friend listed second. About 70% of the students could be matched to the first-listed friend. Overall, 85% of the subjects were matched to a friend. In Year 2, a shortened form of the questionnaire was given to all 9th-grade students in the junior high school and to about half of the 12th graders in the high school. Scheduling problems with the school, unfortunately, made it impossible to include all the 12th graders in the second year. Each semester, half of the 12th graders were in a required English class; since our follow-up was done in the spring, only half of the 12th grade sample was available. Subjects
The sample consisted of 309 predominantly middle-class students in a midwestern suburban school system. These were subjects who were matched to a friend in Year 1 and were also present in Year 2. This sample was 49.1% male, 50.9% female, and 96.2% white. The majority of the nonwhite students were Oriental. The sample contained 122 subjects who were in the 8th grade and 187 who were in the 1 lth grade in Year 1. The Year 1 subjects who could not be matched to a friend did not differ from those who could be matched on sex, grade, ethnic background, or on the smoking variables (multivariate F = 0.6394; p = .699). The subjects who were present in both Year 1 and Year 2 differed significantly (multivariate F = 13.6768; p = .OOO)from those who were present only in Year 1 in grade (more 1 lth graders were not matched). With grade controlled, the subjects who were present in both years did not differ from those who were present only in Year 1 in amount smoked or in the proportion of friends who smoke. Measures
Two items were used from the questionnaire to measure adolescent cigarette smoking. The first item, “How many cigarettes did you smoke last week?” provided a quantitative measure of smoking. The second item was used only to dichotomize subjects into two stages of smoking: those who had never smoked and those who had ever smoked. The saliva sample analyses of the subjects were not used in the present analyses as an index of smoking. This is because saliva thiocyanate is not sensitive enough, particularly in adolescents, to reliably detect individual differences in cigarette smoking (Pechacek et al., 1984). However, collecting the sample probably served as a bogus pipeline (Jones & Sigall, 1971) to increase the accuracy of the subject’s report of his or her smoking. Peer influence was measured in two ways. First, based on Cohen (1983), peer influence (PIl) was defined as the difference between the Year 1 self-reported smoking of the person that the subject named as one of their two best friends and that of the subject (friend smoking-subject smoking). Thus, if the friend smokes more than the subject in Year 1, we would predict an increase in smoking from Year 1 to Year 2 for the subject if peer influence was operating. Second, peer influence (P12) was measured by the response to a question asking the subject to indicate, on a 5-point scale, what proportion of his or her friends smoked. RESULTS
The variables were first examined for mean differences in level as a function of grade and sex. Accordingly, a multivariate analysis of variance (MANOVA) was performed to examine grade, sex effects on the quantity smoked, the proportion of friends who smoke, and the
24
KATHRYN
Table
A. URBERG,
CHIEN-HUNG
CHENG,
and SHIANG-JEOU
1, Hierarchical influence
SHYLr
regressions predicting Year 2 smoking from Year I smoking, (PII), and its interactlons with grade, sex. and smoking change F
R? Change
IV Year 1 smoking Sex Grade Smoke stage Peer inf. by grade Peer inf. by sex
.8483
Change
Sig. Change
569.809
,000
.0155 .0243 .0323
4.307 22.210 34.053
.006 ,000 ,000
Simple Correlation
.84X3 .I355 .2366 .2297 - .37x3 - .4800
peer
Beta .84X3 .083X .0880 .0356 - .5815 - .2505
Overall: Multiple R = .8916. R’ = .7949, Adj R’ = .7X83. F(8. 272) = 119.63. 17 = .OOO
difference between friend smoking and subject smoking (PIl). The overall tests for grade (F(3, 306) = 15.93, p = .OOO), and sex (F(3, 306) = 5.89; p = ,000) were significant. The univariate analyses were examined to locate the significant effects. Univariate analyses showed significant grade effects for amount smoked (F = 18.11; df = 1,308; p = ,000) and proportion of friends who smoked (F 47.33, df = 1, 308; p = .OOO).The univariate analyses showed significant sex effects only for the proportion of friends who smoke (F = 14.00; df = 1, 308; p = .OOO.) The means indicate more smoking for 1 lth (M = 10.37) than 8th (M = .95) graders, and a greater proportion of friends who smoke for 1 lth (M = 2.41) than 8th (M = 1.74) graders. The sex effect indicates that girls (M = 2.27) reported having a larger proportion of friends who smoked than did boys (M = 1.91). Correlations between subject smoking and friend smoking were as follows: 8th graders, r = .0857; 1 lth graders, r = .4745; (n = 309). The correlation between PI1 and PI2 was r = - .0149. Difference
between friends
as a measure
of peer irlfluence
The next analyses used hierarchical regression to predict Year 2 subject smoking from Year 1 smoking; grade, sex, and stage of smoking (entered as a group); PI1 ; the interactions of PI1 with grade, sex, and stage of smoking (entered in that order). Year 1 smoking was entered first to control for the stability of smoking from Year 1 to Year 2, thus, any significant R2 change attributed to other variables is accounting for change in smoking from Year 1 to Year 2. Grade, sex, and stage of smoking were entered as a group, since we know there are differences in amount smoked as a function of these variables. Of greatest interest are the interactions of peer influence with grade, sex, and stage of smoking. Interactions that account for a significant increment in R* would indicate that peer influence is operating differentially with respect to that variable. The significant beta weights from this analysis are shown in Table 1. The peer influence variable was significantly related to change in smoking only in its interactions with grade and sex. Accordingly, the regression was run separately for the four grade and sex groups. The results of these analyses are as follows: 8th grade males, B = .7261, p = .4165; Sth-grade females, B = - .0310, l3 = - .0272; 1 lth-grade males, B = .1021, l3 = .0843; llth-grade females, B = - .0679, l3 = - .0416. Since within-group variance on the variables of interest is likely to be different for 8th compared to 1lth graders and for males compared to females. the unstandardized (B) as well as the standardized @) regression coefficients are reported. When differences between groups are of interest, the recommendation is to compare unstandardized coefficients. When the differences between different variables within a single group are of interest, the
Peer influence
25
Table 2. Hierarchical regressions predicting Year 2 smoking fromYear 1 smoking, proportion of smoking friends (PI2). and its interactions with grade, sex, and smoking stage
IV Year 1 smoking PI2 PI2 by grade
R2 Change
F Change
Sig. Change
Simple Correlation
.7270 .0163 .0065
593.823 14.109 5.908
.Ocil ,008 .016
.8526 .4694 .4520
Beta .8526 .I494 .2980
Overall: Multiple R = .87225, R* = .7608. Adj. RZ = .7519, F(8, 256) = 85.8856, p = .OOO.
recommendation is to compare the standardized beta coefficients (Cohen & Cohen, 1975; Kenny, 1979). Note that smoking (or nonsmoking) behavior is highly stable from Year 1 to Year 2. Most subjects were nonsmokers in year 1 who remained nonsmokers in Year 2. Proportion offriends who smoke as a measure of peer influence The analyses described above were repeated using the proportion of friends who smoke (P12) as the measure of peer influence rather than the difference between the smoking of the subject and that of their best friend (PIl). The results of these regressions for Year 2 smoking can be seen in Table 2. Only significant beta weights are shown. Peer influence and the interaction of grade with peer influence accounted for significant increments of variance. Accordingly, the regressions were run separately for 8th and 11th graders. These results show that peer influence appeared to be stronger in 1 lth graders (B = 6.180, l3 = .1785) than 8th graders (B = 2.024, p = .1445). Comparison of PI1 and PI2 Each of the peer influence variables has been shown to be a significant predictor of Year 2 smoking. In order to examine the relative strength of the two measures and to determine if they share variance in common, two further regressions were run. In the first analysis, Year 1 smoking, sex, and grade were entered as a block, PI1 and its interactions with sex and grade were entered as a second block, and PI2 and its interactions with sex and grade were entered as the final block. In the second analysis, PI2 and its interactions with sex and grade were entered before PI1 but after Year 1 smoking, sex, and grade. Since stage of smoking had no significant effects in earlier analyses, it was omitted from these analyses. The results of these analyses can be seen in Table 3. The results show that the amount of variance accounted for is about the same for each peer influence variable, regardless of the order of entry into the regression equation. This suggests that they are accounting for quite independent portions of variance in prediction. This is consistent with the very low correlation between the two influence variables. PI1 accounts for about three times more variance than does P12. DISCUSSION
This study examined peer influence in adolescent cigarette smoking using two measures of peer influence. One is a new measure that is relatively unconfounded with selective association. For this measure, only differences between friends in their smoking behavior were assumed to predict changes for either adolescent in their smoking behavior. Using this measure, peer influence was found to interact with both grade and sex to predict changes in cigarette smoking. Looking at the effect separately by grade and sex showed that 8th-grade boys appeared to be the most influenced by their peers of any of the four groups.
26
KATHRYN
A. URBERG.
CHIEN-HUNG
CHENG,
and SHIANG-JEOU
SHYU
Table 3. Hierarchical regressions comparing PI1 and its interactions with grade and sex to PI2 and its interactions with grade and sex in prediction Year 2 smoking
IV Analysis 1 Year 1 smoking, grade, sex PI1 and Interactions PI2 and interactions Analysis 2 Year 1 smoking, grade, sex PI2 and Interactions PI1 and interactions
R2
F
Change
Change
Sig. Change
.7423
212.192
.0555
Simple Correlation
Beta
.OOO
.8526
.8324
19.937
.cQO
.3428
.4232
.0167
6.451
,000
.4520
.2728
.7423
212.192
,000
.8526
.X324
.0164
4.952
,002
.4520
.2858
.0557
21.531
,000
.3428
.4357
Overall (for both analyses): Multiple R = .90255, F(9, 256) = 102.3323, p = ,000.
R’ = .8145,
Adj. R2 = .8067.
When the proportion of friends who smoke was used as the measure of peer influence, as has been done in most previous studies, the results were different. There was a significant peer influence by grade interaction that suggested that the older group was more influenced by peers than the younger group. Thus, the two measures of peer influence indicate opposite grade effects. The existance of grade differences in peer influence in adolescent cigarette smoking is currently a matter of dispute. Three recent studies, all using the proportion of friends who smoke as the measure of peer influence, found different grade effects. Krosnick and Judd’s (1982) cross-sectional study found that the path from peer influence to smoking was stronger for 14-year-olds (probably 8th graders) than it was for 1 1-year-olds (probably 5th graders). Ary, Biglan, Gallison, Wiessman, and Severson’s longitudinal study (1983) found peer influence to be stronger for 9th and 10th graders than 7th graders for initial nonsmokers only. In contrast, Chassin, Presson, Sherman, Montello, and McGrew’s (1986) longitudinal study found little evidence of grade differences. They attributed the difference between their finding and that of Krosnick and Judd to the longitudinal methodology that enabled them to minimize the confounding of selective association with peer influence. This was done by analyzing only data from the adolescents who were initial never smokers or initial triers. Thus, any selective association as a result of being a regular smoker was removed by not having regular smokers in the sample. To understand these somewhat conflicting results, it is necessary to consider the peer influence variable used. First, the proportion of friends who smoke variable is confounded with both selective association and the results of past influence. Thus, it may or may not index any difference between the adolescents’ smoking and that of their friends. Only for initial never-smokers would the proportion of smoking friends indicate unambiguously that there was a difference between the smoking behavior of the adolescents and that of their friends. Second, the variable represents influence from a wide group of friends that probably includes casual acquaintances as well as close friends. Third, both the proportion of friends who smoke and the number of adolescents who smoke were shown in this sample to increase with age. Except for the Chassin et al. study (1986). which found no grade effect. the studies
Peer influence
27
(including the present one) using the proportion of friends who smoke as the index of influence conclude that the older group is more susceptible to influence than the younger group. This suggests that the mutual confounding of friend smoking and subject smoking with age may account for the findings. In contrast, the discrepancy between adolescent smoking and friend smoking used in the present study was not a function of age and, thus, is free of this confounding. The reasons, then, for the differences in the results using the two measures of peer influence may be due (a) to the confounding of the proportion of friends who smoke variable (P12) with selective association, (b) to its confounding with age, or (c) to the fact that this variable measures influence from a different source than the peer influence variable (PIl). We believe that defining peer influence as a difference in behavior between the adolescent and his or her friend more unambiguously measures peer influence than does the proportion of friends who smoke; thus, we conclude that early adolescents, especially boys, appear to be more susceptible to influence from peers than are older adolescents. This conclusion is compatible with the findings of several previous studies on conformity. Bemdt (1979), using hypothetical situations, asked children from the 3rd to 12th grades whether or not they would go along in a situation proposed by peers. For the three types of behavior examined, the peak in conformity to peers occurred in early adolescence. For antisocial behavior, boys were more apt to be conforming than girls. Bixenstine, DeCorte, and Bixenstine (1976) also reported boys were more willing to go along with mischief or antisocial behavior than girls. Recent reviews of smoking research have suggested that the causes of the onset of smoking may be different than the causes of maintaining or increasing smoking (Flay et al., 1983: Leventhal & Cleary, 1980). There is little support in the present study for this view. Stage of smoking did not interact with either measure of peer influence, suggesting that there was no difference in the impact of peer influence for those moving from no smoking to some smoking compared to those moving from some smoking to a changed level of smoking. Although the theoretic argument for differential causes is convincing, to this point there is little empiric evidence as to what the differential causes might be. Ary et al. (1983), Chassin et al. (1986), and Chassin, Presson, & Sherman (1984) all report evidence of different predictors of smoking as a function of smoking stage. However, there is little convergence in the predictors found to be associated with each stage. The correlations found here in Year 1 smoking behavior between adolescent and best friend indicate the initial similarity for 8th and 11th graders. The insignificant correlation between friend and adolescent smoking for 8th graders suggests that, at this age, smoking behavior is not itself a variable on which friends are selected. Smoking is not a common behavior at this age in this sample, and smokers are likely to have nonsmoking friends. The higher correlation for smoking behavior within friendship pairs for 11 th graders is likely to be a result of both selective association on the basis of cigarette smoking, or some variable correlated with it, and the cumulative effects of peer influence. While selective association was not directly examined here, a review of the literature on friendship selection by Epstein (1989) suggests that similarity is likely to play a more important role in friendship selection in adolescents compared to elementary-school-age children. In conclusion, we argue that peer influence is more clearly measured by PIl, based on the difference between the smoking of the adolescent and that of their best friend, than by P12, the percentage of friends who smoke. Based on this measure, the younger boys appeared more influenced by the smoking of their best friend than younger girls or older adolescents. In fact, the 8th graders in this sample exemplify the conditions that Cohen believes are conducive to peer influence. That is, smoking is not important enough to serve as a basis for friendship selection, and friendships are relatively heterogeneous with respect to smoking
28
KATHRYN
A. URBERG,
CHIEN-HUNG
CHENG,
and SHIANG-JEOU
SHYU
behavior. For the 1 lth graders, where there is more initial homogeneity, evidence of peer influence from the best friend.
there was less
REFERENCES Ary, D., Biglan, A., Gallison, C., Weissman, W., & Severson, H. (1983). Longitudinal prediction of the onset and change of adolescent smoking. In W.F. Forbes, R.C. Frecker, and D. Nostbakken (Eds.), Proceedings ofthe Fifth World Conference on Smoking and He&h. Winnipeg: Canadian Council on Smoking and Health. Bemdt. T. (1979). Developmental changes in conformity to peers and parents. De~elopmenral Psychology, 15, 608-616. Bixenstine, V.E., Decorte, M.S., & Bixenstine, B.A. (1976). Conformity to peer-sponsored misconduct at four grade levels. Developmental Psychology, 12, 226236. Chassin, L., Presson, C.C.. & Sherman, S. (1984). Cigarette smoking and adolescent psychosocial development. Basic and Applied Social Psychology,
5. 295-3 15
Chassin. L., Presson. C., Sherman, S., MontelIo, D., & McGrew, J. (1986). Changes in peer and parent influence during adolescence: Longitudinal versus cross-sectional perspectives on smoking initiation. Developmenral Psychology,
22, 327-334.
Cohen, J. (1977). Sources of peer group homogeneity. Sociolog? of Education, 50, 227-241. Cohen, J. (1983). Commentary: The relationship between friendship selection and peer influence. In J.L. Epstein & N. Karweit (Eds.), Friends in school. New York: Academic Press. Cohen, J., & Cohen, P. (1975). Applied mulfiple regressionicorrelcrtion analysis for the behavioral sciences. New York: Wiley & Sons. Epstein, J. (1989). The selection of friends. In T. Bemdt & G. Ladd (Eds.). Peer relafionships in child development. New York: Wiley. Evans, R.I. (1982). Smoking deterrence: Prospects, concerns. Paper presented at American Psychological Association Meetings. Washington, DC. Flay, B., d’Avemas, J., Best, J.A., Kersell, M., & Ryan, K. (1983). Cigarette smoking: Why young people do it and ways of preventing it. In P. McGrath & P. Firestone (Eds.), Pediatric and adolescent behavioral medicine. New York: Springer-Verlag. Jessor, R., & Jessor, S. (1977). Problem behavior and psycho-social development: A longitudinal study of youth. New York: Academic Press. Jones, E., & Sigall. H. (1971). The bogus pipeline: A new paradigm for measuring affect and attitude. Psychological
Kandel,
Psychology,
Kandel,
Bulletin,
D. (1978a).
76. 349-364.
Similarity
in real-life
adolescent
friendship
pairs.
Journal
of Personaliv
and
Social
36, 306312.
D. (1978b).
Homophily,
84, 427436. D. (1979). Correlation
selection.
and socialization
in adolescent
friendships.
American
Journal
of
Sociology,
Kenny, and causality. New York: Wiley & Sons. Krosnick. J.A., & Judd, C.M. (1982). Transitions in social influence at adolescence: Who induces cigarette smoking? Developmental Psychology, 18, 359-368. Leventhal, H., & Cleary, P. (1980). The smoking problem: A review of the research and theory in behavioral risk modification. Psychological Bulletin, 88, 370408. Pechacek, T.F.. Murray, D.M., Luepker, R.V., Mittelmark. M.B.. Johnson, C.A., & Shultz, J.M. (1984). Measurement of adolescent smoking behavior: Rationale and methods. Journal of Behavioral Medicine, 7, 17-25. Sherman. S.J., Presson, C.C., Chassin, L.. Corty. E.. & Olshavsky, R. (1983). The false consensus effect in estimates of smoking prevalence: Underlying mechanisms. Personali~ cmd Social Psychology Bulletin, 9, 197-208.
1991 Copyright
0306-4603/91 $3.00 + .OO Q 1991 Pergamon Press plc
GRADE CHANGES
IN PEER INFLUENCE ON ADOLESCENT CIGARETTE SMOKING: A COMPARISON OF TWO MEASURES
KATHRYN
A. URBERG,
CHIEN-HUNG
CHENG,
and SHIANG-JEOU
SHYU
Wayne State University Abstract - The effects of peer influence on adolescent cigarette smoking were investigated in longitudinal study of 309 white, middle-class subjects in the 8th and 11th grades. Subjects provided data on their smoking behavior, the proportion of their friends who smoked, and the identity of their best friend. Data from the person named as best friend was used to measure peer smoking, rather than the adolescent’s perceptions of the friend’s smoking. Peer influence was defined as the difference between the subject’s smoking behavior and that of their best friend. This definition (PIl) minimized the confounding of peer influence with selective association. The effects of peer influence on change in smoking behavior were found to be stronger for 8th than 11th graders and for boys than girls. When the proportion of friends who smoke (PI2) was used as the measure of peer influence, the effects of peer influence were stronger for 1 lth than 8th graders. The relative merits of the two measures are discussed, and the argument is made that the difference between friend and adolescent smoking is a more appropriate measure of peer influence than the proportion of friends who smoke.
Since peer smoking is the single best predictor of adolescent smoking (Flay, d’Avemas, Best, Kersell, & Ryan, 1983), peer influence is generally considered to be the major cause finding, a causal of adolescent cigarette smoking. From this basically correlational relationship has been inferred. The assumption is usually made that the entire relationship between peer and adolescent smoking is due to peer influence, although little is known about how the influence may be actually transmitted. Modeling and peer pressure are the transmission pathways most often mentioned in the literature. However, some part of the relationship between peer and adolescent smoking may be due to selective association. That is, to the extent that selection of a friend is based on similar smoking status, the observed relationship between peer and adolescent smoking would be accounted for by selective association rather than by peer influence. Cohen (1983) points out that, depending on the degree of similarity, peer influence may act either to produce stability or change. If peers are similar on a dimension such as cigarette smoking, peer influence would result in stability or no change in the current behavior. However, if peers were discrepant with respect to smoking behavior, there would be pressure for one or both of the adolescents to change their behavior in the direction of the other. Selective association with peers who are perceived to be like the self is well documented in the sociology and social psychology literature (Cohen, 1977; Epstein, 1989; Kandel, 1978a, 1978b). Quite broadly based similarities between friends can occur because many traits and behaviors are not independent. For example, adolescents who smoke are also more likely than the average adolescent to drink, to be sexually active, and to have values that do not focus on school and the family (Jessor & Jessor, 1977). The more important the characteristic is to the adolescent, the more likely it is to serve as a basis for friendship
Research for this study was supported by NIH grand HD18425 awarded to Kathryn Urberg. We wish to thank Lori Violante for assistance in data gathering. We gratefully acknowledge helpful discussions with Thomas Bemdt, Laurie Chassin, Steven Sherman, and Clark Presson on these issues. Requests for reprints should be sent to Kathryn Urberg, Department of Psychology, Wayne State University, Detroit, MI 48202. 21
22
KATHRYN
A. URBERG,
CHIEN-HUNG
CHENG,
and SHIANG-JEOU
SHYU
selection. However, Cohen (1983) points out that some forms of similarity are too unimportant to result in friendship choice, so that on less important dimensions, a considerable degree of heterogeneity may be present among close friends. He also reports his earlier finding that peer conformity on a characteristic was negatively related to the importance of that characteristic for selection of friends. He concludes that it may be largely on the aspects of a relationship considered too unimportant for peer selection that the pressures for homogeneity are most effective in producing change. Two studies have examined the relative contributions of peer influence and selective association to similarity among adolescent friends (Cohen, 1977; Kandel, 1978b). Both concluded that selective association and influence each accounted for about 50% of the similarity, although Cohen found some characteristics affected by association but not by influence and vice versa. In most studies of adolescent smoking, the study design has not permitted the separation of the effects of peer influence from the effects of selective association. This has almost certainly resulted in overestimates of the amount of peer influence. Cohen’s (1983) discussion of peer influence suggests a way that a longitudinal study could be used to examine peer influence relatively uncontaminated with the effects of selective association. The initial difference between the adolescent and the peer on the variable of interest, according to Cohen’s view of influence, should serve as a measure of the strength of the pressure for one or both of the adolescents to change on that variable. Thus, since only diflerences between the adolescent and peer in smoking behavior are assumed to predict changes over a period of time in smoking behavior, any similarity as a result of selection on that variable does not contribute to predicting behavior change. This variable is only relatively unconfounded with selective association, since if selection for friends is made on some, as yet unidentified, variable that is causally related to cigarette smoking, peer influence would still be confounded to an extent with selective association. Another factor that has probably also contributed to the overestimation of the amount of peer influence in adolescent smoking is the common practice of using the subject’s estimates of the number of their friends who smoke as the criteria of peer smoking. Several studies (Evans. 1982; Sherman, Presson, Chassin, Corty, & Olshavsky, 1983) have documented the fact that adolescent smokers overestimate their friends’ smoking. To eliminate this source of bias in the results, data should be available from the person that the adolescent designates as a best friend so that the friends’ self-reports of how much they smoke can be used instead of, or in addition to, the report of the subject as to how much their friends smoke. This study is designed to examine the effects of peer influence on adolescent cigarette smoking when potentially confounding influences such as the influence of selective association and overestimation of friend smoking are minimized. For comparison purposes, data analysis will also be done with a frequently used measure of peer influence, the proportion of friends who smoke. Grade and sex differences in susceptibility to influence will also be examined. METHOD
Procedure
In Year 1 of the study, a questionnaire was administered to all 8th graders in a junior high school and all 1 lth graders in the high school fed by that junior high school. A passive consent form was used. About 1% of the parents refused permission for their child to participate, and another 1% of the students refused to participate on the day of the study or turned in unusable forms. The questionnaire was given in school during class time, at which time a saliva sample was also collected. The students were told that an analysis of the saliva
Peer influence
23
sample would reveal how much they had smoked in the past week. Subjects were also asked to write down the names of their two best friends in that grade in their school. The data for each student was matched to that of the first-listed best friend. If no match could be made to the friend listed first, an attempt was made to match to the friend listed second. About 70% of the students could be matched to the first-listed friend. Overall, 85% of the subjects were matched to a friend. In Year 2, a shortened form of the questionnaire was given to all 9th-grade students in the junior high school and to about half of the 12th graders in the high school. Scheduling problems with the school, unfortunately, made it impossible to include all the 12th graders in the second year. Each semester, half of the 12th graders were in a required English class; since our follow-up was done in the spring, only half of the 12th grade sample was available. Subjects
The sample consisted of 309 predominantly middle-class students in a midwestern suburban school system. These were subjects who were matched to a friend in Year 1 and were also present in Year 2. This sample was 49.1% male, 50.9% female, and 96.2% white. The majority of the nonwhite students were Oriental. The sample contained 122 subjects who were in the 8th grade and 187 who were in the 1 lth grade in Year 1. The Year 1 subjects who could not be matched to a friend did not differ from those who could be matched on sex, grade, ethnic background, or on the smoking variables (multivariate F = 0.6394; p = .699). The subjects who were present in both Year 1 and Year 2 differed significantly (multivariate F = 13.6768; p = .OOO)from those who were present only in Year 1 in grade (more 1 lth graders were not matched). With grade controlled, the subjects who were present in both years did not differ from those who were present only in Year 1 in amount smoked or in the proportion of friends who smoke. Measures
Two items were used from the questionnaire to measure adolescent cigarette smoking. The first item, “How many cigarettes did you smoke last week?” provided a quantitative measure of smoking. The second item was used only to dichotomize subjects into two stages of smoking: those who had never smoked and those who had ever smoked. The saliva sample analyses of the subjects were not used in the present analyses as an index of smoking. This is because saliva thiocyanate is not sensitive enough, particularly in adolescents, to reliably detect individual differences in cigarette smoking (Pechacek et al., 1984). However, collecting the sample probably served as a bogus pipeline (Jones & Sigall, 1971) to increase the accuracy of the subject’s report of his or her smoking. Peer influence was measured in two ways. First, based on Cohen (1983), peer influence (PIl) was defined as the difference between the Year 1 self-reported smoking of the person that the subject named as one of their two best friends and that of the subject (friend smoking-subject smoking). Thus, if the friend smokes more than the subject in Year 1, we would predict an increase in smoking from Year 1 to Year 2 for the subject if peer influence was operating. Second, peer influence (P12) was measured by the response to a question asking the subject to indicate, on a 5-point scale, what proportion of his or her friends smoked. RESULTS
The variables were first examined for mean differences in level as a function of grade and sex. Accordingly, a multivariate analysis of variance (MANOVA) was performed to examine grade, sex effects on the quantity smoked, the proportion of friends who smoke, and the
24
KATHRYN
Table
A. URBERG,
CHIEN-HUNG
CHENG,
and SHIANG-JEOU
1, Hierarchical influence
SHYLr
regressions predicting Year 2 smoking from Year I smoking, (PII), and its interactlons with grade, sex. and smoking change F
R? Change
IV Year 1 smoking Sex Grade Smoke stage Peer inf. by grade Peer inf. by sex
.8483
Change
Sig. Change
569.809
,000
.0155 .0243 .0323
4.307 22.210 34.053
.006 ,000 ,000
Simple Correlation
.84X3 .I355 .2366 .2297 - .37x3 - .4800
peer
Beta .84X3 .083X .0880 .0356 - .5815 - .2505
Overall: Multiple R = .8916. R’ = .7949, Adj R’ = .7X83. F(8. 272) = 119.63. 17 = .OOO
difference between friend smoking and subject smoking (PIl). The overall tests for grade (F(3, 306) = 15.93, p = .OOO), and sex (F(3, 306) = 5.89; p = ,000) were significant. The univariate analyses were examined to locate the significant effects. Univariate analyses showed significant grade effects for amount smoked (F = 18.11; df = 1,308; p = ,000) and proportion of friends who smoked (F 47.33, df = 1, 308; p = .OOO).The univariate analyses showed significant sex effects only for the proportion of friends who smoke (F = 14.00; df = 1, 308; p = .OOO.) The means indicate more smoking for 1 lth (M = 10.37) than 8th (M = .95) graders, and a greater proportion of friends who smoke for 1 lth (M = 2.41) than 8th (M = 1.74) graders. The sex effect indicates that girls (M = 2.27) reported having a larger proportion of friends who smoked than did boys (M = 1.91). Correlations between subject smoking and friend smoking were as follows: 8th graders, r = .0857; 1 lth graders, r = .4745; (n = 309). The correlation between PI1 and PI2 was r = - .0149. Difference
between friends
as a measure
of peer irlfluence
The next analyses used hierarchical regression to predict Year 2 subject smoking from Year 1 smoking; grade, sex, and stage of smoking (entered as a group); PI1 ; the interactions of PI1 with grade, sex, and stage of smoking (entered in that order). Year 1 smoking was entered first to control for the stability of smoking from Year 1 to Year 2, thus, any significant R2 change attributed to other variables is accounting for change in smoking from Year 1 to Year 2. Grade, sex, and stage of smoking were entered as a group, since we know there are differences in amount smoked as a function of these variables. Of greatest interest are the interactions of peer influence with grade, sex, and stage of smoking. Interactions that account for a significant increment in R* would indicate that peer influence is operating differentially with respect to that variable. The significant beta weights from this analysis are shown in Table 1. The peer influence variable was significantly related to change in smoking only in its interactions with grade and sex. Accordingly, the regression was run separately for the four grade and sex groups. The results of these analyses are as follows: 8th grade males, B = .7261, p = .4165; Sth-grade females, B = - .0310, l3 = - .0272; 1 lth-grade males, B = .1021, l3 = .0843; llth-grade females, B = - .0679, l3 = - .0416. Since within-group variance on the variables of interest is likely to be different for 8th compared to 1lth graders and for males compared to females. the unstandardized (B) as well as the standardized @) regression coefficients are reported. When differences between groups are of interest, the recommendation is to compare unstandardized coefficients. When the differences between different variables within a single group are of interest, the
Peer influence
25
Table 2. Hierarchical regressions predicting Year 2 smoking fromYear 1 smoking, proportion of smoking friends (PI2). and its interactions with grade, sex, and smoking stage
IV Year 1 smoking PI2 PI2 by grade
R2 Change
F Change
Sig. Change
Simple Correlation
.7270 .0163 .0065
593.823 14.109 5.908
.Ocil ,008 .016
.8526 .4694 .4520
Beta .8526 .I494 .2980
Overall: Multiple R = .87225, R* = .7608. Adj. RZ = .7519, F(8, 256) = 85.8856, p = .OOO.
recommendation is to compare the standardized beta coefficients (Cohen & Cohen, 1975; Kenny, 1979). Note that smoking (or nonsmoking) behavior is highly stable from Year 1 to Year 2. Most subjects were nonsmokers in year 1 who remained nonsmokers in Year 2. Proportion offriends who smoke as a measure of peer influence The analyses described above were repeated using the proportion of friends who smoke (P12) as the measure of peer influence rather than the difference between the smoking of the subject and that of their best friend (PIl). The results of these regressions for Year 2 smoking can be seen in Table 2. Only significant beta weights are shown. Peer influence and the interaction of grade with peer influence accounted for significant increments of variance. Accordingly, the regressions were run separately for 8th and 11th graders. These results show that peer influence appeared to be stronger in 1 lth graders (B = 6.180, l3 = .1785) than 8th graders (B = 2.024, p = .1445). Comparison of PI1 and PI2 Each of the peer influence variables has been shown to be a significant predictor of Year 2 smoking. In order to examine the relative strength of the two measures and to determine if they share variance in common, two further regressions were run. In the first analysis, Year 1 smoking, sex, and grade were entered as a block, PI1 and its interactions with sex and grade were entered as a second block, and PI2 and its interactions with sex and grade were entered as the final block. In the second analysis, PI2 and its interactions with sex and grade were entered before PI1 but after Year 1 smoking, sex, and grade. Since stage of smoking had no significant effects in earlier analyses, it was omitted from these analyses. The results of these analyses can be seen in Table 3. The results show that the amount of variance accounted for is about the same for each peer influence variable, regardless of the order of entry into the regression equation. This suggests that they are accounting for quite independent portions of variance in prediction. This is consistent with the very low correlation between the two influence variables. PI1 accounts for about three times more variance than does P12. DISCUSSION
This study examined peer influence in adolescent cigarette smoking using two measures of peer influence. One is a new measure that is relatively unconfounded with selective association. For this measure, only differences between friends in their smoking behavior were assumed to predict changes for either adolescent in their smoking behavior. Using this measure, peer influence was found to interact with both grade and sex to predict changes in cigarette smoking. Looking at the effect separately by grade and sex showed that 8th-grade boys appeared to be the most influenced by their peers of any of the four groups.
26
KATHRYN
A. URBERG.
CHIEN-HUNG
CHENG,
and SHIANG-JEOU
SHYU
Table 3. Hierarchical regressions comparing PI1 and its interactions with grade and sex to PI2 and its interactions with grade and sex in prediction Year 2 smoking
IV Analysis 1 Year 1 smoking, grade, sex PI1 and Interactions PI2 and interactions Analysis 2 Year 1 smoking, grade, sex PI2 and Interactions PI1 and interactions
R2
F
Change
Change
Sig. Change
.7423
212.192
.0555
Simple Correlation
Beta
.OOO
.8526
.8324
19.937
.cQO
.3428
.4232
.0167
6.451
,000
.4520
.2728
.7423
212.192
,000
.8526
.X324
.0164
4.952
,002
.4520
.2858
.0557
21.531
,000
.3428
.4357
Overall (for both analyses): Multiple R = .90255, F(9, 256) = 102.3323, p = ,000.
R’ = .8145,
Adj. R2 = .8067.
When the proportion of friends who smoke was used as the measure of peer influence, as has been done in most previous studies, the results were different. There was a significant peer influence by grade interaction that suggested that the older group was more influenced by peers than the younger group. Thus, the two measures of peer influence indicate opposite grade effects. The existance of grade differences in peer influence in adolescent cigarette smoking is currently a matter of dispute. Three recent studies, all using the proportion of friends who smoke as the measure of peer influence, found different grade effects. Krosnick and Judd’s (1982) cross-sectional study found that the path from peer influence to smoking was stronger for 14-year-olds (probably 8th graders) than it was for 1 1-year-olds (probably 5th graders). Ary, Biglan, Gallison, Wiessman, and Severson’s longitudinal study (1983) found peer influence to be stronger for 9th and 10th graders than 7th graders for initial nonsmokers only. In contrast, Chassin, Presson, Sherman, Montello, and McGrew’s (1986) longitudinal study found little evidence of grade differences. They attributed the difference between their finding and that of Krosnick and Judd to the longitudinal methodology that enabled them to minimize the confounding of selective association with peer influence. This was done by analyzing only data from the adolescents who were initial never smokers or initial triers. Thus, any selective association as a result of being a regular smoker was removed by not having regular smokers in the sample. To understand these somewhat conflicting results, it is necessary to consider the peer influence variable used. First, the proportion of friends who smoke variable is confounded with both selective association and the results of past influence. Thus, it may or may not index any difference between the adolescents’ smoking and that of their friends. Only for initial never-smokers would the proportion of smoking friends indicate unambiguously that there was a difference between the smoking behavior of the adolescents and that of their friends. Second, the variable represents influence from a wide group of friends that probably includes casual acquaintances as well as close friends. Third, both the proportion of friends who smoke and the number of adolescents who smoke were shown in this sample to increase with age. Except for the Chassin et al. study (1986). which found no grade effect. the studies
Peer influence
27
(including the present one) using the proportion of friends who smoke as the index of influence conclude that the older group is more susceptible to influence than the younger group. This suggests that the mutual confounding of friend smoking and subject smoking with age may account for the findings. In contrast, the discrepancy between adolescent smoking and friend smoking used in the present study was not a function of age and, thus, is free of this confounding. The reasons, then, for the differences in the results using the two measures of peer influence may be due (a) to the confounding of the proportion of friends who smoke variable (P12) with selective association, (b) to its confounding with age, or (c) to the fact that this variable measures influence from a different source than the peer influence variable (PIl). We believe that defining peer influence as a difference in behavior between the adolescent and his or her friend more unambiguously measures peer influence than does the proportion of friends who smoke; thus, we conclude that early adolescents, especially boys, appear to be more susceptible to influence from peers than are older adolescents. This conclusion is compatible with the findings of several previous studies on conformity. Bemdt (1979), using hypothetical situations, asked children from the 3rd to 12th grades whether or not they would go along in a situation proposed by peers. For the three types of behavior examined, the peak in conformity to peers occurred in early adolescence. For antisocial behavior, boys were more apt to be conforming than girls. Bixenstine, DeCorte, and Bixenstine (1976) also reported boys were more willing to go along with mischief or antisocial behavior than girls. Recent reviews of smoking research have suggested that the causes of the onset of smoking may be different than the causes of maintaining or increasing smoking (Flay et al., 1983: Leventhal & Cleary, 1980). There is little support in the present study for this view. Stage of smoking did not interact with either measure of peer influence, suggesting that there was no difference in the impact of peer influence for those moving from no smoking to some smoking compared to those moving from some smoking to a changed level of smoking. Although the theoretic argument for differential causes is convincing, to this point there is little empiric evidence as to what the differential causes might be. Ary et al. (1983), Chassin et al. (1986), and Chassin, Presson, & Sherman (1984) all report evidence of different predictors of smoking as a function of smoking stage. However, there is little convergence in the predictors found to be associated with each stage. The correlations found here in Year 1 smoking behavior between adolescent and best friend indicate the initial similarity for 8th and 11th graders. The insignificant correlation between friend and adolescent smoking for 8th graders suggests that, at this age, smoking behavior is not itself a variable on which friends are selected. Smoking is not a common behavior at this age in this sample, and smokers are likely to have nonsmoking friends. The higher correlation for smoking behavior within friendship pairs for 11 th graders is likely to be a result of both selective association on the basis of cigarette smoking, or some variable correlated with it, and the cumulative effects of peer influence. While selective association was not directly examined here, a review of the literature on friendship selection by Epstein (1989) suggests that similarity is likely to play a more important role in friendship selection in adolescents compared to elementary-school-age children. In conclusion, we argue that peer influence is more clearly measured by PIl, based on the difference between the smoking of the adolescent and that of their best friend, than by P12, the percentage of friends who smoke. Based on this measure, the younger boys appeared more influenced by the smoking of their best friend than younger girls or older adolescents. In fact, the 8th graders in this sample exemplify the conditions that Cohen believes are conducive to peer influence. That is, smoking is not important enough to serve as a basis for friendship selection, and friendships are relatively heterogeneous with respect to smoking
28
KATHRYN
A. URBERG,
CHIEN-HUNG
CHENG,
and SHIANG-JEOU
SHYU
behavior. For the 1 lth graders, where there is more initial homogeneity, evidence of peer influence from the best friend.
there was less
REFERENCES Ary, D., Biglan, A., Gallison, C., Weissman, W., & Severson, H. (1983). Longitudinal prediction of the onset and change of adolescent smoking. In W.F. Forbes, R.C. Frecker, and D. Nostbakken (Eds.), Proceedings ofthe Fifth World Conference on Smoking and He&h. Winnipeg: Canadian Council on Smoking and Health. Bemdt. T. (1979). Developmental changes in conformity to peers and parents. De~elopmenral Psychology, 15, 608-616. Bixenstine, V.E., Decorte, M.S., & Bixenstine, B.A. (1976). Conformity to peer-sponsored misconduct at four grade levels. Developmental Psychology, 12, 226236. Chassin, L., Presson, C.C.. & Sherman, S. (1984). Cigarette smoking and adolescent psychosocial development. Basic and Applied Social Psychology,
5. 295-3 15
Chassin. L., Presson. C., Sherman, S., MontelIo, D., & McGrew, J. (1986). Changes in peer and parent influence during adolescence: Longitudinal versus cross-sectional perspectives on smoking initiation. Developmenral Psychology,
22, 327-334.
Cohen, J. (1977). Sources of peer group homogeneity. Sociolog? of Education, 50, 227-241. Cohen, J. (1983). Commentary: The relationship between friendship selection and peer influence. In J.L. Epstein & N. Karweit (Eds.), Friends in school. New York: Academic Press. Cohen, J., & Cohen, P. (1975). Applied mulfiple regressionicorrelcrtion analysis for the behavioral sciences. New York: Wiley & Sons. Epstein, J. (1989). The selection of friends. In T. Bemdt & G. Ladd (Eds.). Peer relafionships in child development. New York: Wiley. Evans, R.I. (1982). Smoking deterrence: Prospects, concerns. Paper presented at American Psychological Association Meetings. Washington, DC. Flay, B., d’Avemas, J., Best, J.A., Kersell, M., & Ryan, K. (1983). Cigarette smoking: Why young people do it and ways of preventing it. In P. McGrath & P. Firestone (Eds.), Pediatric and adolescent behavioral medicine. New York: Springer-Verlag. Jessor, R., & Jessor, S. (1977). Problem behavior and psycho-social development: A longitudinal study of youth. New York: Academic Press. Jones, E., & Sigall. H. (1971). The bogus pipeline: A new paradigm for measuring affect and attitude. Psychological
Kandel,
Psychology,
Kandel,
Bulletin,
D. (1978a).
76. 349-364.
Similarity
in real-life
adolescent
friendship
pairs.
Journal
of Personaliv
and
Social
36, 306312.
D. (1978b).
Homophily,
84, 427436. D. (1979). Correlation
selection.
and socialization
in adolescent
friendships.
American
Journal
of
Sociology,
Kenny, and causality. New York: Wiley & Sons. Krosnick. J.A., & Judd, C.M. (1982). Transitions in social influence at adolescence: Who induces cigarette smoking? Developmental Psychology, 18, 359-368. Leventhal, H., & Cleary, P. (1980). The smoking problem: A review of the research and theory in behavioral risk modification. Psychological Bulletin, 88, 370408. Pechacek, T.F.. Murray, D.M., Luepker, R.V., Mittelmark. M.B.. Johnson, C.A., & Shultz, J.M. (1984). Measurement of adolescent smoking behavior: Rationale and methods. Journal of Behavioral Medicine, 7, 17-25. Sherman. S.J., Presson, C.C., Chassin, L.. Corty. E.. & Olshavsky, R. (1983). The false consensus effect in estimates of smoking prevalence: Underlying mechanisms. Personali~ cmd Social Psychology Bulletin, 9, 197-208.
