- Email: [email protected]
Peer influence and selection effects on adolescent smoking
Drug and Alcohol Dependence 109 (2010) 239–242
Contents lists available at ScienceDirect
Drug and Alcohol Dependence journal homepage: www.elsevier.com/locate/drugalcdep
Short communication
Peer influence and selection effects on adolescent smoking Myong-Hyun Go, Harold D. Green Jr., David P. Kennedy, Michael Pollard, Joan S. Tucker ∗ RAND Corporation, Santa Monica, CA 90407, USA
a r t i c l e
i n f o
Article history: Received 4 September 2009 Received in revised form 17 December 2009 Accepted 17 December 2009 Available online 13 January 2010 Keywords: Smoking Adolescence Friendship networks Selection Sociology
a b s t r a c t Background: Studies showing that adolescents are more likely to smoke if they have friends who smoke typically infer that this is the result of peer influence. However, it may also be due to adolescents choosing friends who have smoking behaviors similar to their own (i.e., selection). One of the most influential studies of influence and selection effects on smoking concluded that these processes contribute about equally to peer group homogeneity in adolescent smoking (Ennett and Bauman, 1994). The goal of this study was to conduct a partial replication of these findings. Methods: Data are from 1223 participants in the National Longitudinal Study of Adolescent Health. Spectral decomposition techniques identified friendship cliques, which were then used as the unit of analysis to examine influence and selection effects over a one-year period. Results: Non-smokers were more likely to become smokers if they initially belonged to a smoking (vs. non-smoking) group, and smokers were more likely to become non-smokers if they initially belonged to a non-smoking (vs. smoking) group, indicating an influence effect on both initiation and cessation. Further, group members who changed groups between waves were more likely to select groups with smoking behavior congruent to their own, providing evidence of a selection effect. Conclusions: While our results generally replicate the group analyses reported by Ennett and Bauman (1994), they suggest that peer influence and selection effects on adolescent smoking may be much weaker than assumed based on this earlier research. © 2010 Elsevier Ireland Ltd. All rights reserved.
1. Introduction It is well-established that adolescents are more likely to smoke if they have friends who smoke (Kobus, 2003). This association is typically interpreted as evidence of a peer influence effect, while ignoring the possibility that adolescents choose friends who are similar to themselves in terms of smoking (i.e., selection effect). However, both of these tendencies may contribute to homogeneity in adolescent peer group smoking behavior. To understand the extent to which influence vs. selection accounts for this homogeneity, it is critical to use longitudinal data to examine changes in peer group affiliations and smoking behavior. Relatively few studies have evaluated peer influence and selection effects through the use of longitudinal friendship network data, where each network member reports on his/her own smoking behavior (Alexander et al., 2001; Kirke, 2004; Mercken et al., 2009; Urberg et al., 1997). One of the first and most influential of these studies was conducted by Ennett and Bauman (1994), who followed a cohort of 9th grade students in five North Car-
∗ Corresponding author at: RAND Corporation, Health, 1776 Main Street, PO Box 2138, Santa Monica, CA 90407-2138, USA. Tel.: +1 310 393 0411x7519; fax: +1 310 260 8160. E-mail address: [email protected] (J.S. Tucker). 0376-8716/$ – see front matter © 2010 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.drugalcdep.2009.12.017
olina schools from 1980 to 1981. One strength of this study was the innovative use of group membership to investigate peer influence and selection, employing spectral decomposition techniques (Richards, 1995) to identify clusters of friends (i.e., friendship cliques), which became the unit of analysis. The advantage of this approach is that it allows the use of standard statistical techniques. When observations are known to be correlated, as is the case for these network data, the use of standard statistical techniques that assume independence of errors (e.g., OLS) may bias results (Calvo-Armengol et al., 2005; Steglich et al., 2004). Most studies that estimate the effect of peer influence on individual behavior suffer from the problem of autocorrelation; they control for the ego’s friends’ behavior only, without also controlling for the behavior of the friends of these friends and so forth. Ennett and Bauman’s approach addressed this autocorrelation problem by treating group membership as a categorical factor. Members of the same group (or clique) by definition are highly connected with each other compared to those who are outside of the group. As a result, groups can be treated as natural bounds of autocorrelation where the peer effects are limited to the members of the same group. Ennett and Bauman’s heavily cited study concluded that both peer influence and selection contribute moderately, and about equally, to peer group homogeneity in adolescent smoking. To the best of our knowledge, the present study is the first to use
240
M.-H. Go et al. / Drug and Alcohol Dependence 109 (2010) 239–242
the same spectral decomposition technique to investigate whether their results for smoking groups (cliques) replicate in a nationally representative sample. 2. Methods 2.1. Participants Data come from the National Longitudinal Study of Adolescent Health, a survey of adolescents who were initially recruited in grades 7–12 during the 1994–1995 school year (Bearman et al., 1997). Using a stratified sampling design, 80 schools in the U.S. were randomly selected based on a combination of factors such as region, urbanicity, racial composition, and size. Sixteen of these schools were selected for the so-called “saturated” sample: the entire school’s student body was interviewed, creating a nearly complete socio-centric school network. We restricted our sample to adolescents who completed the one-year follow-up (Wave 2), had complete information on the study variables, and were classified as “group members”1 (see below). The final sample of N = 1223 was an average of 15.5 years, 68% non-Hispanic white, and 48% male. Information on the Add Health design and longitudinal data is available elsewhere (Harris et al., 2009). 2.2. Smoking behavior Participants were classified as a “smoker” if they reported smoking at least four days in the past 30 days. This level of smoking roughly corresponds to weekly smoking, is the median number of smoking days per month for smokers in the entire dataset, and has been previously used to define regular smoking among adolescents (Tucker et al., 2003).2 Initiators were those who were a lifetime non-smoker at Wave 1, but were classified as a smoker at Wave 2, quitters were classified as a smoker at Wave 1, but did not smoked at all in the past 30 days at Wave 2. Twenty-five percent of the sample was classified as a “smoker” at baseline, with this group consuming an average of 5.5 cigarettes per month. 2.3. Group membership To measure group effects, we determined: (a) individuals’ social positions in the school social network; and (b) whether the group was a smoking or non-smoking group. Individuals’ social positions fell into three broad categories: group members, liaisons, and isolates. Group members were those with 50% or more ties directed to members of the same group, liaisons were nodes that connect one or more groups with other groups or nodes, and isolates were those with at most one tie with the rest of the network (see Ennett and Bauman, 1993). In addition, group members were assigned a unique group identifier. NEGOPY software (Richards, 1995) was used to determine social position, optimized when we combine information on the number of connections with the strength of those connections. For each nominated friend, adolescents indicated whether they had engaged in each of five activities with this friend during the past week (e.g., talked on the phone, “hung out” after school). We calculated a tie strength score by summing the number of endorsed items (mean = 2.5 at Wave 1 and 2.8 at Wave 2), and then transformed the scores by giving more weight to those who shared more activities. Prior to valuing the strengths of the friendship ties, we imputed missing and/or unobserved ties (e.g., a participant not nominating anyone as a friend, but being nominated as a friend by several others) as 0/1 using the Preferential Attachment algorithm (Huisman, 2007). We used the number of received nominations of each individual scaled by the total number of ties as the imputation probability, based on the idea that the odds of these participants nominating someone else as their friend should increase in proportion to the number of times they are nominated as friends by others. Having determined individuals’ social position in the school social network, we classified the groups as either “smoking” or “non-smoking.” Following Ennett and Bauman’s (1994) approach, we defined a “smoking group” as a group that included at least one smoker and a “non-smoking group” as a group that did not include any smokers. The focal adolescent’s own smoking behavior was not used to determine group type. As a result, the group label (i.e., smoking/non-smoking) is not universally applied to everyone in a given group, but rather depends on the presence/absence of smokers other than the individual in question. For instance, a group containing a single smoker is a non-smoking group for the smoker but a smoking group for the other group members.
1 Although the NEGOPY analyses also identified adolescents in the social categories of liaisons and isolates, these were excluded from the present analyses because of our primary interest in understanding group (clique) effects on smoking. Thus, these analyses do not fully replicate those of Ennett and Bauman (1994), which included examination of liaisons and isolates. 2 Ennett and Bauman (1994) defined current smokers as those who reported smoking cigarettes at the time of the interview and had smoked one or more packs of cigarettes in their lifetime. All adolescents with 9 parts per million or higher carbon monoxide in their breath also were classified as current smokers.
We also classified groups as “surviving” based on the proportion of group members that were members during both waves. Following Ennett and Bauman’s definition, we defined surviving groups as those that retained at least 50% of members between waves. Identifying surviving groups allowed us to control for changes in group membership from Wave 1 to Wave 2. We refer to the “reduced sample” in analyses that are restricted to surviving groups, and to the “full sample” in analyses that include both the surviving and non-surviving groups.
3. Results 3.1. Influence We used crosstabs to test the two influence hypotheses for smoking initiation and cessation, respectively: (1) Wave 1 nonsmokers are more likely to become smokers by Wave 2 if they belong to a smoking (vs. non-smoking) group at Wave 1; and (2) Wave 1 smokers are more likely to become non-smokers by Wave 2 if they belong to a non-smoking (vs. smoking) group at Wave 1. Results for initiation were consistent with an influence effect (see top of Table 1): among initial non-smokers in the full sample, those belonging to a smoking group at Wave 1 were about 1.5 more likely than those in a non-smoking group to start smoking by Wave 2 (OR = 1.48, 95% CI = 1.03–2.15). However, this association was not found in the reduced sample. Results for cessation were also consistent with an influence effect (see bottom of Table 1): initial smokers in the full sample who belonged to a non-smoking group at Wave 1 were about twice as likely to be non-smokers by Wave 2 compared to initial smokers who belonged to a smoking group at Wave 1 (OR = 2.12, 95% CI = 1.10–4.06). This association was even stronger in the reduced sample (OR = 15.7, 95% CI = 2.65–93.4). We infer from these results that there is stronger evidence for peer influence in the case of smoking cessation than smoking initiation. In addition, we measure the strength of association using the contingency coefficient (Barton and Huynh, 2003). This is an index of association strength that is scaled by the sample size (small effect = 0.10, medium effect = 0.30, large effect = 0.50; Cohen, 1988). The contingency coefficients were 0.09–0.10 for the two initiation influence effects, and 0.18 for the full sample cessation influence effect, indicating that these effects are small but statistically significant. In contrast, the contingency coefficient of 0.68 for the reduced sample cessation influence effect indicates that there is a strong influence effect on cessation among smokers in surviving nonsmoking groups. Note that Ennett and Bauman (1994) reported contingency coefficients of 0.12 (full sample) and 0.32 (reduced sample) for initiation, and 0.04 (full sample) and 0.35 (reduced sample) for cessation. 3.2. Selection We evaluated the hypothesis that group members who changed groups between waves would select groups with smoking behavior congruent to their own. As shown in Table 2, there was a general tendency for both initial smokers and non-smokers to join a smoking group by Wave 2. However, in the full sample, initial smokers were more likely than non-smokers to join a smoking group by Wave 2, and non-smokers were more likely than smokers to join a non-smoking group by Wave 2 (OR = 1.95, 95% CI = 1.35–2.83). This general pattern of selection was also found in the reduced sample, although it was not statistically significant. The contingency coefficients of 0.14 and 0.16 indicate that the selection effect is weak in both samples (note that Ennett and Bauman reported coefficients of 0.40–0.48 for these analyses). The non-significant selection effect in the reduced sample should be taken with a grain of salt, however: members of surviving groups are by definition less likely to switch groups, and this leads to weaker selection effects in the reduced sample.
M.-H. Go et al. / Drug and Alcohol Dependence 109 (2010) 239–242
241
Table 1 Peer influence on initiation and cessation: crosstab of Wave 1 clique membership type (smoking vs. non-smoking) with smoking status at Wave 2 (smoker vs. non-smoker), by Wave 1 smoking status (in percentages). Among Wave 1 non-smokers
Type of Wave 1 clique Surviving only
All
Wave 2 smoking behavior
Smoking
Non-smoking
Smoking
Non-smoking
Smoker N Non-smoker N Total N
9.4 5 90.6 48 100.0 53
13.9 17 86.1 105 100.0 122
18.0 92 82.0 418 100.0 510
12.9 52 87.1 351 100.0 403
2 = 4.47, p = 0.03 OR = 1.48 (1.03, 2.15) Contingency coefficient = 0.10
2 = 0.68, p = 0.41 OR = 0.64 (0.22, 1.84) Contingency coefficient = 0.09 Among Wave 1 smokers
Type of Wave 1 clique Surviving only
All
Wave 2 smoking behavior
Smoking
Non-smoking
Smoking
Non-smoking
Smoker N Non-smoker N Total N
77.8 21 22.2 6 100.0 27
18.2 2 81.8 9 100.0 11
68.9 184 31.1 83 100.0 267
51.2 22 48.8 21 100.0 43
2 = 5.23, p = 0.02 OR = 2.12 (1.10, 4.06) Contingency coefficient = 0.18
2 = 11.62, p < 0.001 OR = 15.75 (2.65, 93.46) Contingency coefficient = 0.68
Table 2 Peer selection: crosstab of Wave 1 smoking status (smoker vs. non-smoker) by type of Wave 2 clique joined (smoking vs. non-smoking) (in percentages). Wave 1 smoking status Surviving only
All
Type of Wave 2 clique joined
Smoker
Non-smoker
Smoker
Non-smoker
Smoking N Non-smoking N Total N
62.5 10 37.5 6 100.0 16
50.0 27 50.0 26 100.0 53
84.9 231 15.1 41 100.0 272
74.3 548 25.7 190 100.0 738
2 = 12.83, p < 0.001 OR = 1.95 (1.35, 2.83) Contingency coefficient = 0.16
2 = 0.66, p = 0.41 OR = 1.60 (0.51, 5.05) Contingency coefficient = 0.14
4. Discussion These results add to a growing literature suggesting that both influence and selection contribute to peer group smoking homogeneity (e.g., Hall and Valente, 2007; Hoffman et al., 2007; Kirke, 2004; Mercken et al., 2007). Our results provide support for an influence effect on smoking initiation and cessation, with stronger evidence in the latter case. Although Ennett and Bauman (1994) similarly found that initiation was more likely if adolescents belonged to a smoking group, they did not find that smokers were more likely to quit if they belonged to a non-smoking group (perhaps due to smaller sample for their analysis). Similar to Ennett and Bauman, we found that influence and selection are about equally associated with group homogeneity. However, our effect sizes were generally smaller than those reported in this earlier study. The one exception where we found a large effect was for peer influence on cessation in the reduced sample, suggesting that peer influence may have a strong impact on cessation if the composition of the peer group is relatively stable over time. Study limitations include reliance on self-reported smoking, use of a school-based sample, a somewhat different definition of cur-
rent smoker than used by Ennett and Bauman, and the inability to rule out alternative, third variable explanations for our results. Nonetheless, the generally weaker effects in the present study compared to Ennett and Bauman (1994) serve as a cautionary note to both researchers and practioners to not overestimate the role of peers in adolescent smoking. Role of funding sources Funding for this study was provided by grant 16RT-0169 from the California Tobacco-Related Disease Research Program (TRDRP) of the University of California. TRDRP had no role in the study design, collection, analysis or interpretation of the data, writing the manuscript, or the decision to submit the paper for publication. Contributors Mr. Go was primarily responsible for the design of the study, conducted the data analysis, and wrote the first draft of the manuscript. Drs. Green, Kennedy, Pollard, and Tucker assisted with
242
M.-H. Go et al. / Drug and Alcohol Dependence 109 (2010) 239–242
the design of the study and provided feedback on drafts of the manuscript. Conflict of interest All authors declare that they have no conflicts of interest. Acknowledgements This research uses data from Add Health, a program project designed by J. Richard Udry, Peter S. Bearman, and Kathleen Mullan Harris, and funded by a grant P01-HD31921 from the National Institute of Child Health and Human Development, with Cooperative funding from 17 other agencies. Special acknowledgment is due Ronald R. Rindfuss and Barbara Entwisle for assistance in the original design. Persons interested in obtaining data files from Add Health should contact Add Health, Carolina Population Center, 123 W. Franklin Street, Chapel Hill, NC 27516-2524 ([email protected]). References Alexander, C., Piazza, M., Mekos, D., Valente, T., 2001. Peers, schools, and adolescent cigarette smoking. J. Adolesc. Health 29, 22–30. Barton, K.E., Huynh, H., 2003. Patterns of errors made by students with disabilities on a reading test with oral reading administration. Educ. Psychol. Meas. 63, 602–614. Bearman, P.S., Jones, J., Udry, J.R., 1997. The National longitudinal study of adolescent health: research design. Available at: http://www.cpc.unc.edu/projects/ addhealth/design.html.
Calvo-Armengol, A., Antoni & Patacchni, E., Zenou, Y., 2005. Peer effects and social networks in education and crime. CEPR Discussion Papers. CEPR. Cohen, J., 1988. Statistical Power Analysis for the Behavioral Sciences, 2nd ed. Lawrence Erlbaum Associates, Hillsdale, NJ. Ennett, S.T., Bauman, K.E., 1993. Peer group structure and adolescent cigarette smoking: a social network analysis. J. Health Soc. Behav. 34, 226–236. Ennett, S.T., Bauman, K.E., 1994. The contribution of influence and selection to adolescent peer group homogeneity: the case of adolescent cigarette smoking. J. Pers. Soc. Psychol. 67, 653–663. Hall, J.A., Valente, T.W., 2007. Adolescent smoking networks: the effects of influence and selection on future smoking. Addict. Behav. 32, 3054–3059. Harris, K.M., Halpern, C.T., Entzel, P., Tabor, J., Bearman, P.S., Udry, J.R. Available at: http://www.cpc.unc.edu/projects/addhealth/design (accessed 20 August 2009). Hoffman, B.R., Monge, P.R., Chou, C.P., Valente, T.W., 2007. Perceived peer influence and peer selection on adolescent smoking. Addict. Behav. 32, 1546–1554. Huisman, M., 2007. Imputation of missing network data: some simple procedures. In: International Sunbelt Social Network Conference XXVII, Corfu, Greece. Kirke, D.M., 2004. Chain reactions in adolescents’ cigarette, alcohol and drug use: similarity through peer influence or the patterning of ties in peer networks? Soc. Networks 26, 3–28. Kobus, K., 2003. Peers and adolescent smoking. Addiction 98, 37–55. Mercken, L., Candel, M., Willems, P., de Vries, H., 2007. Disentangling social selection and social influence effects on adolescent smoking: the importance of reciprocity in friendships. Addiction 102, 1483–1492. Mercken, L., Candel, M., Willems, P., de Vries, H., 2009. Social influence and selection effects in the context of smoking behavior: changes during early and mid adolescence. Health Psychol. 28, 73–82. Richards, W.D., 1995. The NEGOPY Network Analysis Program. Department of Communications, Simon Fraser University, Burnaby, BC. Steglich, C., Snijders, T.A.B., Pearson, M., 2004. Dynamic Networks and Behavior: Separating Selection from Influence. Working Paper. Interuniversity Center for Social Science Theory and Methodology. Tucker, J.S., Ellickson, P.L., Klein, D.J., 2003. Predictors of the transition to regular smoking during adolescence and young adulthood. J. Adolesc. Health 32, 314–324. Urberg, K.A., Degirmencioglu, S.M., Pilgrim, C., 1997. Close friend and group influence on adolescent cigarette smoking and alcohol use. Dev. Psychol. 33, 834–844.
Contents lists available at ScienceDirect
Drug and Alcohol Dependence journal homepage: www.elsevier.com/locate/drugalcdep
Short communication
Peer influence and selection effects on adolescent smoking Myong-Hyun Go, Harold D. Green Jr., David P. Kennedy, Michael Pollard, Joan S. Tucker ∗ RAND Corporation, Santa Monica, CA 90407, USA
a r t i c l e
i n f o
Article history: Received 4 September 2009 Received in revised form 17 December 2009 Accepted 17 December 2009 Available online 13 January 2010 Keywords: Smoking Adolescence Friendship networks Selection Sociology
a b s t r a c t Background: Studies showing that adolescents are more likely to smoke if they have friends who smoke typically infer that this is the result of peer influence. However, it may also be due to adolescents choosing friends who have smoking behaviors similar to their own (i.e., selection). One of the most influential studies of influence and selection effects on smoking concluded that these processes contribute about equally to peer group homogeneity in adolescent smoking (Ennett and Bauman, 1994). The goal of this study was to conduct a partial replication of these findings. Methods: Data are from 1223 participants in the National Longitudinal Study of Adolescent Health. Spectral decomposition techniques identified friendship cliques, which were then used as the unit of analysis to examine influence and selection effects over a one-year period. Results: Non-smokers were more likely to become smokers if they initially belonged to a smoking (vs. non-smoking) group, and smokers were more likely to become non-smokers if they initially belonged to a non-smoking (vs. smoking) group, indicating an influence effect on both initiation and cessation. Further, group members who changed groups between waves were more likely to select groups with smoking behavior congruent to their own, providing evidence of a selection effect. Conclusions: While our results generally replicate the group analyses reported by Ennett and Bauman (1994), they suggest that peer influence and selection effects on adolescent smoking may be much weaker than assumed based on this earlier research. © 2010 Elsevier Ireland Ltd. All rights reserved.
1. Introduction It is well-established that adolescents are more likely to smoke if they have friends who smoke (Kobus, 2003). This association is typically interpreted as evidence of a peer influence effect, while ignoring the possibility that adolescents choose friends who are similar to themselves in terms of smoking (i.e., selection effect). However, both of these tendencies may contribute to homogeneity in adolescent peer group smoking behavior. To understand the extent to which influence vs. selection accounts for this homogeneity, it is critical to use longitudinal data to examine changes in peer group affiliations and smoking behavior. Relatively few studies have evaluated peer influence and selection effects through the use of longitudinal friendship network data, where each network member reports on his/her own smoking behavior (Alexander et al., 2001; Kirke, 2004; Mercken et al., 2009; Urberg et al., 1997). One of the first and most influential of these studies was conducted by Ennett and Bauman (1994), who followed a cohort of 9th grade students in five North Car-
∗ Corresponding author at: RAND Corporation, Health, 1776 Main Street, PO Box 2138, Santa Monica, CA 90407-2138, USA. Tel.: +1 310 393 0411x7519; fax: +1 310 260 8160. E-mail address: [email protected] (J.S. Tucker). 0376-8716/$ – see front matter © 2010 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.drugalcdep.2009.12.017
olina schools from 1980 to 1981. One strength of this study was the innovative use of group membership to investigate peer influence and selection, employing spectral decomposition techniques (Richards, 1995) to identify clusters of friends (i.e., friendship cliques), which became the unit of analysis. The advantage of this approach is that it allows the use of standard statistical techniques. When observations are known to be correlated, as is the case for these network data, the use of standard statistical techniques that assume independence of errors (e.g., OLS) may bias results (Calvo-Armengol et al., 2005; Steglich et al., 2004). Most studies that estimate the effect of peer influence on individual behavior suffer from the problem of autocorrelation; they control for the ego’s friends’ behavior only, without also controlling for the behavior of the friends of these friends and so forth. Ennett and Bauman’s approach addressed this autocorrelation problem by treating group membership as a categorical factor. Members of the same group (or clique) by definition are highly connected with each other compared to those who are outside of the group. As a result, groups can be treated as natural bounds of autocorrelation where the peer effects are limited to the members of the same group. Ennett and Bauman’s heavily cited study concluded that both peer influence and selection contribute moderately, and about equally, to peer group homogeneity in adolescent smoking. To the best of our knowledge, the present study is the first to use
240
M.-H. Go et al. / Drug and Alcohol Dependence 109 (2010) 239–242
the same spectral decomposition technique to investigate whether their results for smoking groups (cliques) replicate in a nationally representative sample. 2. Methods 2.1. Participants Data come from the National Longitudinal Study of Adolescent Health, a survey of adolescents who were initially recruited in grades 7–12 during the 1994–1995 school year (Bearman et al., 1997). Using a stratified sampling design, 80 schools in the U.S. were randomly selected based on a combination of factors such as region, urbanicity, racial composition, and size. Sixteen of these schools were selected for the so-called “saturated” sample: the entire school’s student body was interviewed, creating a nearly complete socio-centric school network. We restricted our sample to adolescents who completed the one-year follow-up (Wave 2), had complete information on the study variables, and were classified as “group members”1 (see below). The final sample of N = 1223 was an average of 15.5 years, 68% non-Hispanic white, and 48% male. Information on the Add Health design and longitudinal data is available elsewhere (Harris et al., 2009). 2.2. Smoking behavior Participants were classified as a “smoker” if they reported smoking at least four days in the past 30 days. This level of smoking roughly corresponds to weekly smoking, is the median number of smoking days per month for smokers in the entire dataset, and has been previously used to define regular smoking among adolescents (Tucker et al., 2003).2 Initiators were those who were a lifetime non-smoker at Wave 1, but were classified as a smoker at Wave 2, quitters were classified as a smoker at Wave 1, but did not smoked at all in the past 30 days at Wave 2. Twenty-five percent of the sample was classified as a “smoker” at baseline, with this group consuming an average of 5.5 cigarettes per month. 2.3. Group membership To measure group effects, we determined: (a) individuals’ social positions in the school social network; and (b) whether the group was a smoking or non-smoking group. Individuals’ social positions fell into three broad categories: group members, liaisons, and isolates. Group members were those with 50% or more ties directed to members of the same group, liaisons were nodes that connect one or more groups with other groups or nodes, and isolates were those with at most one tie with the rest of the network (see Ennett and Bauman, 1993). In addition, group members were assigned a unique group identifier. NEGOPY software (Richards, 1995) was used to determine social position, optimized when we combine information on the number of connections with the strength of those connections. For each nominated friend, adolescents indicated whether they had engaged in each of five activities with this friend during the past week (e.g., talked on the phone, “hung out” after school). We calculated a tie strength score by summing the number of endorsed items (mean = 2.5 at Wave 1 and 2.8 at Wave 2), and then transformed the scores by giving more weight to those who shared more activities. Prior to valuing the strengths of the friendship ties, we imputed missing and/or unobserved ties (e.g., a participant not nominating anyone as a friend, but being nominated as a friend by several others) as 0/1 using the Preferential Attachment algorithm (Huisman, 2007). We used the number of received nominations of each individual scaled by the total number of ties as the imputation probability, based on the idea that the odds of these participants nominating someone else as their friend should increase in proportion to the number of times they are nominated as friends by others. Having determined individuals’ social position in the school social network, we classified the groups as either “smoking” or “non-smoking.” Following Ennett and Bauman’s (1994) approach, we defined a “smoking group” as a group that included at least one smoker and a “non-smoking group” as a group that did not include any smokers. The focal adolescent’s own smoking behavior was not used to determine group type. As a result, the group label (i.e., smoking/non-smoking) is not universally applied to everyone in a given group, but rather depends on the presence/absence of smokers other than the individual in question. For instance, a group containing a single smoker is a non-smoking group for the smoker but a smoking group for the other group members.
1 Although the NEGOPY analyses also identified adolescents in the social categories of liaisons and isolates, these were excluded from the present analyses because of our primary interest in understanding group (clique) effects on smoking. Thus, these analyses do not fully replicate those of Ennett and Bauman (1994), which included examination of liaisons and isolates. 2 Ennett and Bauman (1994) defined current smokers as those who reported smoking cigarettes at the time of the interview and had smoked one or more packs of cigarettes in their lifetime. All adolescents with 9 parts per million or higher carbon monoxide in their breath also were classified as current smokers.
We also classified groups as “surviving” based on the proportion of group members that were members during both waves. Following Ennett and Bauman’s definition, we defined surviving groups as those that retained at least 50% of members between waves. Identifying surviving groups allowed us to control for changes in group membership from Wave 1 to Wave 2. We refer to the “reduced sample” in analyses that are restricted to surviving groups, and to the “full sample” in analyses that include both the surviving and non-surviving groups.
3. Results 3.1. Influence We used crosstabs to test the two influence hypotheses for smoking initiation and cessation, respectively: (1) Wave 1 nonsmokers are more likely to become smokers by Wave 2 if they belong to a smoking (vs. non-smoking) group at Wave 1; and (2) Wave 1 smokers are more likely to become non-smokers by Wave 2 if they belong to a non-smoking (vs. smoking) group at Wave 1. Results for initiation were consistent with an influence effect (see top of Table 1): among initial non-smokers in the full sample, those belonging to a smoking group at Wave 1 were about 1.5 more likely than those in a non-smoking group to start smoking by Wave 2 (OR = 1.48, 95% CI = 1.03–2.15). However, this association was not found in the reduced sample. Results for cessation were also consistent with an influence effect (see bottom of Table 1): initial smokers in the full sample who belonged to a non-smoking group at Wave 1 were about twice as likely to be non-smokers by Wave 2 compared to initial smokers who belonged to a smoking group at Wave 1 (OR = 2.12, 95% CI = 1.10–4.06). This association was even stronger in the reduced sample (OR = 15.7, 95% CI = 2.65–93.4). We infer from these results that there is stronger evidence for peer influence in the case of smoking cessation than smoking initiation. In addition, we measure the strength of association using the contingency coefficient (Barton and Huynh, 2003). This is an index of association strength that is scaled by the sample size (small effect = 0.10, medium effect = 0.30, large effect = 0.50; Cohen, 1988). The contingency coefficients were 0.09–0.10 for the two initiation influence effects, and 0.18 for the full sample cessation influence effect, indicating that these effects are small but statistically significant. In contrast, the contingency coefficient of 0.68 for the reduced sample cessation influence effect indicates that there is a strong influence effect on cessation among smokers in surviving nonsmoking groups. Note that Ennett and Bauman (1994) reported contingency coefficients of 0.12 (full sample) and 0.32 (reduced sample) for initiation, and 0.04 (full sample) and 0.35 (reduced sample) for cessation. 3.2. Selection We evaluated the hypothesis that group members who changed groups between waves would select groups with smoking behavior congruent to their own. As shown in Table 2, there was a general tendency for both initial smokers and non-smokers to join a smoking group by Wave 2. However, in the full sample, initial smokers were more likely than non-smokers to join a smoking group by Wave 2, and non-smokers were more likely than smokers to join a non-smoking group by Wave 2 (OR = 1.95, 95% CI = 1.35–2.83). This general pattern of selection was also found in the reduced sample, although it was not statistically significant. The contingency coefficients of 0.14 and 0.16 indicate that the selection effect is weak in both samples (note that Ennett and Bauman reported coefficients of 0.40–0.48 for these analyses). The non-significant selection effect in the reduced sample should be taken with a grain of salt, however: members of surviving groups are by definition less likely to switch groups, and this leads to weaker selection effects in the reduced sample.
M.-H. Go et al. / Drug and Alcohol Dependence 109 (2010) 239–242
241
Table 1 Peer influence on initiation and cessation: crosstab of Wave 1 clique membership type (smoking vs. non-smoking) with smoking status at Wave 2 (smoker vs. non-smoker), by Wave 1 smoking status (in percentages). Among Wave 1 non-smokers
Type of Wave 1 clique Surviving only
All
Wave 2 smoking behavior
Smoking
Non-smoking
Smoking
Non-smoking
Smoker N Non-smoker N Total N
9.4 5 90.6 48 100.0 53
13.9 17 86.1 105 100.0 122
18.0 92 82.0 418 100.0 510
12.9 52 87.1 351 100.0 403
2 = 4.47, p = 0.03 OR = 1.48 (1.03, 2.15) Contingency coefficient = 0.10
2 = 0.68, p = 0.41 OR = 0.64 (0.22, 1.84) Contingency coefficient = 0.09 Among Wave 1 smokers
Type of Wave 1 clique Surviving only
All
Wave 2 smoking behavior
Smoking
Non-smoking
Smoking
Non-smoking
Smoker N Non-smoker N Total N
77.8 21 22.2 6 100.0 27
18.2 2 81.8 9 100.0 11
68.9 184 31.1 83 100.0 267
51.2 22 48.8 21 100.0 43
2 = 5.23, p = 0.02 OR = 2.12 (1.10, 4.06) Contingency coefficient = 0.18
2 = 11.62, p < 0.001 OR = 15.75 (2.65, 93.46) Contingency coefficient = 0.68
Table 2 Peer selection: crosstab of Wave 1 smoking status (smoker vs. non-smoker) by type of Wave 2 clique joined (smoking vs. non-smoking) (in percentages). Wave 1 smoking status Surviving only
All
Type of Wave 2 clique joined
Smoker
Non-smoker
Smoker
Non-smoker
Smoking N Non-smoking N Total N
62.5 10 37.5 6 100.0 16
50.0 27 50.0 26 100.0 53
84.9 231 15.1 41 100.0 272
74.3 548 25.7 190 100.0 738
2 = 12.83, p < 0.001 OR = 1.95 (1.35, 2.83) Contingency coefficient = 0.16
2 = 0.66, p = 0.41 OR = 1.60 (0.51, 5.05) Contingency coefficient = 0.14
4. Discussion These results add to a growing literature suggesting that both influence and selection contribute to peer group smoking homogeneity (e.g., Hall and Valente, 2007; Hoffman et al., 2007; Kirke, 2004; Mercken et al., 2007). Our results provide support for an influence effect on smoking initiation and cessation, with stronger evidence in the latter case. Although Ennett and Bauman (1994) similarly found that initiation was more likely if adolescents belonged to a smoking group, they did not find that smokers were more likely to quit if they belonged to a non-smoking group (perhaps due to smaller sample for their analysis). Similar to Ennett and Bauman, we found that influence and selection are about equally associated with group homogeneity. However, our effect sizes were generally smaller than those reported in this earlier study. The one exception where we found a large effect was for peer influence on cessation in the reduced sample, suggesting that peer influence may have a strong impact on cessation if the composition of the peer group is relatively stable over time. Study limitations include reliance on self-reported smoking, use of a school-based sample, a somewhat different definition of cur-
rent smoker than used by Ennett and Bauman, and the inability to rule out alternative, third variable explanations for our results. Nonetheless, the generally weaker effects in the present study compared to Ennett and Bauman (1994) serve as a cautionary note to both researchers and practioners to not overestimate the role of peers in adolescent smoking. Role of funding sources Funding for this study was provided by grant 16RT-0169 from the California Tobacco-Related Disease Research Program (TRDRP) of the University of California. TRDRP had no role in the study design, collection, analysis or interpretation of the data, writing the manuscript, or the decision to submit the paper for publication. Contributors Mr. Go was primarily responsible for the design of the study, conducted the data analysis, and wrote the first draft of the manuscript. Drs. Green, Kennedy, Pollard, and Tucker assisted with
242
M.-H. Go et al. / Drug and Alcohol Dependence 109 (2010) 239–242
the design of the study and provided feedback on drafts of the manuscript. Conflict of interest All authors declare that they have no conflicts of interest. Acknowledgements This research uses data from Add Health, a program project designed by J. Richard Udry, Peter S. Bearman, and Kathleen Mullan Harris, and funded by a grant P01-HD31921 from the National Institute of Child Health and Human Development, with Cooperative funding from 17 other agencies. Special acknowledgment is due Ronald R. Rindfuss and Barbara Entwisle for assistance in the original design. Persons interested in obtaining data files from Add Health should contact Add Health, Carolina Population Center, 123 W. Franklin Street, Chapel Hill, NC 27516-2524 ([email protected]). References Alexander, C., Piazza, M., Mekos, D., Valente, T., 2001. Peers, schools, and adolescent cigarette smoking. J. Adolesc. Health 29, 22–30. Barton, K.E., Huynh, H., 2003. Patterns of errors made by students with disabilities on a reading test with oral reading administration. Educ. Psychol. Meas. 63, 602–614. Bearman, P.S., Jones, J., Udry, J.R., 1997. The National longitudinal study of adolescent health: research design. Available at: http://www.cpc.unc.edu/projects/ addhealth/design.html.
Calvo-Armengol, A., Antoni & Patacchni, E., Zenou, Y., 2005. Peer effects and social networks in education and crime. CEPR Discussion Papers. CEPR. Cohen, J., 1988. Statistical Power Analysis for the Behavioral Sciences, 2nd ed. Lawrence Erlbaum Associates, Hillsdale, NJ. Ennett, S.T., Bauman, K.E., 1993. Peer group structure and adolescent cigarette smoking: a social network analysis. J. Health Soc. Behav. 34, 226–236. Ennett, S.T., Bauman, K.E., 1994. The contribution of influence and selection to adolescent peer group homogeneity: the case of adolescent cigarette smoking. J. Pers. Soc. Psychol. 67, 653–663. Hall, J.A., Valente, T.W., 2007. Adolescent smoking networks: the effects of influence and selection on future smoking. Addict. Behav. 32, 3054–3059. Harris, K.M., Halpern, C.T., Entzel, P., Tabor, J., Bearman, P.S., Udry, J.R. Available at: http://www.cpc.unc.edu/projects/addhealth/design (accessed 20 August 2009). Hoffman, B.R., Monge, P.R., Chou, C.P., Valente, T.W., 2007. Perceived peer influence and peer selection on adolescent smoking. Addict. Behav. 32, 1546–1554. Huisman, M., 2007. Imputation of missing network data: some simple procedures. In: International Sunbelt Social Network Conference XXVII, Corfu, Greece. Kirke, D.M., 2004. Chain reactions in adolescents’ cigarette, alcohol and drug use: similarity through peer influence or the patterning of ties in peer networks? Soc. Networks 26, 3–28. Kobus, K., 2003. Peers and adolescent smoking. Addiction 98, 37–55. Mercken, L., Candel, M., Willems, P., de Vries, H., 2007. Disentangling social selection and social influence effects on adolescent smoking: the importance of reciprocity in friendships. Addiction 102, 1483–1492. Mercken, L., Candel, M., Willems, P., de Vries, H., 2009. Social influence and selection effects in the context of smoking behavior: changes during early and mid adolescence. Health Psychol. 28, 73–82. Richards, W.D., 1995. The NEGOPY Network Analysis Program. Department of Communications, Simon Fraser University, Burnaby, BC. Steglich, C., Snijders, T.A.B., Pearson, M., 2004. Dynamic Networks and Behavior: Separating Selection from Influence. Working Paper. Interuniversity Center for Social Science Theory and Methodology. Tucker, J.S., Ellickson, P.L., Klein, D.J., 2003. Predictors of the transition to regular smoking during adolescence and young adulthood. J. Adolesc. Health 32, 314–324. Urberg, K.A., Degirmencioglu, S.M., Pilgrim, C., 1997. Close friend and group influence on adolescent cigarette smoking and alcohol use. Dev. Psychol. 33, 834–844.