Whereas our initial models tested the relationship between interdependent substance use behavior, they assumed that these effects are symmetric: that is, usage of one substance equally increases or decreases usage of another substance. In our next set of models, we relax this assumption and test whether usage of one substance increases behavior of another substance or decreases behavior , or both . These models were estimated separately as the combined model exhibited extreme collinearity. As shown in Table 3, there is a significantly positive creation function from marijuana use to drinking in both samples, implying that respondents’ marijuana use increased their odds of drinking initiation. Thus, one unit higher marijuana use made a nondrinker 62% and 60% more likely to start drinking rather than stay as a non-drinker at the next time point in Sunshine High and Jefferson High, respectively. On the other hand, the endowment function from marijuana use to drinking is not statistically significant at either school, implying that marijuana use does not affect the likelihood of stopping drinking behavior. The impact of marijuana use on smoking behavior differs across the two schools. We detect a statistically significant creation function in Sunshine High: a one unit increase in marijuana use increases the odds 62% that adolescent non-smoker will initiate smoking rather than stay as a non-smoker. There was no evidence of a statistically significant endowment function in Sunshine High. On the other hand, the pattern is reversed in Jefferson High with a statistically significant endowment function but a statistically insignificant creation function. Thus, in Jefferson High although marijuana use does not impact respondent’s likelihood of smoking initiation, one unit higher marijuana use made smokers 27% more likely to stay as smokers rather than quit smoking at the next time point.To understand the magnitude of these effects , we engaged in a small simulation study in which we omitted some of the effects from the SAB model shown in Table 2 and assessed the consequences for the level of substance use behavior in the schools. That is,grow rack we changed a particular parameter value from the one estimated in the model to zero, and then simulated the networks and behaviors forward 1000 times.

We then assessed the average level of smoking, drinking, and marijuana use in the network at the end of the simulation runs. To save space, we only present the results for Sunshine High; see S2 File for the Jefferson High results, which were similar.The highest level of smoking is observed when we set to zero the influence effect of friends on smoking behavior, as the percentage of non-smokers drops from 72% in the original model to 63%, and the percentage of heavy-smokers increases from 11% to 18% . The pattern was similar in Jefferson High, with analogous values of 48% to 42%, and 31% to 35%. This corroborates the findings in previous simulation research that peer influence has a protective effect on smoking and drinking adoption. The lowest levels of smoking are observed in the hypothetical scenario in which marijuana use has no effect on one’s own smoking behavior, as the percentage of non-smokers rises from 72% to 81%, and the percentage of heavy-smokers decreases from 11% to 5%. The analogous values in Jefferson High were 48% to 54%, and 31% to 25%. Regarding drinking behavior, we see that the effect of one’s own marijuana use is particularly important as setting this effect to zero results in a decrease in drinking behavior . In the scenario of no effect of marijuana use on drinking behavior the percentage of nondrinkers rises from 50% to 59% and the percentage of heavy drinkers falls from 13% to 7%. The analogous values in Jefferson High were 35% to 42% and 16% to 10%. It is notable that setting the influence effect of friends’ drinking on one’s own drinking behavior to zero reduces drinking somewhat . In Jefferson High, the number of heavy drinkers rises from 16% to 20%. For marijuana usage, very pronounced strong effects are observed for friends’ influence . Setting this influence effect to zero results in a sharp decrease in non-marijuana users from 62% to 47%, and a parallel large increase in heavy users from 19% to 32%. In Jefferson High, the analogous values were 61% to 43% and 18% to 33%. In sum, when the effect from marijuana use to cigarette use is turned off, more non-smokers and fewer heavy-smokers are expected in both schools. When the peer influence effect with regard to each substance use is turned off, fewer non-users and more heavy-users of each substance are expected in both schools.

In the scenarios in which we set other parameters to zero, the simulation results indicated that the substance use distribution was not altered in either school.Overall, our findings indicate some evidence of sequential substance use, as adolescent marijuana use increased subsequent smoking and drinking behavior in our two school samples. Whereas some existing research has found evidence that marijuana use leads to use of these substances, an important contribution of our study was simultaneously taking into account the substance use behavior of adolescents’ peer networks and other social processes occurring in networks. We found that marijuana use resulted in more smoking and drinking in both samples. Our findings are partially consistent with Pearson et al. , who found that that marijuana users smoked cigarettes more over time. Our findings are suggestive that marijuana use increases both alcohol and cigarette use. In addition, we made a distinction between whether interdependent substance use going from marijuana to cigarettes and alcohol results in initiation, cessation, or both. We found that marijuana use resulted in drinking initiation in both samples, and smoking initiation in Sunshine High. In contrast, marijuana use decreased the likelihood of smoking cessation in Jefferson High. Previous literature suggests that alcohol use is not a prerequisite for the initiation of marijuana use and the effect of alcohol use on the onset of marijuana use has declined while that of marijuana use on the onset of alcohol use has increased since 1965 , and our findings are consistent with this prior literature. Moreover, we tested cross-substance influence effects, which assessed whether the substance use behavior of one’s friends on a particular substance affected an individual’s own use of the other two substances. We found no evidence that such effects exist in our samples. We did, however, find peer influence effects for each specific substance, which is consistent with multiple past studies. Note, however, that whereas one implication is that having more friends who use marijuana, for example, results in greater marijuana use behavior on the part of the individual, another implication is that having more friends who do not use marijuana results in less marijuana use behavior. This relative symmetry of influence effects is sometimes overlooked when interpreting influence results, and our simulation results confirmed that this influence effect is in fact more likely to have a negative effect on substance use behavior.

These results are similar to an earlier simulation study that found that increasing the amount of peer influence in two high schools diminished school level smoking and drinking behavior . These results are consistent with theoretical insights from the Dynamic Social Impact Theory, which would predict that youth in friendship networks would adopt the same substance use behaviors through peer influence pathways, likely through social proximity and consolidation of youths’ attitudes and behaviors in adolescent networks. This highlights that the presumption that influence effects will always increase behavior is not necessarily accurate. In fact, we might expect that the dominant norms in a context will drive the direction of influence effects: in a school with little substance use, the greater number of non-users will push adolescents towards non-use, whereas in a school with high levels of substance adolescents are more likely pushed towards greater use. Given the complexity of our agent-based network models,greenhouse grow tables we demonstrated the relative magnitude of the effects by combining a small-scale simulation with a strategy in which we constructed hypothetical models that set certain key effects to zero and simulated the networks and behaviors forward. A key finding was that in a simulated world in which one’s own marijuana use did not affect smoking or drinking behavior, there would be a notable decrease in overall levels of smoking and alcohol usage in these schools, even controlling for the complexity of these models. We also saw that marijuana use operates as a mechanism between friends’ marijuana use and one’s own smoking and drinking behavior, as adolescents’ use of marijuana is impacted by their friends’ marijuana use, and this then affects the adolescent’s level of cigarette and alcohol use. Furthermore, one of the strongest effects detected was the influence effect of friends’ marijuana usage, as this has a particularly strong relationship to adolescents’ own marijuana use. Our findings highlight the importance of understanding interdependence in the use of multiple substances in adolescence, particularly those which operate through peer influence effects within friendship networks. Another notable finding was that depressive symptoms increased smoking behavior in Jefferson High. This high school has a relatively high average level of substance use compared to Sunshine High. Perhaps in a social milieu with a high average level of drug use, adolescents reporting higher levels of depressive symptoms may be more likely to display higher levels of cigarette smoking as compared to those who report lower level of depressive symptoms, given that past studies link depression and adolescent smoking. There are some limitations to note in this study. First, the time lags between the two sets of waves are not equal . Although it is preferable to have equal time periods, we performed a post hoc time heterogeneity test to ensure that the co-evolution of substance use behaviors and friendship networks was not significantly different across the three waves, or two time periods. Second, our SAB model specification is data intensive and can only be estimated for the two large schools among the 16 saturated schools in Add Health which are feasible for this type of analysis.

This limits generalizability and does not allow assessing why the interdependent effect from marijuana use to smoking is different across the two schools. Third, we had indirect information about marijuana use at time one, for a large percentage of the sample. Using this indirect information allowed us to avoid discarding a large amount of information at t1, however with a relatively small amount of potentially misclassified cases. Fourth, while the data are relatively old, we are aware of no evidence that the mechanisms of in person friendship formation, as captured in these Add Health network data,have changed significantly since the mid-nineties. In the current study, friendship networks were constructed through name generator items instead of real-time communication technology such as cell phone use. While future studies are needed to leverage existing technology such as cell phone usage for collecting adolescent social network data, these in person network data are likely still meaningful. Moreover, research suggests that cell phones help reinforce and reproduce existing social roles and structures rather than alter them. That said, future studies are needed to collect nationally representative contemporary data from US adolescents and investigate how the findings herein would be different if such technology was considered.Our findings have important implications for future studies. First, our findings suggest both feasibility and merit in exploring concurrent or sequential substance use behaviors across multiple time periods. Interdependence in substance use should be studied within one single model framework with multiple simultaneous on-going processes to reduce the risk of over-estimation of each process due to the auto correlation among them. Second, further explication of the interdependent effects from marijuana use to smoking and drinking is a useful direction for future research. Third, given smoking rates among adolescent youth have decreased significantly since the mid-1990s, more recent data are required to test whether our findings from these two Add Health large schools can be replicated in future research. Our findings also have practical implications for health behavior change interventions targeting adolescent substance use. Moreover, other research indicates that social networks can be leveraged for health behavior change interventions and may even be superior to non-network based interventions . Peer network based interventions targeting adolescent substance use might address the possibility that marijuana use increases alcohol and cigarette use.