Table 2.

Characteristics of the longitudinal included studies.

Data source (country)	Authors (year)	Study details	Outcome	Type of network(s): (friendship ties, family ties, best friend ties, etc)	Social network measure(s) (Peer influence effects on cannabis use and peer selection effects related to cannabis use)	Social network analysis, software and built models
Add Health (USA)	Wang et al. (2018) (37)	3 waves n = 3,128 Aged 12–17 48% female	Cannabis last month into 3 levels: 0 = “never,”1 = “1–10times,”2 = “more than 10 times”	2 school friendship networks Friendship ties: nominations of maximum 5 male best friends and 5 female best friends, within the same school (not necessarily the same grade).	Peer Influence effects on cannabis use: Indegree (own popularity; received friendship nominations) Peer influence cannabis use Number of friends who smoked Number of friends who drank Peer Selection effects on cannabis use: Similarity of cannabis use	SNA (SABM)/R Siena For each school, three-wave SAB model was built. Each model had three behaviour functions, each modelling the dynamic of one behaviour: alcohol use, tobacco use and cannabis use. In this review, we just focused on cannabis. In the behavior equations, peer influence effects were measured as the sum of negative absolute difference between ego’s and alters’ behavior averaged by ego’s out-degree. In the network equation, they included endogenous network effects (e.g., reciprocity) and homophily selection effects for each substance use behavior as well as additional covariates.
	Schaefer (2018) (38)	3 waves n = 1,373 Mean age: 15.56 48.87% female	Cannabis last month was recoded into dichotomous measure (1 = yes cannabis use, 0 = no cannabis use)	2 school friendship networks Friendship ties: nominations of maximum 5 male best friends and 5 female best friends, within the same school (not necessarily the same grade).	Peer Influence effects on cannabis use: Friend’s cannabis use (i.e., average similarity of cannabis use, that is, adolescent’s preference to have similar cannabis use than their friends, where the total influence of the friends is the same regardless of the number of friends) Peer Selection effects on cannabis use: Ego’s cannabis use Alter’s cannabis use Similarity cannabis use Ego’s cannabis use x R alter R ego x alter’s cannabis use *R refers to ego’s risk factors	SNA (SAOM)/ R Siena Four R Siena models were built for both schools. The difference between the models is that each model provided an illustration of the SAOM for one risk factor of cannabis use. Risk factors are: M1 = Family Connectedness, M2 = School belonging, M3 = Grade point average (GPA), M4 = Religiosity and M5 = Self-control. Changes in the friendship network were modelled with two functions: a rate function that determines which actor is given the chance to change a tie, and a friend selection function that determines which change a chosen actor makes. Change in a given behavior was modeled with two functions: a rate function to specify how often individuals are given the chance to change their behavior, and a behavior function that includes predictors of behavior change. They used a multigroup method to estimate one pooled model for the two schools.
	De la Haye et al. (2015) (39)	3 waves n = 1,612 Mean age: 16.4 48.87% female	A dichotomous measure of lifetime of cannabis use was computed at each wave where 1 = ever used cannabis, with changes from 0 to 1 in history of use between waves capturing cannabis initiation.	16 school friendship networks Friendship tie: nominations of maximum 5 male best friends and 5 female best friends, within the same school grade.	Peer Influence effects on cannabis use: Friends lifetime cannabis use Peer Selection effects on cannabis use: Any history of cannabis use and current cannabis use of ego Any history of cannabis use and current cannabis use of alter Same history or same current cannabis use	SNA (SABM)/R siena 4.0 For each school, various R Siena models were built. Baseline models (M1) included effects of history of cannabis use and covariates, but not risk factors, in predicting the friendship network and history of cannabis use. Phase 2 models (M2) added effects of current (last month) cannabis use on friendship choices. Phase 3 models tested for effects of each of the risk factors on friendship choices. Phase 3 included all parameters from M2 model, + the three new effects (ego, alter, similar) of the risk factor on friendship selection, and the effect of the risk factor on change in history of cannabis use (i.e., cannabis initiation). Final models (M3) included risk factors that were found to significantly predict friendship choices and/or history of marijuana use in phase 3 alongside parameters included in M1 and M2. Therefore, M1 = friendship network and history of cannabis use dynamics; M2 = friendship network, history of cannabis use, and current cannabis use dynamics; M3 = friendship network, cannabis use, and risk factor dynamics. Only significant effects on friendships or cannabis initiation, independent of other risk factors, were retained in M3.
	Vogel et al. (2015) (40)	3 waves n = 7,754 Mean age: 15.2 55% female	Cannabis last month was used as dichotomous measure (1 = yes cannabis use, 0 = no cannabis use)	109 school friendship networks Friendship tie: nominations of maximum 5 male best friends and 5 female best friends, within the same school grade.	Peer Influence effects on cannabis use: Own popularity (indegree; number of friendship nominations an individual received on the friendship roster) Peer substance use (alcohol and tobacco) Network centrality (the number of ties the respondent has weighted by the number of ties of those to whom he/she sends and received nominations) Popularity x density; popularity x school connectedness; popularity x school drug use (drug use is referred to alcohol and tobacco)	Multilevel logistic regression models on Stata/preconstructed measures from secondary data that can be found on the Add health website. Three hierarchical logistic regression models were performed: model 1 included the main effects of the socio-demographic characteristics, individual risk factors, and peer-network characteristics on self-reported cannabis use. Second model introduced the school level covariates (network density, connectedness, and normative drug culture) and model 3 included three interactions of school context (popularity x density, popularity x connectedness, popularity x school drug use).
	Tucker et al. (2014) (41)	3 waves n = 1,612 Mean age: 16.4 47.3% female	Cannabis last month into 4 levels: 0 = none, 1 = 1–3 times, 2 = 4–11 times, 3 = 12–32 times, and 4 = 33 times or more.	2 schools friendship networks Friendship tie: nominations of maximum 5 male best friends and 5 female best friends, within the same school grade.	Peer Influence effects on cannabis use: These effects were measured through interactions (e.g., friend’s cannabis use x reciprocity). Friends’ cannabis use Reciprocity (both participants nominated each other as a friend) Friend popularity (friend indegree: the total number of friendship nominations received by a nominated friend) Popularity difference of respondent (the difference in number of friend nominations received, i.e., indegree) Peer Selection effects on cannabis use: Ego’s cannabis use Alter’s cannabis use Squared alter cannabis use Same cannabis use	SNA - SABM/R Siena For each school, 3 R Siena models were build. The only difference between the 3 models is that each one of the models included one specific interaction with peer influence, either friends’ cannabis use (influence) × Friendship reciprocity (model 1-M1), Friends’ cannabis use (influence) × friend popularity (model 2-M2) or Friends’ cannabis use (influence) × popularity difference (model 3-M3).
	De la Haye et al. (2013) (42)	3 waves n = 1,612 Mean age: 16.4	Cannabis last month into 4 levels: 0 = none, 1 = 1–3 times, 2 = 4–11 times, 3 = 12–32 times, and 4 = 33 times or more. They also created a dichotomous measure of lifetime use at each wave (where 1 = had ever used cannabis, with changes from 0 to 1 in lifetime use between waves capturing initiation)	2 school friendship networks Friendship tie: nominations of maximum 5 male best friends and 5 female best friends, within the same school grade and out-school friends, but finally they only included friends were also survey respondents	Peer Influence effects on cannabis use: Friends’ cannabis use (last month/lifetime) Peer Selection effects on cannabis use: Ego’s cannabis use Alter’s cannabis use Squared alter cannabis use (on last month) Same cannabis use	SNA, SABM/R Siena Two R Siena models were estimated to examine associations of adolescent friendships with (1) cannabis initiation, and (2) frequency of past month cannabis use. Each model includes effects predicting the evolution of the friendship network (friend selection effects) and effects predicting cannabis use (cannabis effects). For cannabis initiation, friend influence was tested with two effects: the effect of having friends who had ever used cannabis in their lifetime, and the effect of having friends who had used cannabis in the past month.
PROSPER (Promoting School-Community-University Partnerships to Enhance Resilience) (USA)	Osgood et al. (2014) (43)	5 waves n =  9,500 Mean age: 6th through 9th grades 51.5% female	Cannabis last month into 5 levels: from 1 = “Not at all” to 5 = “More than once a week.”	27 school friendship networks	Peer Influence Effects: Indegree (own popularity; friendship nominations an individual received, i.e., number of other students who named the respondent)	Multi-level logistic regression model/a routine programmed themselves (SAS)
				Friendship tie: nominations of maximum 2 best friends and 5 additional friends from their current school grade. 368 school grade cohort friendship networks.	“Group members” were distinguished as (1) core member vs. (2) peripheral members of the group (meaning that removing a single friendship link would be sufficient to separate them from the main portion of the group), while “non-group members” are (1) isolates (who did not send or receive friendship nominations to anyone else in the grade network or who shared ties with one person who was disconnected from the rest of the network, forming an isolated dyad), (2) liaison (who had ties to members of two or more groups) and (3) other non-members (students who were not defined as members of a group, a liaison, or an isolate). All participants are core group members except those labelled as peripheral group members, liaison and isolate. Core member was treated as the reference to test the others. Friend’s cannabis use Outdegree (number of friendship nominations an individual made) Reach (reach of an adolescent to others in the networks through pathways of ties)	A three-level logistic regression was performed. Five waves of data (level 1) as nested within individual respondents (level 2-stable individual differences in substance use), who are in nested within the school district cohorts that define the social networks (level 3-unexplained differences among social networks in rates of substance use). For the group detection they used a variant of Moody’s CROWDs routine, which is similar in form to other algorithms designed to search for groups by maximizing modularity scores.
Context of Adolescent Substance Use Study (USA)	Ennett et al. (2006) (44)	5 waves n = 5,104 Mean age: sixth (35.9%), seventh (33.1%), and eighth (31.0%) graders 65.5% female	Cannabis last 3 months: from 0 to 10 or more times. *Because responses were skewed toward never and infrequent use, a binary variable was formed for each that contrasted adolescents who reported any days/times of use in the last 3 months with those who reported none.	26 separate networks from 13 schools. Friendship tie: nominations of maximum 5 best friend within the same school grade.	Peer influence effects: Social embeddedness: Reciprocity Neighborhood density (number of friendship ties present among friends/alters divided by the total number of possible ties) Out nominations (out-degree)	Three-level hierarchical generalized linear models/SAS IML was used to calculate all measures except two: betweenness centrality and Bonacich power centrality, which were calculated by UCINET (Version 6).
					Social position: (1) Group member (who shared most of their friendship ties with each other and where the removal of one member of the group would not cause the group to be disconnected, (2) Isolate (one or no friendship ties) and (3) Bridge (those with friendship ties to adolescents who were members of different groups, but who were not themselves members of any group) *The three social positions were measured by two dummy-coded variables with group member as the reference Social status: Normed indegree (own popularity, measured by number of friendship nominations received by ego divided by the number of possible friendship nominations) Reach centrality (incoming ties only) Betweenness centrality (the possibility to the adolescent can control flows of information or norms by serving as a gatekeeper between peers, and can connect peers from different parts of the network who are not directly connected to each other) Bonacich power centrality (centrality of the friends with whom ego is linked) Social proximity to cannabis users: Best friend cannabis user (having a best friend who reported recent use) No. neighborhood cannabis users Distance to cannabis user (low coding for nearest user friends or nearest user friend’s friends)	A three-level hierarchical generalized linear model was performed. The three levels were time nested within adolescents nested within networks. Data were arranged in a cohort sequential design with adolescent age. For each network variable, they presented the exponentiated b coefficient predicting the starting point of cannabis use at the different ages (i.e., the age 11 odds ratio). They included the main effect of the network variable and the interaction between the network variable and age. Moody’s CROWDS algorithm for identifying peer groups was also used to measure adolescents’ group position in the network as a group member, bridge, or isolate.

*Means relevant clarifications.