**4. Results**

Means, standard deviations, and correlations are presented in Table 1.

In order to perform the network analyses, data was processed in the open-source statistical programming language R (R Core Team 2013, Vienna, Austria). Networks were generated by aggregating individual perceptions within each group, in each simulation, at four points in time. Networks visualization was run using the qgraph package from R (Epskamp et al. 2012) and the igraph package (Csardi and Nepusz 2006). In order to weight the degree of collaboration for each stakeholder nested in each simulation, we used the scores ranging from 0 to +5 where 0 represents absence of collaboration and +5 a very collaborative relationship (scores between −5 and 0 were excluded as they were illustrative of conflictual relations and not collaboration.). The aggregated networks obtained were in the form of weighted undirected networks with multiple ties. Closeness and betweenness indices (Freeman 1979) were computed with the centrality function from qgraph packages. This function computes and returns betweenness and closeness indices between all pairs of nodes in the graph with a tuning parameter of α = 1. When α = 1, the outcome is the same as the one obtained with the classical Dijkstra's algorithm (Dijkstra 1959; Opsahl et al. 2010).


As we collected data in four successive waves during the simulation, we could explore sequential mediation models (model 6) using the Process Macro (Preacher and Hayes 2008, Preacher and Hayes 2008). As network indicators were computed based on aggregated networks at the group level, we entered group size as a covariate in the analyses. We then estimated sequential mediation paths from trust self-enhancement as evaluated at Time 1 (expectations) to subsequent centrality scores in the four time lags. The results of the mediation analysis for betweenness as an indicator of network centrality are presented in Table 2 and summarized in Figure 1.


**Table 2.** Overview of the mediation effects estimated in our analyses.

Note: TSE—trust self-enhancement, CT1—centrality at Time 1, CT2—centrality at Time 2, CT3—centrality at Time 3, CT4—centrality at Time 4.

**Figure 1.** Results of the sequential mediation model for collaboration betweenness. Note: SE—self-enhancement, CollBet—collaboration betweenness, T1 = Time 1, T2 = Time 2, etc., \* *p* < 0.05; \*\* *p* < 0.01; path coefficients are non-standardized coefficients reported from the most complete model, i.e., the model in which all previous variables in the mediation chain are included.

The full sequential mediation chain from trust self-enhancement expectations, to collaboration betweenness at Time 1, then at Times 2, 3, and 4 is significant. The indirect effect (trust self-enhancement → collaboration betweenness at T1 → collaboration betweenness at T2 → collaboration betweenness at T3 → collaboration betweenness at T4) was positive and significant, the effect size was 0.13 (SE = 0.12), with a 95% confidence interval (CI) of 0.01–0.70, and because the confidence interval did not contain zero, we can conclude that the indirect effect was positive and significant as hypothesized. In other words, trust self-enhancement had a positive influence on the perceived betweenness at the end of the simulation, by sequentially increasing betweenness at Time 1, then Time 2, then Time 3 and Time 4. However, the results of the sequential mediation revealed an additional significant indirect effect. This indirect effect led from trust self-enhancement expectations at Time 1 to betweenness at Time 2, then Time 3, then Time 4, thus estimating the effect of trust expectations on network centrality as estimated after the inter-group interactions commence. This indirect effect was, however, negative: −0.12 (SE = 0.10), 95% CI [−0.55, −0.01], and as the confidence interval did not include zero, the effect was considered significant. In other words, trust self-enhancement negatively predicted the betweenness at the end of the simulation, by sequentially decreasing betweenness at Times 2 and 3.

We used a similar bootstrapping procedure to estimate the sequential mediation effects from trust expectations to network closeness. The results of the mediation analysis for the closeness centrality indicator are presented in Table 2 and summarized in Figure 2.

**Figure 2.** Results of the sequential mediation model for collaboration closeness. Note: SE—self-enhancement, CollClo—collaboration closeness, T1 = Time 1, T2 = Time 2, etc.; † *p* < 0.10, \*\* *p* < 0.01; path coefficients are non-standardized coefficients reported from the most complete model, i.e., the model in which all previous variables in the mediation chain are included.

As indicated by the path coefficients presented in Figure 2, the full sequential indirect effect (trust self-enhancement → collaboration closeness at T1 → collaboration closeness at T2 → collaboration closeness at T3 → collaboration closeness at T4) was positive. The indirect effect was 0.02 (SE = 0.01), 95% CI [0.01, 0.05], and as the confidence interval did not include zero, we can conclude that the effect was significant. The sequential mediation analysis for closeness did not reveal any other significant indirect effects; therefore, we can conclude that the indirect effect for closeness was aligned with our hypothesis. A full summary of all mediation effects estimated with model 6 in the Process Macro (Preacher and Hayes 2008) is presented in Table 2.
