Methods: Data consisted of clinical records of 353,294 affirmations exchanged between 4,634 residents of four Midwestern TCs over an eight year period. 12.7% of all participants were female and 87.3% were male. Men and women were segregated into separate residential units. 72.1% of all residents were European-American, 16.7% African American, 4% Latino and the rest unknown. Each affirmation was treated as a directed tie in a social network. Fifty-five two-week time series of successive daily networks were drawn. The network time series were analyzed using the RSiena network time series analysis program. The program utilizes a stochastic actor-oriented model in which Markov Chain Monte Carlo simulation is used to minimize the distance between the estimated parameters and target parameters defined by the actual data (Snijders & Ripley, 2010). The analyses were combined using the meta-analytic methodology of Snijders and Baerveldt (2003).
Results: Residents showed a statistically significant tendency to reciprocate affirmations (B = .72, SE = .04) and a more modest but statistically significant tendency to close transitive triads (B = .08, SE = .01). Residents who received an affirmation one day were more likely to receive one on the next (B = .08, SE = .006) but less likely to give one (B = -.05, SE = .008). They also showed a statistically significant tendency to affirm peers who entered the facility at a similar time (B = .74, SE = .09). Homophily, in this case the tendency to affirm others of the same race, played a small but statistically significant role in network formation (B = .10, SE = .02). All of these relationships were statistically significant at the p < .001 level.
Implications: The results of this analysis demonstrate that participation in therapeutic communities arises to a great extent from interaction between peers. It will therefore be largely unpredictable by variables that can be measured before program entry, such as motivation or projected likelihood of recidivism. Since participation is itself a powerful predictor of outcomes, this suggests that there are limits to our ability to increase program success using predictors measured before admission. On the other hand, it also raises the possibility that clinicians could leverage the social networks of TC residents to improve treatment. For instance, the strong tendency to reciprocate pushups might be used to overcome resident conflict, or use the tendency of residents to affirm those who arrived at roughly the same time to build groups of cooperating peers.