Abstract:
Online mental health communities enable people to seek and provide support, and there is growing evidence showing the efficacy of participation in these communities to help cope with mental health distress. However, what factors of peer support lead to favorable psychosocial outcomes for individuals is less clear. Using a dataset of over 300K posts by ~39K individuals on an online community TalkLife, we present a longitudinal causal inference study to investigate the effect of several factors, such as adaptability, diversity, immediacy, and nature of support. Unlike typical causal inference studies that focus on the effect of each treatment, we focus on the outcome and address the reverse causal question of identifying treatments that may have led to the outcome, drawing on case-control studies in epidemiology. Specifically, we define the outcome as an aggregate of affective, behavioral, and cognitive psychosocial change and identify Case (most improved) and Control (least improved) cohorts of individuals. Considering supportive responses from peers as treatments, we evaluate the differences in the responses received by Case and Control individuals, per matched clusters of similar individuals. We find that effective support include complex language factors such as diversity, adaptability and language style, but simple indicators such as the quantity, immediacy or emotionality of support are not causally relevant. Our work bears methodological and design implications for online mental health platforms, and has the potential to guide suggestive interventions for peer supporters on these platforms.
