As shown in Fig. 2, we regressed outcomes on each mediator (ISS and SNI) and PAS. Regressions were specified in a way consistent with MLT such that SNI was the first mediator and ISS was the second mediator. ISS was regressed on SNI and both mediators were regressed on PAS (MacKinnon, 2000). This was strictly a theory-based specification; we could not establish temporal ordering owing to the single cross-sectional assessment of PAS, SNI, and ISS at W2, which were concurrently assessed with the past-year psychiatric outcomes.
