Aim Three
To test whether topic use was related to narrative valence, we used a regression model of sentiment scores with the topic proportion estimates as predictors, as well as sample, NSSI history, emotional dysregulation, and the second- and third-order interactions among sample, NSSI history, and emotion dysregulation. By comparing this model to a model without the four topic proportions, we found that the topics were not significantly related to narrative valence, F (3, 196) = 0.63,p = 0.595, R2 = .009. Only NSSI history was a significant predictor of narrative valence; participants with a history of NSSI used significantly more negatively-valenced language in their narratives than those without a history of NSSI,Mdiff = -0.21 (SE ) = 0.08, t (196) = -2.64, p = .009, 95% CI : [-0.37, -0.05].