Aim One
Goodness-of-fit metrics suggested that a four-topic model was optimal
(see Supplemental Table 1). The most strongly representative words
(measured by term score11Because some words are highly probable
across all or several topics, word probabilities alone are not always
sufficient to interpret a topic. Term scores emphasize words thatuniquely represent each topic by downweighting words that are
highly probable across many topics.; Blei & Lafferty, 2009)
associated with each topic are shown in Table 1. Face validity of the
topics were assessed third and last author. Topic 1 corresponded to
discussion of daily difficulties (i.e., tax, staff, homeless); Topic 2
corresponded to discussion regarding family members (i.e., dad, mom,
sister); Topic 3 corresponded to the use of adjectives often used as
verbal fillers (i.e., like, really, things); and Topic 4 corresponded to
discussion of friendships and romantic relationships (i.e., friends,
talking, meet, dating).