> The honesty of the status updates written by the participants was assessed following the approach introduced by Newman, Pennebaker, Berry, and Richards (2003) using LIWC. Their analyses showed that liars use fewer first-person pronouns (e.g., I, me), fewer third-person pronouns (e.g., she, their), fewer exclusive words (e.g., but, exclude), more motion verbs (e.g., arrive, go), and more negative words (e.g., worried, fearful; Newman, Pennebaker, Berry, & Richards, 2003). The explanation was that dishonest people subconsciously try to (1) dissociate themselves from the lie and therefore refrain from referring to themselves, (2) prefer concrete over abstract language when referring to others (using someone’s name instead of “he” or “she”), (3) are likely to feel discomfort by lying and therefore express more negative feelings, and (4) require more mental resources to obscure the lie and therefore end up using less cognitively demanding language, which is characterized by a lower frequency of exclusive words and a higher frequency of motion verbs. Equation and usage rates in this study are summarized in Table 2.

> Newman et al. (2003) achieved up to 67% accuracy when detecting lies, which was significantly higher than the 52% near-chance accuracy achieved by human judges. Their approach has been successfully applied to behavioral data (Slatcher et al., 2007) and to Facebook status updates (Feldman, Chao, Farh, & Bardi, 2015). Other studies have since found support for these LIWC dimensions as being indicative of lying and dishonesty (Bond & Lee, 2005; Hancock, Curry, Goorha, & Woodworth, 2007; see meta-analyses by DePaulo et al., 2003 and Hauch, Masip, Blandón-Gitlin, & Sporer, 2012).

So it's all purely from linguistic analysis of the status updates. Seems like a stretch to me but I wouldn't doubt the overall result is true.