Table 5.

For selected tokens, strata with most significant increase and decrease in likelihoods of posting to SW. We show the top distinguishing tokens for each

	Increase likelihood of MH → SW			Decrease likelihood of MH → SW
treat. token	propensity (strata #)	effect	distinguishing historical words	propensity (strata #)	effect	distinguishing historical words
baby	0.03–0.04(5)	+44%	me_on, problem, mind, week, was_in	0.00–0.01(2)	−39%	wait, okay, thanks, re, her
have_sex	0.01–0.02(4)	+43%	instead_of, months, isn, well, a_girl	0.03–0.05(6)	−53%	months, dating, isn, such, there
medication	0.06–0.09(6)	+46%	seem_to, its, back, myself_and, seem	0.02–0.03(3)	−51%	but_then, not_to, d, i_never, every-thing
own	0.30–0.36(7)	+37%	all_i, can_be, we, of_this, this	0.45–0.59(9)	−32%	stop, my_own, yourself, you_can, needs
relationship_with	0.00–0.01(2)	+44%	spent, finally, my_mind, etc, think_it	0.13–0.17(9)	−36%	had_a, good, i_as, m_not, do_i
stressed	0.08–0.09(9)	+44%	i_do, as, i_hate, by, i_m	0.02–0.04(3)	−33%	there_and, but_they, deal_with, ve_had, took
upset	0.13–0.16(8)	+58%	but_i, some, a, haven_t, we	0.08–0.11(6)	−28%	living, is_this, with_the, i_started, sorry