Table 3.

Results of Bayesian multilevel models estimating change from pretraining to posttraining among all participants (model 1) and the difference between the 2 conditions in change from the first to second laboratory visits (model 2).

Outcome variable		Pre-post training effect (model 1), estimatea	Pre-post training effect (model 1), 95% CIb	Visit×condition interaction (model 2), estimate	Visit×condition interaction (model 2), 95% CI
Total sentence units utteredc		−1.88	−2.21 to −1.54	−1.97	−2.47 to −1.46
Behavior categoriesd
	Restatement	16.44	7.92 to 34.81	28.50	6.89 to 119.1
	Influencing	0.09	0.04 to 0.16	0.14	0.07 to 0.28
	Open-ended questions	3.63	2.46 to 5.42	1.40	0.59 to 3.22
	Closed-ended questions	1.15	0.84 to 1.55	0.63	0.36 to 1.07
	Self-disclosure	0.05	0.01 to 0.16	0.09	0.03 to 0.25
	Sympathy	0.86	0.59 to 1.23	1.08	0.59 to 1.95
	Other	0.51	0.36 to 0.71	0.64	0.36 to 1.13
Adherence scoree		1.64	1.39 to 1.9	1.90	1.48 to 2.32
Perceived session helpfulnesse
	CSRSf task reactions	0.96	0.69 to 1.26	1.00	0.41 to 1.56
	CSRS relationship reactions	0.28	−0.06 to 0.62	0.20	−0.38 to 0.77

^aEstimate: mean of estimated posterior distribution.

^b95% CI: 95% credibility interval.

^cCount variable with Poisson link function; coefficients are in log units.

^dBinomial or Bernoulli distributed variables with logistic link function; coefficients are in odds ratio units. For model 1, the estimate represents the relative odds of the behavior at posttraining compared with pretraining; for model 2, the estimate represents the degree to which the relative odds of a behavior at visit 2 compared with visit 1 were higher or lower in the group that underwent immediate training.

^eMetric variables with identity link function; coefficients are in standardized units (ie, SDs).

^fCSRS: Crowdsourcing Mental Health Session Reaction Scale.