Table 2.
Results of machine learning models predicting FIS scores from transcripts
| FIS domain | R2 | Spearman’s rho | p | % human ICC |
|---|---|---|---|---|
| Verbal fluency | .01 | .27 | < .001 | .31 |
| Hope and positive expectations | .24 | .53 | < .001 | .60 |
| Persuasiveness | .18 | .47 | < .001 | .54 |
| Emotional expression | .00 | .31 | < .001 | .34 |
| Warmth, acceptance, and understanding | .16 | .46 | < .001 | .52 |
| Empathy | .18 | .47 | < .001 | .55 |
| Alliance bond capacity | .15 | .47 | < .001 | .52 |
| Alliance rupture-repair responsiveness | .13 | .47 | < .001 | .54 |
| Total score | .19 | .48 | < .001 | .52 |
Note: FIS = Facilitative Interpersonal Skills task (Anderson et al., 2009); R2 = cross-validated R2 that allows for negative values; p = p-value for Spearman’s rho; % human ICC = machine learning model performance relative to human coders’ intraclass correlation (ICC[3,8]).