Initiation of Drink Reduction and Baseline Condition Equivalence
A repeated measures ANOVA indicated no significant change in drinking across the nine weeks prior to randomization (p > .39). Initial descriptive statistics of demographics yielded no significant differences between treatment groups (). Similarly, the selected markers for drinking severity were equivalent across conditions on all the variables except total number of alcohol dependence criteria (F(2, 86) = 3.673, p < 0.03). A Bonferroni post hoc test was performed and MI was found to have a trend mean difference of .95 more alcohol dependence criteria than SOMI (p = .051). Neither MI nor SOMI significantly differed from the SC condition. Examination of suggests that participants in the MI condition tended to have higher scores on several other drinking severity measures when compared to the other conditions including mean drinks per drinking day, ADS, and the percentage of participants reporting prior treatment, even though these differences were not statistically significant. As noted above, this appeared to be the result of two MI participants who were outliers on severity of drinking measures.
Baseline Characteristics of Study Sample (N = 89)
Condition Fidelity and Discrimination
As predicted, scores on the three MITI scales were 4.0 and above across the treatment conditions. There were no significant differences in Empathy (t(10) = −1.387, ns) or Autonomy/Support (t(10) = −1.536, ns), but MI sessions scored significantly higher on Collaboration (t(10) = −5.0, p < .01) than SOMI. Examination of coding suggested that Collaboration in MI was higher because therapists were given additional credit for MI directive activities (e.g., completing a change plan) that were proscribed in SOMI. As hypothesized, highly significant differences were found between conditions on Evocation, Direction, and Activities. Among the global rating scales Evocation and Direction, MI demonstrated a mean well above 4.0 (Evocation: M = 4.47, SD = .58; Direction: M = 4.56, SD = .59), indicating MI competency, where as SOMI was significantly lower at p < .001 (Evocation: M = 1.97, SD = .56; Direction: M = 2.08, SD = .75). As expected, MI demonstrated a significantly higher mean score per session on Activities (MI: M = 2.3, SD=1.3, SOMI: M = .05, SD = .235; t(17.98)= −6.79, p < .001).
There were no significant differences across conditions on the therapeutic bond (MI: M = 2.7, SD = .32; SOMI: M = 2.6, SD = .31) with ratings in both conditions approaching the maximum anchor (3 = very much). Conditions differed significantly on ratings of directiveness regarding drinking (t(53) = 5.4, p < .001). MI participants rated the amount of therapist guidance around drinking between “some” and “pretty much” (M = 1.8, SD = .76), whereas guidance was rated in SOMI between “not at all” and “some” (M = .66, SD = .76). SOMI participants rated Session 2 on average between “pretty helpful” and “very helpful” (M = 4.8, SD = .83) and MI participants rated it as very helpful (M = 5.0, SD = .93).
Condition Effects on Drinking Outcome at End Treatment and One Month Follow-up
There was a significant reduction in drinking from baseline to the last month of the treatment period, but there were no significant differences in outcomes across conditions. Condition effects were in the opposite of the predicted direction with SOMI reducing drinking the most, followed by MI, and then SC. However, condition differences accounted for less than 1% of the variance. Baseline drinking significantly predicted (β = .53, SE = .064, p < .001) drinking during the outcome period. Most of the reduction occurred within the first two weeks after randomization, and was then maintained for the remainder of the follow-up period.
We conducted several post-hoc analyses in order further elucidate the findings. First, we considered outcome measures including drinks per drinking day and number of negative consequences. None of the condition analyses of the secondary outcome measures were significant (all ps > .40 and none were in the expected direction). Next we examined whether there might be a problem severity by treatment condition interaction such that MI was more effective for those with higher severity. This interaction was not significant (p > .6). Follow-up analysis indicated that participants in MI and SOMI continued to significantly reduce their drinking from end treatment (M = 21.3, SD = 13.7) to posttreatment (M = 16.2, SD = 11.9), but there were no significant differences in posttreatment drinking across MI (M = 16.9, SD = 13.9) or SOMI (M = 15.2, SD = 9.2).
Post-hoc Analyses of Condition Differences on Early Drinking Outcomes
While hypotheses were not supported at end or post-treatment, regression analyses indicated condition differences were in the expected direction during the two weeks following randomization, although differences were not statistically significant (p > .5). For example, during the first week after randomization, participants in the MI condition reduced their drinking (M=2.6, SD=10.9), while participants in SOMI and SC showed no change. While participants across conditions significantly reduced their drinking in the two weeks after randomization, participants in the MI condition reduced drinking more than those in SOMI and SC. We computed effect sizes for the difference in reduction of drinking across conditions for early change and found d=.27 for MI relative to SOMI and d=.41 for MI relative to SC.
Formal Mediation Analysis
presents mediation analysis results for Session 1 and 2. As noted above, condition did not significantly predict outcome for the week following session 1, but the observed condition effect was in the predicted direction (Step 1). MI relative to SOMI predicted significantly greater change talk (Step 2). Greater change talk predicted reduced drinking at a trend level (p < .06) (Step 3). There was a reduction in the unstandardized regression co-efficient (Step 4) for condition from Step 1 (B = 1.59) to Step 3 (B = −0.98) indicating that change talk mediates the relationship between condition differences and outcome in session 1. Results for session 2 indicate change talk was significantly greater in MI relative to SOMI (Step 2). However, change talk did not significantly predict week 2 outcome (p > .3). Thus, Step 3 was not supported.
Baron and Kenny analysis of mediation