Data Preparation

Data from the TLFB were aggregated to compute the average number of drinks per week and average BAC variables. BAC was calculated using the formula outlined by Matthews and Miller (1979), which adjusts for gender and body weight. Average BAC was calculated by applying this formula to the data for each of the 14 TLFB days for baseline and follow-up, and then averaging across BACs for drinking days to obtain mean BAC scores for baseline and follow-up time points. Heaviest BAC was calculated by using this formula with data from the two open-ended questions about the heaviest drinking day, at baseline and follow-up. Estimated BAC is an approximation of BAC and therefore subject to some error due to lack of control over conditions affecting individual rates of absorption and metabolism (

Davies & Bowen, 2000). However, estimated BAC and BAC have been found to be significantly correlated (

Carey & Hustad, 2002;

Hustad & Carey, 2005).

Summary statistics were generated to evaluate the distributions of variables. To correct for non-normality due to positive skew, the following variables were square-root transformed: average drinks per week (baseline and follow-up), average BAC (baseline and follow-up) and heaviest BAC (baseline and follow-up).

Comparisons across Assessment Mode

Drinking data collected from the TLFB assessment at follow-up were compared with drinking data collected from the DD, since these data reflect the same time period. Time 2 TLFB and DD data were found to be highly correlated (average drinks per week:

*r* = 0.96,

*p* < 0.001; average BAC:

*r* = 0.88,

*p* < 0.001). Although heaviest BAC was calculated from open-ended questions, this variable could also be derived from the TLFB and DD data. To assess consistency of report, heaviest BAC variables were created from the TLFB at Time 2 and the DDDs. These data revealed that the heavy BAC variable derived from open-ended questions was highly and significantly correlated with the TLFB and DD heaviest BAC variables (

*r* = 0.89,

*p* < 0.001;

*r* = 0.89,

*p* < 0.001). This same relationship emerged when the TLFB and DD BAC variables were correlated with each other (

*r* = 0.83,

*p* < 0.001). Since the three derived heaviest BAC variables were all highly correlated, it was decided to use the heaviest BAC variable derived from the open-ended questions, as this is most often reported in the literature (

Murphy et al., 2004; Collins et al., 2002; Larimer et al., 2001). These analyses indicated that the variables derived from the TLFB were highly correlated with the variables derived from the DD. Since the TLFB was administered in a controlled setting and was assessed at both baseline and follow-up, the TLFB variables will be used in the primary analyses. However, analyses were also examined with the DD variables and no differences were found for any of the outcomes. All results in this manuscript will refer to TLFB variables.

Descriptive Analyses

Chi-square and one-way analysis of variance (ANOVA) tests were performed to compare the three groups on demographic variables assessed at baseline (see ). The three groups did not differ on any demographic variables (gender, age, year in school, race, residence, and Greek membership), or on baseline drinking characteristics (see ). However, a group difference was present at baseline on one of the strategy use variables. One-way ANOVA revealed a significant group difference for Strategies While Drinking (*F*[2,174] = 3.22, *p* = 0.04), with the SM group having a higher score at baseline compared to the AR and SI groups.

| **Table 2**Drinking characteristics and drinking control strategy use, mean (SD) |

Correlations of the three subscales of the Strategy Questionnaire were as follows: Selective Avoidance and Strategies While Drinking (*r* = 0.48, *p* < 0.001); Selective Avoidance and Alternatives (*r* = 0.45, *p* < 0.001); Strategies While Drinking and Alternatives (*r* = 0.33, *p* < 0.001). Because the subscales shared only 11-23% of variance, separate analyses were run on each.

The relationship between strategy use and alcohol use at baseline was examined by correlating the three strategy use variables (Selective Avoidance, Strategies While Drinking, Alternatives) with the three drinking variables (average number of drinks per week, average BAC, and heaviest BAC). Average drinks per week was inversely correlated with Selective Avoidance and Alternatives scores (*r* = -0.18, *p* < 0.05; *r* = -0.18, *p* < 0.05, respectively), but positively correlated with Strategies While Drinking score (*r* = 0.18, *p* < 0.05). Correlation analyses examining the relationship of average BAC and strategy use revealed significant negative correlations for Selective Avoidance scores (*r* = -0.17, *p* < 0.05) and a significantly positive correlation for Strategies While Drinking scores (*r* = 0.19, *p* < 0.05). Alternatives scores were not significantly correlated with average BAC. Heaviest BAC displayed fewer relationships with strategy use compared to the other drinking variables, with only Selective Avoidance scores inversely correlated with heaviest BAC (*r* = -0.15, *p* < 0.05).

Compliance with Behavioral Instructions

All participants randomized into the study completed follow-up assessments. A percent change score in average drinks per week from baseline to follow-up was derived from the TLFB data. In total, 68% of the AR group participants decreased their alcohol use by some amount over the follow-up period. A percent change in strategy use from baseline to follow-up score was derived from the Strategy Use Questionnaire data. Overall, 84% of the SI participants increased their strategy use by frequency or number over the follow-up period. In comparison, 50% of the SM group reduced their alcohol use by some amount, and 55% of the SM group increased their strategy use by frequency or number.

Alcohol Use Outcomes

A 3 (group: AR, SI, SM) × 2 (time: baseline, follow-up) ANOVA using the square-root transformed average drinks per week variables was conducted. There was no main effect for group (*F*[2,174] = 1.64, *p* = 0.20), but there was a main effect for time *(F*[1,174] = 8.60, *p* = 0.004). Results indicated a significant group X time interaction (*F*[1,174] = 4.87, *p* = 0.008), as shown in . To explore the interaction, change scores were created to investigate differences in average number of drinks per week by subtracting baseline from follow-up raw values. The three change score variables showed non-normality in their distributions that could not be corrected by transformation. A one-way ANOVA comparing the three groups on change in average drinks per week was non-significant (*F*[2,174] = 2.53, *p* = 0.08). The mean change scores (and standard deviations) for average drinks per week for the three groups were as follows: AR = -3.69 (1.05), SI = 0.09 (1.31), SM = -1.71 (1.24). Only the AR change score differed significantly from zero (*t* = -3.50, *p* < 0.001). Post hoc pairwise comparisons revealed that the AR group reduced average drinks per week more than the SI group (*p* < 0.05), but not significantly more than the SM group.

The effect of group on intoxication level was assessed by examining average and peak BAC scores. A 3 (group: AR, SI, SM) × 2 (time: baseline, follow-up) ANOVA using the square-root transformed average BAC variables showed non-significant main effects of group (*F*[2,172] = 0.52, *p* = 0.59) and time (*F*[1,157] = 0.31, *p* = 0.58), and a non-significant group X time interaction (*F*[2,171] = 2.60, *p* = 0.08). A one-way ANOVA comparing the three groups on change in average BAC indicated no significant differences in change scores (*F*[2,171] = 1.24, *p* = 0.29), see . The mean change scores (and standard deviations) for average BAC for the three groups were as follows: AR = -0.01 (0.02), SI = -0.001 (0.03), SM = -0.01 (0.03); again only the AR group's change score differed significantly from zero (*t* = -3.21, *p* = 0.002).

A 3 (group: AR, SI, SM) × 2 (time: baseline, follow-up) ANOVA using the square-root transformed heaviest BAC variables revealed non-significant main effects of group (*F*[2,173] = 0.27, *p* = 0.77) and time (*F*[1,172] = 0.02, *p* = 0.89), and a significant group X time interaction (*F*[2,172] = 5.85, *p* = 0.004). A one-way ANOVA comparing the three groups on change in heaviest BAC showed significant differences in change (*F*[2,172] = 5.93, *p* = 0.003; see ). The mean change scores (and standard deviations) for heaviest BAC were as follows: AR = -0.02 (0.01), SI = 0.002 (0.01), SM = 0.03 (0.01). Both the AR and SM groups' change scores differed significantly from zero (AR: *t* = -2.10, *p* = 0.04; SM: *t* = 2.72, *p* = 0.008), however the SM group increased in heaviest BAC from baseline to follow-up, whereas the AR group decreased. Post hoc pairwise comparisons indicated that the AR group decreased in heaviest BAC significantly more than the SM group (*p* < 0.05). However the AR group did not differ from the SI group, and the SI group did not differ from the SM group.

Strategy Use Outcomes

Three separate 3 (group: AR, SI, SM) × 2 (time: baseline, follow-up) ANOVAs were conducted using the Selective Avoidance, Strategies While Drinking, and Alternatives variables. There was a significant main effect of time for the Selective Avoidance variable (*F*[1,174] = 40.90, *p* < 0.001) and a significant main effect of time for the Alternatives variable (*F*[1,174] = 8.48, *p* < 0.01). The main effect for Selective Avoidance was qualified by significant group X time interaction (*F*[2,174] = 7.12, *p* < 0.001); another significant group X time interaction emerged for Strategies While Drinking scores (*F*[2,174] = 6.77, *p* = 0.002). The group X time interaction for Alternatives scores was not significant (*F*[2,174] = 1.08, *p* = 0.34). represents these results graphically.

To further explore these findings, change scores were created for the Selective Avoidance, Strategies While Drinking, and Alternatives variables by subtracting baseline from follow-up values. One-way ANOVAs indicated significant differences in change for Selective Avoidance (*F*[2,174] = 6.85, *p* < 0.001) and Strategies While Drinking (*F*[2,174] = 6.77, *p* = 0.002) scores, but no significant difference in change for Alternatives scores (*F*[2,174] = 1.08, *p* = 0.34).

Selective Avoidance Pairwise comparisons on the Selective Avoidance change scores revealed that both the SI and AR groups significantly increased in strategy use compared to the SM group (*p* < 0.05), with no differences between the AR and SI groups. The mean change scores (and standard deviations) for Selective Avoidance for the three groups were as follows: AR = 3.13 (0.69), SI = 4.08 (0.80), SM = 0.46 (0.62), with both the AR and SI groups' change significantly different from zero (AR: *t* = 4.57, *p* <0.001; SI: *t* = 5.12, *p* <0.001).

Strategies While Drinking Examination of the means for Strategies While Drinking scores revealed that the SI group increased in strategy use (*M* = 3.02, *SD* = 1.09), the AR group showed relatively no change (*M* = -0.02, *SD* = 0.97), and the SM group decreased in strategy use scores (*M* = -2.07, *SD* = 0.85). The mean change scores (and standard deviations) for Strategies While Drinking for the three groups were as follows: AR = -0.02 (0.97), SI = 3.02 (1.1), SM = -2.07 (0.85), with both the SI and SM groups' change significantly different from zero (SI: *t* = 2.77, *p* = 0.007; SM: *t* = -2.45, *p* = 0.02). Pairwise comparisons of the means of the change scores indicated a difference between the SI and AR group, and the SI and SM group (*p* < 0.05), but no difference between the AR and SM groups.

Alternatives Examination of the means for Alternatives scores revealed that the SI group increased in strategy use (*M* = 1.54, *SD* = 0.60), whereas the AR group and the SM group showed relatively no change (AR: *M* = 0.45, *SD* = 0.49; SM: *M* = 0.77, *SD* = 0.53). The mean change scores (and standard deviations) for Alternatives for the three groups were as follows: AR = 0.45 (0.49), SI = 1.54 (0.60), SM = 0.77 (0.53). Change in Alternatives was only significantly different from zero for the SI group (*t* = 2.55, *p* = 0.013). Moreover, pairwise comparisons of the Alternatives change scores indicated no significant differences between any of the groups.

Comparison across strategy type To address the possibility that SI and AR participants differentially changed across strategy types, additional within-subjects analyses were conducted. All strategy scores were standardized to allow comparison across strategy type, and change scores were compared across strategy type, within each experimental group separately. No significant differences in magnitude of change were found across the three strategy types for either the SI group (*F*[2, 177] = 0.75, *p* = 0.47) or the AR group (*F*[2, 177] = 0.47, *p* = 0.46).

Relationship between Change in Strategy Use and Change in Alcohol Use

The three strategy use change scores were correlated with the three alcohol use change scores and no significant relationships emerged (*r* ranged from .02 – -.13). We also looked at these relationships separately for the two instruction groups. The correlations between the three strategy use variables and the three alcohol use variables were non-significant for both instruction groups (AR: *r* ranged from .01 – -.17; SI: *r* ranged from .01 – -.16).