Data Analyses
Data analyses were performed with SPSS software† and an alpha level of .05 on an intention-to-treat basis. The last observation carried forward was used to impute data missing at reassessment; a score of 3 (“unchanged”) was allocated for missing global change measures. Data were checked for normality and homogeneity of variance prior to analyses. Baseline comparability between groups was determined with independent t tests (because most data were normally distributed) or chi-square tests. Mean scores at 6 weeks were compared between groups by use of univariate analysis of variance; baseline scores were included as covariates to control
for any group differences at baseline. Effect sizes were calculated; effect
sizes of .2 were regarded as small, those of .5 were regarded as medium, and those of .8 were regarded as large. A comparison of the numbers of participants showing improvement between groups was made with chi-square tests, and odds ratios (OR) (with 95% confidence intervals [CI])
were calculated. Outcomes at 6 and 12 weeks in the intervention group were compared by use of paired t tests.