Using GPOWER, an a priori sample size of 82 was determined to provide 80%
power to detect medium effects of .30 using two-tailed tests to calculate Pearson’s
correlations and .15 to build regression models with up to three predictors at an
alpha of .05 (Faul & Erdfelder, 1992). Cases with more than two missing items per
scale were not included in the analysis. All composites were allowed to have one or
two missing items per case because the items were so similar as to be redundant,
consistent with the recommendation of Moss-Morris et al. (2002).