After reviewing and coding each study, it became clear that a variety of both dependent variables and metrics were reported across the studies. To make the outcome variables comparable while retaining the greatest number of studies in our database, we used the point-biserial correlation as our effect size given that most of the other reported measures could be converted to it (see Glass et al. 1981; Hunter et al. 1982). From this we retained a total of 132 observations from 42 of the 51 studies. Our inclusion rate is generally consistent with previous meta-analyses (see e.g., Compeau and Grewal 1999; Krishna et al. 2002). One of the principal investigators and a graduate student independently coded the data in each article. Coding consistency between the two was 94%. The few discrepancies were rectified through discussion and subsequent re-coding.