We took a facet rather than a domain measurement approach
because we believe facets are more proximal to observable and
malleable behavior, and are valid predictors of educational and
work success (e.g., Lounsbury et al., 2003; Paunonen & Ashton,
2001; Peterson et al., 2006). One drawback to this approach is that
our statistical model for early high school GPA used many individual
predictor variables (26), which increases the probability of
significant findings that are simply due to chance. Related to this
issue, we observed two regression coefficients in our regression
model that were statistically significant, but in an unexpected
negative direction. Future research should explore the relative
predictive effects of PSFs along different levels of a hierarchy
(ranging from behaviors to broad domains) to better understand
developmental differences and key moderators of educational
achievement, persistence, and academic risk.