H as a measure of complexity of social information processing

Many studies have used H (a measure of unpredictability derived from information theory) to quantify the complexity of descriptions of persons across multiple roles. Interpreting these studies is problematic, though, because H confounds unpredictability across roles (which is typically the construct of interest) and unpredictability within roles (which is simply a function of the proportion of traits endorsed). The need to control for unpredictability within roles was highlighted by 3 demonstration studies in which controlling for unpredictability within roles eliminated relations between well-being and H. I also show how, controlling for unpredictability due to the number of traits endorsed and number of roles described, H provides a unique measure of role dependence and independence. However, H does not measure the type of role overlaps that would predict "spillover effects" between roles; therefore, I recommend an alternative index of role similarity for future research on spillover effects.