A generalizability approach to assessing the effects of ethnicity and training on observer ratings of family interactions

This exploratory investigation assessed the utility of generalizability analyses for investigating race of coder and race of family member ('target') as contributors to bias in observer ratings of family interaction processes. Thirty behavioral scales were rated on three occasions during an initial 5-week (100-hour) training period. African-American and European-American coders observed videotaped interactions occurring in one African-American and one European-American parent-child dyad. For each scale, levels of rater bias and rater agreement were examined over time. Although most scales showed decreasing levels of bias with training (as expected), some did not. For scales showing a main effect for coder race, European-American coders rated targets more favorably than did African-American coders. For scales susceptible to coder race by target race interactions, coders tended to favor other-race rather than same-race targets. Suggestions for applying the generalizability approach in future family interaction research are presented.