Tests for the comparability of a causal model of depression under two conditions of interviewing

Although many, if not most, surveys have an explanatory purpose, comparisons of the results obtained through telephone and in-person interviews have generally been focused on descriptive outcomes only. We assessed interview method effects on parameter estimates for a latent variable causal model of age, sex, socioeconomic status, illness, and depression. The data were derived from interviews with a community sample of adults randomly assigned to either telephone or in-person interviews. We found that the measurement model for five constructs was highly similar for the two groups. Differences appeared in the estimation of causal paths among these constructs, but were not statistically significant. In both groups, the largest single effect was the impact of illness on depression. A comparison of factor means also revealed only minor method effects. In general, the two interview methods appeared to yield highly comparable estimates of the causal model, which suggests that relations among variables are stable across methods of interviewing.