Analytic strategies for data with nonindependent observations : data transformation, structural equation modeling, and information-theoretic criteria

In studying the family, researchers tend to avoid using measures from multiple family members, as doing so violates the fundamental assumption of interdependence between observations. When multiple measures are used, the analysis is typically comparative, such as comparing the marital satisfaction of husbands and wives, comparing academic achievement between siblings. While these approaches may be quite appropriate for particular research questions, it may also be the case that family researchers tend to fashion research questions to suit the analytic techniques available to them. This dissertation presents alternative analytic techniques to analyze data in the presence of observation interdependence. Three primary issues are addressed. First, the effect of observation interdependence on analytic results is explored, by simulating data with known parameters and increasing levels of observation interdependence at varying sample sizes. The data is analyzed and tabulated to reflect trends which appear in the estimates of regression parameters, sample variance, and the accuracy with which the correct model is produced in regression analysis. Second, five alternative techniques are tested using both the analysis of simulated data and real data from the National Survey ofChildren, to determine their effectiveness in regaining the analytic power lost in the presence of observation interdependence. Third, the author examines whether substantive results change with the use of different sampling strategies. A study conducted by Harrisand Manner (1996), using NSC data, is replicated and the results are compared when using multiple measures per family (with correction for observation interdependence),measures from the oldest child, from the youngest child, and from one target child chosen randomly. The author argues the importance of using multiple measures per family while correcting for the analytic impact of observation interdependence.