Multiple Perspectives on Family Differentiation: Analyses by Multitrait Multimethod Matrix and Triadic Social Relations Models

Abstract

This article illustrates two statistical analysis procedures that provide options for handling multiple perspectives on same variable. We used Differentiation in Family System Scale to assess boundary maintenance in from three members' perspectives. The first procedure, multitrait multimethod matrix with correlated uniquenesses, suggested that best model was one in which six dyadic subscales formed distinct factors using three perspectives. The second procedure, triadic social relations model, provided further evidence of agreement. Mothers' views tended to be most shared, whereas fathers' were least shared. As one person exhibited differentiated behavior, other also tended to elicit same behavior. Key Words: data analysis, differentiation, multiple perspectives, multitrait multimethod matrix. This article illustrates two statistical analysis procedures that allow for multiple perspectives of members on same variable. Recently, Bartle-Haring and Gavazzi (1996) demonstrated utility of using confirmatory factor analysis with multiple-perspective data for family-level variables. The objective of our investigation is to extend this line of work by again using a confirmatory factor analysis with triadic-level data (three members of same commenting on how members treat self, how self treats others, and how other members treat other members) rather than dyadic-level data (two or more members of a commenting on how members treat self and how self treats others). We demonstrate a second analysis technique, triadic social relations modeling. Both techniques can incorporate multiple perspectives on same construct and measure how much raters agree about particular construct of interest while controlling for rater bias. In all too many studies, information we have on families is from one person's perspective only. Although information from one informant is valuable, such a perspective may be idiosyncratic view of one individual, rather than consensual reflection of what a particular is. With a single informant, researcher cannot assess extent of agreement between members or extent to which they share a reality. Teachman, Carver, and Day (1995) and Hauser (1988) have suggested models that include paired data that partition variance into individual and common factors with sibling data. Others have demonstrated that three members' perspectives on variables can also be used and that it provides useful information (Cole & Jordan, 1991; Martin & Cole, 1993). Having information from more than one person in a seems essential if researcher is to understand the family as a family, rather than as just one person's perspective of that system (Sabatelli & Bartle, 1995). Before delving into details of these statistical methods, we need to understand importance of theory in applying them. The rationale for using these statistical techniques must come from theoretical conceptualization of constructs of interest. If construct of interest is not defined as a system-level property or at least as a relational property, these techniques may be irrelevant. Thus, to demonstrate utility of these techniques, we outline theoretical conceptualization of construct of interest, in this case differentiation. THEORETICAL CONCEPTUALIZATION OF DIFFERENTIATION Family System Differentiation As a marker of distance regulation in systems, differentiation construct is considered an important component of functioning. Bowen (1976, 1978) defines differentiation at individual level and family-system level. At a system level, differentiation manifests itself as family's tolerance for individuality or difference and its tolerance for intimacy or belongingness (Anderson & Sabatelli, 1990). …