Dyadic Data Analysis

Nonindependence

The concept of nonindependence is the most fundamental to dyadic analysis. Nonindependence occurs when the scores of two dyad members are statistically related. In heterosexual couples, for example, husbands and wives are typically similar in many respects, including education, attitudes, and personality characteristics. Nonindependence can be positive, reflecting similarity between dyad members, or negative, reflecting dissimilarity. Negative nonindependence occurs less frequently but can be expected if there is compensation within dyads (e.g., the more I do, the less my partner needs to do) or competition between members (e.g., the happier I am with how I performed, the less happy my partner is with his or her performance).

Nonindependence can be assessed when members are distinguishable or indistinguishable. When members are distinguishable and the variable measured in both members is continuous, the Pearson correlation between the members’ scores is often used to measure nonindependence. When members are indistinguishable, nonindependence can be assessed by the intraclass correlation coefficient, which can be calculated using one-way analysis of variance or multilevel modeling (MLM), or the pairwise correlation, which requires a data structure known as pairwise or double-entry structure.

Dyadic Models

The dyadic data are analyzed using specific techniques because the group size is only two. In the last 3 decades, a wide range of models have been developed to study dyads, with two models dominating the field: the actor–partner interdependence model (APIM) and the dyadic growth curve model. Importantly, most dyadic models require that the same set of variables is measured in both members.

Figure 1 displays the APIM for two variables, X and Y, both measured in member A and member B, which might be caregiver and patient. This model allows researchers to predict a person’s [Page 554]outcome by the person’s own predictor and the partner’s predictor. The path from the person’s predictor X to that same person’s outcome Y is called the actor effect and the path from the partner’s predictor X to the person’s outcome Y is called the partner effect. With distinguishable members, there are two actor effects and two partner effects, one for each type of dyad member. With indistinguishable members, there is only one actor effect and one partner effect.

...