Dyadic Data Analysis

Violations of Independence

Observations or units are sometimes connected, and those connections often come in pairs. Typically, statistics instructors stress the importance of collecting independent data points when constructing statistical tests. Independence makes statistical estimation and inference more tractable, but it may not be realistic in settings involving dyads and groups. Even in basic statistics, there are at least three common violations of the independence assumption in social science studies, two of which are generally well understood. Two data points can be related (i.e., not independent) because they are taken from the same person on the same variable at two points in time. Repeated-measures analyses, time series, and latent growth curve models handle violations of independence due to time. Two data points can be related because they are collected on two different variables from the same person. Correlations, multiple regression, factor analysis, and structural equation modeling are all well-known methods for handling violations of independence due to multiple variables from the same person.

What characterizes these two well-known methods for modeling violations of independence is an appreciation that the violation of independence can illuminate the underlying mechanism. Time series analyses can illuminate how participants grow or change at different rates; correlations between variables can illuminate the underlying factor structure of multiple variables.

The third type of violation of the independence assumption is less well known. Two data points can be related because they are taken from two individuals who are related in some way and can be considered members of the same dyad. For example, the researcher may ask both the husband and the wife how satisfied each is with their marriage. Those two data points are not conceptually independent from each other and so are probably not statistically independent. If the marriage is devoid of intimacy, both partners are likely to report low satisfaction. If the marriage is awash in intimacy, both partners are likely to report high marital satisfaction. Cases where one partner is satisfied and the other not are also interesting. The point is that because both individuals share a common dyadic experience—the marriage—their evaluations of [Page 472]that experience may be related. Each person serves both as a member of the dyad, or the context, as well as an evaluator of the dyad. If these two individuals are called rater 1 and rater 2, it is easy to see that the two raters can be correlated because they are judging the same object.

...