Commentary: Dyadic analyses of family data.

It is an honor and privilege to be asked to comment on the 12 papers in this special issue of the Journal of Pediatric Psychology on Family Assessment. In most of the papers in this issue, the family member who was assessed was a parent whose child had been diagnosed with a serious chronic condition. It should be clear that when a family member is assessed that measurement reflects not only the respondent but also reflects the other family members, the respondent’s relationship to the other family members, and the whole family. Consider the study by Knafl et al. (2009) in which they asked 579 parents, 414 mothers, and 165 fathers (father is used throughout this paper, even though in some cases the father may not be the biological father of the child), of a child with a chronic medical condition to complete a 53-item Family Management measure which measured six dimensions: Child’s Daily Life, Condition Management Ability, Condition Management Effort, Family Life Difficulty, Parental Mutuality, and View of Condition Impact. If we consider the Parental Mutuality dimension (e.g., ‘‘I am pleased with how my partner and I work together to manage our child’s condition’’), this scale reflects on the respondent, the respondent’s partner, the child, the respondent–partner relationship, respondent– child relationship, the partner–child relationship, and the family. My own area of expertise is not in the assessment of families per se, but in the analysis of data when measures are used to study families. I have primarily focused on the fundamental social unit in the study of families and that unit is the dyad. Almost all of what I know about dyadic data analysis is contained in the book Dyadic Data Analysis (Kenny, Kashy, & Cook, 2006). I believe that research in this area would benefit by thinking about the data in terms of dyads and in this short note, I outline some possibilities. As I hope to show, by using dyadic analysis, one can learn more about what is occurring in families. Before I begin my discussion of dyadic data analysis, I wish to make one suggestion about assessment. Many of the articles in this issue assess multiple dimensions. For instance, I mentioned earlier that Knafl et al. (2009) assessed six dimensions of Family Management. An important, but sometimes forgotten, issue in assessment is discriminant validity. The question of discriminant validity is whether the two scales correlate too highly, making them not two distinct constructs but just one. The complicated way to assess discriminant validity is to perform a confirmatory factor analysis and determine whether the latent variables correlate too highly (i.e., greater than .85). An alternative and simpler way to do so is as follows: First, correlate the two scales. Second, divide that correlation by the square root of the product of the two scales’ reliabilities. (Normally, the reliability coefficient used would be a Cronbach alpha coefficient.) Third, make sure that the corrected correlation is not too large, e.g., less than .85. Because, most of the studies in this issues conducted factor analyses on the entire set of items (not just the items separately for each scale) most of scales discussed in this paper have good discriminant validity. For instance, Palmer et al. (2010) conducted separate confirmatory factor analyses for three scales for mothers and fathers and found decent convergent validity; however, one correlation was relatively large, being .788. Researchers are generally aware today of issues of nonindependence in family data. It can be problematic to analyze family data with individual as unit because significance testing results will likely be wrong because the two dyad members’ responses are dependent. Violation of the independence assumption can sometimes lead to too liberal a significance test (which is well recognized), but it can sometimes lead to too conservative a test, a test that is under-powered. One strategy to handle nonindependence (e.g., employed by Benzies et al., 2010; Berlin, Davies,