Identifying the Missing Pieces in the Study of Families

The study of families has largely focused on mothers and children despite assertions that more research on fathers is needed (Phares, 1992). One explanation is that mothers have traditionally performed the majority of care-giving duties and therefore may influence child outcomes to a greater degree than fathers. Moreover, some scientists believe that mothers are easier to recruit for participation in research than fathers. Many authors have criticized such explanations with theoretical and empirical evidence suggesting that fathers make important contributions to child outcomes and may not be more difficult to recruit than mothers (e.g., Phares, Fields, Kamboukos, & Lopez, 2005).

In recent years, a few researchers have altered their methodology by using samples of mothers and fathers within the same family. Such research has typically treated maternal and paternal characteristics as separate, or statistically unrelated, predictors of child outcomes. However, in light of evidence that characteristics of parents may be interdependent (Barnett, Deng, Willoughby, & Cox, 2008), there are a number of problems with treating maternal and paternal characteristics as separate predictors of child outcomes within families. Understanding these problems should encourage graduate students to recognize the shortcomings of traditional statistical techniques and apply more advanced techniques to their research when necessary.

Problems With Modeling Maternal and Paternal Characteristics Separately

Given that mothers and fathers from the same family are typically romantic partners who are embedded within the same familial environment, researchers cannot assume that data collected from these mothers and fathers are independent or unrelated. Unfortunately, traditional statistical techniques, such as analysis of variance (ANOVA) and linear regression, assume that the data are independent. Testing nonindependent data as independent also biases p values (Kenny, Kashy, & Cook, 2006), resulting in potential ambiguity in the interpretation of the findings. Researchers should seek alternative statistical techniques when working with samples of mothers and fathers within families to avoid making false assumptions and generating biased statistics. In addition, using alternative statistical techniques can allow for an assessment of how maternal characteristics and paternal characteristics may be interrelated — a “missing piece” in the study of families.

Alternative Statistical Techniques

A variety of statistical techniques are well suited to evaluating the relationships between maternal and paternal characteristics (or any nonindependent data more generally). In their review, Kenny et al. (2006) highlight data-organization techniques and research designs (e.g., actor-partner interdependence models, social-relations models, etc.) that target appropriate modeling of nonindependent or dyadic data. Such techniques use the dyad, rather than the individual, as the unit of analysis by combining characteristics of two individuals into one testable unit. For example, in a sample of 200 parents in which mothers and fathers within the same family participate, the testable sample size would be 100 because characteristics of mothers and fathers within the same family would be combined into one testable unit. It is important to note, however, that depending on the sample size and strength of the effects, the power to detect significant effects tends not to be reduced when using the dyad rather than the individual as the unit of analysis (Kenny, Cashy, & Cook, 2006).

Utilizing the models highlighted by Kenny et al. (2006) makes it possible to evaluate the interdependence of maternal and paternal characteristics — for instance, how maternal and paternal negative affect are associated with both maternal and paternal harsh/negative parenting behavior and, in turn, child behavior problems (see Figure 1). In the model highlighted in Figure 1, associations between the reported negative affect of one dyad member and the harsh/negative parenting behavior of the same dyad member are referred to as “actor effects”; associations between the reported negative affect of one dyad member and the harsh/negative parenting behavior of the other dyad member are referred to as “partner effects.” Additionally, in Figure 1, maternal and paternal negative affect are independent variables (IVs), whereas harsh/negative parenting behavior and child behavior problems are dependent variables (DVs). Direct effects of maternal and paternal negative affect on child behavior can also be included, but to increase clarity, they were not included here.

Figure 1. An actor-partner interdependence model of maternal and paternal negative affect and parenting behavior, and child outcomes. NA=negative affect; H/N=harsh/negative; e_m=residual for maternal H/N parenting; e_p= residual for paternal H/N parenting; e_c=residual for child behavior.

Models of dyadic data, such as the model presented in Figure 1, are typically evaluated using Structural Equation Modeling (SEM) or Multilevel Modeling (MLM) software. SEM and MLM offer an advantage over traditional regression techniques in that they allow for simultaneous evaluation of multiple equations within one model. When testing the associations in Figure 1, traditional techniques would treat the association between maternal negative affect and maternal harsh/negative parenting behavior as statistically separate from the association between paternal negative affect and maternal harsh/negative parenting behavior, even though such equations may be interrelated. SEM and MLM model both equations simultaneously, resulting in findings that may reflect “real-world” phenomena more accurately. For example, correlational analyses may show that both maternal and paternal harsh/negative parenting behavior are significantly associated with child behavior problems. However, modeling simultaneously may suggest that the harsh/negative parenting behavior of mothers is significantly associated with child behavior problems, whereas paternal harsh/negative parenting behavior is not. Each of these hypothetical findings would result in vastly different interpretations of the data.

Learn to Embrace ‘Complicated’ Statistics During Graduate Training

Graduate students may shy away from learning to use statistical techniques that are not required by their department because they think the techniques are too complicated for them to master. This assumption is an avoidance strategy associated with a fear of failure. As Lisa Diamond indicated in a previous Student Notebook article (Vaughn-Blount, 2008), graduate school is a time to take risks, make mistakes, and, most importantly, learn from mistakes by asking for guidance. Graduate students should not be afraid to learn advanced statistical techniques (e.g., SEM) and ask for help when needed. In this manner, students can build the statistical skills necessary for becoming successful researchers. Further, increased knowledge of various statistical techniques may also enhance students’ theory-building skills because alternative statistical techniques provide new ways to conceptualize and analyze data.

Conclusion

Since the relationships among family members’ characteristics are complex, scientists who study family dynamics should employ statistical strategies that differ from traditional methods (e.g., SEM, MLM). Although the focus of the current article is on familial relationships, the information discussed is also applicable to the study of other nonindependent data (e.g., workplace environment). It is important that graduate students recognize the benefits of learning and utilizing advanced statistical techniques in their own research. A solid understanding of statistical techniques can help graduate students from all fields of psychology conduct more accurate and useful research.