couplebeige.gif (19185 bytes)

 

 

 

 

Causal Analysis1

custdiv3.gif (1619 bytes)

Most research is directed to understanding causal relationships. Consider the following propositions: Conflict in marriage erodes marital satisfaction; dissatisfaction with the sexual aspects of the marriage foretells more general disenchantment with the spouse; similarity between newlyweds in gender-role attitudes and leisure interests accounts, in part, for their later marital satisfaction. These statements are of the "if-then" category, suggesting a causal linkage between variables. Since most of you will create hypotheses of a causal nature, you will need to consider a number of issues pertaining to making inferences about causality.

The logic of causal analysis and the problems involved in establishing causal linkages are discussed at length by Cook and Campbell (1979). These authors, following the lead of epistemologists such as John Stuart Mill, identify three key criteria for inferring a cause and effect relationship: (a) covariation between the presumed cause(s) and effect(s); (b) temporal precedence of the cause(s); and (c) exclusion of alternative explanations for cause-effect linkages. These criteria are well known among social scientists, and proscriptions like "correlation does not imply causation" and "the cause must precede its presumed effect" are commonly   heard. Nonetheless, it is the rare investigation of marriage relationships that seriously attempts to satisfy all three of the above requirements for causal inference2.

The resources and care necessary to carry out research of a causal nature on relationships make such work extremely difficult. First, either a longitudinal research design, an experimental method, or prior knowledge that a presumed cause antedates its presumed effect is required for conclusions to be reached regarding causation. Even so, demonstrating that A causes B does not preclude the possibility that B also causes A - as shown, for example, in the reciprocal causal relationship between the extent to which spouses exchange criticism and their marital dissatisfaction, or between one partner's self-esteem and the tolerance of power in their marriage. Thus, the importance of longitudinal design like that used in the PAIR Project is particularly important when a researcher wishes to assess the causal links between two variables that may have mutual influence.

Second, the causes of both "relationship properties" (such as conflict) and "subjective conditions" (such as feelings of satisfaction and love) frequently depend upon  simultaneous, or temporally sequenced, operation of other causes. Suppose, for example, that one is interested in the impact of differences in gender-role attitudes on conflict between noncohabiting premarital partners as compared to spouses. Such differences may have a less strong effect on premarital conflict than on marital conflict because courting couples may be less inclined to confront differences openly; in addition, such differences may become more evident and important after partners set up a joint household. If this reasoning is correct, differences in gender-role attitudes would be only weakly related to conflict between noncohabiting premarital partners but strongly related to conflict between married pairs. Thus, the operation of a particular cause (dissimilarity in gender-role attitudes) depends upon the presence of other conditions, such as the partners' willingness to express their dissatisfaction and the relevance of dissimilarity to their activity patterns.

A third problem is tied to the fact that studies of close relationships tend to be nonexperimental. Researchers have examined primarily the covariation of presumed causes and effects as these are found "in nature." Because the presumed causes are not manipulated experimentally, it is extremely difficult to rule out plausible alternative explanations of the linkages found between the variables. When individuals   or groups differing in the level of the presumed cause are compared, they inevitably differ in ways other than in regard to the causal variable of interest. Thus, for example, if you are interested in establishing the linkage between similarity in religious background and level of marital conflict, you need to recognize that couples who are similar in religion are also apt to be similar on an assortment of other characteristics. Some of these "extraneous" differences may contribute to the strength of the association between similarity in religion and level of conflict. Statistical procedures that can completely control for this problem do not exist, and designs that disentangle competing causes are often difficult to implement.

To tie the consideration elucidated above together, suppose we carry out an investigation in which we assess the gender-role attitudes of men and women prior to their first meeting. After some of these people marry each other, we examine the association between the degree of similarity of their gender-role attitudes and the level of overt marital conflict. Such a study could demonstrate covariation between the presumed cause and effect and could establish the temporal precedence of the putative cause. If the partners' similarity remained constant over time while their amount of conflict changed, there would be no concern that conflict somehow causes the dissimilarity. It still would be very difficult to rule out competing causal explanation, however, for the amount of conflict. Gender-role attitudes presumably vary with the individuals' membership in various subpopulations (e.g., social class, location of residence) and the socialization experiences contributing to these attitudes may systematically affect other individual characteristics relevant to conflict. thus, if partners have similar gender-role attitudes, they are likely to be similar in many other ways that contribute to marital harmony; the opposite may occur for those who differ in gender-role attitudes. For example, gender-role attitudes may covary with style of conflict resolution so that partners who differ in attitudes may also differ in the way they handle disagreements. Thus, a conflict might remain unresolved because of attitudes and behavior that are not caused by, but rather covary with, gender-role attitudes. Alternatively, people who choose a partner with dissimilar attitudes in a fundamental area may be generally insensitive or conflict-prone. To rule out such alternative explanations of the effect, it would be necessary for each possible cause to be validly and reliably measured so that its status as a causal candidate could be assessed.

We have not attempted to discuss here the wide range of threats to valid causal inference (the reader is referred to Cook & Campbell, 1979 for this), but rather to suggest the logic underlying causal inference. It is rare for research on close relationship to approach, in a systematic fashion, the problem of ruling out plausible alternative explanations of the observed effects. Far more commonly, researchers merely demonstrate association and speculate about causation. This is perhaps inevitable, given the complex ways in which causes are intertwined and act upon features of close relationships. The best social scientists can hope for are probabilistic statements about causes, with most of these statements being highly circumscribed. Nonetheless, a science of close relationship requires that researchers incorporate the search for causal linkages into their designs. In the next section, we'll discuss how to begin that process.

backbut3.gif (1373 bytes)tocbut2.gif (2101 bytes)nextbut3.gif (1377 bytes)

custdiv3.gif (1619 bytes)

1. Adapted from T.L. Huston and E. Robins (1982), Conceptual and methodological issues in studying close relationships, Journal of Marriage and the Family, 44, 901-925.

2. Most conceptions of causality assume a temporal ordering between cause and effect. Systems conceptions of causality do not differ from more positivistic views with regard to matters of temporal sequencing. As Kelly and McGrath (1988) note:

The system position does not permit time to run backwards any more than does the standard positivistic causal position. It simply downplays the effects of simple causal chains and simple causal orderings, insisting not so much on an atemporal causation as on a causation that is multivariate (A, B, and C together affect D) and multidirectional (A may affect B ay one moment, whereas B may affect A at another). The key feature of this systems view is the causal interdependence among multiple systemic forces, all of which act mutually and more or less simultaneously on each other, rather than acting unidirectionally and in turn.

 

custom1.gif (1802 bytes)

This page created & maintained by Shanna Smith, ella@utxsvs.cc.utexas.edu