Causation in Sociology Today

A Revised View

Bibliographic Data

ID2517470
AuthorsDavid M Klein (0000-0002-1956-3984, corresponding author)
Year1987
Volume5
Issue1
Pages19
Publication date1987-01-01
Peer ReviewedYes
Open AccessNo
TypeARTICLE
VenueSociological Theory (JOURNAL)
Journal identifiersISSN: 0735-2751 • E-ISSN: 1467-9558
PublisherJSTOR (PUBLISHER)
DOI10.2307/201989
OpenAlexW2331947225
LanguageEN
Citations received5
References cited12

statements of a theory, whether they imply causality or not, are never subject to empirical demonstration in the form of an airtight proof. Hence, Gibbs asks too much of causally structured theories, and his own scheme of constructs, concepts, and referentials leaves much room open for questions of empirical relevance. Nevertheless, we do want clear, reasonable, and potentially agreeable criteria for making causal inferences. This issue is addressed below. Criteria For Causality Several attempts have been made over the years to identify the form which sociological data must take to warrant causal inferences about them (e.g., Blalock, 1961; Hirschi and Selvin, 1967; Selltiz, et al., 1959; Stinchcombe, 1968). These efforts focus on three criteria, commonly known as constant conjunction, temporal ordering, and lack of spuriousness. None of the three is unproblematic, so they deserve reexamination and clarification. In addition, other criteria will be proposed, criteria which appear to be consistent with claims sometimes expressed, implied, or debated. It is important to remember that the inferences based on meeting these criteria only add to the plausibility of causal relationships asserted in theories. They provide no ironclad guarantee because causation is a metatheoretical construct, and no amount of empirical data ever proves beyond a doubt the adequacy of a theory, whether it is causally formulated or not. 1. The concepts must be variables (principle of variation). Sometimes one encounters the statement that norms, meanings, social stratification, and the like cause social behavior or some other phenomenon. Such statements are, technically speaking, misuses of the notion of causation, because constants cannot have effects. Similarly, it is technically incorrect to assert that constants have causes. Substantive concepts such as norms may vary, of course, but only if conceptualized in terms of some property which varies. If the statement that norms have causes or effects can be interpreted to mean that the existence of norms is a variable property, so that sometimes they exist and sometimes they do not, then normativity is a variable which can enter into causal statements. Ordinarily, however, a causal theorist would be concerned with other properties associated with norms, such as the strength of their enforcement, their importance, or the degree of consensus about them. The same reasoning applies to any substantive concept in the sociologist's arsenal. The principle of variation also rules out mere existence statements or descriptions from causal assertions. To claim that all social organizations have systems of stratification, for example, may be true or false. It may even be one of a set of existence statements in a theory, and hypotheses 22 This content downloaded from 40.77.167.45 on Fri, 27 May 2016 06:34:27 UTC All use subject to http://about.jstor.org/terms CAUSE, LAW AND PROBABILITY may be devised to test for the degree of support for the claim. But, the claim by itself makes no causal assertion. Only when two or more existence statements are said to be linked by a causal relationship can it be argued that the principle of variation has been followed. The existence of X is then linked causally to the existence of Y. 2. The variables should vary over time (principle of alterability). Social phenomena may vary in the sense that different social units display different values of a variable at the same point in time. In such cases it could be argued that the postulated differences have a causal effect, at least when those differences can be argued to result in differences in another variable at a later time. But, such reasoning is not in accord with the experimental ideal normally held up as the model of causation. Instead, the more restricted position taken here is that causation implies that changes in one variable are responsible for changes in another. We ordinarily want to know what happens when a variable changes or what produces that change. If variables do not change in value but merely differ from one social setting to another, we can perhaps accept this as a weak condition for the demonstration of causality, but surely evidence based on changes over time is to be preferred and is a stronger condition. One reason for preferring that theories and empirical evidence speak to temporal variation in substantive concepts is the widespread belief in human agency. Whether or not social units can produce the effects they intend to produce is an important issue. Even when they cannot create desired effects but the effects happen due to as yet uncontrollable forces, we ordinarily suppose that knowledge about this permits us to respond in useful ways, whether by adjustment of social behavior to the situation or by fatalistic resignation. An even stricter requirement than temporal variation might be applied. It could be demanded that the variables are capable of and demonstrated to both increase and decrease over time. Hence, for example, evidence about the assertion that dual-earner marriage causes divorce might require the observation of declines in the divorce rate when wives leave the paid labor force. The principle of alterability should be understood as subject to a sliding scale of judgment. Differences between social units on variables measured at different times represent the weakest demonstration of causality. Changes in one direction of X shown to result in changes in one direction of Y is an intermediate demonstration. Finally, changes in both directions of X shown to result in changes in both directions of Y constitutes the strongest demonstration. 3. The relationship between variables must approach a constant conjunction (principle of orderliness). This is one of the standard criteria for causality and one to which Gibbs objects. It is indeed difficult to establish firm decision rules about the stability of convariances. This is so both within specific studies as well as across a series of studies dealing with essentially the same causal assertion. Any such rule will be arbitrary. It does not seem advisable, however, to expect or demand that conjunctions between variables be absolutely constant. At least it is possible to apply statistical tests to determine the degree to which covariance stability is a matter of chance. 4. The relationship must be stronger specified conditions (principle of veracity). As Gibbs rightly notes, the observation of perfect correlations is virtually unknown to sociologists. Even correlations much beyond .50 often raise the suspicion that the variables are really measures of the same underlying concept, so that assertions of causality in such cases would be artifactual. Causal modelers may even have a tendency to claim a causal relationship when partial correlation or regression coefficients are quite small, as long as they are statistically significant by accepted standards. The key here, however, is the phrase under specified conditions. While sociologists may quite often prefer universally true assertions or laws unfettered by a host of qualifications, especially in their theories, research is hardly ever able to provide such elegant certainty. Blalock and the other causal modelers are, therefore, providing an important service when they remind us that our theories should be explicit about their assumptions. When assumptions about the conditions which theories are supposed to hold are not specified, it becomes difficult or impossible to interpret the meaning of empirical evidence offered. In a similar vein, Cohen and his colleagues (Cohen, 1980; Walker and Cohen, 1981) have recently emphasized the importance of including scope conditions in all sociological theories. The hope shared by causal modelers and DN theorists like Cohen is that strong relationships are more likely to be observed when the social circumstances which they are expected to hold are included in our theories. These remarks do not settle in any final sense the issue of how strong a relationship needs to be to qualify for treatment as a causal relationship. Again, a sliding scale of judgment seems most practical. The larger the statistical association between two variables, the stronger the evidence for a causal relationship between the two. Statistical significance or the number of observations upon which the relationship is based must also be taken into consideration, but 23 This content downloaded from 40.77.167.45 on Fri, 27 May 2016 06:34:27 UTC All use subject to http://about.jstor.org/terms

Causation · Epistemology · Historical sociology · Social science · Sociology · Sociology of knowledge · Philosophy · Qualitative Comparative Analysis Research