Using Qualitative Comparative Analysis to Study Causal Complexity

Health Services Research, Dec, 1999 by Charles C. Ragin

In case-oriented research, investigators study the ways in which the different aspects of cases fit together Within each case, and they make sense of each case separately (Ragin, earlier in this issue). Although this approach is rich in detail, it highlights individual cases at the expense of knowledge about cross-case patterns. This strategy differs from variable-oriented research, where the key concern is for patterns observable across cases, not the specificity of individual cases. In this article, I bring these two kinds of concerns together. The key to bridging them is a "configurational" view of cases. In this view cases are understood in terms of the aspects they combine, as different configurations of set memberships. Rather than viewing cross-case patterns through the lens of "relationships between variables," the researcher compares and contrasts configurations. In my presentation of this approach, I emphasize the study of the different combinations of causal conditions that are sufficient for a sel ected outcome. However, the configurational approach is not limited to the study of causation or causal complexity. It is relevant to any investigation where the number of cases is small enough to permit some degree of familiarity with each case, yet large enough to warrant an interest in cross-case patterns.

First, I explain my approach to causal complexity. This discussion builds on the contrast between studies that focus exclusively on cases displaying a specific outcome and a search for common antecedent conditions, on the one hand, and studies that allow for the possibility that the same outcome can follow from different combinations of conditions, on the other. Next, I present Qualitative Comparative Analysis (QCA) (Drass and Ragin 1992), an analytic technique designed specifically for the study of cases as configurations of aspects, conceived as combinations of set memberships. I illustrate QCA with an analysis of hypothetical data on health maintenance organizations (HMOs). In this analysis, the problem is to identify the different combinations of conditions linked to high rates of staff turnover. Finally, I sketch general issues in the use of QCA.

CAUSAL COMPLEXITY AND CASE-ORIENTED RESEARCH

A common strategy in comparative case-oriented research is to study a small to moderate number of cases in which a specific outcome has occurred. Usually, this design involves a search for antecedent conditions shared by all (or virtually all) instances of the outcome, with an eye to understanding how these conditions fit together to produce the outcome. Cross-case commonalities identified by the investigator provide the basis for constructing a general account of how the outcome comes about. For example, a researcher might study several HMOs that experienced a very high turnover of physicians due to defection to other forms of practice. Common antecedent conditions exhibited by these HMOs, which might include "speed-up" in the patient flow, increased management oversight of referrals to medical specialists, and so on, would contribute to a general understanding of the forces generating the outcome.

Although simple and straightforward, this case-oriented research design is far from problem-free. The most obvious problem is that the investigator's confidence in the causal conditions that he or she has identified increases as the number of instances of the outcome increases. The greater the number of cases examined (e.g., instances of HMOs with defection-related turnover), the more impressive the fact that they share common antecedent conditions (e.g., speed-up, increased oversight, and so on). But as the number of cases increases, so does the difficulty of knowing cases well, making it impossible to become familiar enough with each case to make sound judgments about causally relevant features. Besides, as the number of cases increases, the likelihood that they will share causally relevant features declines. "More cases" almost always means "more heterogeneity."

Another problem with this design is the fact that it is useful only for identifying necessary conditions (Dion 1998). When selecting on instances of an outcome, it is not possible to assess the sufficiency of causal conditions, at least not in any direct or systematic manner. To assess sufficiency, the researcher must select on instances of the causal condition (or on instances of the relevant combination of causal conditions), not on instances of the outcome. A cause is sufficient if it is invariably (or almost invariably) followed by the outcome; it is necessary if it is present in all instances of the outcome. (See Ragin in this issue; other common problems with the study of the antecedent conditions shared by instances of an outcome are addressed in Corner 1995; Collier and Mahoney 1996.)

These common problems are not the primary concern of this article. Rather, the focus is on the problem of causal complexity: the fact that many of the outcomes that interest social scientists often result from several different, non-overlapping combinations of conditions. For example, several different combinations of conditions may spark "defection-related turnover" in HMOs, and no single antecedent condition may be common to all or even to most instances of the outcome. When causation is this complex--probably the rule and not the exception in the study of social phenomena--then no single causal condition is either necessary or sufficient.