Dear Pekka,

QCA, just like all other methods of causal data analysis, depends on you as the researcher having some prior theory/intuition regarding exogenous/endogenous factors, at least to a certain extent. Otherwise, you couldn't specify what to analyze. I am convinced that, at least for social science research, you'll always find existing theories or at least easily importable theories from neighboring or in other ways related areas.

Once you have such a factor frame, there's no reason not to take a data-mining approach, and rank subset factor frames (the analytically feasible combinations of exogenous/endogenous factors from among all possible combinations) according to the neatness of the QCA solutions they produce (e.g., prefer solutions with high consistency and high coverage and low number of alternative models).

Best wishes,

Alrik

I agree with Alrik there there are two distinct aspects of QCA in this regard. First, it does require some prior theory about which causal factors to enter in the model. Second, it does not require testable hypotheses about the configurations that it produces.

Thank you for your replies Alrik and David, this is very helpful.

As a follow-up question: what about using QCA as a validation technique in a qualitative study? In other words, using tentative conclusions/inferences as previous "theory".

Dear Pekka, are you asking whether QCA can be used to test hypotheses derived from the results of previous studies?

Dear Alrik, not exactly: instead of previous and finalized studies one would use hypotheses / findings derived through other means of analysis within a study (i.e. using "theory in the making" for defining the conditions). For example, assume that findings from a non-QCA interview study I have conducted suggests that condition X leads to outcome Y (and there is no previous theory to support that X would be a relevant condition). Would QCA be suitable for testing the robustness of these findings? I guess in such case contradictory conditions would need to be solved through substantive case knowledge only.

I hope I am making sense here :).

Dear Pekka, it is perfectly fine to use factors that you may have only started to consider as possibly relevant, and for which there's scant, if any, prior theoretical support (in the 1960s, many health professionals still saw no connection whatsoever between smoking and lung cancer, for example). There's really nothing in QCA that requires the selection of only factors for which a significant body of existing theory claims that they are relevant.

Regarding so-called "contradictions", you shouldn't worry too much, either. The rule is simple: Given two different yet equally sized sets of explanatory factors, prefer the one set for which the evidence is clearest, that is, the one for which inclusion (consistency) / coverage scores for the final solution are highest.

Excellent, thank you so much!

Since QCA gives its results in configurations, it seems to me that you would need hypotheses in the same form, which is rather different from something as simple as X leads to Y.

Good point David, X leads to Y is too simple an example.

Just a follow-up remark on David's comment: David is absolutely right that QCA outputs configurations as results. However, for testing a configurational hypothesis, assuming you have one, you do, in fact, not need QCA at all.

For example, assume your hypothesis to be A + B -> C [in the following, "¬" means "not", "*" means "and" and "+" means "or"]. You can rewrite this as ¬(A + B) + C, because, in Boolean algebra, X -> Y is defined as ¬X + Y. Negating the hypothesis thus yields ¬(¬A*¬B + C), which can be transformed into A + B * ¬C. Since, unlike in linear algebra, both "*" and "+" distribute over each other in Boolean algebra, A + B * ¬C can be rearranged to A*¬C + B*¬C. If your data are such that A*¬C + B*¬C is true more often than A + B -> C, you can reject your hypothesis. Otherwise, there's evidence in support of A + B -> C.

I have a more complex example with an equivalence hypothesis in the attached paper. Theoretically, you can, of course, also conceive even more complex configurational hypotheses, involving also complex disjunctions of complex conjunctions as part of equivalences, but the conclusion will not change: QCA's purpose is not to test configurational hypotheses.

Best wishes,

Alrik

Exploratory research is effective in laying the groundwork that will lead to future studies. you can use qualitative comparative analysis for exploratory research.

While it's certainly preferred to derive QCA conditions from theory, you can also use QCA for exploratory research. This can be useful when you have a larger research project and want to narrow down your conditions in a first step or want to find out which conditions combine to bring about the outcome you're interested in.

For the tasks of defining conditions and solving contradictions in your truth table it would not necessarily make a difference whether you're taking an exploratory or theory-based approach. That being said, having theory will make it much easier to interpret your eventual results.