1. Winship and Morgan (1999: 660) argue that the language of „treatment“and „control“variables is generally applicable: „In almost any situation where a researcher attempts to estimate a causal effect, the analysis can be described, at least in terms of a thought experiment, as an experiment.“A more direct implication of using experimental language, which Winship and Morgan do not discuss in detail, is the restriction that „the treatment must be manipulable“(1999: 663, fn. 2). Citing Holland (1986), they argue that it makes no sense to talk about the causal effect of gender or any other nonmanipulable individual trait alone. One must explicitly model the manipulable mechanism that generates an apparent causal effect of a nonmanipulable attribute“(1999: 663, fn. 2).
2. In any event, counterfactual regression procedures have been developed for application to individual-level data and are feasible only when (1) there is a very large N, and (2) it is plausible a priori that each case could be in either the control or the treatment group (see Winship/ Morgan 1999). Also, these procedures, like conventional statistical analyses, remain linear and additive, so they do not examine problems of limited diversity and matched cases directly. An attempt to address limited diversity, or „the curse of dimensionality,“with Boolean logit and probit regression is offered by Braumoeller (2003).
3. This may or may not be the only pathway to having a generous welfare state. The focus here is simply on the evaluation of this pathway, with its four combined conditions.
4. There can be other, unspecified combinations of causal conditions linked to outcome Y in this example. There is no assumption that this is the only combination linked to the outcome (Y).
5. Note that methodological discussions of counterfactuals often assume a non-configurational variant of the „difficult“form, as in Fearon (1996: 39): „When trying to argue or assess whether some factor A caused event B, social scientists frequently use counterfactuals. That is, they either ask whether or claim that ‘if A had not occurred, B would not have occurred.’“