James Fearon, “Counterfactuals and hypothesis testing in political science”, World Politics 43 (1991), pp. 169-195.

Main Argument: Counterfactuals make claims about events that did not actually occur. Such propositions are necessary and fundamental in political research to assess hypotheses, esp. in small-N research designs. Fearon advocates all working within a scientific frame, favoring openness of procedure, internal coherence of argument, good measurement for variables, increasing attempts to unravel context, assiduous concern for valid causal inferences and rewards for replication.
== Notes: ==
Two strategies outside experimental control for assessing cause C of event E:
*  Using counterfactual imaging if C didn’t occur, or
*  Looking for an actual case that resembles the case in question expect for C being absent
Problems in Using Counterfactuals:
*  Negative degrees of freedom: can’t have negative DoF, so there is no choice but to add or create more cases: either a counterfactual case (or cases) that never actually existed or actual cases [172]
*  Risk in adding cases of systematic bias (often from failure to meet ceteris paribus assumption)
*  Can be used in large-N study, but most consider this too risky
*  Quasi-experiments make counterfactual that if values were different, the error terms would not differ systematically
*  Have to assume that explanatory variables and the errors (the other causes) are uncorrelated
Potential difficulties in using counterfactuals:
1. How to distinguish definition of causality from “if A had not occurred, B would not have occurred”… is this causation or does A have to lead to B in a regular fashion (excludes accidental occurrences as ‘conditions’ instead of causes; Cleopatra’s Nose: causality should not be defined in terms of counterfactuals like P2, that A satisfies P2 does not imply that A is a cause of B
2. Precision of estimates: arguments about the relative importance of possible causes become arguments about the relative plausibility of different counterfactual scenarios
3. Elster’s legitimacy problem with counterfactuals –> counterfactual thought experiments undertaken to assess a causal hypothesis is not legitimate if we have a theory saying that the counterfactual antecedent could not have happened.  Stronger our theories, less need to use counterfactuals because theory disproves ‘other world’
a. does changing the IV necessitate the changing of other variables? Can you conceive of alternate world where only IV is different?
Counterfactual Argument In Practice:
*  necessity of counter factual argument for justifying causal claims in small-N settings
*  when degrees of freedom in the actual world are negative, a causal claim requires argument about counterfactual cases for its justification (or addition of other actual cases) [180]
*  counterfactuals are most likely to be found performing confirmatory work
*  can use for N=1 (necessary) and N=+1
*  In N=+1: Researchers with more than one actual case are not logically compelled to use the counterfactual strategy to justify a causal claim, as long as they do not have more independent variables than cases (less one), or two or more independent variables that vary together (perfect multicollinearity)
*  In this case, application of statistical methods either would fail to yield estimates of causal effects or would yield wildly imprecise estimates [186]