There exists a well-developed literature on case selection strategies within qualitative methods (e.g. Gerring, 2007). However, few of the guidelines are directly applicable when we are interested in investigating causal mechanisms linking an explanatory variable (X) with an outcome (Y) using process-tracing methods. This paper offer for guidelines for case selection in Process-tracing. The paper proceeds in two parts.
First, the paper contend that case selection strategies should vary depending upon the type of Process-tracing that is being undertaken. ‘Theory-testing Process-tracing’ involves testing whether a causal mechanism exists in a research area where a well-established empirical correlation between a given X and Y has been found (e.g. democracy and peace), or testing whether a well-developed theoretical conjecture actually exists in reality. Theory building involves analysis where we know X and Y, but are in the dark regarding the mechanism linking the two. ‘Explaining outcome Process-tracing’ involves using the available evidence to work backwards from a known outcome to uncover the causal mechanism(s) that produced the outcome.
Second, using examples drawn from recent scholarship, we develop the logic of case selection strategies in three variants of Process-tracing. In the theory-testing variant, we contend that cases should be selected based upon the likely presence of the hypothesized causal mechanism; in other words selected on both X, Y and the presence of the conditions necessary for the hypothesized causal mechanism to be present. In theory-building case selection strategies depend upon whether the purpose is to: 1) uncover a mechanism between X and Y, or 2) whether we are attempting to build a theory when we know the outcome but are unsure about what mechanism(s) made it happen. Regarding ‘explaining outcome Process-tracing’, we argue that cases should be selected purely on the outcome; directly contradicting the otherwise prevalent advice (especially KKV) that selection on the dependent variable is problematic.
