Case Studies and Theory Development in the Social Sciences - Alexander L. George [115]
As for measurement error, case study research is less prone to some kinds of measurement error because it can intensively assess a few variables along several qualitative dimensions, rather than having to quantify variables across many cases. Similarly, probabilistic causal mechanisms and the potential for omitted variables pose difficult challenges and limits to all research methods, but they do not necessarily invalidate the use of single case studies. The inductive side of process-tracing may identify potential omitted variables through the intensive study of a few cases, and single case studies have changed entire research programs when they have impugned theories that failed to explain their most-likely cases.443
In before-after research designs, discussed in Chapter 8, the investigator can use process-tracing to focus on whether the variable of interest was causally linked to any change in outcome and to assess whether other independent variables that change over time might have been causal. In Donald Campbell’s and Julian Stanley’s terms, the potential confounding variables of greatest interest in a before-after design are maturation effects (the effects of a unit maturing from one developmental stage to another) and the effects of history (exogenous changes over time).444 For example, in Andrew Bennett’s comparison of the Soviet decision to intervene in Afghanistan in 1979 to the Soviet withdrawal from that country in 1989, he needed to look at several variables that had changed in the intervening decade. In particular, it was essential to use process-tracing to assess the respective roles of changes in Soviet leaders’ views on the use of force, changes in the Soviet government (such as Mikhail Gorbachev’s political reforms), and changes in Soviet interactions with other actors (such as the emergence of a U.S. policy of providing aid to the Afghan rebels). Process-tracing evidence in this study indicated that U.S. aid to Afghan rebels likely delayed a Soviet withdrawal, but made a more complete withdrawal more likely. Soviet democratization had little effect because it largely took place after 1989, and changes in Soviet ideas fit both the specifics and timing of the Soviet withdrawal and associated Soviet policies.445
We have emphasized the use of process-tracing to develop and refine many theories that are not yet capable of generating testable predictions about causal processes and outcomes. Such a procedure need not degenerate into an atheoretical and idiosyncratic enterprise. When a researcher uncovers a potential causal path for which there is no pre-existing theory, there are several possible approaches for converting this atheoretical finding into an analytical result couched in terms of theoretical variables. For example, deductive logic or study of other cases may suggest a generalizable theory that includes the novel causal path. If so, it may be possible to specify and operationalize that new theory and assess it by means of a plausibility probe involving other cases. Or the novel causal path may be identified as an exemplar of an existing theory that the investigator had overlooked or had thought to be irrelevant. The newly identified causal process may then contribute to the evaluation of the existing theory. Finally, it is possible that the novel causal path may have to remain ungeneralizable and unconnected to a useful theory for the time being.
The Limits of Process-Tracing
There are two key constraints on process-tracing. Process-tracing provides a strong basis for