Individual Submission Summary
Share...

Direct link:

Download

What Is Distinctive About Process Tracing?

Sun, October 3, 6:00 to 7:30am PDT (6:00 to 7:30am PDT), TBA

Abstract

It is commonly accepted that the logic behind process tracing in political science is given by Bayes rule (Fairfield and Charman 2017). But Bayes rule is arguably the logic behind *all* good scientific inference, be it quantitative or qualitative, contrary to what “frequentists” about quantitative methods claim (Howson and Urbach 1989). And advocates of process tracing claim that process tracing is a qualitative method whose logic is distinct from that of standard quantitative methods (Brady and Collier 2010). So what then is this distinctive logic? Given three assumptions, I argue that there are only three ways in which process tracing could possibly distinguish itself from standard quantitative methods:
1) Provide a distinct way of using within-case evidence to learn about the “propensities” attached to a single case;
2) Provide some distinct principles / ways in which detailed causal hypotheses (about a single case) make predictions about the propensities attached to that case;
3) Provide a distinct way of leveraging cross-case evidence when discriminating between causal hypotheses about a single case.

I then contend that (1) and (2) do not seem to be what process tracers do. In relation to (2), for example, process tracers do not seek to deny (or to provide alternatives to) quantitative principles such as the principle of the common cause, or the principle of the common effect, or the principle of faithfulness. This leaves only (3) as the way in which process tracing can distinguish itself. This is ironic, because the methodologists who have done most to clarify and to defend the logic behind process tracing usually insist that process tracing is a “within case” method, thus playing down the reliance of process tracing upon cross-case evidence (Beach and Pedersen 2016).

The assumptions I need to make this argument are Bayes rule itself, plus:
- David Lewis’ Principal Principle (1980): For any set of variables, given perfect knowledge of the propensities governing these variables (in a particular case), all other available information is redundant to predicting in advance the values that these variables will take (in this particular case). This principle thus relates credences (rational degrees of confidence) to propensities (probabilities as features of the world).
- Causation can be inferred, but it cannot be directly observed.

These assumptions are minimal. To make this argument, I will not need to take a position on:
- the nature of variable causation (Interventionist / Difference Making / Mechanistic / Productive / Energy Transfer / Primitive)
- whether the cases one examines are governed by the *same* propensities, and the *same* causal relations between variables (Homogeneity / Heterogeneity)
- what hypotheses about variable causation imply about propensities (see above).

Thus this argument still applies even if (as qualitative methodologists suggest) there are two distinct types of causation in the world, one type “difference making” that is best studied by quantitative methods, and another type “mechanistic production” that is best suited to being studied by qualitative methods.

I then illustrate this abstract argument with a concrete suggestion about how process tracing might leverage cross-case evidence in a distinctive way: process tracing makes a much more subtle and sophisticated set of “causal homogeneity” assumptions than standard quantitative methods do. This is not obvious because methodologists have played down the reliance of process tracing on background knowledge about other cases, I suggest.

The upshot is that at a very high level of abstraction there is indeed (as King Keohane and Verba suggest) only one logic of causal inference, but that at a lower level of abstraction there are multiple logics of causal inference.

Author