‘Rigorous evidence’? Yes — and totally useless!
from Lars Syll
So far we have shown that for two prominent questions in the economics of education, experimental and non-experimental estimates appear to be in tension. Furthermore, experimental results across different contexts are often in tension with each other. The first tension presents policymakers with a trade-off between the internal validity of estimates from the “wrong” context, and the greater external validity of observational data analysis from the “right” context. The second tension, between equally well-identifed results across contexts, suggests that the resolution of this trade-off is not trivial. There appears to be genuine heterogeneity in the true causal parameter across contexts.
These findings imply that the common practice of ranking evidence by its level of “rigor”,
without respect to context, may produce misleading policy recommendations …
Despite the fact that we have chosen to focus on extremely well-researched literatures,
it is plausible that a development practitioner confronting questions related to class size, private schooling, or the labor-market returns to education would confront a dearth of well-identified, experimental or quasi-experimental evidence from the country or context in which they are working. They would instead be forced to choose between less internally valid OLS estimates, and more internally valid experimental estimates produced in a very different setting. For all five of the examples explored here, the literature provides a compelling case that policymakers interested in minimizing the error of their parameter estimates would do well to prioritize careful thinking about local evidence over rigorously-estimated causal effects from the wrong context.
Randomization — just as econometrics — is basically a deductive method. Given the assumptions (such as manipulability, transitivity, separability, additivity, linearity, etc.) these methods deliver ‘deductive’ inferences. The problem, of course, is that we will never completely know when the assumptions are right. And although randomization may contribute to controlling for confounding, it does not guarantee it, since genuine ramdomness presupposes infinite experimentation and we know all real experimentation is finite. And even if randomization may help to establish average causal effects, it says nothing of individual effects unless homogeneity is added to the list of assumptions. Real target systems are seldom epistemically isomorphic to our axiomatic-deductive models/systems, and even if they were, we still have to argue for the external validity of the conclusions reached from within these epistemically convenient models/systems. Causal evidence generated by randomization procedures may be valid in “closed” models, but what we usually are interested in, is causal evidence in the real world we happen to live in.