Keane (2010) Structural vs. Atheoretic Approaches to Econometrics

Summary

This essay lays out the sources of conflict between the "structural" and "experimentalist" (atheoretic) camps in econometrics and argues that the popular contrast between them is a false dichotomy. The experimentalist critique holds that structural work rests on too many assumptions to be credible, and that natural-experiment / instrumental-variable methods deliver causal effects with far fewer. Keane counters that all empirical work relies heavily on a priori assumptions; the real difference is not the number of assumptions but the extent to which they are made explicit. Structural modeling makes its assumptions visible and its estimates interpretable — a prerequisite, he argues, for scientific progress and for the counterfactual/policy questions that atheoretic methods cannot answer.

Key Claims

• The false dichotomy. "All econometric work relies heavily on a priori assumptions. The main difference between structural and experimental… approaches is not in the number of assumptions but the extent to which they are made explicit." Experimentalist studies also rest on strong assumptions — they are just implicit.
• The experimentalist program. Following Angrist-Krueger (1999), the atheoretic approach "puts front and center the problem of identifying causal effects from specific events or situations" — natural experiments that generate exogenous variation, exploited via instrumental variables to estimate the effect of $X$ on $Y$ without a behavioral model.
• Hidden assumptions of IV/natural experiments. Instrument validity (exogeneity and the exclusion restriction) is untestable and theory-laden; the estimand is typically a local average treatment effect (LATE) — the effect for the subpopulation whose behavior the instrument shifts — not a structural parameter, and its external validity (extrapolation to other populations, policies, or ranges of $X$) is an implicit, usually undefended, assumption.
• Why structure is needed. Only a structural model with explicit preferences/technology/constraints can answer counterfactual and policy questions — the effect of a policy not yet tried, or over a range outside the natural experiment — because it identifies interpretable behavioral parameters rather than one local causal contrast.
• A defense of theory. "The 'simple' path that eschews theory as the basis for empirical work has almost never led to scientific progress"; explicitness of assumptions, not their absence, is what makes estimates interpretable and cumulative.

Concepts Introduced or Extended

• Structural Econometrics
• Structural Identification

Entities Mentioned

• Michael P. Keane

Quotes

"Critics of the structural approach often assert that it produces results that rely on too many assumptions to be credible… Here, I argue that this is a false dichotomy. All econometric work relies heavily on a priori assumptions."

"The 'simple' path that eschews theory as the basis for empirical work has almost never led to scientific progress."

My Take

Keane's reframing is the durable contribution: the debate is not "assumptions vs. no assumptions" but "explicit vs. implicit assumptions," and once you accept that, the experimentalist claim to epistemic higher ground weakens — an exclusion restriction is every bit an assumption as a utility function, just less visible and often less defended. The strongest part of the argument is about estimands: a well-identified LATE answers a narrow question (the effect for compliers, in this setting), whereas policy analysis usually needs a parameter you can extrapolate, which requires structure. It pairs directly with the same author's shorter Journal of Economic Perspectives piece (Keane 2010, "A Structural Perspective on the Experimentalist School"). The fair counter, which the credibility-revolution literature presses, is that explicit structural assumptions can be wrong in ways that are hard to check, so transparency about assumptions is not the same as correctness — which is why the wiki's prior-based identification idea (make the assumptions explicit and carry their uncertainty into the answer) is the natural synthesis.