Identification & What Makes the Result Tight (aejmic-identification)
AEJ: Micro is theory-first, so "identification" here is two things. For pure theory it means: which assumptions are doing the work, and how tight/robust the mechanism is. For structural and experimental work it means the standard data-to-object mapping. Pick the branch.
When to trigger
- (Theory) A referee asks whether the result is a knife-edge artifact of one assumption
- (Theory) You cannot say cleanly which primitive drives the comparative static
- (Structural) Parameters are estimated but it is unclear what in the data identifies them
- (Experimental) The estimand or the assumptions behind the treatment effect are not pinned down
Branch A: Pure theory — what makes the result tight
The AEJ: Micro bar is that the reader sees exactly which assumption is load-bearing and how far the mechanism extends.
-
Decompose the assumptions. For each substantive assumption, ask: is the result false without it, weaker without it, or unchanged (then it was WLOG — say so)? The result is "tight" when you can name the assumption that breaks it.
-
Comparative statics as identification. Show the sign/magnitude of the key comparative static and what primitive drives it (single-crossing? a supermodularity? a curvature condition?). Monotone-comparative-statics tools (Topkis, Milgrom–Shannon) make the driver explicit.
-
Necessity, not just sufficiency. Where you can, show the assumption is necessary (a counterexample when it fails), not merely sufficient — this is what makes a characterization tight.
-
Robustness of the mechanism (then hand to
aejmic-robustness for full extensions): does the result survive a small perturbation of the information structure, the timing, or the type distribution?
Branch B: Structural / empirical IO
-
Name what identifies each parameter. Tie parameters to specific data features / moments; argue identification from the model's structure, not "the estimator converged."
-
Targeted vs. untargeted moments; report a sensitivity/informativeness measure so readers see which data move which parameters.
-
Estimation regularity: objective (MLE/GMM/MSM), starting values, tolerances, multi-start; Monte Carlo recovery of known parameters.
-
Counterfactual validity: argue the estimated parameters are policy-invariant enough for the counterfactual (Lucas critique).
- For reduced-form companions, use design-appropriate diagnostics (pre-trends, first-stage strength, density tests) and report SEs, not asterisks.
Branch C: Experimental (theory-grounded)
-
Design maps to the model: each treatment isolates a model primitive or prediction; state the estimand.
-
Pre-registration in a recognized registry where applicable; report deviations; include instructions/transcripts.
- Randomization balance; attrition (Lee bounds if differential); multiple-hypothesis adjustment; external-validity scope.
Execution bridge (StatsPAI / Stata MCP)
Estimate and audit the identification claim, don't only argue it. Full map:
execution-with-mcp. AEJ: Micro spans applied and structural micro; the chain below is for the reduced-form / causal lane — structural estimation uses the field's own solvers.
-
detect_design → recommend → fit with as_handle=true → audit_result to list
the checks the design still owes.
-
Staggered DiD:
callaway_santanna / sun_abraham + bacon_decomposition +
honest_did_from_result (the pre-trend test is low-power, Roth 2022).
-
IV:
effective_f_test + an anderson_rubin_ci (valid under weak instruments),
not a 2SLS t-stat alone.
-
RDD:
rdrobust (bias-corrected) + rddensity / mccrary_test for manipulation.
-
OVB:
oster_delta / sensemakr — how strong a confounder would have to be.
Report the economic magnitude; route the full battery to the appendix; keep every
number reproducible. A run end-to-end (synthetic data, real returns) is in the
JF execution walkthrough. If StatsPAI/Stata are not connected, adapt the
vendored resources/code/ skeleton and flag any unverified number.
Checklist
Anti-patterns
- (Theory) A result whose driving assumption is never identified — "it just works"
- (Theory) Claiming a characterization is tight without a counterexample when the assumption fails
- (Structural) "The estimator converged" presented as identification
- (Structural) A counterfactual on calibrated parameters with no policy-invariance argument
- (Experimental) No pre-registration or no stated estimand; significance asterisks instead of SEs
Worked vignette (illustrative)
A matching paper proves stability is preserved under a new preference domain. A referee suspects it rides on a substitutability condition. The AEJ: Micro answer names it: "Substitutability is load-bearing — without it, Example 3 exhibits an empty core; with the weaker 'unilateral substitutes' condition the existence result survives but uniqueness fails." That sentence makes the result tight: the necessary assumption is named, and the cost of relaxing it is shown.
Output format
【Branch】theory / structural / experimental
【What makes it tight / data-to-object】one sentence
【Load-bearing assumption(s) or identifying moments】[...]
【Tightness evidence】counterexample-on-failure / sensitivity+Monte Carlo / balance+estimand
【What it does NOT establish】[...]
【Next step】aejmic-robustness (extensions/edge cases)