Robustness, Extensions & Edge Cases (aejmic-robustness)

Robustness, Extensions & Edge Cases (aejmic-robustness)

When to trigger

• The main result is proved but referees will ask "does it survive [relaxation]?"
• You have many possible extensions and must decide which belong in the paper
• A knife-edge or boundary case is unaddressed
• (Applied) The empirical/experimental result needs a robustness battery

What robustness means at AEJ: Micro

For a theory paper, robustness is about the mechanism's reach: which relaxations preserve the result, which break it, and which boundary cases need care. AEJ: Micro values knowing the edges of a result as much as the result. For structural/experimental work, it is the standard robustness battery. The discipline is the same: every extension must earn its place — it either broadens the contribution or defends a load-bearing assumption flagged in aejmic-identification.

Theory extensions — the menu (include only what earns its place)

• Relax a substantive assumption: continuum vs. finite types, asymmetric vs. symmetric players, correlated vs. independent values. Show the qualitative result survives or pin down where it changes.
• Alternative solution concept / refinement: does the result hold under a coarser or finer equilibrium notion? If it is concept-specific, say so.
• Perturbations: small changes to the information structure, timing, or commitment level (full → partial). Continuity/upper-hemicontinuity arguments belong here.
• Boundary and knife-edge cases: tie-breaking, measure-zero events, corner solutions — handle explicitly, do not hand-wave.
• Negative extensions are informative: an extension that fails and explains why sharpens the contribution and pre-empts a referee.

Applied / experimental robustness

• Alternative specifications/estimators; sensitivity to grids, tuning, and seeds (structural/simulation).
• Placebo / falsification; multiple-testing adjustment; subsample stability.
• Report as SEs / coverage sets, never significance asterisks.

The "earns its place" test

Before adding an extension, ask which of two jobs it does. If it does neither, cut it.

1. Broadens the contribution — the result now covers a setting readers care about (continuum types, dynamics, asymmetry) that the base model excluded.
2. Defends a load-bearing assumption — it answers the specific "is this knife-edge?" objection that aejmic-identification flagged.

An extension that merely re-derives the base result under a cosmetic re-parameterization fails the test and dilutes the paper.

Placement discipline

• Core extensions that change the reading: main text. Supporting extensions: online appendix. Do not bury a result-defining extension in supplementary material, and do not pad the main text with extensions that add nothing.
• A negative extension that explains a boundary of the result often belongs in the main text precisely because it sharpens the contribution; a routine confirmation belongs in the appendix.

Execution bridge (StatsPAI / Stata MCP)

Run the battery, don't just enumerate 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.