geb-identification-strategy

Identification Strategy — Assumptions, Results & Proofs (geb-identification-strategy)

When to trigger

• The theorem is (you believe) true but the proof is hard to follow or audit
• You are unsure which assumptions are doing the work versus merely convenient
• A reviewer may ask "does this hold without assumption A / with infinite players / in continuous time?"
• For experimental work: the design does not cleanly separate competing strategic theories

What "identification" means at a theory journal

GEB is a game-theory journal, so for theoretical papers the "identification strategy" is the logical architecture of the result, not a causal-inference design. The bar is correctness, tightness, and generality, with proofs a referee can verify. Because the Editor in Charge routes the paper to an anonymous Advisory Editor and expert referees, the argument must withstand specialists who will reconstruct each step. Treat this skill in two branches.

Branch A — Theory (equilibrium, mechanism design, learning)

• Map the assumptions. List every assumption; for each, state why it is needed and what breaks without it. Separate substantive assumptions from normalizations.
• State results precisely. Each theorem/lemma/proposition self-contained: domain (class of games), hypotheses, conclusion. Number them for cross-reference.
• Establish generality and limits. Finite vs. infinite players/actions, complete vs. incomplete information, static vs. dynamic; give the most general statement the proof supports and note where it fails (a clean counterexample at the boundary is valued).
• Proof exposition. Lead with the idea before the algebra; signpost the construction; relegate long calculations to an appendix; ensure every invoked result (fixed-point theorem, separating hyperplane, etc.) has its hypotheses checked. Tightness: is the bound achieved? Is uniqueness or only existence shown?
• Verify computationally as illustration, not proof. Use a solver (e.g., Gambit) to confirm worked examples and equilibria, but the written proof must stand alone.

Branch B — Behavioral / experimental identification

• Design separates theories. Treatments must distinguish the competing strategic hypotheses (e.g., level-k vs. QRE vs. equilibrium), not just reject a null.
• Power and unit of observation. Justify sample size per cell; the independent unit is the session, not the subject — cluster inference accordingly.
• Confounds and incentives. Salient incentives, no demand effects, randomization across roles/order; pre-register the hypotheses where possible.
• Structural discipline. If estimating a behavioral model, state identification of its parameters.

Anti-patterns

• A proof that is a wall of algebra with no stated idea or roadmap
• Unstated assumptions silently used mid-proof
• Claiming generality the proof does not reach (e.g., asserting an infinite-game result proved only for finite games)
• Numerical examples presented as if they were a proof
• An experiment that rejects a null but cannot tell which strategic theory wins

Identification pass for Games and Economic Behavior

Use this as a second-pass capability check. First lock the primitives, equilibrium concept, comparative statics, and proof or experiment boundary; then test whether the manuscript addresses game theorists who ask what the model teaches beyond a clever example.

• Primary move: State identifying variation, identifying assumption, falsification, sensitivity, and the table/figure that would convince a skeptical referee.
• Decision ledger: return claim / evidence / blocker / next edit rows so the next pass can patch the manuscript directly.
• Neighbor test: compare against JET for theory abstraction, Theoretical Economics for compact theory contribution, Experimental Economics for experiment-first designs; if the neighboring outlet has the stronger audience claim, recommend re-routing before polishing.
• Verification floor: before submission-ready advice, re-open resources/official-source-map.md for volatile rules and name the one unresolved fact that could change the recommendation.

Output format

【Branch】theory / behavioral-experimental
【Main result】one-line statement (domain → conclusion)
【Assumptions】[each: needed because ... / fails without ...]
【Generality】most general statement supported; known limits
【Proof exposition】idea-first? steps signposted? invoked theorems checked? [Y/N each]
【Tightness】bound achieved / uniqueness vs. existence
【(Behavioral) design】separates theories? session-clustered? powered? [Y/N each]
【Next step】geb-data-analysis or geb-tables-figures