From experimental-economics-skills
Derives testable predictions from behavioral or game-theoretic models for experimental economics manuscripts. Generates pre-specified hypotheses for pre-analysis plans.
How this skill is triggered — by the user, by Claude, or both
Slash command
/experimental-economics-skills:expecon-theory-modelThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
- The design exists but the paper never states *which model predicts what* in each treatment
At a method-defined journal, the model earns its place by producing the predictions your treatments adjudicate. You do not need a new theorem (that is GEB). You need a transparent map: given this game and these parameters, model A predicts X in treatment T1 and Y in T2; model B predicts the reverse. Build that map in four steps.
expecon-identification).expecon-robustness/expecon-identification for identification; here, just state which moments would identify it.State which solution concept your benchmark uses, because the experiment will test the behavioral departure from it. Nash or subgame-perfect equilibrium gives a sharp point to reject; QRE (quantal response) builds in noise and is often the more honest benchmark for interior behavior; level-k or cognitive-hierarchy is the right benchmark when iterated reasoning is the phenomenon (beauty contests, guessing games). The choice is not cosmetic: a "deviation from Nash" can be fully explained by QRE noise, so if your contribution is that subjects depart systematically from the rational benchmark, anchor on QRE and show the residual the noise model cannot absorb. Name the concept, justify it in one sentence, and tie each treatment's prediction to it.
A gift-exchange labor experiment wants to test whether reciprocity, not just selfish play, drives effort. Standard preferences predict minimum effort regardless of wage (the benchmark, numeric: e = 1). A reciprocity model predicts effort rising in the wage; inequity aversion also predicts a wage–effort link but flattens once payoffs equalize. The two diverge at high wages: reciprocity keeps climbing, inequity aversion plateaus. So the design adds a high-wage treatment precisely where the predictions split, and pre-specifies H1 (effort increases in wage) and H2 (the high-wage marginal increase is positive under reciprocity, ≈0 under inequity aversion). Now a single contrast adjudicates.
Not every ExpEcon paper has a game-theoretic model; some test a decision-theoretic or measurement claim (risk attitudes, time preferences, ambiguity, belief updating). The same discipline applies: state the functional form whose parameter you elicit (e.g., CRRA utility, (quasi-)hyperbolic discounting), the prediction each rival specification makes across treatments, and the moments that identify the parameter. A treatment that shifts elicited present bias only under one discounting model, and not under another, is your discriminating test. Make that explicit rather than reporting a parameter as if its model were uncontested.
The flagship rewards predictions, not page count of algebra. A half-page derivation that yields a sharp, signed, treatment-indexed prediction is worth more than a five-page model whose comparative statics the experiment never tests. If a proof is needed, it goes in an appendix; the main text carries only the predictions the design adjudicates. If your model implies a prediction you did not build a treatment to test, either add the treatment or cut the prediction — unused theory invites the "this is a theory paper" objection without earning the credit.
The PAP is where these predictions become commitments, so write them in testable form now. For each hypothesis, specify: the outcome variable (and how it is constructed from raw choices), the comparison (which treatments, which direction), the test and the unit (session/matching-group), and the decision rule (what result confirms vs. rejects). Distinguish the single primary confirmatory test from secondary ones. A prediction you cannot phrase as "outcome Y is higher in treatment T than C, tested by [test] at the group level, p<α" is not yet operational — sharpen it here before it reaches expecon-robustness, where an unspecified prediction becomes an uncorrected fishing expedition.
The deliverable of this stage is a small table the rest of the pack consumes: for each treatment, the prediction of every candidate model, with the cells where they diverge highlighted. expecon-identification uses it to confirm the contrast that produces the divergence is clean; expecon-robustness uses the primary signed hypothesis to set the powered comparison; expecon-tables-figures plots the predicted vs. observed pattern. If you cannot fill that table — if some treatment has no distinct prediction from any model — that treatment is not yet earning its place and should be cut or re-specified before any data are collected.
【Journal】Experimental Economics (ESA method flagship)
【Skill】expecon-theory-model
【Verdict】pass / sharpen / reroute
【Game】players / actions / info / ECU payoffs + conversion / matching / horizon
【Benchmark prediction】standard-preferences point (numeric)
【Rival predictions】model → treatment-indexed prediction (where they diverge)
【Pre-specified hypotheses】H1, H2… (signed, treatment-indexed)
【Confusion check】how the design separates mechanism from comprehension
【Next skill】expecon-identification
npx claudepluginhub brycewang-stanford/awesome-journal-skills --plugin experimental-economics-skillsCreates a behavioral or bounded-rationality model for a JEBO manuscript. Matches model scope to the paper's need, from a testable prediction to a full agent-based simulation.
Builds economic mechanisms and derives signed, falsifiable predictions for JAR manuscripts using information economics, contracting, and disclosure theory.
Frames whether a research question is a method-defined fit for Experimental Economics manuscripts and selects the minimal treatment contrast for the design.