Jebo Identification

Identification Strategy (jebo-identification)

When to trigger

• A lab/field experiment's treatment, incentives, or comprehension are not pinned down
• An effect could be an experimenter-demand effect rather than the claimed behavioral mechanism
• An observational behavioral claim rests on OLS + controls, or TWFE on staggered timing
• A study uses deception, or runs many treatments, and the inference/ethics implications are unaddressed
• You are unsure the design isolates the behavioral mechanism, not a confound

The JEBO identification bar

JEBO judges identification through a behavioral-mechanism lens: the design must isolate the psychological or institutional channel the paper claims, not merely produce a significant difference. Because JEBO treats experimental design as a first-class identification branch alongside observational causal designs, "identification" means different things by branch — pick the branch and make the channel transparent. Inference must match the design (clustering at the level of randomization or assignment; few-cluster corrections).

Branch A: Lab / online experiment (the JEBO core)

• Incentive compatibility: payoffs must make truthful/effortful behavior the dominant strategy for the elicited object (e.g., BDM, strategy method, incentivized beliefs). State the mechanism and the stakes.
• No-deception norm: the experimental-economics convention is no deception; if you deviate, justify it and expect scrutiny — many referees treat deception as disqualifying for an incentivized study.
• Comprehension & attention checks: report comprehension quizzes, control questions, and how failures were handled (drop / re-instruct), so the effect is not confusion.
• Experimenter demand: rule out demand effects — neutral framing, between-subject where within-subject would cue the hypothesis, obfuscated objectives, or an explicit demand-treatment (e.g., Mummolo–Peterson / de Quidt-style bounds).
• Randomization & balance: show balance on observables; report the randomization procedure and unit.
• Multiple treatments: if several treatment arms, plan the comparisons and correct for multiplicity (see jebo-robustness).
• Pre-registration: pre-register the design and primary analysis (AEA RCT Registry / AsPredicted / OSF) and report deviations. (JEBO does not currently mandate it — 待核实 — but referees increasingly expect it.)

Branch B: Field experiment

• ITT vs. LATE/TOT stated; randomization unit and stratification described; spillovers and SUTVA addressed.
• Attrition examined and bounded (Lee bounds if differential); compliance documented.
• Ethical clearance / consent noted; external-validity scope stated (the field setting's generality).

Branch C: Observational behavioral empirics

• DID / event study with staggered adoption: move beyond TWFE (Callaway–Sant'Anna, Sun–Abraham, de Chaisemartin–D'Haultfœuille); show clean event-study leads; Goodman-Bacon decomposition.
• IV: strong first stage (effective F); weak-IV-robust sets (Anderson–Rubin) when needed; defend exclusion in institutions/theory + falsification.
• RDD: density/manipulation test (McCrary / Cattaneo–Jansson–Ma); local-linear, data-driven bandwidth, bias-corrected robust CIs.
• The behavioral interpretation must be argued, not assumed — the design identifies an estimate; the mechanism connecting it to a behavioral channel needs its own evidence (see jebo-theory-model).

Branch D: Agent-based / simulation

• Document the data-generating process and behavioral rules; set and report seeds; show the result is not an artifact of grid/tuning choices (see jebo-robustness).