Theory of Change & Mechanism (jpam-theory-building)

JPAM is empirical, but the best policy papers are not atheoretical. The "theory" here is a theory of change: a transparent logic that says why the policy lever should move the outcome, through what mechanism, for whom, and under what conditions it would (or would not) transfer to another jurisdiction or scale. This frame is what turns a single estimate into evidence a policymaker elsewhere can use, and it disciplines the cost-benefit and heterogeneity analysis downstream.

When to trigger

• Specifying the logic model / theory of change before or alongside the design
• A reviewer asked "what's the mechanism?" or "why would this generalize?"
• Deciding which heterogeneity and mediation analyses are theory-driven (not fishing)
• Framing scope conditions for external validity and scale-up

Building the theory of change

1. Lever → mechanism → outcome. Write the causal chain explicitly: the policy changes X (incentive, constraint, information, price, access), which operates through mechanism M, producing outcome Y.
2. Behavioral or institutional micro-foundation. Say who responds and why — incentives, liquidity, salience, capacity, compliance, take-up. Ground it in the relevant econ / PS / PA theory.
3. Predicted heterogeneity. Whom should the effect be larger or smaller for? Specify before estimation so subgroup results read as tests, not data-mining.
4. Scope conditions for transfer. What features of this setting (administrative capacity, market structure, population, complementary policies) does the effect depend on? This governs external validity and scale-up claims.
5. Unintended effects and general equilibrium. Name plausible offsetting or spillover responses a policymaker would worry about (displacement, crowd-out, behavioral adaptation).

What the theory buys (JPAM-specific)

• It tells policymakers in other jurisdictions whether the result should travel.
• It pre-commits the heterogeneity and mediation analyses, protecting them from fishing critiques.
• It specifies which costs and benefits enter the cost-benefit analysis and for whom (distribution).
• It distinguishes "the program worked here" from "this kind of program works through this mechanism."

Checklist

• An explicit lever → mechanism → outcome chain written down
• A behavioral/institutional micro-foundation for who responds and why
• Predicted heterogeneity stated before estimation
• Scope conditions for transfer / scale-up named
• Plausible unintended effects, spillovers, or GE responses flagged
• Mechanism tied to specific, testable observable implications

Anti-patterns

• A bare estimate with no mechanism — "the program raised Y" with no account of why
• Post hoc storytelling that fits whatever heterogeneity the data happened to show
• Claiming broad generalizability with no stated scope conditions
• Ignoring unintended consequences a policymaker would immediately ask about
• Borrowing a formal model that adds notation but no testable, policy-relevant implication

Worked micro-example (illustrative)

A team evaluates a tax-credit expansion. The bare version says "the credit raised employment." The JPAM theory-of-change version writes the chain — credit raises the after-tax return to work (mechanism: labor-supply incentive) → larger response for the population facing the steepest participation tax (predicted heterogeneity: single parents near the phase-in) → effect depends on local labor demand and on take-up via tax-filing (scope conditions) — and flags a plausible unintended effect (employers capturing part of the credit through wage adjustment). Each link is then a testable implication the design and heterogeneity analysis must address, and the scope conditions tell a policymaker in another state whether the result should travel. (Fields/numbers illustrative.)

Calibration anchors (hedged)

• The "theory" JPAM wants is a transparent logic model, not necessarily a formal model; added notation must buy a testable, policy-relevant implication.
• Scope conditions are a strength, not a hedge: they are how a single evaluation informs decisions elsewhere, which is JPAM's whole purpose.
• Pre-specify heterogeneity and mechanism tests so they read as confirmatory, not as fishing.

Output format

【Lever → mechanism → outcome】the causal chain in one line
【Micro-foundation】who responds and why
【Predicted heterogeneity】subgroups specified ex ante
【Scope conditions】what the effect depends on for transfer
【Unintended effects】spillovers / GE / displacement flagged
【Next】jpam-research-design

Supplementary resources

• ../../resources/exemplars/library.md — JPAM papers whose mechanism makes the result portable
• ../../resources/external_tools.md — mediation / mechanism estimation packages