Jhe Theory Model

Theory and Model Craft (jhe-theory-model)

When to trigger

• A reduced-form health-policy estimate needs a model to be interpreted as welfare, demand, or a structural parameter
• An insurance/selection paper needs the demand-and-selection model to map data to risk preferences or adverse selection
• A provider paper needs a payment-response model to predict how DRG/P4P/capitation reshapes behavior
• A counterfactual (a coverage expansion, a price change, an optimal cost-sharing rule) requires a model to extrapolate beyond the estimated variation

How JHE uses models

At JHE a model rarely stands alone; it earns the right to interpret an estimate or run a counterfactual. The bar is that the model be tight enough to discipline the data and realistic enough to match the institution. JHE's house demand-and-selection and provider-incentive models are the lingua franca — use them rather than reinventing. The decisive questions: does the model identify the object you claim (an elasticity, a selection parameter, a welfare quantity)? Are its parameters policy-invariant for the counterfactual you run (the Lucas critique bites hard for coverage and payment reforms)? Does it respect institutional facts — risk adjustment, network rules, mandatory-coverage floors — that change the equilibrium?

Model archetypes and what each must deliver

In practice most JHE papers sit in the demand-selection or provider-incentive rows; the production and market rows appear when health is the input (human capital) or when the regulated equilibrium itself is the object. Pick the row that matches your contribution and resist borrowing apparatus from another row that the data cannot support.

Discipline checks

1. State the object the model identifies and tie each structural parameter to the data feature that pins it.
2. Match the institution. If risk adjustment, a coverage mandate, or a network constraint is present in the data, it must be in the model — a model that ignores the binding institutional rule will not survive review.
3. Defend policy-invariance for any counterfactual: argue the estimated parameters are not themselves functions of the policy you change.
4. Report what the model assumes away and whether the result is robust to relaxing it (e.g., homogeneous vs. heterogeneous risk preferences).
5. Keep the model proportionate. A reduced-form paper needs only enough theory to interpret the sign and magnitude; do not bolt on a structural model the data cannot identify.

Checklist

• The object the model identifies is named (elasticity / selection parameter / welfare quantity)
• Each structural parameter is tied to the identifying data feature
• Binding institutional rules (risk adjustment, mandates, networks) are in the model, not assumed away
• Policy-invariance defended for every counterfactual
• Welfare claims rest on the model, not on a reduced-form coefficient alone
• What the model assumes away is stated and stress-tested
• Model complexity is proportionate to what the data can identify
• The chosen model archetype matches the contribution, not borrowed from another subfield

Anti-patterns

• A model whose equilibrium ignores a binding institutional rule the data clearly reflects
• A structural model whose parameters the data cannot actually identify ("calibration in disguise")
• Reading a reduced-form coefficient as welfare with no model linking them
• A demand/selection model that ignores risk adjustment or the mandate that shapes the real market
• Running a counterfactual that changes a policy the parameters secretly depend on
• A heavy structural apparatus where a one-equation interpretive frame would do
• Borrowing a model from another health-econ subfield without checking the data can identify it