Unintended consequences: modeling behavioral responses

A policy doesn’t act on the world — it acts on people, who then act back.

The idea

Every rule changes someone’s incentives, and they respond: they substitute one behavior for another, evade the rule, get the cost capitalized into a price, or hit the number while missing the point (Goodhart’s law). The first-order effect is what you intended. The second and third orders are what actually happens.

So the discipline is a single habit: after “this policy will…”, always ask who adapts, how, and what the headline metric quietly misses.

How it works

Model the response in four moves. For each affected group, ask which door they walk through:

1. Name the groups the rule touches.  Firms, owners, workers, buyers.

2. For each group, pick the cheapest response:
   comply      do what was intended            (rare to be the whole story)
   substitute  swap to an untaxed / unbanned near-equivalent
   evade       keep the behavior, dodge the rule
   exit        leave the market entirely        (e.g. price ceiling ->
                                                 landlords withdraw units)

3. Trace orders of effect.
   1st order  the intended, measured change     -> the headline metric
   2nd order  the behavioral response
   3rd order  the response's effect on the real goal

4. Compare two metrics.
   official target   what the rule literally measures
   true goal         what you actually wanted
   gap = official - true  =  "what the headline misses"

Capitalization is the sly one: everyone can fully comply and the goal still fails, because a subsidy or tax gets absorbed into a price when supply can’t respond. Goodhart is its cousin: the moment a measure becomes a target, effort flows to the measure, not the goal.

When to use it

Reach for this when…	The limit
A policy targets a behavior people can cheaply change or route around.	Responses are estimates — you’re reasoning about direction and size, not forecasting exactly.
Success is judged by a single, easy-to-move metric.	Not every policy backfires; some responses are small or even helpful. Don’t assume the worst reflexively.
Supply or capacity is fixed in the short run (housing, permits, slots).	Second-order size depends on elasticities you may not know precisely.

Watch out for

• Judging by the target metric alone. “Vacancies fell 70%” can coexist with “no one new is housed.” Always pair the official number with the true goal.
• Assuming compliance equals success. With capitalization, everyone obeys and the benefit still leaks to sellers. Cheating isn’t required for a policy to miss.
• Forgetting the near-substitute. Ban or tax one thing and demand flows to the closest unbanned option — often no better, sometimes worse.
• Ignoring who has the cheapest exit. The group that can most easily leave or evade shapes the outcome; model them first.
• Stopping at second order. The interesting damage — enforcement costs, new markets, distributional shifts — often lives at the third order.

Worked example

An interviewer asks: “A city wants more homes lived in, so it taxes vacant units. Evaluate it.” A strong answer doesn’t stop at “vacancies will fall.” It names the group (owners), the cheapest response (evade — register a relative, use the unit a few nights, fake a lease), and then separates the two metrics: recorded vacancy drops sharply, but lived-in supply barely moves, and the city inherits an audit bill. The candidate closes with what would change the verdict: strong occupancy verification and taxing on assessed value rather than self-reported use. That’s the whole move — who adapts, how, and what the headline misses.