Binary Choice Models

Binary choice models estimate the probability of a yes-or-no outcome such as working, defaulting, or buying.

Binary choice models estimate the probability of an outcome that can take only two values, such as employed versus unemployed, default versus no default, or purchase versus no purchase.

Core setup

In applied econometrics, the usual idea is that a person or firm has an unobserved latent payoff:

y* = Xb + e

We do not observe y* directly. We observe only whether it crosses a threshold:

  • y = 1 if y* > 0
  • y = 0 otherwise

That structure makes binary choice models natural for decisions where the observed result is discrete even if the underlying incentives are continuous.

Main model types

  • The logit model uses a logistic cumulative distribution for the error term.
  • The probit model uses a normal cumulative distribution.
  • The linear probability model is simpler, but it can predict probabilities below 0 or above 1 and usually has heteroskedastic errors.

In most economics applications, logit and probit are preferred because they keep predicted probabilities in the valid 0 to 1 range.

How economists interpret them

The coefficient itself is not usually the marginal effect on probability. Instead, economists often compute:

  • marginal effects,
  • odds ratios in logit applications,
  • predicted probabilities for representative cases.

That matters for policy work. A coefficient may show the sign of an effect, but the economically meaningful question is often: “How much does the probability change?”

Typical uses

Binary choice models are common in labor economics, public economics, finance, health economics, and industrial organization. Examples include labor-force participation, benefit take-up, loan default, entry decisions, and technology adoption.

Knowledge Check

### What kind of dependent variable is handled by a binary choice model? - [x] One with two possible outcomes - [ ] One that is always normally distributed - [ ] One measured only in dollars - [ ] One with no random error > **Explanation:** Binary choice models are built for yes-or-no, success-or-failure, and similar two-outcome variables. ### Why are logit and probit often preferred to a linear probability model? - [x] They keep predicted probabilities between 0 and 1 - [ ] They remove the need for data - [ ] They make every coefficient a direct probability change - [ ] They assume no threshold behavior > **Explanation:** Logit and probit respect the bounded nature of probabilities, which is a major advantage over a simple linear specification. ### In many applications, what do economists report after estimating a binary choice model? - [x] Marginal effects or predicted probabilities - [ ] Only the sample mean - [ ] Only the t-statistic for the intercept - [ ] A histogram of the residuals and nothing else > **Explanation:** The practical interpretation usually focuses on how explanatory variables change the probability of the event occurring.