17 Limited Dependent Variables

Many economic outcomes are not continuous and unbounded. Whether a person works, buys a car, defaults on a loan or migrates is a yes-or-no choice. Expenditure on a durable good is zero for most households in a given year and positive for the rest. Hours of work are observed only for those who work. The number of doctor visits or patents is a count. Linear regression handles such variables badly, and this section develops the models designed for them: probit and logit for binary choices, the Tobit model for censored outcomes, Heckman’s correction for sample selection, and Poisson regression for counts. All are estimated by maximum likelihood, using the theory of Section 9.

Questions on this section

Stuck on something here? Ask below and it stays attached to this topic.