Partial Identification in Monotone Binary Models : Discrete Regressors and Interval Data

We investigate inference in semi-parametric binary regression models, y = 1(x¯ +v+² > 0) when ² is assumed uncorrelated with a set of instruments z, ² is independentof v conditionally on x and z, and the conditional support of ² is su¢ciently smallrelative to the support of v. We characterize the set of observationally equivalentparameters ¯ when interval data only are available on v or when v is discrete. Whenthere exist as many instruments z as variables x, the sets within which lie the scalarcomponents ¯k of parameter ¯ can be estimated by simple linear regressions. Also, inthe case of interval data, it is shown that additional information on the distribution ofv within intervals shrinks the identification set. Namely, the closer to uniformity thedistribution of v is, the smaller the identification set is. Point identification is achievedif and only if v is uniform within intervals.