Georgetown University

Combining Large (but Coarse) Datasets with Fine (but Small) Datasets.

Email address: pjc22@georgetown.edu

Abstract:
Much empirical microeconomics uses data from a single small-sample source, although a larger dataset conditioning on a coarser set of covariates is often available. For example, a cross-sectional sample on the conditional distribution of interest is typically analysed in isolation; but it may be augmented with census data containing nearly exact information on the marginal distribution of the variable of interest. This marginal distribution information allows for sharper inference than that afforded by solely analysing the small-sample dataset. The marginal distribution places nonparametric bounds on the value assumed by a parameter of the conditional distribution. These bounds tighten the classical statistical inference about this parameter. The bounds are deterministic when the sample from the marginal distribution is arbitrarily large (as in a census); for a finitely large sample the bound is stochastic. In either case, the bounds from the marginal distribution information may substantially sharpen inference.

PDF file of paper: Not available.

Session: Econometric Methods I

Time: Saturday, 7 July, 8am - 9:30am

Room: B