Bootstrap with Clustering in Two or More Dimensions

Abstract

We propose a bootstrap procedure for data that may exhibit clustering in two or more dimensions. We use insights from the theory of generalized U-statistics to analyze the large-sample properties of statistics that are sample averages from the observations pooled across clusters. The asymptotic distribution of these statistics may be non-standard if there is no clustering in means. We show that the proposed bootstrap procedure is (a) point-wise consistent for any fixed data-generating process (DGP), (b) uniformly consistent if we exclude the case of clustering without clustering in means, and (c) provides refinements for any DGP such that the limiting distribution is Gaussian.