Abstract:
|
Interest in clustered data has risen sharply in empirical economics in the past 15 years or so, partly because a popular computer package has made it easy to compute "clustered" standard errors for regression models estimated by either least squares or instrumental variables. In many fields, where independence assumptions are hard to justify, it is now customary to use these "clustered" standard errors rather than the "heteroskedasticity-robust" ones that have been widely used since the 1980s. This is particularly important when sample sizes, and cluster sizes, are large. In such cases, inference based on heteroskedasticity-robust standard errors, or ones that are clustered at too fine a level, can be extremely misleading. Some recent developments in econometrics include: studies of several forms of the wild bootstrap and wild cluster bootstrap; attempts to deal with the severe inferential problems that can arise when all of the "treated" observations belong to just a few clusters; advances in the asymptotic theory of cluster-robust inference; methods for testing the appropriate level of clustering; and methods for dealing with two-way and multi-way clustering.
|