Coverage estimation of bootstrap confidence regions in R2
Minhajuddin, Abu Taher M.
MetadataShow full item record
Confidence sets are constructed in almost any statistical analysis. Both parametric and nonparametric approaches can be taken to construct a confidence set for a parameter [theta]. In the one dimensional case, a disjoint interval is essentially the only form available for a confidence set, and construction of such an interval has been extensively studied. For the multiparameter case, the problem is not that simple, as the question of shape of the set arises and very few alternatives to the normal approximation method are available. One alternative is to use bootstrap theory to construct such confidence sets for parameter vector. These methods are yet to be studied empirically for comparison purposes in terms of coverage accuracy. This thesis represents an effort to focus on the empirical performance of some of the bootstrap methods available for construction of confidence sets for a vector parameter.