Jones, M. C.
|DOI (Digital Object Identifier) Link:||http://dx.doi.org/10.1016/0167-7152(94)00183-9|
|Google Scholar:||Look up in Google Scholar|
In two recent papers in this journal, Kanazawa (1993a, b) has described asymptotic equivalences between smoothing parameters minimising certain risk functions for histograms and kernel density estimators, respectively. In the current note, we (i) describe some well-known heuristic equivalences (known also to Kanazawa, but not mentioned in either paper) which explain why results are as they are, and (ii) put forward the more usual opposing views to Kanazawa's claim that his results “force us to entertain the possibility that the [Hellinger distance], not the conventional [integrated squared error], be the standard distance-based measure of discrepancy in density estimation”.
|Item Type:||Journal Article|
|Copyright Holders:||1995 Elsevier Science B. V.|
|Keywords:||Akaike's information criterion; Hellinger distance; histogram; Kernel density estimator; Kullback-Liebler loss; weighted integrated squared error|
|Academic Unit/Department:||Mathematics, Computing and Technology > Mathematics and Statistics
Mathematics, Computing and Technology
|Depositing User:||Sarah Frain|
|Date Deposited:||19 Apr 2011 13:52|
|Last Modified:||18 Jan 2016 10:05|
|Share this page:|