CFACT Seminar


Wilcoxon, It's Not Just for i.i.d. Anymore



Presented by

Robert M. Baskin, CFACT

Date/Location
Monday, September 22, 2014
10:30 a.m. – Noon
CFACT 5th Floor Conference Room



Abstract

In September 2013 a seminal paper by Thomas Lumley (created R survey) and Alastair Scott (Rao-Scott correction), entitled “Two-sample1 rank tests under complex sampling”, changed the decades old question of “Why can’t you do standard nonparametric rank sum tests with complex survey data?” to “What software is available to do nonparametric rank sum tests with complex survey data?”. In this presentation from the bleeding edge of survey statistics research, a short introduction to Wilcoxon-Mann-Whitney tests (WMW) will be given. WMW will be compared to standard t-tests, both for i.i.d. data. Some insight as to why it has been so difficult to deal with rank-sum tests in the setting of complex surveys will be attempted although this part is fairly technical. Examples of WMW with MEPS data using the R survey package will be given. The presentation will finish with some open questions for research in this area.

1 -sample, in the traditional language of nonparametric statistics, refers to what survey statisticians would call two ‘domains’.