[rkward-devel] wilcoxon tests

[Stefan Rödiger]

On Monday 12 December 2011 19:47:56 Thomas Friedrichsmeier wrote:
> Hi,
> 
> On Monday 12 December 2011, meik michalke wrote:
> > looking at the details sections of both ?wilcox.test and
> > ?exactRankTests:::wilcox.exact, the two functions used by the dialogs
> > "Wilcoxon Test" and "Wilcoxon Exact Test", they seem to differ only in
> > the handling of ties. may i merge the two into one dialog? i find the
> > current split into two separate dialogs a little confusing, given that
> > both do exact tests under certain conditions. both functions have almost
> > identical signatures and both return an object of class htest.
> 
> sounds reasonable to me. Stefan, you created those plugins. Do you agree,
> too?

No objection to that

> 
> > btw, i also added support for tests against costants to the t-test
> > plugin, is that ok?
> 
> This looks ok to me for the moment. I corrected a typo, and the initial
> value of the constant.
> 
> However in the long run, there is a bunch more features that we should add:
> - Testing an arbitrary number of variables against each other or against a
> constant
> - Support for data in long format (single outcome variable and grouping
> factor). Using all levels or selected levels of the grouping factor(*).
> - Various modes of alpha correction
> 
> I think having all of that inside a single plugin will be too much, so we
> will have to split all that into two or more plugins. Once at that point,
> I think it may make sense to restore one "very simple" t-test plugin,
> which may or may not look like the one without support for testing against
> constants.
> 
> BTW, while testing your addition, I noted some more details that could be
> improved:
> - "Paired sample" and "Assume equal variances" should probably be on the
> same tab. Having mutually exclusive options on different tabs has some
> potential for confusion.
> - "Print confidence interval" should be a checkable frame, instead of a
> checkbox.
> Should I go ahead and do that, or are you still working on the plugin?
> 
> Regards
> Thomas
> 
> (*) And I know, that's another use case for a <valueselector>.