[EMAIL PROTECTED] (Herman Rubin) wrote in message news:<[EMAIL PROTECTED]>... > In article <[EMAIL PROTECTED]>, > Glen <[EMAIL PROTECTED]> wrote: > >I assume that what [EMAIL PROTECTED] (seenu) intended to ask was: > >> Is pearson's chi-square for testing independence in two-way > >> contingency tables a non-parametric statistical test? > > >The answer depends on who you ask, or on how you define > >"nonparametric". (Some definitions will exclude it.) > > >It is widely, but not universally, regarded as one; many > >textbooks on nonparametric statistics include it. > > This is a poor way of putting it. A non-parametric test, > more precisely a distribution-free test, is one where > the test properties are unchanged under reasonable changes, > usually taken to be continuous transformations, of the > variables. This is not possible in two-way tables. > > For two-way contingency tables, the marginals definitely > affect the properties of the test for independence.
Herman has obliged me by posting almost exactly what I expected him to. If you take the definition of 'nonparametric test' as one where "the test properties are unchanged under continuous transformations", then the answer is as Herman indicates. This is a fairly sensible definition. This is why I said "some definitions will exclude it". Nonetheless, many books on nonparametrics /include/ it, and a few even attempt to provide a justification for doing so. So, as I pointed out, it depends on who you ask, and how they define it. Glen . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
