jim clark wrote: > > Hi > > On 18 Nov 2003, Sharon Lane-Getaz wrote: > > > I have ordinal data that I collected by grading a statistics > > project based on an established rubric. The data is being > > used to measure students' increased conceptual understanding. > > Using the inappropriate t-tests to measure a difference in > > means from year to year, the data show statistically > > stignificant increases. > > > However, a nonparametric approach would be more appropriate > > (and more conservative.) Can someone point me in the > > direction of a non parametric test that could be conducted on > > Occasionally (i.e., every decade or two) the debate over > parametric statistics and level of measurement emerges from the > depths, where it should stay buried. Use your parametric stats > and hope that you don't have to run it by someone misled by > Stevens' long-ago misguided paper on the issue. If you want some > justification, search for references to John Gaito.
Agreed. The bottom line, in my experience is that if the assumptions of the parametric procedure are reasonable (i.e., there is no evidence that there are serious violations) then the standard parametric tests should be preferred. If not, so-called non-parametric tests may be suitable alternatives (especially for small samples), but other options such as transformations or other parametric models should also be considered. * Interval level of measurement is not an assumption of the tests - for example Stevens' work post-dates the Student t. I also think that the levels of measurement issue conflates at least two different issues leading to major confusion. Thom . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
