Different Results Using Parametric and Nonparametric Tests

[strumpeace]

Hi, all.

I'm stuck on a problem and hope someone can help.

After running a one-way ANOVA on SPSS, I found that the Levene stat showed that the assumption of homogeneity of variance had been violated. So instead of using the ANOVA table, I ran the Welch and Brown-Forsythe tests, both of which showed significance.

I could just leave it at that, but I want to delve a bit deeper.

Does it make sense here to run a nonparametric test (Kruskall-Wallis, to be exact)? I figured that since one of the assumptions of parametric tests had been violated, I should look at the K-W test.

So I ran the K-W test, and it does not show significance. Thus, I have two different tests telling me two different things (parametric test showing significance, nonparametric test showing nonsignificance).

Am I completely misunderstanding the purpose of these tests? If not, how do I reconcile different findings?

[Dason]

What are the sample sizes like?

[krytellan]

As above, listing some more details would help.

Second, I applaud you for being aware enough to know when your assumptions had been violated and taking proper alternative steps. Kudos.

My third thing is not of much specific help to your problem, but worth mentioning. The idea behind the parametric tests needing to meet certain assumptions is that in the absence of these conditions, some tests can become slightly, highly, or completely unreliable. As such, it's not surprising that a nonparametric test might provide a different result than its parametric counterpart. Nonparametrics do not require normality and deal with the exact data at hand, and this may be (not definitely, maybe) why you have two different results. You should be inclined to trust the K-W test, sample size notwithstanding (and even more so perhaps with different sized groups), because it is proposed to be designed for samples situations such as yours.

[strumpeace]

Thanks so much for the quick replies.

The total sample size is 273. One independent variable (categorical), one dependent variable (continuous).