4
$\begingroup$

I am writing a paper comparing quantitative values across $5$ groups. I have performed an ANOVA in SPSS 29 and have requested from the software the Levene statistic to judge if there is homogeneity in variance between the $5$ groups.

The software has reported a Levene statistic of means, medians, a test based on means adjusted with df, and the trimmed mean.

How do I know which Levene statistic to use? Some are significant and some aren't, so it would alter which post-hoc tests I use. Do I have to run a Kolmogorov-Smirnov test first to know if my data is skewed, or is there a quicker way? And what do trimmed mean and adjusted with df mean?

$\endgroup$
1
  • 2
    $\begingroup$ A piece of advice: If software is going to offer you more than one statistic (which any good help should indicate and which you will be sure to have read before using the function), choose which one you plan to use before you see the output. Making the choice after you see output is problematic, given that the choice may be influenced by the results, or may appear to others to have been influenced by them. Either way, that's less than ideal. $\endgroup$
    – Glen_b
    Commented Dec 17, 2023 at 9:03

1 Answer 1

5
$\begingroup$

None of them? Homogeneity of variance usually doesn't matter, and if there's any concern, it's easy to deal with.

If you do a test that does not assume homogeneity of variance, you get the 'right' answer, whether you violated it or not. This is my favorite paper on this: https://bpspsychub.onlinelibrary.wiley.com/doi/abs/10.1348/000711004849222.

Your questions on normality have come up a lot on this site. Short answer: You almost certainly shouldn't do them.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.