Interpretting Levene's test (Stata)

Lukan27

New Member
#1
I looked this up quite a bit, and found alot about it, but I'm still not entirely sure if I got it right. I just recently got an assignment back with some needed corrections; this one is about Levene's test and t-test. My lecturer wants me to do a Levene's test (robvar in Stata), and I get these results:

Code:
W0  =  3.1939829   df(1, 757)     Pr > F = 0.07430973

W50 =  3.2780661   df(1, 757)     Pr > F = 0.07060797

W10 =  3.3437674   df(1, 757)     Pr > F = 0.06785331
As I have understood so far, the higher the p-value (Pr > F in Stata), the higher the odds that observations will deviate from the total (or mean?) (~3.27). My lecturer only wants me to do a t-test if it passes the Levene test, of course. So, if I only accept 95% confidence intervals (0.05), it would fail the Levene test (just barely), right? Indicating just a little too much variance?

I've read that you need to look for W0, but what about the others, what do they mean?
 
Last edited:

Lukan27

New Member
#2
For future Googler's, I can now verify that this Levene's test actually passes. If p < 0.05 (95% CI), then it fails. So the higher the value, the more equal the variance.

W50 and W10 still makes no sense on my behalf.
 

Dason

Ambassador to the humans
#3
For future Googler's, I can now verify that this Levene's test actually passes. If p < 0.05 (95% CI), then it fails. So the higher the value, the more equal the variance.
Well that's not entirely correct. The null hypothesis is that the variances are equal. A higher p-value means you have less evidence against the null but that doesn't mean that the null is true.

I have no idea what the different W__ stand for though.

One thing I'm wondering about is what your sample sizes look like. Your denominator degrees of freedom seem to be large enough that it really doesn't matter if the variances are that equal and using a Welch correction would more than alleviate any problems.
 

Lukan27

New Member
#4
I see.. I better get a better grip at this before writing anything more. :shakehead

The sample size is 759.

A higher p-value means you have less evidence against the null but that doesn't mean that the null is true.
Will this count for a chi-square test too? Say, if P > 0.05, then it does not neccesarily mean it's dependent (if say, 0.130)?
 
Last edited:

Dason

Ambassador to the humans
#5
I'm assuming you're talking about a chi-square test of independence and it sounds like you have it mixed up actually. Typically we take the null to be that there is independence so if you fail to reject then you don't have evidence of dependence. That doesn't mean that the random variables are independent - just that we don't have evidence that they aren't.

No for most cases, especially Levene's, if you fail to reject the null people just take that to mean that the null is true. This isn't actually the case (like I pointed out) but for the sake of an analysis it's not the worst thing in the world because you have to make assumptions somewhere.