The poor English makes this almost unreadable. Please address to make this article useful.

— Preceding unsigned comment added by 130.63.39.82 (talk) 15:31, 27 October 2020 (UTC)[reply]

Good god, in the name of everything useful, would someone please define the variables?!

Antagonistrex 00:26, 26 February 2007 (UTC)[reply]

Under 0.05?[edit]

If I read the article right, if my results turns out to .02 for example, something is "wrong"? Ie. statistically significant? Is that correct? --194.255.112.22 21:38, 31 October 2007 (UTC)[reply]

Jason Crowthers example[edit]

I may be wrong, but isn't levenes test performed using a one way anova of the absolute residual values - not squared residuals? Sureley this will make a huge difference to the final p value. —Preceding unsigned comment added by 130.95.32.127 (talk) 04:48, 26 November 2007 (UTC)[reply]

Distribution[edit]

The text suggests that the test statistic W follows an F-distribution, although nowhere any assumption of normality has been made. So I guess W is only approximately F-distributed. Madyno (talk) 17:51, 9 May 2017 (UTC)[reply]

The current preference is to just use Welch's test[edit]

As the current article says, testing for homogeneity of variance first and using its results to determine whether to use the Student's t or the Welch-Satterthwaite modified t test will change the alpha from what the researcher selected. As demonstrated by Zimmerman (https://pubmed.ncbi.nlm.nih.gov/15171807/), the preferred approach now is to 1) not determine the test used based on an initial test for equal variances but also 2) to always use Welch's t test. You lose very little power, since in the case of equal variances Welch's t test reduces to the original Student's t, but in all cases Welch's test preserves the researchers selected alpha. I will change this article to reflect that. SteveOuellette (talk) 15:32, 3 April 2024 (UTC)[reply]