[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: RE: Wech One-way ANOVA

From   Steven Samuels <>
Subject   Re: st: RE: Wech One-way ANOVA
Date   Sun, 18 May 2008 21:47:36 -0400

David is partly right.

Welch's test uses separate standard deviations to compute the denominator of the t-statistic, and, as David says, this is the same denominator as that computed by the robust option in -reg-. However - reg- with the robust option uses the usual d.f. for the test (n-2), but Welch used a corrected d.f.



Welch BL. 1938. The significance of the difference between two means when the population variances are unequal. Biometrika 29:350–62.

Available for download at: reprint/29/3-4/350? ijkey=aba51fa995b725950195c70c7157f8c706bb62ad&keytype2=tf_ipsecsha

Graeme D. Ruxton The unequal variance t-test is an underused alternative to Student's t-test and the Mann–Whitney U test Behavioral Ecology 2006 17(4):688-690
available for download at: content/full/17/4/688#FD4

On May 16, 2008, at 5:48 PM, Verkuilen, Jay wrote:

I'm wondering if anybody knows a welch option in ANOVA. Stata runs a welch test in ttest. But, I was not able to find the same option in oneway or anova.<<
I discovered an interesting little fact a few months ago, which is in retrospect not so surprising:

If you set up the independent samples t-test as a dummy variable regression and use robust standard errors, you get the Welch- corrected t-test. (Try a test case yourself to verify; it's been a long year, I may be misremembering.) So you could always set up your one-way ANOVA as the equivalent dummy variable regression (using xi or, better, xi3) and use robust.


*   For searches and help try:

© Copyright 1996–2015 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index