Statalist The Stata Listserver

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

Re: st: Mean Test

From   Richard Goldstein <>
Subject   Re: st: Mean Test
Date   Sat, 21 Jan 2006 12:05:36 -0500

To me the logical flaw in your suggested procedure (using the results of a test of variances to decide which version of the t-test to use) is that you are taking a probabalistic result and treating it as though it were absolute. In general, you can find literature on this flaw under the name "pretest estimator bias".

Rich Goldstein

Rijo John wrote:

I guess there is no concensus on using ttest with the welch's
correction when the variances are different. But I have been seeing
(at least so far) people using it like that. Now I dont know what to
do.  As I understand, when the variances are unequal one should do a
"Satterthwaite's Approximation F-test". The SAS's T-test procedure
reports both the equal variance and unequal variance cases in one
t-test command itself. But in stata (I guess) we need to explicitly
state the unequal option in the ttest. Which then essentially means
that we have to first test for the differences in variance and then
accordingly decide whether to put the unequal option or not. Then the
only difference that I see from my first post is using
"Satterthwaite's Approximation F-test" instead of welch's correction.

It would be nice if someone could kindly clarify this for me. If I am
right here can someone tell me how to recode my initial program so as
to incorporate three categories?


Rijo.M.John,Research Scholar
Indira Gandhi Institute of Development Research,
Film City Road, Goregaon East,
Mumbai, India-400065.
contact: (+91)9892412476

*   For searches and help try:

© Copyright 1996–2021 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index