Statalist


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

Re: Re: st: R: ttest and log transformation


From   "Carlo Lazzaro" <carlo.lazzaro@tiscalinet.it>
To   <statalist@hsphsun2.harvard.edu>
Subject   Re: Re: st: R: ttest and log transformation
Date   Sun, 28 Sep 2008 15:31:26 +0200

Dear Rich,
Your Stata syntax is definitely too advanced for me so I shy away to judge
it over and I take the liberty to refer you to the one sketched in my
yesterday's example.
As far as r(t) bootstrapping is concerned, the only remark that seems to me
noteworthy to add is to prepare your pre-bootstrap data in order to make the
compared samples have equal means (please, see my yesterday' example and
Stata Manual (my release is) 9.2 [R] A-J -bootstrap-), whereas, as stated by
Efron B, Tibshirani RJ. An Introduction to the Bootstrap. New York: Chapman
& Hall, 1993: 223 "there is non compelling reason to assume equal variances"
in "considering a bootstrap hypothesis test form comparing the two means (in
the example the mean survival times for 7 treated mice vs 9 untreated mice
were compared (pag. 11 of the cited textbook); hence, the small sample issue
should be properly tackled via the bootstrap procedure). Some skewness
problems may arise with the bootstrap CIs (particularly with the percentile
ones)of your bootstrap estimate: however, Stata can offer you 3 other
different CIs to minimize this drawback (please, see -bootstrap- and
-bootstrap postestimation- in Stata Manual [R] A-J.

Kind Regards and enjoy your Sunday, 

Carlo


-----Messaggio originale-----
Da: owner-statalist@hsphsun2.harvard.edu
[mailto:owner-statalist@hsphsun2.harvard.edu] Per conto di Richard Harvey
Inviato: domenica 28 settembre 2008 13.33
A: statalist@hsphsun2.harvard.edu
Oggetto: Re: st: R: ttest and log transformation

Hi ,

Carlo..thanks for your reply. My main problem is the skewness and small
sample size. In the summary stats I posted the N is large as it is for the
whole sample  but when I analyse subsamples there are some every small
samples. i.e less than 20.

The bootstrap seems like a good idea.  Can I do something  as simple as

bootstrap r(t) reps(1000) saving(c:\), ttest var, by(catvar)  unpaired
unequal

or is it  something more involved as below?

bootstrap r(mean) if catvar=="cat1", reps(1000):sum var matrix mu_1=e(b)
matrix sterrsq_1=e(V) bootstrap r(mean) if catvar=="cat2", reps(1000):sum
var matrix mu_2=e(b) matrix sterrsq_2=e(V) scalar Z=((mu_1[1,1]-
mu_2[1,1])/sqrt(sterrsq_1[1,1]+ sterrsq_2[1,1])) scalar
p=(1-normal(abs(z)))*2 di "z-value: "[Z] di "p = "[p]

thanks very much for your help

regards
rich

<<the previous thread has been snipped due to a C++ buffer overrun warning
signal that appeared when Carlo tried to reply. Sorry for that>>


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index