Re: st: Chi-Square produces consistent Type I error for large sample

[Steve Samuels <sjsamuels@gmail.com>]

Maria,

I agree with Richard. 

That said, I suspect that you are using the Uncorrected chi square, which is known to reject too often. Use the Design-based F p-value instead. Also, if you have -svyset- your data incorrectly (with the wrong PSU, for example), this could also lead to spuriously small p-values.


If you want further help I suggest that you follow the FAQ guideines (show exactly what you typed and what Stata did): describe your data;  show us the -svyset-  statement; the results of -svydes-; the -svy: tab-statement and results for an association that appears to be "too significant".


Steve


On Nov 29, 2011, at 5:55 PM, Davydenko, Maria wrote:

Hello all,

I'm having trouble interpreting the output from Stata's survey-based
tab/chi-square analysis. I have a panel data sample of ~33,000
observations and am trying to test for variation in one categorical
var due to another categorical var. As far as I know, the best way to
do this is using chi-square. Here's the trouble: everything I test
using svy:tab comes up significant to a p-level of 0.004 or lower. If
I randomly select a smaller sample size, the p-value gradually gets
larger.

Is there something I'm missing here? Any help would be much appreciated.

Thanks!

Maria

-- 
Maria Davydenko
Master in Public Policy Candidate 2013
Harvard Kennedy School
+1 (907) 301-5339 | maria_davydenko@hks13.harvard.edu

www.givology.org
Give to learn, learn to give
*
*   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/


*
*   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/