Re: st: VEC: t- instead of z-statistics

[Barbara Engels <engels.ba@gmail.com>]

I am just wondering why you mainly find t-statistics in published empirical work. I have a supersmall sample of 30 observations, hence the z-statistics do not seem to be worth a lot to make any statistical inferences. Is there anything I can do besides reporting the z-statistics and pointing to the sample size problem?
 
El 22.07.2011, a las 09:34, John Luke Gallup escribió:

> I am puzzled why -svy: logit- report t-statistics, since -logit- without svy: reports z-statistics.  A mechanical answer may be that the variance of the errors, which is part of the t- and z-statististic formulas, are calculated differently in each case.  The variance, sigma squared, is calculated with a factor of 1 / (N-K) when calculating a t-statistic, but sigma squared is often calculated with a factor of 1 / N for a z-statistic.  The two formulas for sigma squared are both consistent, but the first is unbiased.
> 
> The main point is that there is no good reason to think that a t-statistic is better than a z-statistic unless you are using a linear estimator and you believe that your errors are actually normal.  Otherwise you only know the asymptotic properties of the estimator. The t-statistic and the z-statistic are both asymptotically normal, but have an unknown small sample distribution.
> 
> John
> *
> *   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/