re: st: Corr2data questions

[David Airey <david.airey@vanderbilt.edu>]

> 1. The corr2data command is handy if, say, there is a published 
> analysis that includes the means, correlations and sds, and you want 
> to replicate or modify the work (e.g. add or drop variables). I do 
> this in some classroom exercises. At the same time, you have to 
> remember that these are not the original data, and you are very 
> limited in what you can do, e.g. you can't analyze subsets of the 
> data, compute interaction terms, etc. All you can do is basic 
> correlational and regression analysis with no modifications of the 
> data (correct?). If I had to invent a term, I would call a data set 
> created by corr2data a pseudo-replication of the original data, but is 
> there a standard term already in use?
>
> 2. Is there any reason the N for the corr2data command has to be the 
> same as in the original data? I did a little experiment where I 
> created a data set with 200,000 cases and ran a regression. I then 
> created a 2nd data set with N = 200 and ran a regression specifying 
> fw=1000. Results were virtually identical. Anyway, if the original 
> data set was monstrous, this might be a way of saving disk space and 
> computing time.
WRT #1, I thought about using corr2data to create data sets that 
violated assumptions in a reliable way, but I did not get very far! I 
could not find examples to help me. I wanted small data sets with the 
variance-covariance structure a certain type, and then to submit these 
to the wrong, ok, good, and best models, to see what happened. Maybe 
the idea was wrongheaded.

An alternative to #2 is just to sample (help sample) from your 
monstrous data set, work up your .do files on the sample, then try the 
.do files on a larger sample or the whole data set.

-Dave

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/