# Re: st: weights in multivariate analysis

 From Ernest Berkhout <[email protected]> To [email protected], "'[email protected]'" <[email protected]> Subject Re: st: weights in multivariate analysis Date Tue, 30 Mar 2004 14:06:15 +0200

```At 11:34 30/03/2004, Koeppen, Katja wrote:

```
```I have a question concerning the use of weights in multivariate analysis.
Because of the oversampling of a particular subgroup in my data set, I had
to use a given weighting factor. I estimated a multiplicative
intensity-regression model. I know that I cannot say anything about my
t-statistic now, because the weighted number of cases is different than the
unweighted number of cases. Tests of significance are not valid anymore
because the standard errors are biased. I know that there are procedures in
SAS and SPSS to normalize the weights. I don't know about any function inn
STATA that would allow to adjust the weights. That's why I adjusted the
weighting factor by dividing the weight variable by the mean of the weight
variable. The relative values of the weights do not change but they are
adjusted so that the mean weight in the sample is 1, and the sum of weights
equals the number of cases. Now, I think, the p-values can be used for
interpretation because they are no longer biased. Is this procedure correct?
Or is there any other method how to adjust weights in STATA? And does
anybody know about some literature concerning this topic?
```
Katja,

have you already checked the manual entry on weights, especially [U] 23.16? You might find some Stata explanation in there. It looks as if you're procedure is correct, although in Stata it can be accomplished even more easily by using analytic weights. Maybe the following link can be helpful as well: http://www.stata.com/support/faqs/stat/supweight.html

Ernest Berkhout
SEO Amsterdam Economics
University of Amsterdam

Room 3.08
1018 WB Amsterdam
The Netherlands

tel.:+ 31 20 525 1657
fax:+ 31 20 525 1686
http://www.seo.nl
===========================
A statistician: someone who insists