Using pweight with pwcorr

[Tom Prosser]

Dear forum,
I'm using Stata 17 and working with several surveys of individual US and UK voters. Normally when working with survey data, one would use the pweight option. With regressions, I do this. However, pwcorr won't accept pweight and will only allow aweight. Is using aweight an acceptable alternative? In much older advice, I read that aweight and pweight yield the same point estimates for correlations. However, I worry that things have changed with newer versions of Stata. Any advice would be appreciated immensely,
Tom

[Jeff Pitblado (StataCorp)]

The FAQ How can I estimate correlation coefficients and their p-values for complex survey data? might be of interest for you. It depends on what you want to do with the correlations, such as testing against zero, and if you care about missing value patterns in your data.

Since you have Stata 17, you might also be interested in the FAQ How to build customizable tables of correlation coefficients for complex survey data?.

[Tom Prosser]

Thanks Jeff, it's very good of you to reply and I'll read the links. Primarily, I'm interested in the strength/direction of relationships, rather than testing against zero. I'm interested in general trends and not extremely concerned about missing value patterns. What difference do these goals make? Best and thanks, Tom

[Tom Prosser]

The FAQ How can I estimate correlation coefficients and their p-values for complex survey data? might be of interest for you. It depends on what you want to do with the correlations, such as testing against zero, and if you care about missing value patterns in your data.

Since you have Stata 17, you might also be interested in the FAQ How to build customizable tables of correlation coefficients for complex survey data?.

Hi Jeff,
I tried the svy: sem command in the article and found that bivariate correlations are the same as the result for pwcorr using aweight. But when I include three or more variables in the svy: sem command, the results become different. Why is this? Do they become partial covariances, i.e. controlling for the other variables in the model?
Best and thanks,
Tom

[Tom Prosser]

Hi Jeff,
I tried the svy: sem command in the article and found that bivariate correlations are the same as the result for pwcorr using aweight. But when I include three or more variables in the svy: sem command, the results become different. Why is this? Do they become partial covariances, i.e. controlling for the other variables in the model?
Best and thanks,
Tom

UPDATE - Apologies, I now see that the difference reflects listwise v pairwise deletion.