tetrachoric correlations with noninterger weights

[Burt Rembrandt]

I need to compute a correlation matrix of binary variables. The tetrachoric command only allows fweights, however, my dataset contains noninteger weights. Does anyone know how to proceed?
Thanks!

[Clyde Schechter]

I don't think you can do this with the -tetrachoric- command. But it might be possible to do this with -cmp- (which is a user-written program you can get from SSC). I'm not sure, but it's worth a try.

[Burt Rembrandt]

Thanks! It works with the cmp command.

[Christina Misunas]

I actually just posted the same question about tetrachoric correlation with non-integer weights. Can you please share your syntax for using the cmp command?

[Markus Stratmann]

I am currently searching for the same answer. Could someone help by sharing how to solve this using the cmp command?

[Joseph Coveney]

I am currently searching for the same answer. Could someone help by sharing how to solve this using the cmp command?

You can use official Stata gsem to compute tetrachoric correlation coefficients and their standard errors. And gsem allows iweights and pweights.

version 19

clear *

quietly webuse familyvalues
keep RS074 RS075
rename *, lower

tetrachoric rs074 rs075

*
* Begin here
*
gsem ///
    (rs074@1 <- F1, probit) ///
    (rs075@1 <- F2, probit), ///
        variance(F1@1 F2@1) ///
        nocnsreport nodvheader nolog

lincom _b[/:cov(F1, F2)] / 2 // <= here

exit