For -svy- prefixed estimation, one usually wants the -subpop- option
documented at -help svy-. E.g.
g byte male=(sex==1) if sex<.
svy, subpop(male): reg BMI age
On Mon, Oct 26, 2009 at 2:12 PM, Nick Cox <[email protected]> wrote:
> I leave all -svy- questions to the savvy, but there's apparently a simple misconception here.
>
> You can get separate regressions by using -by:- as a prefix command. Expecting -by()- and -over()- to work as options is not surprising if you've ingested graphics syntax, but the logic is different for regressions and the syntax certainly is.
>
> by <whatever> : <somecommand>
>
> produces separate results.
>
> <somecommand>, by(<whatever>)
>
> produces composite results. At least it should be something like that.
>
> Nick
> [email protected]
>
> CHEN HSINJEN
>
> My aim is to do stratified regression analysis while accounting for
> the complex sampling design, i.e. to estimate the relationship between
> BMI and age, separately in women and men.
> However, the "reg" and "glm" commands do not allow "over", or even "by", option.
>
> For example, the following statement is not allowed.
>
> svy: reg BMI age, over(sex)
>
> Thus, we may want to solve this by using "if":
>
> svy: reg BMI age if sex == 1 *(for men)
> svy: reg BMI age if sex == 2 *(for women)
>
> But, the problem is, when sex == 2, some strata have only one cluster
> (or some clusters have only one observations.) And then, the S.E. is
> not estimated by stata.
>
> How can I get unbiased estimate of regression coefficient and its S.E.
> under this circumstance?
> Is this solvable in 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/
