weighting in gsem

[Janka Vanschoenwinkel]

I estimate the following model

irrigation_dummy = croptype_dummy + other variables
croptype_dummy = irrigation_dummy + other variables

• irrigation_dummy has 2 categories (0 and 1)
• croptype_dummy has multiple categories (1, 2, 3 ...)

In STATA I type

gsem (irrigation_dummy0 <- ts1 ts2 ts3 ts4 P1[newid]@1  , logit)(croptype_dummy <- ts1 ts2 ts3 ts4    P1[newid]@1 , mlogit) [pweight=SYS02]

However, there is something wrong with the weigthing part. The simultaneous system that I estimate should be weighted, but I get the following error message:

Fitting full model:

 mopt__check_by_weight():  3204  <tmp>[0,0] found where scalar required
        mopt__validate():     -  function returned error
            _moptimize():     -  function returned error
           Mopt_maxmin():     -  function returned error
                 <istmt>:     -  function returned error
r(3204);

The weights tell me how many observations one observation in my sample represent.

There are no zeros in the weighing variable, so I do not know what is going wrong.

Any suggestions?

Thank you very much!

[Marcos Almeida]

Hello Janka,

The weights tell me how many observations one observation in my sample represent.

I gather you should use fweights instead of pweights.

[Richard Williams]

I suspect errors like that show up when Stata hasn't really anticipated this particular problem, so you just get a message that isn't very helpful.

Like Marcos says, maybe you should be using fweights anyway. If fweights are appropriate but problems persist you could expand the data set (see -help expand-).

You may want to run the model without weights just to confirm that the weighting is causing the problem.

You can try simplifying the model, doing one equation at a time.