www.cmm.bristol.ac.uk/forum

Got it. Thanks so much.
Diego

Chris,

One more question: Can I get first-stage results from either your .ado code or the ivmlwin macro? I want to report my F statistic and make sure the instrument(s) chosen is(are) good. Maybe there's a postestimation command in runmlwin that will give me what I need.

Diego

Hi Diego,

I don't think that is implemented currently.

However, the first stage estimates will simply be the same as ones you would obtain in a standard IV analysis using the -regress- command.

I am guessing that the F statistic relates to testing whether the instrument(s) have sufficient explanatory power

I suppose a potential problem that you will run into here is that the first stage regression has ignored the dependency in the data and so your F statistic will have spurious precision?

Best wishes

George

Dear George,

I am trying to use two instruments x1w and x2w for the endogenous variable yw.

The response variable say y depends on the endogenous variable yw and other two regressors x1 and x2 (not instrumented)

If I give the command

iv cons x1w x2w x1 x2: runmlwin y cons yw x1 x2, level1(l1id: cons) ...

It will give the error "Variable lists do not match". The stata iv.ado file accepts only one instrument per instrumented variable whereas I understood that the original mlwin file was accepting more than one instrument.

Finally, is it correct to list in the variable set after iv also the non instrumented controls x1 and x2?

Thanks

The iv.ado file is a direct translation of the MLwiN macro, which as far as I could see only allowed a one-to-one correspondence between the variables. If you are aware of a way of setting up more than one instrument per instrumental variable in the original MLwiN macro I would be interested in seeing how this is done and could try to adapt the Stata ado version to match. The ado version is entirely in human-readable form, so anyone is of course free to make adaptations themselves.

Yes, it is correct to include the non-instrumented variables after iv. The length of variable list after iv must match the number of predictors after the runmlwin command, and the corresponding positions determine which variables are matched to each other.