2 step least square regression

[ADou]

Hi Experts,

Now I have the following 2SLS model: I have 25 year dummy variables and 48 industry dummies, which are shown yr1-yr24 and ind1-ind47.

proc syslin data=robtest1 2sls first;
endogenous lev;
instruments bcg_mtrnoint logpps size logavg mtb pci yr1-yr24 ind1-ind47;
model wage = logpps lev size logavg mtb pci yr1-yr24 ind1-ind47;
run;

When I run it, the log shows WARNING: The model is not of full rank. Least Squares solutions for parameters are not unique. Certain statistics will be misleading. A reported degree of freedom of 0 or B means the estimate is biased.

And my 1st ad 2nd stage results show "intercept" with DF "0". What does the warning mean and how do I correct my model? Thanks a lot!!!

[Reeza]

For the dummy variables, did you remember to drop one? Ie if you have 25 years, you have 24 dummy variables?

[ADou]

Yes. That's why I have 24 year dummies and 47 industry dummy variables.

[Reeza · Accepted solution]

Unfortunately, that error does mean you likely have two variables that are exactly the same or exactly opposite or some form of a linear combination. You can try dropping sets of variables iteratively until you figure out which ones.

[ADou]

Thank you! I drop all my industry dummies and the warning disappeared. Does that mean among the industry dummies, there are the same or opposite variables?

[Reeza]

It could also mean that industry and year dummies overlap somehow, but I would first start by verifying my industry dummies were created correctly.

[ADou]

Thank you very much! I'll try that.