Regression results: omitted standard errors and other statistics

[Ali Taleb]

Hi Everyone,

I am trying to estimate the impact of a policy on an outcome that contains 17% zeros (zeros are true values and cannot be dropped). However, when I want to do a specification with "specific-linear trends" the regression takes more than 1hr to report the results (I have 13484 obs and the size is 95M) but at the end, the standard error and the p-values, t-test, and the confidence intervals show nothing at all (only dots)


the code I am using is the following

#delimit;
xi i.zipcode, noomit;
forvalues i = 1(1)1441{;
generate lineartrend_`i' = _Izipcode_`i' * year;
};
#delimit;
set matsize 10000;
xi: ppmlhdfe Dep_Var Policy i.year lineartrend_* 1 X2 X3 X4...... Xn...., 
absorb(zipcode) vce(cluster zipcode);

Any help is appreciated,

Best
Ali

[Phil Bromiley]

You'll increase your chances of a useful answer by following the FAQ on asking questions. Providing specific code and output (not your half code) helps.

You are running 1440 plus variables plus 1441 fixed effects? So, you're running roughly 3000 parameters on 13000 observations. That is really a bit much - no wonder it it taking so much time. It also explains why you're not getting results. I suspect you can do this with the Stata provided routines for panel poisson. You're estimating too many things relative to your observations. You also don't need the xi in recent versions of Stata.

By the way, for getting things set up, it is sometimes easier and faster to run xtreg or other linear models - they'll illustrate problems and get there more quickly than the non-linear estimator. After you've got it set right, then do the poisson. Whether poisson is the right tool is not clear - you might consider tobit.

[Carlo Lazzaro]

Ali:
as an aside to Phil's helpful comment, please also note that, as per FAQ, you are kindly requested to mention that you used a community-contributed programme, such as -ppmlhdfe- by Correia, Guimarães, Zylkin . Thanks.

[Ali Taleb]

Phil and Carlo,

Thank you for your reply. I will consider your advice.

Best
Ali