spxtregress with unbalanced panel data

[Maria Ribeiro]

Dear Statalist members,

How would be possible to use spxtregress with unbalanced panel data?

I'm currently using a sample of firms for 3 years, with the coordinates for each one, and without assuming a shapefile. However from one year to the other, there are certain firms that enter in the sample, and others that exit. According with STATA 15 material, I should consider the last year matrix since it is assumed that location is the same in both years.

My problem is that imposing a balanced panel of firms that exist in the 3 years lead to a reduction of 20% in my sample, which is in fact relevant.

Thank you in advance.

[Sebastian Kripfganz]

Unfortunately, the answer is: You cannot use spxtregress with an unbalanced panel because it would require a different spatial weights matrix for every year. As far as I know there is also no other Stata command that would be a feasible alternative.

There are also more theoretical concerns about unbalanced panels due to sample selection. In such spatial panels, this raises the question how to standardize the weights matrices for the different years in a consistent way such that the weights remain comparable across years.

[Greg Worthington]

I've read that, theoretically, you can use an NT x NT matrix in which there's a block diagonal with each block in the diagonal being the weighting matrix for a given year and it's spatial units. However, you for sure can't use it if your firm while absent from the data. I emailed the contact for this paper and they informed me concerning my data which involves a whole population of schools in a city and it's surrounding area. Schools open up and close during the brief time period in my study, which means they're not actually "missing" since they were non-existent during the times in which they weren't in the data. It's also important that the outcomes of your model not be related to the data in your matrix in such a fashion that the outcomes can affect the data matrix values.