Understanding type of panel

[Kapoor Manisha]

Hi everyone,

I have individual level data for 20 regions for 18 years. The question is to examine the impact of x on y (binary variable). However, there is no individual id. Basically the data looks like this:

region	year	x	y	age	gender
1	1993	20	1	20	1
1	1993	26	0	25	1
1	1993	12	1	40	1
1	1994	13	1	21	0
1	1994	20	1	30	1
2	1993	25	0	25	1

It is not the same individuals that are tracked every year. My questions are:

• Is this still a panel despite me not knowing anything about individuals?
• I have run a regression using the following command: logit y x gender age i.region i.year, vce (cluster region) Is this the correct way to include region and year fixed effects?
• Should I define it to be a panel data using xtset and then run the xtreg command. I read in a different post that when you have multiple observations under a particular region and year that might not be the right way.
• What if I want to run a non-parametric regression on this? Will having dependent variable and some of the controls are binary impact anything?

[Jeff Wooldridge]

This isn't a panel data set. It's a pooled cross section. You can turn it into a "pseudo" panel by computing averages within each (region,year) pair, but there's no real need for that. What you have done is correct except you might not need to cluster at the regional level because your key variables -- gender and age -- vary mostly at the individual level.

BInary y causes no problems for nonnparametric regression but discrete explanatory variables do. In fact, all of your variables are pretty discrete. I doubt nonparametric will do much for you. Put age^2 and maybe interact age and gender, and maybe include year interactions.

JW