Time-fixed effects regression: groups are omitted

[Tim Vodegel]

Hi, I am trying to do a time-fixed effects regression using xtreg. Although I have data from 48 countries in my dataset (and 9 years of data per country), only 18 of those are used in the actual regression (as displayed in: number of groups). I noticed that this probably has to do with the control variables which often include some missing values.

With so few countries I will never get any significant coefficients in my regression. Therefore I have 2 questions:

1. Why does Stata only take 18 of the 48 countries into consideration?

2. How can I get the other countries to be included as well? Does that mean that there cannot be missing values in any of the variables?

The regression I am trying to do is:
xtreg gdpgr lreg_assetshare lgdppc gov linfl popgr school ltrade lunempl lcredit life ldebt legal i.year, fe

I included a link to my dataset: https://www.dropbox.com/s/4ais1ra4mi...02.3%20v12.dta

Thanks in advance, some advice on this would help me a lot!

[FernandoRios]

Hi Vodegel,
First of all, for future questions dont forget to register with your real name. its part of the Forum policy.
Second, Stata, as any other statistical program I know, will only use observations for which there is full information across the variables of interest. Otherwise, what can it do with missing values? Assume there is a latent variable? Impute it? Assign the mean?. There is certainly ways to go around missing information in order not to loose all the data, however you need to decide and include any of those techniques before estimating the Panel model.
In your specific model. School and ldebt are the ones with the largest share of missing data. Dropping them from your model increases the sample to include 33 countries. Whether this variables are substantial or not for your research question is something you have to decide, as well as what to do with the missing data.
Hope this helps
Fernando