What type of model to use for TSCS

[Karel Novak]

Dear all,

I would like to ask you a question: What type of regression model do you think is the most appropriate to use for my type of data?

The dataset I work with is TSCS. Cases in the dataset are elections in a given country in a given year (for example: Bulgaria 2001; Bulgaria 2005; Denmark 1994) and the data are unbalanced and with gaps. The number of groups (a group is a country) is larger than number of T (number of obs is around 175, number of groups is 30 and average obs. per group is about 5.7).

I've read so far number of articles in my field of study where authors were using same or very similiar type of data, but unfortunately there is no consensus.

So far I've came up with this: -xtreg IV DV DV DV, fe vce(robust)-, since I have strong assumption about autocorrelation and unit heterogeneity being present. Yet because of the character of the data, I am currently not able to test the autocorrelation via xtserial. I've also seen authors using -regress IV DV DV DV, vce(cluster)- where there is no autocorrelation but heteroscedasticity remains. The problem is that when I compare those two commands the results are quite different, and I have no robust criteria which can me help to decide.

Thank you very much for any help.

[Phil Bromiley]

You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.
If you believe the intercepts differ across panels, then xtreg or xtregar is the way to go. You can use robust standard errors that take care of serial correlation in xtreg. I suspect most election behaviors have some strong stable national differences.

[Phil Bromiley]

Note that xtreg automatically gives you a test whether the intercepts (fixed or random effects) differ.