Region dummies in fixed effects panel analysis

[Sarah Thompson]

Hello, I am very new to both statalist and statistics in general, so I apologise in advance for my very elementary question!

I am working on a dissertation focused on the impacts of financial remittances by using panel data for ~160 developing countries over a nearly 30 year period.

I am following the design of Azizi (2018) quite closely (particularly his fixed effects models), however, do not understand some of his steps. For example, he uses region rather than country dummies (because many countries only have one year of data available, meaning they would get dropped if using country-specific effects), but I'm not sure how to actually do this.

I first thought that meant using "xtset region time" rather than "xtset country time". However, this didn't work (error message = repeated time values in panel). The other thing I tried was making dummy variables for each of the six regions I'm studying (e.g., Sub-Saharan Africa; Latin America and the Caribbean, etc.) and including each of these as variables in the following regression:

. xtreg [dependent variable] Var1 Var2 Var3 Var4 Var5 europeasia latinamcarib mideastnorthaf northamerica southasia subsahafrica asiapacific, fe

However, this didn't work either as it omitted all the region variables for collinearity.

Any suggestions on how I can account for these region dummies?

Note: this is actually one of MANY methodology questions I have as I work on my dissertation so if there is some kind soul that would be willing to chat a little bit more with me to help me figure things out, it would be greatly appreciated!!!

[Jeff Wooldridge]

Sarah: Azizi included regional dummies which means no country-level dummies. So just use the -regress- command and include the regional dummies. Be sure to cluster your standard errors at the country level to ensure no serial correlation. Generically,

reg y i.region x1 ... xk, vce(cluster countryid)

There should also be some control for aggregate time effects. The most flexible is to include a full set of year dummies:

reg y i.region i.year x1 ... xk, vce(cluster countryid)

I should add that I'd be a bit suspicious of an analysis that does not use country fixed effects. Countries with a single year essentially contain no information unless you assume the country heterogeneity is uncorrelated with everything in the model -- which is what Azizi does.