I am currently working on a research by comparing two countries. For each country I have 20 years of data. If I merge the two countries to do regression, could it give me a legitimate panel data? In other words, can I have a panel data with two countries and 20 years?
In principle, I think that I would include a country dummy and I would use time-series techniques... Given that you have two countries, you cannot use clustered standard errors (the most straightforward way of dealing with serial correlation). If you have the suspect that the data could be AR(1) or similar (eg, your dependent variable has been used somewhere else and people reports that) you could use the Newey-West estimator (available in Stata, for instance).
I hope it helps. I used this approach here:
Article The impact of the legalisation of abortion on birth outcomes...
I think that having just 2 units x 20 obs each one could rais some estimation nuances. Nevertheless, and depending on the nature and other characteristics of your data, you could follow the suggestion of Jose-Ignacio. I also would run the regression as a panel data with fixed effects. But in the end, what will sign you the best path is the data and kind of variables with you are working.
Please look at the comments that others gave on a similar question.
https://www.researchgate.net/post/What_is_the_ideal_size_for_study_in_a_panel_data#:~:text=Some%20researchers%20tell%20us%20that,%2D200%20observations%20(cases).
I would think that this is more appropriate for a SUR type analysis. Theory may allow you to make some cross country constraints on the two equations. Make your model as parsimonious as theory will allow.
- Badges
- Science topic
- Similar topics
- Social Psychology
- Attitude