Hello,
I am regressing an equation specified as follows
GDP per capita_it= bo + b1 Human capital_it + b2 Openness_it + .....+ u_it
the regression is based on unbalanced panel data for 97 countries for 30 years (but averaged for 5-years period). Human capital is proxied by mean years of schooling.
Now, my question is, when I did the Hausman test, the appropriate model is the Random Effect. In fact, the regression coefficients with RE are quite significant, unlike the FE where none is significant. please also note that I drop outlier countries.
On the other hand, from a theoretical point of view, it is argued that there is endogeneity in the above model (eg. Mincerian equation "ability effect" on individual income other than years of schooling). therefore, from a theoretical perspective, I should use FE as it accounts for such a problem.
As RE gives me a consistent result, I am intending to adopt the RE model. But, am wondering, if there is a loophole that I can argue that RE is still okay to use even with the presence of endogeneity beside its advantage of the model extension by allowing the inclusion of time-invariant regressors. Would you please kindly suggest me your thoughts on that? Thank you!
This is very interesting paper for Prof Kelvyn Jones:
Explaining Fixed Effects: Random Effects Modeling of Time-Series Cross-Sectional and Panel Data
Article Explaining Fixed Effects: Random Effects Modeling of Time-Se...
In your case I suggest not to report the results of FE or RE model solely. If you suspect endogeneity you should try to use IV models for panel data as well as dynamic models. In my opinion, a good way of reporting results in panel data regression is to start with OLS and then use other models (FE, RE, IV, dynamic model, adding trend) and see how the results change. You can well show the results of all these models and discuss them.
Best, Blazej
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression