Ingenuously, I think random effects as a mean term of fixed effects and pooled-ols.
Fixed effects focus only in the between cross-sections factor, while pooled-ols accounts for within. Mostly, FE presents interesting results and solves the problem of many papers, but in my case I'm avoiding it because I need this cross-sectional "within" factor, it's a point of my countries unit.
My first naive logic is: Let's use a dummy for each observation (my case, countries) to control fixed effects in a random effect model. It may not seem obvious to me at this first time that I lost my degrees of freedom and my RE models became a little useless.
My point is, can we control at least partially FE in a RE model? My next attempt is cluster my dummies (rich countries, poor countries, region) to recover my degrees of freedom and create a statistically suitable model. Is it a good attempt? Or else, new directions that doesn't involve dummies?
I'm using an ordered probit model, directions in this model are useful.
Best,
These papers should help - they discuss a multilevel random effects within and between model that retains the countries as entities, gets the same results as fixed effects for time-varying variables and allows you to put in cross-sectional time-invariant variables:
https://www.researchgate.net/publication/233756428_Explaining_Fixed_Effects_Random_Effects_Modeling_of_Time-Series_Cross-Sectional_and_Panel_Data
and
https://www.researchgate.net/publication/299604336_Fixed_and_Random_effects_making_an_informed_choice
Article Explaining Fixed Effects: Random Effects Modeling of Time-Se...
Article Fixed and Random effects models: making an informed choice
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics