Maize trade in Southern Africa
Fixed and Random Effects
Two ways of handling the things about a country pair that never change. The choice is a judgement, and it is worth stating out loud.
Panel data lets you absorb the permanent quirks of each country pair. What it does not do is tell you what to assume about them, and there are two standard answers.
Fixed effects treats each country pair as having its own permanent quirk and makes no assumption about where that quirk came from. It is the cautious choice. It costs you the ability to say anything about characteristics that never change, and in a trade model that includes distance itself, which is usually the thing you most wanted to measure.
Random effects treats those quirks as if they were drawn from a common distribution. It is more efficient and it keeps distance in the model. In exchange it assumes the quirks are unrelated to everything else in the equation.
I used random effects. SADC members differ from one another in culture, religion, political philosophy and geography in ways that are reasonably treated as scattered rather than systematic.
What it cannot do
Neither option is neutral and neither is provable. This is a judgement about the character of what you could not measure, made before you see the answer, and it is exactly the kind of judgement a reader is entitled to disagree with.
The equation
random effects: εijt = cij + uijt, with cij assumed unrelated to xijt
- cij
- the pair’s permanent quirk, now folded into a composite error rather than estimated
- uijt
- the ordinary year-to-year noise
- the assumption
- that cij is unrelated to the things being measured
Fixed effects makes the opposite choice: it estimates a separate cij for every pair and assumes nothing about it. The price is that anything which never changes, distance included, drops out of the model entirely.
Further reading
- Hausman, J. (1978). "Specification Tests in Econometrics", Econometrica. Read it.The test for whether the random effects assumption survives contact with your data.
- Kennedy, P. A Guide to Econometrics.The most plain-spoken treatment of this particular choice anywhere.
From An analysis of maize trade in the Southern African Development Community, MSc (Agricultural Economics), University of Pretoria, 2012. Read it in UPSpace.