Assumptions for Panel data models

[Jon Ko]

Hi,

I'm trying to grasp the basic assumptions behind panel data regression models in order to understand which assumptions I have to check for my model. Does the following table (based very roughly on Woolridge, 2013) capture the essential assumptions correctly. I do understand that the assumptions are formulated not very mathematically, but I'm looking more for a high level overview as opposed to mathematical definitions.

OLS cross-sectional	OLS time series	FE	RE
Linearity in parameters	Linearity in parameters	Linearity in parameters	Linearity in parameters
Random sampling		Random sampling	Random sampling
No perfect collinearity	No perfect collinearity	No perfect collinearity	No perfect collinearity
Zero conditional mean	Zero conditional mean	Zero conditional mean	Zero conditional mean
Homoskedasticity	Homoskedasticity	Homoskedasticity	Homoskedasticity
	No autocorrelation	No autocorrelation	No autocorrelation
	Normality	Normality
		Independent variables change over time
			Expected value and variance of unobserved effects uncorrelated with independent variables

Thanks a lot for your help in advance!

[Carlo Lazzaro]

Jon:
as the topic you're interested in is quite broad, your question should probably be addressed to an econometric workshop on panel data analysis.
That said, you may fing some useful details skimming through -xtreg- entry in Stata .pdf manual.

[Phil Bromiley]

Carlo is correct that this is a very general question rather than a Stata question. However, I think you have some obvious errors in your table. These techniques generally assume assume random errors and for some (maybe all?) tests assume normality of the error term, but the RHS variables themselves do not have to be random. FE assumes fixed values for the panel effects while random effects assumes the panel effects are drawn from a distribution (see xtreg documentation). Many forms of homoskedasticity and autocorrelation are easily handled either by direct techniques (e.g., ivreg or xtivreg) or by using robust error variance estimates.

[Jon Ko]

Thanks for your help... It seems like every textbook uses slightly different assumptions...

[Carlo Lazzaro]

Jon:
as others might benefit from the reference you quoted, do you mind to provide its full details (as per FAQ)? Thanks.

[Jorrit Gosens]

You might also have a look at Hoechle, 2007 (here) for different panel data techniques and the propoerties of their resulting errors