In carrying out panel data regression analysis, it is required that Hausman Specification Test be carried out to choose from Fixed Effect or Random Effect estimation approaches. Another theory holds that Breusch-Pagan Lagrange multiplier (LM) test for panel data is also required to choose between Random Effect estimation and Pooled Effect Estimation.
Which of the preliminary tests should come first? Are these tests the final determinants of which estimation approach to deploy?
Hi, LM test for random effects versus OLS, LM test for fixed effects versus OLS, and Hausman test for fixed versus random effects model. This is how I look at the analysis.
Hardeep Singh Mundi Thanks a lot. This is helpful. I know how to conduct LM test for random effect in both Stata and Eviews. Can you put me through on how I can conduct LM test for fixed effect (using Stata and/or Eviews)?
I use R, i don't know Stata. I am sure there are sufficient YouTube videos for this.
Hardeep Singh Mundi Okay. I will get R tomorrow, then. There is no video on YouTube covering how to conduct LM test for fixed effect in Stata (or Eview).
I will just have to get R tomorrow to enable me pull through. thanks for the assistance so far, sir.
Hardeep Singh Mundi
I truly appreciate this offer for assistance. Thanks a bunch. Will surely reach out to you if I encounter any challenge along the line.
You should first consider the economic theory/common sense underlying your model.
Fixed or Random effects impose a lot of homogeneity on your data. Given the effects, the beta coefficients are the same for each member of the panel. Only the effects vary.
Random effects estimates are efficient if the effects are not correlated with the explanatory variables. If the effects are correlated with the explanatory variables Random effects estimates are not consistent
A Fixed effects estimate is consistent whether the effects are or are not correlated with the explanatory variables.
One should always consider what unobserved values are the source of the effects.
The null of the Hausman test is that the fixed and floating effects estimators provide similar estimates of the coefficients. The Hausman statistic is a measure of the difference between the two sets of coefficients. If the measure is large one can reject the null and use a fixed effects estimator. If the distance is not large enough to reject the null you can not conclude that the random effects is appropriate. Such a conclusion is not in accordance with the statistical theory of hypothesis testing. Your sample size may be to small to reject the null. You should give consideration to correlations between the effects and the explanatory variables and to the differences in the estimates of the coefficients.
In panel estimation, the effects are the same for all members of the panel. There is more than one test that might be used and I would recommend that you consult your econometric software manual. You should also consider the economics of any possible unobserved heterogeneity. It is always better to have economic answers to questions such as yours.
John C Frain Thanks a great deal, Professor. This is helpful.
However, aside a recourse to economic theory, other statisticians have suggested that economic theory should be combined with statistical analysis in determining the choice among Fixed Effect/Random Effect/Pooled Effect.
The above formed part of the major reason I needed clarification on "What are the preliminary tests that will determine the choice among fixed effect regression, random effect regression and pooled effect regression?"