DESPERATION Random and Fixed Effect on Panel Data

[confused_economist]

Hi,

I am having too many problems with my project at the moment, partly because I am not an econometrician and have only started doing econometris very recently.

I am a possibly very simple problem that I hope someone could really help me with, I am so stuck and have literally given up hope on econometrics, but before I do so, I would be grateful for any help.

What I have done so far:

1) Specify my model
2) Select with fixed effect on cross section
3) View --> Fixed/Random Effect Testing --> Redundant Fixed Effect - Likelihood Test
4) It comes back with 0.0000 for the p-value

WHAT DOES THIS MEAN? sorry for "screaming", I searched the Eviews handbook both part I and II and there is no answer.

Then I continues to do this:

1) Specify my model
2) Select with random effect on cross section
3) View --> Fixed/Random Effect Testing --> Correlated Random Effects - Hausman Test
4) On some of the models it comes back with 0.0000 or sometimes 1.0000 for the p-value

Again what does this mean? Does this mean the model should be fixed or random??

What is the null and what is the alternative?

I am super grateful for any help, please......

[confused_economist]

Hi, I am using Eviews 6 if that helps.....

[EViews Glenn]

Case 1:
Test evaluates the joint significance of the fixed effects. A zero p-value indicates that the effects are significant.

Case 2:
You are testing the random effects model against the fixed. Under the null, both are consistent estimators and the random effects model is efficient. Under the alternative, only the fixed is consistent. A zero p-value indicates the results for the two models are significantly different from each other, indication that assumptions required for the random effects model to be valid are violated. A p-value of 1 indicates the opposite.

[confused_economist]

Thanks Glenn,

Let me get this clear, it I have zero p-value from the likelihood test it means fixed effects is significant, and if I recieve a 1 for my hausman test, it means random effects is valid.

Would this not be a contradiction, and what should I do in this situation?

Thanks

[EViews Glenn]

No contradiction. They're testing different things. The pval=1 in the Hausman test simply indicates that there is no evidence that the random effects estimates are invalid. The fixed effects estimates are valid under both the null and the alternative.

[confused_economist]

Thanks Glenn,

Would thi smean even if I recieve a p-value of 1 in Hausman test, but a p-value of 0 in the fixed effect likelihood test, I could still continue to specify my model with Fixed effect?

Many Thanks

[EViews Glenn]

Yes. In principle, the RE is more efficient, but the FE is fine. In fact, even if you don't reject the null that the FE are statistically significant, the FE are fine (though not efficient).

[confused_economist]

Thanks again Glenn, that really clear things. Very grateful and happy!

[miki1zg]

Can someone help me with this? How do I decide should I use fixed or random effects? Which p-value should I look at?