warning! panel data

[cocoabutter]

Hello!

I have been running the Hausman test to decide between random effects or fixed effects for my panel
The result is that the random effects model cannot be rejected, however, at the bottom panel of the test it estimates fixed effects version as well, but with the message :WARNING coefficent covariance matrix is of reduced rank. What does this mean please?
Should I be worried?

Also, if I try to use White correlation terms, I end up with the same warning. My panel is being run on households by the way.

Please help!

[EViews Glenn]

Some of the estimated matrices can be of reduced rank, generally when one of the dimensions of your panel is small. If, for example, you try to do a White cross-section robust covariance and you have fewer periods than cross-sections, the estimated covariance is, by construction, reduced rank. More generally it means that test statistics should be interpreted with some caution, so we try to warn you.

[cocoabutter]

Ah, ok I see what you mean now. Many thanks.

Also, do I have to be worried about clustering effects when using a panel?
It seems to me that random/fixed effects only controls for unobserved heterogeneity at the level of the households ( which is the units my data is in).
I see that it has been recommended to "rearrange your data within the panel so it is also identified by cluster".
How does one go about this within the format given by Eviews?

[EViews Glenn]

You may want to make a copy of the workfile page for this since we're going to restructure your workfile. It's not essential, but...

Define a variable that identifies your cluster. Structure as an undated panel using that identifier. EViews will ask if you want to make an additional index series. This is simply a variable that is going to identify the observation within the cluster (1, 2, 3, ...). Go ahead and do so.

You'll now have a panel data set where the cross-section is defined by your clustering variable.

Note that if your original cross-section identifier, households in your case, are nested within clusters, then the original analysis which controls for the finer level of classification will also control for the clusters (though perhaps inefficiently).