Panel data of only 2 companies?

[Jeff Stu]

Hi everyone,

I am reading this paper (not published yet) about the effects of eight dependent variables (X) on Y. The paper uses panel data on a pretty standard regression analysis (linear, no interactions, etc).

What is intringuing me, is thtat the autor has data on only 2 companies, with 63 observations each.

Is that even plausable? Can one do that? What about freedom degrees? Variance?

I have been reading and doing studies with panel data for some year, but have never seen anything like this.

Right now I am no able to judge if this can be done.

Any help is welcome.

JS

[Clyde Schechter]

Well, if we're talking about the usual panel regression command, -xtreg-, I don't see any problem per se. But there are some limitations.

1. vce(cluster panelvar) is not valid with only two, so should not be used.

2. If random effects were used, the sample of panels is very small and inferences about the population of panels should be viewed with extreme skepticism. The confidence intervals around the variance component estimates will likely be very wide, as they should be.

3. If fixed effects were used, then inferences about the coefficients are within-panel in any case. However, informal inferences about applicability to other panels beyond the two in the model are even more suspect than they would normally be.

4. I think that df and variance (but see 1. above) should be OK.

[Nick Cox]

eight dependent variables (X) on Y

That's independent. Presumably the slip is just a typo, but to my mind one of several objections to the terminology of dependent and independent variables is that many people find it hard to keep straight which is which.

So many good alternatives! I tend to say "outcome" or "response" on the one hand and "predictors" on the other, but there are good arguments for other choices (and "predictor" is also overloaded with another meaning, the prediction from a model).

[Carlo Lazzaro]

Jeff:
what about the -timevar- chosen by Authors (that is, how many data waves are considered in the regression model)?