"Outcome does not vary in any group"

[Shilpa Shetty]

Hi,

I have a balanced panel data comprising 10 independent variables for a period of five years for 120 companies. My dependent variable is binary coded as 0 and 1 and I've been attempting to use logistic regression (xtlogit). When I try running fixed effects model, I get the response as "Outcome does not vary in any group". However, for the random effects, I have got the output. I wish to conduct the Hausman test to choose between the model.

Kindly guide in resolving the issue relating to fixed effects

[Maarten Buis]

The error message tells you that the dependent variable (the outcome) does not change over time. So if a company is a 1 at time 1, you know that it will remain 1 for all subsequent time points.

This could be right or wrong, and we cannot know which one it is. You know what the dependent variable is supposed to measure, how it was measured, and whether or not it should change over time.

If it should change, then that indicated that there was a mistake in the code that prepared the data, and you need to fix that.

If the dependent variable is fixed over time, then there is not much to be gained from doing a panel analysis. A fixed effects model is just not possible. But even random effects model in this case just does not add much. If you wanted to use a panel structure to get closer to a causal effect, then you really need another design.

[Shilpa Shetty]

Thank You for your quick response.
As rightly pointed, the dependent variable does not vary over time. However, the dependent variable varies across my sample companies. Can that be captured in the code? If not, what other design would you recommend?

[Maarten Buis]

That really depends on so many factors

[cengiz hokka]

Dear @Maarten Buis,

I examine the relationship between the board structures of banks and their performance. what I want to ask here is: the board variables I have usually don't vary over time, so I'm considering using random effects rather than fixed effects. Whatever the hausman test shows. This is the right approach isn't it?