EViews.com

I have a panel of 31 chinese provinces, each with different characteristics e.g. no. of educated, population density, amount of FDI etc.

I am trying to run a regression with GDP per capita growth as the dependent variable.

In the literature, they commonly run both a fixed effects and random effects model to serve as a robustness test.

I tried doing this with my data, but it would not allow me to run a fixed effects OLS regression, stating that it is a 'near singular matrix'

Can anyone advise me on what to do?

Thanks

You'll have to explain more what you are trying to do. Your workfile doesn't provide sufficient information.

Thanks for the quick reply glen.

Im trying to run the following regression:

ΔGDP=FDI+Trade balance+Education Level+ ...+μ

Where all the variables are in logs, and it is a panel data set for 31 different provinces, each with their own values for GDP, FDI etc.

Still not enough info. Do you use all of the variables in your specification? Does it estimate if you don't have fixed effects?

By the way, what's the difference between VAR01 or by CELLID? Your panel is structured by both.

Not sure what the difference between VAR01 and CELLID is. I used the stack function and both variables were automatically created.

What is the best way to structure my data in your opinion?

Ah, I see. VAR01 looks to be cross-section identifiers, but when you stacked you didn't provide a period ID so that we had to create CELLID to identify the observations within the cross-section. I would structure as a dated panel with VAR01 as the cross-section ID and ID01 as the period ID.

But that still doesn't answer my previous questions.