Fixed effect estimation Rsquared negative

[Lorenzo Turchi]

Hello everyone,

this is my first post! i'll try to be concise and explain my problem, any help would be more than appreciate
I'm launching a fixed effect regression on a panel dataset, and when launching the regression with the command xtreg i get the R square of the model extremely low, around 0,0239. Then i compute the R square adjusted since the moment i inclued more than one variable, with the command ereturn list, and i get the value of e(r2_a)= -.27 (aproximated). As far as i know also the R squared adjusted should be within 0 and 1...i don't know why and with this result i'm wondering if my model is suitable to draw some conclusion. I attache the stata result of the regression to be more explicative.

Thank to anyone who can help!

[Andrew Musau]

As far as i know also the R squared adjusted should be within 0 and 1

This is not the case, the adjusted R2 can be negative and often is in fixed effects regressions in Stata. If you look at the formula, this will be apparent. I give a suggestion of an alternative calculation of the within adjusted R2 in #9 of the following link which in my opinion is a more sensible way of calculating the statistic.

https://www.statalist.org/forums/for...ted-as-missing

[Lorenzo Turchi]

Thank you so much Andrew, i will launch the regression with the commands you have provided!

Best regards,

Lorenzo

[Phil Bromiley]

It doesn't usually make a difference, but you might consider rescaling your variables so they don't have such extremely small parameter values. Since they are all very very small, you might try rescaling the dv. Alternatively, rescale the ivs. There is a remote possibility that you're hitting some precision problems. In any case, it is much easier to explain a normal sized parameter's influence on a rescaled dv than to work with 10 to the -8 sized parameters.