Hi,
I found this online regarding how to choose between the fixed and random effects.
webuse nlswork.dta xtset idcode year // Fixed effects regression xtreg ln_wage age hours , fe est store fe // Pooled OLS regression reg ln_wage age hours est store pols // Hausman test comparing FE and pooled OLS hausman fe pols, sigmamore // Random effects regression with LM test xtreg ln_wage age hours, re est store re xttest0 // Hausman test comparing FE and RE hausman fe re, sigmamore Well, I did something like it, but now, I have heteroskedasticity and sometimes autocorrelations ( and I think I should include in all the models: RE, FE and OLS ).
iF I include in FE, then I dont get F-test after "xtreg , fe robust" . So I am not sure how I can decide if OLS or FE here.
If I use xtoverid, I get that I should use OLS instead of RE ( but nothing about FE) - this is what i get: Error - saved RE estimates are degenerate (sigma_u=0) and equivalent to pooled OLS-
For Breusch and Pagan Lagrangian multiplier test for random effects, It is clear that it is OLS.
Now, between OLS and FE, how do I decide without F-test? Can I obtain this for the presence of heteroskedasticity in some way, separate than the "xtreg , fe robust" ( as Stata wont give this result to me after running the FE regression.
Thank you,
Alexa
I found this online regarding how to choose between the fixed and random effects.
webuse nlswork.dta xtset idcode year // Fixed effects regression xtreg ln_wage age hours , fe est store fe // Pooled OLS regression reg ln_wage age hours est store pols // Hausman test comparing FE and pooled OLS hausman fe pols, sigmamore // Random effects regression with LM test xtreg ln_wage age hours, re est store re xttest0 // Hausman test comparing FE and RE hausman fe re, sigmamore Well, I did something like it, but now, I have heteroskedasticity and sometimes autocorrelations ( and I think I should include in all the models: RE, FE and OLS ).
iF I include in FE, then I dont get F-test after "xtreg , fe robust" . So I am not sure how I can decide if OLS or FE here.
If I use xtoverid, I get that I should use OLS instead of RE ( but nothing about FE) - this is what i get: Error - saved RE estimates are degenerate (sigma_u=0) and equivalent to pooled OLS-
For Breusch and Pagan Lagrangian multiplier test for random effects, It is clear that it is OLS.
Now, between OLS and FE, how do I decide without F-test? Can I obtain this for the presence of heteroskedasticity in some way, separate than the "xtreg , fe robust" ( as Stata wont give this result to me after running the FE regression.
Thank you,
Alexa