@Ben Lambert why is it that towards the end you say: "in FE alpha_i is some fixed parameter that needs to be estimated by adding individual level dummies" when fixed effects by construction removes these constant parameters i.e. dummies and time invariant constants.
As far as I understand it, the fixed effect alpha_i cannot be measured when using the mean differencing method (which is algebraically equivalent to adding dummies to account for the fixed Effects in your regression). The key is knowing what effect you want to estimate. If it is alpha_i, then you have to include the dummies because as you said, the alpha term drops out with mean differencing, and thus cannot be estimated. But if you want to estimate the effect of another variable (say unemployment rate) in an unbiased manner, you can use the mean differencing to correct the endogeneity problem cause you don’t care about alphas effect anymore.. you’re just looking for unemployment effect
The key I think is to look at it like the fixed effect method of adding dummies to account for whatever “region” has a time fixed effect isn’t necessarily the same methodology of doing mean differencing... but they both produce equivalent results. Just where with mean differencing you’re not going to be able to estimate alphas effect
Thank you so much! You mention at the end that there is a particular test which shows corellation between alpha and independent variable. Could you please tell which one is that and how can I do it in Stata?
I know you can use the code estat endog after your regression to test the null hypothesis that the variables are exogenous. Thus if you get a corresponding f statistic that makes you reject the null, you may have endogenity in model
+NNNN1818 Alpha_i in RE model should be random in the sense that cov(alpha_i, x_i)=0. In FE model, alpha_i is not random (fixed) such that cov(alpha_i, x_i)>
I think in FE, we assume alpha cannot be observed and being time-invariant but for RE we assume it is random then COV [alphas,regressors] =0 but COV [errorterm,errorterm] is not 0.
Could you please explain why random effects model is more efficient than fixed effects model ?
Fantastic video!!!! This clarifies so many questions
@Ben Lambert why is it that towards the end you say: "in FE alpha_i is some fixed parameter that needs to be estimated by adding individual level dummies" when fixed effects by construction removes these constant parameters i.e. dummies and time invariant constants.
As far as I understand it, the fixed effect alpha_i cannot be measured when using the mean differencing method (which is algebraically equivalent to adding dummies to account for the fixed Effects in your regression). The key is knowing what effect you want to estimate. If it is alpha_i, then you have to include the dummies because as you said, the alpha term drops out with mean differencing, and thus cannot be estimated. But if you want to estimate the effect of another variable (say unemployment rate) in an unbiased manner, you can use the mean differencing to correct the endogeneity problem cause you don’t care about alphas effect anymore.. you’re just looking for unemployment effect
The key I think is to look at it like the fixed effect method of adding dummies to account for whatever “region” has a time fixed effect isn’t necessarily the same methodology of doing mean differencing... but they both produce equivalent results. Just where with mean differencing you’re not going to be able to estimate alphas effect
Thank you so much! You mention at the end that there is a particular test which shows corellation between alpha and independent variable. Could you please tell which one is that and how can I do it in Stata?
I know you can use the code estat endog after your regression to test the null hypothesis that the variables are exogenous. Thus if you get a corresponding f statistic that makes you reject the null, you may have endogenity in model
Or maybe estat endogenous
gagash sencani harda oxuyursan ? :D buralarda azerbaycanlilari gormek sevindiricidi tehsilimiz adinaksjdkasjd
@@orxanaliyev894, elə mən də onu yazacaqdım, gördüm sən yazmısan. :D
Jamal Baghirov bes seni gece saat 4 de random effect oyrenmeye kim mecbur edib :(
Ben, you are a legend.
The only thing that bugs me is how you add a "R" between alpha and "i", it's a funny quirk =)
I think there was an error: in either RE or FE, alpha_i is random. Alpha_i is not fixed as described in the video in FE model.
+NNNN1818 i think you are right. it says so in the Cameron's book
+NNNN1818 Alpha_i in RE model should be random in the sense that cov(alpha_i, x_i)=0. In FE model, alpha_i is not random (fixed) such that cov(alpha_i, x_i)>
I think in FE, we assume alpha cannot be observed and being time-invariant but for RE we assume it is random then COV [alphas,regressors] =0 but COV [errorterm,errorterm] is not 0.
comparing to 'what we discussed before' not helpful.
Given that he's uploaded literally everything on RUclips in neat ordered playlists it's simply up to you to work a little.
That is correct. This is literally the best teacher you could possibly get.
Lazy Jane xP