Hi, I have 4 dependent variables, when I run the hausman test for each one of them, 2 of them validate the random effect model, and the 2 others validate the fe model. Now which one to choose since I'm doing a multivariate regression? And on top of that, some of my dummy variables (usefull ones) are ommitted under fixed effect, but I really need to use them.
I did a video on "Why does Stata drop my variables?", which will be published on Monday 1st April 2024 - April Fool's Day Special. Link: ua-cam.com/video/CgHueBDaEnU/v-deo.html If you have four dependent variables, it suggests that you will have four separate models to explain them. Of course, you can have FE or RE as these models are separate. If you do not know which one is actually dependent, then you need to explore causality (or causal order if you have a time dimension). May the Force be with you!
You can use time and firm dummies. However, there is a risk of overfitting. You might get great in-sample model fit - but once you try to predict out-of-sample the quality drops.
The full course is available for USD 3.59 here www.yunikarn.com/p/data-science-using-stata-complete-beginners-course, including datasets and Stata do files. The slides are available as an E-book (129 pages) gerhard-kling-s-school.teachable.com/p/my-downloadable-14033 - combined offer: USD 5.40.
Was eagerly wating for the video. Thanks for sharing.
Thanks for your comment, Sourav! There is more to come on panel data next week
Was very insightful and useful, thanks
Glad it was helpful!
Hi, I have 4 dependent variables, when I run the hausman test for each one of them, 2 of them validate the random effect model, and the 2 others validate the fe model. Now which one to choose since I'm doing a multivariate regression? And on top of that, some of my dummy variables (usefull ones) are ommitted under fixed effect, but I really need to use them.
I did a video on "Why does Stata drop my variables?", which will be published on Monday 1st April 2024 - April Fool's Day Special. Link: ua-cam.com/video/CgHueBDaEnU/v-deo.html If you have four dependent variables, it suggests that you will have four separate models to explain them. Of course, you can have FE or RE as these models are separate. If you do not know which one is actually dependent, then you need to explore causality (or causal order if you have a time dimension). May the Force be with you!
Hi there, just confirming that we can use time dummies but not firm dummies in fixed effects? Thank you.
You can use time and firm dummies. However, there is a risk of overfitting. You might get great in-sample model fit - but once you try to predict out-of-sample the quality drops.
Is it possible to use robust option with FE and RE? Thank you.
In panel data, the standard approach is to use the cluster option in Stata, which also accounts for serial correlation.
The full course is available for USD 3.59 here www.yunikarn.com/p/data-science-using-stata-complete-beginners-course, including datasets and Stata do files. The slides are available as an E-book (129 pages) gerhard-kling-s-school.teachable.com/p/my-downloadable-14033 - combined offer: USD 5.40.