Hello everyone,
i am new here and i need to clarify some question because i write my dissertation.
i use fixed effect model panel data. the regression outcome is different when i use robust standard error. some coefficient values becomes negative and statistically insignificant. i get different results from fixed effect model, year fixed effect and robust standard error.
my question is;
whats the problem here? why is this happening?
second, i need to understand the difference and importance between year fixed effect and robust standard error
Thanks in advance
Dear Uzeyir Hajigadirli ,
With 52 groups, you can safely use clustered standard errors in order to control for both for serial correlation and heteroskedasticity. Keep that in any model.
You must use in all cases year fixed effects, unless you were interested in a variable that only varies over time (in which case, you cannot include the former effects).
An important decision is to choose between fixed effects and random effects. Theoretically, you can use a Hausman test to help you to decide, but, in practice, we almost always employ fixed effects directly. They require much weaker assumptions for consistency.
In principle, I would use and trust on fixed effects + clustered standard errors + time (year) fixed effects.
I hope it helps.
José-Ignacio
Thank you so much José-Ignacio Antón for your answer.
I used Hausman and result shows that i need to use fixed effects. In the pics, i get different results. Why is this happening?
The coefficients must differ among Fe, re and adding year Fe. Actually, Hausman test aims to determine if de and re coefficients are substantially different. Therefore, it is ok what happens.
Clustering only affects standard errors.
sorry i noticed that hasuman result is above 5%. in this case i will choose random effect. in this case, can i still use year fixed effect? José-Ignacio Antón José-Ignacio Antón
Yes. This Hausman is due to individual fixed effects.
I would use Fe anyway in applied work. It is very hard to defend RE even with the non-rejection if the null in Hausman.
I would use Fe (individual , implemented with Fe in Stata), year Fe (using dummies) and clustered se at the individual level.
José-Ignacio Antón when i use FE i get negative impact on ROA but in year FE i get positive. why is this happening? is there something wrong?
No. Use both types of FE at the same time. It is common that results change with Fe (not including them means omitting variables, which can obviously affect the results).
I prefer the within-between version of the RE model for the reasons set out here
Article Explaining Fixed Effects: Random Effects Modeling of Time-Se...
and Article Fixed and Random effects models: making an informed choice
An extensive review of the potential of this approach in economics concluded that it has been "unreasonably ignored" due in part to "disciplinary isolation" of the subject. Preprint MULTILEVEL MODELING FOR ECONOMISTS: WHY, WHEN AND HOW
On robust standard errors see
https://www.researchgate.net/post/When-should-I-use-multilevel-modellings-vs-cluster-robust-standard-errors
on Hausman test, see
https://www.researchgate.net/post/Is_there_some_other_method_other_than_hausman_test_to_decide_which_method_either_fixed_or_random_should_be_adopted_in_panel_data_analysis
- Badges
- Science topic
- Similar topics
- Biological Science
- Biochemistry
- Biochemical Techniques
- Analytical Chemistry
- pH