Individual fixed effects vs. school fixed effects

[TrentJN ZClay]

Hi all,

I am deciding between a school and year fixed effects model and an individual fixed effects model. I wonder which one makes the most sense.
My data have two waves. The outcome variable is continuous and the key predictor is a dummy. It is a nested dataset (individuals nested in schools). What are some of the statistical considerations for deciding which one to use?
One thing that comes to my mind is if there is enough variation in both my outcome and key independent variable across these two waves for an individual fixed effects model. I then did the following checks, but I am unsure what sufficient variable means for my case. I used multiple imputation to fill in the missing values of my predictor, and the imputed values follow the patterns of the observed values.

In addition, what are other things I should check to decide between these two models? Thank you so much!

Dependent variable:
RECODE of
dep_diff | Freq. Percent Cum.
--------------+-----------------------------------
No change. | 888 16.88 16.88
Change exists | 3,325 63.19 80.06
Missing value | 1,049 19.94 100.00
--------------+-----------------------------------
Total | 5,262 100.00


Independent variable:
RECODE of
predictor_
dff | Freq. Percent Cum.
--------------+-----------------------------------
No change | 2,727 51.82 51.82
Change exists | 1,485 28.22 80.05
Missing value | 1,050 19.95 100.00
--------------+-----------------------------------
Total | 5,262 100.00

[Carlo Lazzaro]

Trent:
welcome to this forum.
Whenever you have a nested study design, the first option should be some commands from the -ME- suite according the type of your regressand.

[Joro Kolev]

Given that your subjects do not switch schools, when you control for individual fixed effects, you re controlling for school fixed effects as well. The opposite is not true, if you control for school fixed effects you are not controlling for individual fixed effects such as individual gender and preferences.

So if you can estimate individual fixed effects regression, and it is giving you reasonable estimates, I would stop there.