In panel data, should I include both wave dummies and age?

[selim keskin]

I have four waves of data, which is collected in two years intervals. I use within-between models. I estimate age as random effects specification, so I should include wave dummies as well, right?

Thanks a lot!

[Carlo Lazzaro]

Selim:
there's something I did not understand from your post.
What is your panel identifier?
What do you mean with within-between models?
I doubt that -age- can be the random effect in your model.

[selim keskin]

I am sorry that I did not explicate more. Here are the details.

N = 2500. A probability sample of US individuals.
T = 4.

Within-Between models refer to Schnuck (2013), but also to Allison (2009), who calls this model a "hybrid model". It basically gives FE estimates of some variables (if you include their mean values to the regression) and RE estimates of some other variables.

Including the age variable into the FE model does not make any sense to me, as it would measure a 2-year change for everyone. This is why I estimated my models by including wave dummies and age variables as RE specifications. My question is that including wave dummies actually hampers the age effects (which are, including age square, significant without the wave dummies), but I was not sure that I should exclude wave dummies anyway.

[Carlo Lazzaro]

Selim:
thanks for clarifying.
I figured out that you referred to hybrid model, but I was not really sure about that.
I guess you are intetested in estimating time-invariant coefficient(s) that the-fe- estimator, as expected, wipes out.
Usually It is a good idea to include -i.year- in the right-hand side of your panel data regression equation.
You can the check whether-i.year- is jointly significant via -testparm-.

[selim keskin]

Thanks a lot Carlo!