I'm examining the link between age and an outcome in an unbalanced panel data (each individual responded to the survey 1 to 8 times)
I'm estimating the following fixed effects model:
model <- plm(outcome ~ age + I(age^2), index=c("id","surveyyear"), model="within", data=data)
From what I'm understanding, age in this model is a within-person effect. If so, does it make sense that I have individuals who have only one observation in this dataset? I guess they are not contributing anything to the model?
similarly, given that I'm estimating a quadratic effect, are individuals with fewer than three observations not contributing anything here? (since more than three points are needed to estimate a quadratic effect) In other words, should I get rid of the quadratic term?
I was thinking this model was identical to
model2 <- lm(outcome ~ age + I(age^2) + factor(id), data=data)
so I'm confused when reading all this stuff about demeaning I was supposed to do in FE models.... For example, in Wooldridge (2013; p. 231) it says
"Instead of applying the within transformation to all variables and running lm, we can simply use plm on the original data with the option model = "within.""
Does this mean I should have person mean centered age? Or the outcome prior to running the model?
