GMM or FE? Is my model dynamic ?

[Gabriela Kalibatseva]

Dear all,

I am working with survey panel data covering 32 years and very large amount of individuals taking part in each wave. My depended variable is life satisfaction (measured on scale between 0 and 10). The dependent variables in my model are the lagged income variable, marital status, number of children in the household, sex, age and so on.

I am currently using FE model since I want to control for the unobserved time-invariant characteristics. My main goal is to find if there is self-detection into marriage due to personal characteristics or is marriage making people more happy in general.

However, I am a bit confused at this point if my model is dynamic or not. From what I know, the dependent variable (life satisfaction) needs to be dependent on its own past value in order to consider the model dynamic. Does happiness today depends on how I felt yesterday - doesn't seem very likely. Can you please advise what should be the best approach to apply ? Thank you !

Kind regards,
Gabriella

[Justin Niakamal]

I'm not an expert in this field, but for what it's worth here's my input.

I am a bit confused at this point if my model is dynamic or not

Your model is dynamic if you use a lag of your dependent variable. If you choose to include a lag of your dependent variable then you should use GMM for reasons mentioned here. However, from what I understand estimators such as xtabond are more common with short T (in your case number of years). For large T, pooled mean group estimators are more appropriate. Here's a helpful reference (Stata Journal) for this type of estimator in Stata( xtpmg).

Does happiness today depends on how I felt yesterday - doesn't seem very likely.

You could look into panel unit root tests to see if this is *statistically* true or not -- ie is there persistence in the series.

...and very large amount of individuals taking part in each wave.

As a caveat, I'm not sure how the unbalanced nature of your dataset will affect any of my suggestions.