I have data from three surveys conducted in three different years (2012, 2014, 2016). Each survey was administered to public managers, but not necessarily the same ones (some retired or changed positions) and some managers did not answer all three surveys. Therefore, data cannot be combined in a longitudinal format (if we do, we lose too many obs).
We decided to use a pooled regression model, including a year dummy. Only 2016-dummy is significant (p<0.05).
I have a couple of questions:
there is some overlap across the samples, as few managers might have responded to more than one survey. Does this affect the model in any way?
Should we also present the cross-sectional model for each year? And what would it mean if results change across models? e.g., the coefficient of interest is significant in the model for year 2012 but not in the model for year 2014.
is there anything else that we need to pay attention to? What is the advantage of a pooled regression over a cross-sectional OLS for only one year? Aside from the sample size.