Partial answers:
For 3, you will need to create a time variable that is sequential
without missing numbers, e.g. 1996=1, 2000=2, and so on.
For 4, it depends on whether the time-invariant variables vary across
cross-sections. If they do not then obviously they are not "variables"
and cannot be modeled as such. But assuming that they do vary across
cross-sections, then you have two options: use cross-section unit dummy
variables as a blunt but effective way of absorbing all of the
unobserved, time-invariant, unit-varying effects, or forgo the dummy
variables and directly include your time-invariant, unit-varying
variables in the equation.
The advantage of the first method is that it takes care of "everything"
that is unit-specific & time-invariant, with no need to collect data or
to use up degrees of freedom; the disadvantage is that using unit fixed
effects allows you to say extremely little of theoretical interest
because it is such an aggregated concept.
3. In one dataset, I have 3 years corresponding to 3 US presidential
elections--1996, 2000, and 2004. I constructed the lagged dependent
rather than relying on Stata's time series command (so the lagged
dependent is for 1992, 1996, and 2000). Data are at the county level
for one state. My question is about tsset. Would I do tsset county
year, even though the years are 4-years apart? Or would I create a
time variable so that 1996=1, 2000=2, and 2004=3 then do tsset county
time? The answers differ depending on how I do tsset.
4. What if some of your independent variables don't vary with time
but a dynamic panel model makes sense due to the nature of the
dependent variable? Can the command still be used?
Thanks in advance.
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
