Choice across periods

[Tim Sulls]

In period one, actors can choose as many options as they want from a given set, with no limit. In period two, they have the chance to refine the set, if they want, with most selecting some subset that is small than the original. No new option that was not selected in period one can be chosen in period two.

I want to test whether they have a preference for green options in the first period. Then, I want to test if their preference for green stays the same, increases, or decreases.

set seed 1000
clear all
set obs 25
generate actorid = _n
expand 100
bys actorid : generate choiceid = _n
generate period1chosen = runiform() < 0.3
generate period2chosen = runiform() < 0.2 if period1chosen == 1
replace period2chosen = 0 if period2chosen == .
gen green = (choiceid < 25)

The first analysis is obvious reg period1chosen green.

The second analysis to get at changes in preference is less clear. reg period2chosen green seems wrong.

[Carlo Lazzaro]

Tim:
welcome to this forum.
It would seem a two-wave panel regression with a count regressand. See -xtpoisson, fe-.

[Tim Sulls]

Thanks, Carlos. I will look into xtpoisson but to clarify I want to estimate it in a binary fashion as within "green" there are invariant parameters that need to be controlled for. It is not how many greens were chosen versus non-green. Sorry if I misunderstood

[Carlo Lazzaro]

Tim:
as far as I can get the issue, panel units choose green once (1st wave) and greeen again under a set of constraints (2nd wave).
That's why I thought -xtpoisson- to be a possible way to go.

[Tim Sulls]

Thanks, Carlos. In the second wave, it would be among the units selected in the first wave with the green units being the same between waves. Not sure if that changes the applicability. How would you set up the regression for each wave?

[Carlo Lazzaro]

Tim:
it's seem to me that itìs a matter of how many green whatevers are selected per -panelid- in the first and second wave, respectively.
It this is the case,a conditional fe -xtpoisson- can be the tool to try.

[Tim Sulls]

Thanks, Carlos. I am unfamiliar with this regression. Given the data provided could you offer a suggestion of the actual analysis?

[Carlo Lazzaro]

Tim:
take a look at -xtpoisson- entry, Stata .pdf manual.