Dear Statalist,

I am trying to estimate the effect of a pension reform on work status of poor people in a difference-in-difference model with three time periods.

The reform:
Before 2009 pension eligibility was restricted to exceptional cases only. From 2009 onwards all citizens who were 60+ years old were eligible for the pension.

My data:
I use panel data collected in three waves: 2008, 2010, 2013. In total I have about 2500 observations.
I have five types of individuals in my data set:
Those who are less than 60 years old in 2013; not eligible for pension (id=1)
Those who turn 60 between 2010 and 2013; we observe them to be eligible from 2013 onwards (id=2)
Those who turn 60 between 2008 and 2010; we observe them to be eligible from 2010 onwards (id=3)
Those who are already 60 in 2008; but they are only eligible for pension from 2009 onwards (id=4)
Very few cases who started receiving a pension prior to the reform (id=5)


Code:
* Example generated by -dataex-. To install: ssc install dataex
clear
input float(id year t2010 t2013 y age treat pre_post1 pre_post2)
1 2008 0 0 1 54 0 0 0
1 2010 1 0 0 56 0 0 1
1 2013 0 1 0 59 0 0 1
2 2008 0 0 1 56 1 0 0
2 2010 1 0 0 58 1 0 1
2 2013 0 1 0 61 1 1 1
3 2008 0 0 1 59 1 0 0
3 2010 1 0 1 61 1 1 1
3 2013 0 1 0 64 1 1 1
4 2008 0 0 1 61 1 0 0
4 2010 1 0 1 63 1 1 1
4 2013 0 1 1 66 1 1 1
5 2008 0 0 1 61 1 1 1
5 2010 1 0 1 63 1 1 1
5 2013 0 1 0 66 1 1 1
end

Model:
I aim to regress the following model:

Code:
reg  y   treat ##pre_post   age   t2010   t2013
Question:
I am not sure whether a Diff-in-Diff approach is feasible in the case I presented here. My particular question is, how to define the treatment indicator and the pre_post indicator.
In the data example above I provide two alternative ways to code the pre_post variable, but I am not confident that either of them is correct. Pre_post1 is defined as 1 if an individual in a certain year is 60+ years old, 0 otherwise. Pre_post2 is defined as 1 if a person would have been able to receive a pension in a certain year provided that s/he is/would have been 60+ years old. Or in other words it is equal to 1 when a pension program exists for a particular clientele in a particular year no matter whether the person is already old enough to be eligible; 0 otherwise. The variable treat is defined as 1 across all years for those individuals who are or become eligible at some point during the observation period; 0 otherwise. I include year dummies because I presume there are economic shocks over the years that are unrelated to pension but could possibly effect or outcome variables.

Our case is relatively similar to the one posted here and I tried to take some orientation from Clyde's recommendations there:
https://www.statalist.org/forums/for...o-time-periods

Our situation differs, though, in that assignment to treatment is done by age. It would be great to hear your opinions on whether our situations is feasible for a diff-in-diff model or whether we should alternatively run a fixed effects model of the kind:

Code:
xtset id year
xtreg  y   pre_post1   age   t2010   t2013, fe

Thank you for your replies!
Juliane