Folks,
This is not a matchit question, but I wondered if I could access your wisdom.
Suppose there is a panel dataset, and different units get the
treatment at different dates. That is, there isn't a sharp date on
which the treatment is applied; different units get the treatment on
different dates.
We felt the right estimation strategy would be:
1 Identify the units that never got treated; they're the control pool
2 Walk through time. In each time period, identify the firms who are
treated. Find controls using propensity score matching where
information from that time period (only) is used. Once a control has
been used, flag it so it will never be used again.
3 In this fashion, for each unit that's treated at time t, find a
control at that time t.
4 Shift everything into event time where the treatment date is defined as 0
5 Proceed with the analysis as usual.
Would appreciate your insights.
--
Ajay Shah
ajayshah(a)mayin.org
http://www.mayin.org/ajayshahhttp://ajayshahblog.blogspot.com