--- On Tue, 15/9/09, Wilson, Ian wrote:
> I am undertaking a piece of
> longitudinal research to identify the 'treatment effect' of
> a policy initiative. In textbooks that I have read the
> suggestion is to use an interaction between time and a dummy
> for receiving the treatment or not. In this case since
> only one interaction variable is included to cover multiple
> survey waves am I right in thinking that a linear treatment
> effect has been assumed?
>
> If this is the case could one include a interaction 'dummy'
> for each survey point in time to capture a more dynamic/non
> linear 'treatment effect'?
No, the point of that strategy is that the treatment occurs
at one point in time (wave), so there is a generic change
over time, captured by the linear term, and a discontinuous
change at the point of the treatment, capture by the jump
in the curve as specified by the dummy. If you have dummies
for all waves, there is no way of distinguishing the generic
change over time from the treatment effect.
You are correct however to worry about specifying the generic
change as a linear trend. The alternative is to replace that
with a smooth curve, e.g. a restricted cubic spline, which
you can make with -mkspline-.
There is a nice overview article in the Stata Journal,
covering the various options that are available in Stata for
this type of analysis:
Austin Nichols (2007) Causal inference with observational
data. The Stata Journal, 7(4):507-541.
http://www.stata-journal.com/article.html?article=st0136
Hope this helps,
Maarten
--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany
http://www.maartenbuis.nl
--------------------------
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
