Obtaining TUT estimates with multivalued tretaments

[Michael Howell-Moroney]

Dear Stata listers this query concerns obtaining and gathering results of treatment effects on the untreated (TUTs) with many-valued treatments.

I can do this with binary treatments, no problem. But the multivalued treatment case seems more complex.

Here is where I am on this

Assume that rep78 is the multivalued treatment with control level 1

*** code
sysuse auto

*** What I want is the treatment effect on the untreated (rep78=1), so I type

teffects ra (price mpg) (rep78), atet tlevel(1)

*** The top coefficient (1 vs 2) gives me the TUT for 1 versus 2, but not for 1 versus 3 and so on.
*** In order to obtain the other TUTs, my next step was to estimate separate equations for the TUT at each treatment level

foreach t of numlist 2 3 4 5 {
teffects ra (price mpg) (rep78), atet tlevel(1) control(`t’)
}
*** Notice the pattern. r1vs`t' is always the first ATET estimate in each table. I am assuming that only the r1 contrasts are true TUTs. So here are my queries:

1. In the binary case, it is easy to “flip” treatment and control. But in a multivalued case, does the flipping have to be done treatment by treatment estimating t-1 equations?
2. If I want to report TUTs only for all treatment levels compared to r1, am I correct that I would need to extract the r1vs`t’ coefficients for each treatment level equation?
3. I am at a loss for how to do balance diagnostics in this case such as tebalance summarize, would that need to be done equation by equation as well?

Any suggestions would be welcome!

Cheers,

Michael Howell-Moroney

************************************************** ****
Michael Howell-Moroney, Ph.D.
Professor
Department of Public and Nonprofit Administration
School of Urban Affairs and Public Policy
The University of Memphis
134 McCord Hall
Memphis, TN 38152

Voice: 901.678.2640
Fax: 901.678.2981

http://www.memphis.edu/padm/