treatment effects regression adjustments

[Martha Smith]

Hello everyone,

I am working on the treatment effects regression adjustments (treatment variable is smoking with three levels - never, sometimes, frequently). I ran the unrestricted regress and the ATE values were statistically significant at 10% level of significance, however after I excluded jointly insignificant variables, the ATE for "sometimes vs frequently" became statistically insignificant. Can I still interpret those values?

Also, I would like to show my pomeans aequations results using graphs, however I am unable to find working command. Could you please advise me whether there exists such a command?

Thank you a lot in advance!
Martha

[Jeff Wooldridge]

Martha: As per the FAQ, you'll get better help if you show your commands and what Stata produced. Did you use -teffects ra-- or did you just run a regression with two treatment levels and add the controls?

[Martyna Czapnik]

Hi,

Actually, I have a very similar problem to Martha. I ran treatment effects regression adjustment (-teffects ra-) and one of my ATE turns out to be statistically insignificant even at 90% confidence level. Is it ok if I intepret this model as it is with a note that 2 vs 0 is statistically insignificant? Please see below an example of what I get. In reality, my database consists of over 4000 observations.

.    teffects ra (hrs_wrk freetime age i.relation i.children) (drinking)

Iteration 0:   EE criterion =  4.772e-28  
Iteration 1:   EE criterion =  3.987e-29  

Treatment-effects estimation                    Number of obs     =        179
Estimator      : regression adjustment
Outcome model  : linear
Treatment model: none
------------------------------------------------------------------------------
             |               Robust
     hrs_wrk |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
ATE          |
    drinking |
   (1 vs 0)  |  3.693813   1.675143     2.21   0.027     .4105928    6.977034
   (2 vs 0)  |   3.193974   2.105768     1.52   0.129    -.9332551    7.321202
-------------+----------------------------------------------------------------
POmean       |
    drinking |
          0  |   40.01521   1.352469    29.59   0.000     37.36442      42.666
------------------------------------------------------------------------------

Thank you!
Martyna

[Jeff Wooldridge]

Martyna: I'm pretty sure you can't reject the null that the 1 vs 0 and 2 vs 0 estimates are statistically different, so I might combine treatment levels 1 and 2 into the same category -- if it makes sense. Would that mean drinking versus not drinking?