Hi Statalists,
I am reaching out with some questions regarding the `tebounds’ command. I am wondering how the command can distinguish whether the outcome variable is a positive or a negative outcome (e.g., NOobese or Obese), specifically, when Monotone Treatment Response (MTR) assumption is imposed.
I am using a negative outcome (a dummy indicating if a household is food insecure or not) in my data, and the `tebounds’ output for the MTS+MTR assumptions give me a result [0.000, 0.137]. This is counterintuitive as my MTR assumption is that being in the food assistance program would not hurt the food insecurity status. In other words, it should be something like [a negative LB, 0.000].
To illustrate this concern, I am using the dataset “MT-SBPdictionary.zip” on http://qed.econ.queensu.ca/jae/datasets/millimet001/ to replicate the authors’ results on p.428-429 (https://journals.sagepub.com/doi/pdf...867X1501500205).
Following the authors, I use “NOTobese“ - a positive outcome and assume “negative selection” into the breakfast program. Although the values do not exactly match that presented in the paper, the general picture is consistent. Regarding the MTS+MTR assumptions, the Lower Bound is replaced by 0, implying the program participation will not hurt an individual’s “NOTobese” outcome.
So far, I have no questions about the estimation.
However, I am stuck when using a negative outcome (“Obese”). In the same story, the only thing changes are now I am measuring a “negative outcome” and assuming a “positive selection” into the program, i.e., those selected into the program are more likely to be obese.
Any comments would be appreciated. Thank you!
I am reaching out with some questions regarding the `tebounds’ command. I am wondering how the command can distinguish whether the outcome variable is a positive or a negative outcome (e.g., NOobese or Obese), specifically, when Monotone Treatment Response (MTR) assumption is imposed.
I am using a negative outcome (a dummy indicating if a household is food insecure or not) in my data, and the `tebounds’ output for the MTS+MTR assumptions give me a result [0.000, 0.137]. This is counterintuitive as my MTR assumption is that being in the food assistance program would not hurt the food insecurity status. In other words, it should be something like [a negative LB, 0.000].
To illustrate this concern, I am using the dataset “MT-SBPdictionary.zip” on http://qed.econ.queensu.ca/jae/datasets/millimet001/ to replicate the authors’ results on p.428-429 (https://journals.sagepub.com/doi/pdf...867X1501500205).
keep obese break1 ses
gen NOTobese = 1 if obese == 0
replace NOTobese = 0 if obese == 1
tebounds NOTobese, treat(break1) erates(0) ncells(20) miv(ses) bs reps(100)
Following the authors, I use “NOTobese“ - a positive outcome and assume “negative selection” into the breakfast program. Although the values do not exactly match that presented in the paper, the general picture is consistent. Regarding the MTS+MTR assumptions, the Lower Bound is replaced by 0, implying the program participation will not hurt an individual’s “NOTobese” outcome.
So far, I have no questions about the estimation.
However, I am stuck when using a negative outcome (“Obese”). In the same story, the only thing changes are now I am measuring a “negative outcome” and assuming a “positive selection” into the program, i.e., those selected into the program are more likely to be obese.
tebounds obese, treat(break1) erates(0) ncells(20) miv(ses) bs reps(100)I have a hard time interpreting the result of the MTS+MTR assumptions, [0.000, 0.072]. If we indicate that participating in the program will not hurt an individual’s “Obese” outcome, it is the Upper Bound that should be replaced by 0, if I understood correctly. In other words, the Lower Bound should remain the same as those in the MTS assumption.
Any comments would be appreciated. Thank you!
Comment