Dears,
I want to analyse data from experimental markets. I have 4 different treatments and 3 markets within each treatment.
We introduce a dummy variable for each of our n markets but one. In this way we control for some hydiosincracy in market behavior.
Then I run a tobit model where I have n-1 covariates, and I get my results.
To test pairwise treatments effects on the dependent variable we use lincom. For instance, to test, differences between treatment 1 and 2:
lincom Mkt(1) + Mkt(2) + Mkt(3) - Mkt(4) - Mkt(5) - Mkt(6)
where the first 3 markets belong to Treatment 1 and markets 4...6 belong to Treatment 2. This way I get my results.
if, in order to perform the same comparison, I first drop from the original dataset all the observations which do not belong to treatments 1 and 2 and I run the tobit model on the restricted data and after I run the same lincom as before:
lincom Mkt(1) + Mkt(2) + Mkt(3) - Mkt(4) - Mkt(5) - Mkt(6)
from the lincom test I get the same coefficient as in the non restricted model but different standard errors which, of course, change my p-value.
Which of the two ways is more appropriate? Why?
Many thanks
I want to analyse data from experimental markets. I have 4 different treatments and 3 markets within each treatment.
We introduce a dummy variable for each of our n markets but one. In this way we control for some hydiosincracy in market behavior.
Then I run a tobit model where I have n-1 covariates, and I get my results.
To test pairwise treatments effects on the dependent variable we use lincom. For instance, to test, differences between treatment 1 and 2:
lincom Mkt(1) + Mkt(2) + Mkt(3) - Mkt(4) - Mkt(5) - Mkt(6)
where the first 3 markets belong to Treatment 1 and markets 4...6 belong to Treatment 2. This way I get my results.
if, in order to perform the same comparison, I first drop from the original dataset all the observations which do not belong to treatments 1 and 2 and I run the tobit model on the restricted data and after I run the same lincom as before:
lincom Mkt(1) + Mkt(2) + Mkt(3) - Mkt(4) - Mkt(5) - Mkt(6)
from the lincom test I get the same coefficient as in the non restricted model but different standard errors which, of course, change my p-value.
Which of the two ways is more appropriate? Why?
Many thanks