Forum: help

Cannot calculate average treatment effect from predict [ Reply ]
By: Thai Nguyen on 2020-10-02 15:13

Predict sample selection.JPG (11) downloads

Hi everyone,

I am using sampleSelection package to run an endogenous switching regression model (Tobit 5). However, the average treatment effect on the treated (ATT) and the average treatment effect on the untreated (ATU) calculated from the predict command are different to results from movestay command in Stata. In detail, the predict create a matrix with 4 columns: "E[yo1|ys=0]", "E[yo1|ys=1]", "E[yo2|ys=0]", "E[yo2|ys=1]". When I look at the matrix, the results for column E[yo1|ys=0] and E[yo2|ys=1] are the same (compared to those in Stata), but in E[yo1|ys=1], yo1 is calculated on untreated, not the treated observations. I mean the rows (observations) that we use to calculate E[yo1|ys=1] should be the same as E[yo2|ys=1], only formula is different since they are outcome on the treated. This is the code that I use

df<-as.data.frame(predict( Results, type = "conditional", part="outcome"))

with Results is the object.

I attached a part of the outcome that I have

I dont know why I have an incorrect outcomes. Do you know other ways to predict the ATT and ATU after the switching regression model. Thank you.