d <- read.csv("../naijialiu/Google Drive/My Drive/NaijiaLiu.github.io/Gov_51/Causal/simulated_iv.csv") View(d) ###### Add a type variable to indicate the type of observations d$type <- ifelse(d$treatment_0==1 & d$treatment_1==1, "always-taker", ifelse(d$treatment_0==0 & d$treatment_1==0, "never-taker","complier")) table(d$type) ############ Add a variable to indicate the difference in outcome for everyone d$difference <- d$Y1 - d$Y0 ########## Calculate ATE if were not under IV ate <- sum(d$difference)/nrow(d) ate ######## why is this wrong??? # we are not sure whether there is a randomized assignment of treatments. Hence it is dangerous to utilize difference in means. ######## Calculate treatment effect among compliers d[d$complier==1,] subset(d,d$complier==1) late <- sum(d$difference[d$complier==1])/sum(d$complier==1) late <- sum(d$difference[d$type=="complier"])/sum(d$type=="complier") late ######### Calculate Intention to Treat itt <- sum(d$difference[d$complier==1])/nrow(d) encourage <- sum(d$complier==1)/nrow(d) itt/encourage ### this is also called local average treatment effect (LATE)