Drawbacks on tradition approach to effect heterogeneity
look at one or two heterogeneity variables
overlook relevant variables
spurious findings
Solution to the drawbacks of the traditional approach to heterogeneity effects
pre-analyse plan with correlation for multiple hypothesis testing
simple ML methods
specific ML methods for causal inference
Advantages of DML
included multiple variables at the same time
countinous heterogeneity variables can be modelled non-parametrically (kernel or series regression)
computional attractive
we can use ML to model relationship between X and pseudo outcome
assign treatment to highest CAPO
Last changed7 months ago