The heterogeneous treatment effect estimation procedure
proposed by Imai and Ratkovic (2013)<DOI: 10.1214/12-AOAS593>.
The proposed method is applicable, for example, when selecting
a small number of most (or least) efficacious treatments from
a large number of alternative treatments as well as when
identifying subsets of the population who benefit (or are harmed
by) a treatment of interest. The method adapts the Support Vector
Machine classifier by placing separate LASSO constraints over the
pre-treatment parameters and causal heterogeneity parameters of
interest. This allows for the qualitative distinction between
causal and other parameters, thereby making the variable
selection suitable for the exploration of causal heterogeneity. The
package also contains the function, CausalANOVA, which estimates
the average marginal interaction effects by a regularized ANOVA as
proposed by Egami and Imai (2016+).