FindIt (version 0.5)

INT: Estimating the AMTE, the ATCE and the AMTIE

Description

Estimating the average marginal treatment effect (AMTE), the average treatment combination effect (ATCE) and the average marginal treatment interaction effect (AMTIE) under the assumption of independently randomized factorial treatments. Researchers need to change the baseline condition by relevel the target.data.

Usage

INT(object, target.data, column, dist = "target", base,  sort = TRUE, compare = FALSE, order = 2)

Arguments

object

An output object from FindIt.

target.data

If dist= "target", this data is used to compute the AMTEs, ATCEs and AMTIEs. This data should represent the target population in terms of the treatment distribution and samples. The first column contains outcome variables and the other columns should be factorial treatments of interest.

column

Researchers can explore the AMTIEs within any factor interactions by specifying the names of factors through this argument.

dist

If dist="target", INT computes the AMTEs, ATCEs and AMTIEs for the target population specified by target.data. If dist="sample", it computes them for the data used to fit the model. If dist="unique", it computes them for the unique samples in the data used to fit the model.

base

An optional argument that specifies the baseline condition. This argument can be used only when column is used, and the order of base should follow the order of column.

sort

If sort=TRUE, the AMTE, the ATCE and the AMTIE are sorted in descending order.

compare

If compare=TRUE, INT computes the range of all factor interactions of the order specified with order. Researchers can identify the most important factor interaction for the specified order.

order

INT computes the ATCEs and AMTIEs of this order. Can be 2 or 3.

Value

Range of AMTIE

The range of the estimated AMTIEs

AMTIE

Estimated AMTIEs for treatment combinations

ATCE

Estimated ATCEs and AMTIEs for treatment combinations

Sum of AMTEs

Sum of the estimated AMTEs of each factor and AMTIEs for treatment combinations

Details

INT estimates the AMTEs, ATCEs and AMTIEs under the assumption of independently randomized factorial treatments.

References

Imai, Kosuke and Marc Ratkovic. 2013. ``Estimating Treatment Effect Heterogeneity in Randomized Program Evaluation.'' Annals of Applied Statistics, Vol.7, No.1(March), pp. 443-470. http://imai.princeton.edu/research/files/svm.pdf

Egami, Naoki and Kosuke Imai. 2015. ``Causal Interaction in High-Dimension.'' Working paper. http://imai.princeton.edu/research/files/int.pdf

Examples

################################################### 
## Conjoint Analysis: Causal Interaction.
################################################### 
data(Immigration)

## Not run: 
# ## The SVM classifier with a lasso constraint is estimated.
# ## The model includes all three-way and two-way interactions as 
# ## well as main effects of five factorial treatments.
# 
# ## Run to search for lambdas.
# F.conjoint<- FindIt(model.treat= outcome ~
#                     Education+Gender+Origin+Experience+Plans,
#                     nway=3,
#                     data = Immigration,
#                     type="binary",
#                     treat.type="multiple")
# 
# ## Make the full factorial design matrix as the target population. 
# full <- full.FindIt(F.conjoint)
# 
# ## Compute the predicted potential outcomes for the target population.
# Unifdata <- predict(F.conjoint,newdata=full,sort=FALSE)$data
# ## End(Not run)
## load pre-computed F.conjoint and Unifdata
data(F.conjoint)
data(Unifdata)

## Compute AMTEs, ATCEs, and AMTIEs

## Range of each factor interaction.
## The range of the AMTEs for all factors
compare1 <- INT(F.conjoint,target.data=Unifdata,compare=TRUE,order=1)
compare1

## The range of the two-way AMTIEs for all two-way factor interactions
compare2 <- INT(F.conjoint,target.data=Unifdata,compare=TRUE,order=2)
compare2

## Not run: 
# ## The range of the three-way AMTIEs for all three-way factor interactions
# compare3 <- INT(F.conjoint,target.data=Unifdata,compare=TRUE,order=3)
# compare3
# ## End(Not run)

## Compute AMTIEs within factor interactions.
## Origin x Experience
out.OP <- INT(F.conjoint,target.data=Unifdata,
              column=c("Origin","Experience"),
              base=c("India","No.job"), order=2)
out.OP

## Not run: 
# ## Education x Gender x Origin
# out.EGO <- INT(F.conjoint,target.data=Unifdata,
#                column=c("Education","Gender","Origin"),
#                base=c("No.formal","female","India"), order=3)
# out.EGO
# ## End(Not run)