causalweight (version 1.1.3)

paneltestDML: paneltestDML: Overidentification test for ATET estimation in panel data

Description

This function applies an overidentification test to assess if unconfoundedness of the treatments and conditional common trends as imposed in differences-in-differences jointly hold in panel data when evaluating the average treatment effect on the treated (ATET).

Usage

paneltestDML(y1, y0, d, x, trim = 0.05, MLmethod = "lasso", k = 4)

Value

A list with the following components:

est

Test statistic.

se

Standard error.

pval

P-value.

ntrimmed

Number of trimmed or dropped observations due to propensitiy scores below the threshold trim.

pscore.xy0

Propensity score under unconfoundedness.

pscore.x

Propensity score under conditional common trends.

ATETselobs

ATET based on the selection on observables/unconfoundedness assumption.

seATETselobs

Standard error of the ATET based on the selection on observables/unconfoundedness assumption.

ATETdid

ATET based on difference-in-differences invoking the conditional common trends assumption.

seATETdid

Standard error of the ATET based on difference-in-differences invoking the conditional common trends assumption.

Arguments

y1

Outcome variable in the post-treatment period. Should not contain missing values.

y0

Outcome variable in the pre-treatment period. Should not contain missing values.

d

Treatment group indicator (binary). Should not contain missing values.

x

Covariates to be controlled for. Should not contain missing values.

trim

Trimming threshold for discarding observations with too extreme treatment propensity scores (smaller than trim or larger than 1-trim). Default is 0.05.

MLmethod

Machine learning method for estimating nuisance parameters using the SuperLearner package. Must be one of "lasso" (default), "randomforest", "xgboost", "svm", "ensemble", or "parametric".

k

Number of folds in k-fold cross-fitting. Default is 4.

Details

The test statistic corresponds to the difference between the ATETs that are based on two distinct doubly robust score functions, namely that under unconfoundedness and that based on difference-in-differences under conditional common trends. Estimation in panel data is based on double machine learning and the function supports different machine learning methods to estimate nuisance parameters (conditional mean outcomes and propensity scores) as well as cross-fitting to mitigate overfitting. ATETselobs and ATETdid equals zero.

References

Huber, M., and Oeß, E.-M. (2024): "A joint test of unconfoundedness and common trends", arXiv preprint 2404.16961.

Examples

if (FALSE) {
n=1000
x=data.frame(rnorm(n), rnorm(n))
d=1*(0.5*x[,1]+rnorm(n)>0)
y0=rnorm(n)
y1=0.5*x[,1]+y0+d+rnorm(n)
# report p-value (note that unconfoundedness and common trends hold jointly)
paneltestDML(y1=y1, y0=y0, d=d, x=x)$pval
}