# Bounds on Distributional Treatment Effect Parameters

The joint distribution of potential outcomes is not typically identified under standard identifying assumptions such as selection on observables or even when individuals are randomly assigned to being treated. This package contains methods for obtaining tight bounds on distributional treatment effect parameters when panel data is available and under a Copula Stability Assumption as in Callaway (2017) < https://ssrn.com/abstract=3028251>.

The csabounds package allows the user to obtain tight bounds on distributional treatment effect parameters that depend on the joint distribution of treated and untreated potential outcomes -- the joint distribution is not identified under standard identifying assumptions such as selection on observables or even when experimental data is available. Using the package requires three periods of panel data and the results follow under the Copula Stability Assumption. More details about the method can be found in Callaway (2017).

## Installation

You can install csabounds from github with:

## Example

The following is a simplified example of the heterogeneous effects of job displacement considered in Callaway (2017) using data from the 1979 National Longitudinal Study of Youth.

A subset of the data is contained in the displacements dataset contained in the package. delt.seq contains a vector of values to compute the distribution of the treatment effect for; that is, (F_{Y_{1t}-Y_{0t}|D_t=1}(\delta)). y.seq contains possible values for outcomes.

The first step is to compute the counterfactual distribution of untreated potential outcomes for the treated group, (F_{Y_{0t}|D_t=1}). In this example, I use the Change in Changes method (Athey and Imbens, 2006) which is available in the R qte package. The distribution of outcomes for the treated group in periods (t-1) and (t-2) also needs to be set and is not done automatically using the qte::CiC method.

Once the counterfactual distribution is avialable, one can run the main method to compute bounds on the distribution and the quantile of the treatment effect itself. (Recall the quantile of the treatment effect (QoTT) is significantly more challenging than the quantile treatment effect (QTT) because it depends on the distribution of treated and untreated potential outcomes.) Under the Copula Stability Assumption -- see Callaway (2017) -- one can construct significantly tighter bounds on the distribution and quantile of the treatment effect. These are calculated and plotted next.

# csabounds 1.0.0

• The first version of the package csabounds is available
• Added methods csabounds (the main method in the package)
• method attcpo (for computing the average treatment effect conditional on the previous outcome
• methods ggcsabounds and ggattcpo for plotting the results

# Reference manual

install.packages("csabounds")

1.0.0 by Brantly Callaway, 4 years ago

Browse source code at https://github.com/cran/csabounds

Authors: Brantly Callaway [aut, cre]

Documentation:   PDF Manual