SVMMatch (version 1.1)

SVMMatch-package: Title: Causal effect estimation and diagnostics with support vector machines.

Description

SVMMatch identifies the region of common support between a set of treated and control units in observational data. Using the observations in this region, a set of balancing weights and a treatment effect are estimated. The method, described in Ratkovic (2014), adapts the support vector machine technology in order to estimate these balancing weights, using a Bayesian implementation so as to give uncertainty effects both in treatment assignment and effect estimation.

Details

Package:

SVMMatch

Type:

Package

Version:

1.0

Date:

2015-02-05

License:

GPL (>= 2)

The method implements the matching algorithm through the main function, svmmatch.

A series of diagnostics are implemented. The function balance() assesses the posterior density of covariate imbalance; effect() returns the posterior estimate of the treatment effect; sensitivity() assesses the effect estimate's sensitivity to unobserved confounders; control.overlap() returns the posterior density of number of control observations returned in matching; and treatment.overlap() examines difficult-to-match treated observations.

References

Ratkovic, Marc. 2014. "Balancing within the Margin: Causal Effect Estimation with Support Vector Machines." Working paper.