In josherrickson/epb: Peters-Belson with Prognostic Heterogeneity

title: "Peters-Belson with Prognostic Heterogeneity" author: "Josh Errickson" date: "r Sys.Date()" output: rmarkdown::pdf_document vignette: > %\VignetteIndexEntry{Peters-Belson with Prognostic Heterogeneity} %\VignetteEngine{knitr::rmarkdown} %\VignetteEncoding{UTF-8}

knitr::opts_chunk$set(collapse = TRUE, prompt = TRUE)
library(pbph)

Basic Use

A short example. Lets predict the effect of an afterschool program on an end-of-term exam on the eottest (fake) data.

data(eottest)
mod1 <- lm(test ~ gpa + male, data = eottest, subset = (afterschool == 0))
mod2 <- pbph(mod1, treatment = afterschool, data = eottest)
summary(mod2)
confint(mod2)

The negative coefficient (-.4885) on pred shows evidence that students with lower predicted test score in the absence of treatment are showing a larger treatment effect than those with higher predicted test scores.

Logistic

data(salesdata)
mod1 <- glm(sale ~ experience + previoussales, data = salesdata, 
            subset = (newtechnique == 0), family = binomial)
mod2 <- pbph(mod1, treatment = newtechnique, data = salesdata)
summary(mod2)
confint(mod2)

Clusters

We implement support for clusters by overloading meat (and sandwich) from the sandwich package and adding a cluster argument.

mod1 <- lm(test ~ gpa + male, data = eottest, subset = (afterschool == 0))
mod2 <- pbph(mod1, treatment = afterschool, data = eottest, cluster = class)
summary(mod2)
confint(mod2)