invertRIconfInt: Calculate treatment effect confidence intervals by inverting the randomization test

Description

Using an output object from seqblock or any other matrix or dataframe that includes a treatment and an outcome variable for multiple units, as well as blocking and non-blocking variables for the respective unit(s), invertRIconfInt calculates treatment effect confidence intervals by inverting the randomization inference test.

Usage

invertRIconfInt(dat, outcome.var, tr.var, tau.abs.min = -1, tau.abs.max = 1, 
  tau.length = 10, n.sb.p = 100, id.vars, id.vals, exact.vars = NULL, exact.vals = NULL,
  exact.restr = NULL, exact.alg = "single", covar.vars = NULL, covar.vals = NULL, 
  covar.restr = NULL, covars.ord = NULL, n.tr = 2, tr.names = NULL, assg.prob = NULL,
  seed = NULL, seed.dist, assg.prob.stat = NULL, trim = NULL, assg.prob.method = NULL, 
  assg.prob.kfac = NULL, distance = "mahalanobis", file.name = "sbout.RData", 
  query = FALSE, verbose = TRUE)

Value

A list with elements

ci95: vector of treatment effects within the 95% confidence interval
ci90: vector of treatment effects within the 90% confidence interval
ci80: vector of treatment effects within the 80% confidence interval

Arguments

dat: a matrix or dataframe containing the names and values of the different blocking and non-blocking variables, as well as each unit's treatment assignment and outcome
outcome.var: a string specifying the name of the outcome variable
tr.var: a string specifying the name of the treatment variable
tau.abs.min: lower bound of the range across which the confidence intervals will be computed
tau.abs.max: upper bound of the range across which the confidence intervals will be computed
tau.length: the number of (evenly spaced) possible treatment effects across the range specified by tau.abs.min and tau.abs.max for which location inside or outside the confidence intervals will be computed
n.sb.p: the number of times that sequential blocking will be performed on the dataset
id.vars: see the seqblock documentation
id.vals: see the seqblock documentation
exact.vars: see the seqblock documentation
exact.vals: see the seqblock documentation
exact.restr: see the seqblock documentation
exact.alg: see the seqblock documentation
covar.vars: see the seqblock documentation
covar.vals: see the seqblock documentation
covar.restr: see the seqblock documentation
covars.ord: see the seqblock documentation
n.tr: see the seqblock documentation
tr.names: see the seqblock documentation
assg.prob: see the seqblock documentation
seed: see the seqblock documentation
seed.dist: see the seqblock documentation
assg.prob.stat: see the seqblock documentation
trim: see the seqblock documentation
assg.prob.method: see the seqblock documentation
assg.prob.kfac: see the seqblock documentation
distance: see the seqblock documentation
file.name: see the seqblock documentation
query: see the seqblock documentation
verbose: see the seqblock documentation

Author

Ryan T. Moore rtm@wustl.edu and Jonathan Homola homola@wustl.edu

Details

invertRIconfInt takes a data matrix (or data frame) containing names and values of different blocking and non-blocking variables, as well as each unit's treatment assignment and outcome as input and returns a list of treatment effect confidence intervals.

Apart from specifying the treatment and outcome variable, the user can set all other arguments to seqblock when running invertRIconfInt. The function will then calculate the confidence intervals by employing a method described in Ho and Imai (2006), which inverts Fisher's exact test. The resulting confidence intervals are distribution-free, nonparametric and have accurate coverage probabilities.

References

Moore, Ryan T. and Sally A. Moore. 2013. "Blocking for Sequential Political Experiments." Political Analysis 21(4): 507-523.

Ho, Daniel E., and Kosuke Imai. 2006. "Randomization inference with natural experiments: An analysis of ballot effects in the 2003 California recall election." Journal of the American Statistical Association 101(475): 888-900.

Examples

Run this code

## Create an example data matrix with 50 observations that contains an ID variable, 
##   a dummy variable indicating gender, an age variable (between 18 and 55), a 
##   treatment variable and an outcome variable (between 15 and 20). 
## id <- seq(1,50,1)
## gender <- sample(c(1,2),50,replace=T)
## age <- sample(seq(18,55,1),50,replace=T)
## treat <- sample(c(1,2),50,replace=T)
## out <- treat + sample(seq(15,20,1),50,replace=T)
## data <- cbind(id, gender, age, out, treat)

## Check summary statistics for the created data
## aggregate(out~treat, data, mean)

## Run invertRIconfInt()
## invertRIconfInt(data, outcome.var="out", tr.var="treat", tau.abs.min = -3, 
##  tau.abs.max = 3, id.vars = "id", id.vals = "id", exact.vars = c("gender", "age"), 
##  exact.vals = c("gender", "age"))

Run the code above in your browser using DataLab