Title: Instrumental Variables Estimation for 2^k Factorial Experiments
Version: 0.1.0
Description: Implements instrumental variable estimators for 2^K factorial experiments with noncompliance.
Depends: R (≥ 2.10)
Imports: stats, generics, Formula
Suggests: tibble, dplyr, rmarkdown, knitr
VignetteBuilder: knitr
License: GPL-2 | GPL-3 [expanded from: GPL (≥ 2)]
URL: https://github.com/mattblackwell/factiv
BugReports: https://github.com/mattblackwell/factiv/issues
Encoding: UTF-8
LazyData: true
RoxygenNote: 7.1.1
NeedsCompilation: no
Packaged: 2021-05-20 01:16:30 UTC; mblackwell
Author: Matthew Blackwell ORCID iD [aut, cre], Nicole Pashley [aut]
Maintainer: Matthew Blackwell <mblackwell@gov.harvard.edu>
Repository: CRAN
Date/Publication: 2021-05-21 08:00:05 UTC

Complier covariate profiles

Description

Calculates averages of covariates by compliance group in a 2^K factorial setting.

Usage

compliance_profile(formula, data, subset)

Arguments

formula

one-sided formula to indicate the treatment assignment, treatment uptake, and covariates. The right-hand side of the formula should have three components separated by the | symbol, with the first component containing the K binary treatment variables (treatment assignment), the second component containing the K binary instruments associated with each treatment variable (treatment uptake), and the third giving the covariates to be included in the profile. The order of the variables in the first two parts of the formula must match.

data

a data.frame on which to apply the formula.

subset

subset of the data to pass to estimation.

Value

A list with two objects:

raw_table

a data.frame whose rows represent the covariates and whose columns represent the different compliance groups. Each entry is the estimated mean of the covariate for that compliance group.

std_table

a data.frame similarly structured to raw_table but with the standardized difference between the compilance group means and the overall means in place of the raw means.

Author(s)

Matthew Blackwell

References

Matthew Blackwell and Nicole Pashley (2021) "Noncompliance in Factorial Experiments." Working paper.

Examples


data(newhaven)

cov_prof <- compliance_profile(~ inperson + phone | inperson_rand
  + phone_rand | age + maj_party + turnout_96, data = newhaven)

cov_prof

IV Estimation of 2^K Factorial Design

Description

Estimates principal stratum-specific effects and interactions in a 2^K factorial experiment

Usage

iv_factorial(formula, data, subset, method = "lm", level = 0.95)

Arguments

formula

formula specification of the factorial design with noncompliance. The right-hand side of the formula should have two components separated by the | symbol, with the first component containing the K binary treatment variables and the second component containing the K binary instruments associated with each treatment variable. The order of the variables in the formula must match.

data

A data.frame on which to apply the formula.

subset

subset of the data to pass to estimation.

method

character indiciating if the estimator should be "lm" using the least squares approach (default) or "cmd" to estimate via efficent minimum distance estimator.

level

the confidence level required.

Details

This function estimates treatment effects for 2^K factorial experiments in the face of noncompliance on all factors. A monotonicity assumption is assumed for both treatment-instrument pairs, along with treatment exclusion. See Blackwell (2017) for more details on those assumptions.

The procedure uses iterative generalized method of moments (GMM) to estimate both the proportions of each compliance class (also known as principal strata) and the average potential outcomes within those classes. It also provides estimates of several one-way, joint, and interactive treatment effects within these classes.

Under the above assumptions, the compliance classes are the product of the compliance classes for each treatment-instrument pair. For instance, "cc" is the class that would comply with both treatments, "ca" is the class that would comply with the first treatment and always take the second treatment, and "cn" is the class that would comply with the first treatment and never take the second treatment. Finally, note that treatment effects are only well-defined for compliance classes for which there is compliance on at least one treatment.

Value

A list of class iv_factorial that contains the following components:

rho

vector of estimated compliance class probabilities.

psi

vector of the estimated conditional mean of the outcome within the compliance classes.

vcov

estimated asymptotic variance matrix of the combined rho and psi parameters.

pcafe_est

vector of estimated main effects of each factor among perfect compliers.

pcafe_se

vector of estimated standard errors for the estimated effects in tau.

pcafe_cis

a matrix of confidence intervals for the PCAFE estimates.

level

the confidence level of the returned confience intervals.

Author(s)

Matt Blackwell

References

Matthew Blackwell (2017) Instrumental Variable Methods for Conditional Effects and Causal Interaction in Voter Mobilization Experiments, Journal of the American Statistical Association, 112:518, 590-599, doi: 10.1080/01621459.2016.1246363

Matthew Blackwell and Nicole Pashley (2020) "Noncompliance in Factorial Experiments." Working paper.

Examples

data(newhaven)

out <- iv_factorial(turnout_98 ~ inperson + phone | inperson_rand
  + phone_rand, data = newhaven)

summary(out)

Finite-Sample IV Estimation of 2^K Factorial Design

Description

Estimate main effect IV ratios for 2^K factorial experiments

Usage

iv_finite_factorial(formula, data, subset, level = 0.95)

Arguments

formula

formula specification of the factorial design with noncompliance. The right-hand side of the formula should have two components separated by the | symbol, with the first component containing the K binary treatment variables and the second component containing the K binary instruments associated with each treatment variable. The order of the variables in the formula must match.

data

a data.frame on which to apply the formula.

subset

subset of the data to pass to estimation.

level

the confidence level required.

Details

This function estimates the ratio of the effect of treatment assignment on the outcome to the effect of treatment assignment on treatment uptake in 2^K factorial experiments. The approach uses finite sample asymptotic inference to generate confidence intervals.

Value

A list of class iv_finite_factorial that contains the following components:

tau

a vector of estimated effect ratios for each factor.

tau_cis

a matrix of confidence intervals for each effect ratio. This matrix has 4 columns because it is possible to have disjoint confidence intervals in this method.

tau_y

a vector of the estimated effects on the outcome.

v_tau_y

the estimated sample variances of the effects on the outcome.

tau_d

a vector of the estimated effects on treatment uptake.

v_tau_y

the estimated sample variances of the effects on treatment uptake.

level

the confidence level of tau_cis.

Author(s)

Matt Blackwell

References

Matthew Blackwell and Nicole Pashley (2021) "Noncompliance in Factorial Experiments." Working paper.

Examples

data(newhaven)

out <- iv_finite_factorial(turnout_98 ~ inperson + phone | inperson_rand
  + phone_rand, data = newhaven)

out

joint <- iv_finite_factorial(turnout_98 ~ inperson + phone |
  inperson_rand + phone_rand, data = newhaven)

joint

New Haven Voter Modelization Field Experiment

Description

A data set that contains a subset of the voter mobilzation field experiment analyzed in Gerber and Green (2000).

Format

A data frame with 7865 observations and 6 variables:

ward

ward of the residence.

turnout_98

indicator variable for voting in the 1998 general election.

inperson

indicator variable for if the respondent received in-person canvassing.

phone

indicator variable for if the respondent received phone canvassing.

inperson_rand

indicator variable for if the respondent was randomized to receive in-person canvassing.

phone_rand

indicator variable for if the respondent was randomized to receive phone canvassing.

age

respondent age.

maj_party

indicator variable for if the respondent is a registered Democrat or Republican.

turnout_96

indicator variable for whether or not the respondent voted in the 1996 general election.

Details

Data was cleaned and used in Bowers and Hansen (2009). It was a 2x2 factorial design with noncompliance on both factors. This is the subset of the subjects who lived in a single-person household and were not randomized to receive get-out-the-vote mailers. Blackwell (2017) analyzed the full data adjusting for noncompliance.

Source

https://doi.org/10.7910/DVN/Q6CID7

References

Gerber, Alan S., and Donald P. Green. “The Effects of Canvassing, Telephone Calls, and Direct Mail on Voter Turnout: A Field Experiment.” The American Political Science Review 94, no. 3 (2000): 653. https://doi.org/10.2307/2585837.

Hansen, Ben B., and Jake Bowers. “Attributing Effects to a Cluster-Randomized Get-Out-the-Vote Campaign.” Journal of the American Statistical Association 104, no. 487 (2009): 873–85. https://doi.org/10.1198/jasa.2009.ap06589.

Blackwell, Matthew. “Instrumental Variable Methods for Conditional Effects and Causal Interaction in Voter Mobilization Experiments.” Journal of the American Statistical Association 112, no. 518 (2017): 590–99. https://doi.org/10.1080/01621459.2016.1246363.

Objects exported from other packages

Description

These objects are imported from other packages. Follow the links below to see their documentation.

generics

tidy

Tidy an iv_factorial object

Description

Tidy summarizes information about the components of a model.

Usage

## S3 method for class 'iv_factorial'
tidy(x, conf.int = FALSE, conf.level = 0.95, ...)

Arguments

x

An iv_factorial object produced by a call to iv_factorial()

conf.int

Logical indicating whether or not to include a confidence interval in the tidied output. Defaults to FALSE

conf.level

The confidence level to use for the confidence interval if conf.int = TRUE. Must be strictly greater than 0 and less than 1. Defaults to 0.95, which corresponds to a 95 percent confidence interval.

...

Additional arguments. Not used. Needed to match generic signature only.

Value

A tibble::tibble() with columns:

term

The name of the effect term.

estimand

Which complier effect being estimated.

estimate

The estimated value of the effect.

std.error

The estimated standard error of the effect.

conf.low

Lower bound of the confidence interval for the estimate.

conf.high

Upper bound of the confidence interval for the estimate.

Author(s)

Matt Blackwell

Tidy an iv_finite_factorial object

Description

Tidy summarizes information about the components of a model.

Usage

## S3 method for class 'iv_finite_factorial'
tidy(x, conf.level = 0.95, ...)

Arguments

x

An iv_factorial object produced by a call to iv_finite_factorial()

conf.level

The confidence level to use for the confidence interval. Must be strictly greater than 0 and less than 1. Defaults to 0.95, which corresponds to a 95 percent confidence interval.

...

Additional arguments. Not used. Needed to match generic signature only.

Value

A tibble::tibble() with columns:

term

The name of the effect term.

estimand

Which complier effect being estimated.

estimate

The estimated value of the effect.

ci_1_lower

Lower bound for the first interval of the Fieller confidence regresion for the estimate.

ci_1_upper

Upper bound for the first interval of the Fieller confidence regresion for tshe estimate.

ci_2_lower

Lower bound for the second interval of the Fieller confidence regresion for the estimate. Only non-NA when the confidence region is disjoint.

ci_2_upper

Upper bound for the second interval of the Fieller confidence regresion for the estimate. Only non-NA when the confidence region is disjoint.

Author(s)

Matt Blackwell