Weighting for Covariate Balance in Observational Studies

Generates weights to form equivalent groups in observational studies with point or longitudinal treatments by easing and extending the functionality of the R packages 'twang' for generalized boosted modeling (McCaffrey, Ridgeway & Morral, 2004) , 'CBPS' for covariate balancing propensity score weighting (Imai & Ratkovic, 2014) , 'ebal' for entropy balancing (Hainmueller, 2012) , 'optweight' for optimization-based weights (Zubizarreta, 2015) , 'ATE' for empirical balancing calibration weighting (Chan, Yam, & Zhang, 2016) , and 'SuperLearner' for stacked machine learning-based propensity scores (Pirracchio, Petersen, & van der Laan, 2015) . Also allows for assessment of weights and checking of covariate balance by interfacing directly with 'cobalt'.

WeightIt is a one-stop package to generate balancing weights for point and longitudinal treatments in observational studies. Contained within WeightIt are methods that call on other R packages to estimate weights. The value of WeightIt is in its unified and familiar syntax used to generate the weights, as each of these other packages have their own, often challenging to navigate, syntax. WeightIt extends the capabilities of these packages to generate weights used to estimate the ATE, ATT, and ATC for binary or multinomial treatments, and treatment effects for continuous treatments when available. In these ways, WeightIt does for weighting what MatchIt has done for matching, and MatchIt users will find the syntax familiar.

To install and load WeightIt, use the code below:

install.packages("WeightIt")  #CRAN version
devtools::install_github("ngreifer/WeightIt")  #Development version
library("WeightIt")

The workhorse function of WeightIt is weightit(), which generates weights from a given formula and data input according to methods and other parameters sepcified by the user. Below is an example of the use of weightit() to generate weights for estimating the ATE:

data("lalonde", package = "cobalt")
W <- weightit(treat ~ age + educ + nodegree + married + race + re74 + re75, 
    data = lalonde, method = "ps", estimand = "ATE")
print(W)

A weightit object
 - method: "ps" (propensity score weighting)
 - number of obs.: 614
 - sampling weights: none
 - treatment: 2-category
 - estimand: ATE
 - covariates: age, educ, nodegree, married, race, re74, re75

Evaluating weights has two components: evaluating the covariate balance produces by the weights, and evaluating whether the weights will allow for sufficient precision in the eventual effect estimate. For the first goal, functions in the cobalt package, which are fully compatible with WeightIt, can be used, as demonstrated below:

library("cobalt")
bal.tab(W, un = TRUE)

Call
 weightit(formula = treat ~ age + educ + nodegree + married + 
    race + re74 + re75, data = lalonde, method = "ps", estimand = "ATE")

Balance Measures
                Type Diff.Un Diff.Adj
prop.score  Distance  1.7569   0.1360
age          Contin. -0.2419  -0.1676
educ         Contin.  0.0448   0.1296
nodegree      Binary  0.1114  -0.0547
married       Binary -0.3236  -0.0944
race_black    Binary  0.6404   0.0499
race_hispan   Binary -0.0827   0.0047
race_white    Binary -0.5577  -0.0546
re74         Contin. -0.5958  -0.2740
re75         Contin. -0.2870  -0.1579

Effective sample sizes
           Control Treated
Unadjusted 429.000 185.000
Adjusted   329.008  58.327

For the second goal, qualities of the distributions of weights can be assessed using summary(), as demonstrated below.

summary(W)

Summary of weights:

- Weight ranges:
           Min                                   Max
treated 1.1721 |---------------------------| 40.0773
control 1.0092 |-|                            4.7432

- Units with 5 greatest weights by group:
                                                
             137     124     116      68      10
 treated 13.5451 15.9884 23.2967 23.3891 40.0773
             597     573     411     381     303
 control  4.0301  4.0592  4.2397  4.5231  4.7432

          Ratio Coef of Var
treated 34.1921      1.4777
control  4.7002      0.5519
overall 39.7134      1.3709

- Effective Sample Sizes:
           Control Treated
Unweighted 429.000 185.000
Weighted   329.008  58.327

Desirable qualities include ratios close to 1, coefficients of variation close to 0, and large effective sample sizes.

The table below contains the available methods in WeightIt for estimating weights for binary, multinomial, and continuous treatments using various methods and functions from various packages.

If you would like to see your package or method integrated into WeightIt, or for any other questions or comments about WeightIt, please contact Noah Greifer at [email protected].