Repository files navigation

optweight contains functions to estimate stable balancing weights thatbalance covariates up to given thresholds. It solves a convexoptimization problem to minimize a function of the weights that capturestheir variability (or divergence from a set of base weights). This isthe method described in Zubizarreta(2015), Källberg and Waernbaum(2023), and Wang andZubizarreta (2020).optweight extends the method to multi-category, continuous, andmultivariate treatments and provides a simple user interface andcompatibility with thecobaltpackage for balance assessment. Seevignette("optweight") for a morethorough description of the package’s capabilities.

To installoptweight, use the code below:

#CRAN versioninstall.packages("optweight")#Development versionpak::pkg_install("ngreifer/optweight")

Below is an example of estimating weights withoptweight and assessingbalance on the covariates withcobalt.

library("optweight")library("cobalt")data("lalonde")# Estimate weightsow<- optweight(treat~age+educ+race+nodegree+married+re74+re75,data=lalonde,estimand="ATT",tols=.01,min.w=0)ow

#> An optweight object#>  - number of obs.: 614#>  - norm minimized: "l2"#>  - sampling weights: present#>  - base weights: present#>  - treatment: 2-category#>  - estimand: ATT (focal: 1)#>  - covariates: age, educ, race, nodegree, married, re74, re75

# Information about the weightssummary(ow)

#>                   Summary of weights#> #> - Weight ranges:#> #>         Min                                 Max#> treated   1      ||                       1.   #> control   0 |---------------------------| 5.588#> #> - Units with the 5 most extreme weights by group:#>                                      #>             1     2     3     4     5#>  treated    1     1     1     1     1#>           423   388   226   196   118#>  control 5.27 5.298 5.324 5.479 5.588#> #> #> - Weight statistics:#> #>            L2    L1    L∞ # Zeros#> treated 0.    0.    0.          0#> control 1.663 1.302 4.588     231#> #> - Effective Sample Sizes:#> #>            Control Treated#> Unweighted   429.      185#> Weighted     113.9     185

# Covariate balancebal.tab(ow)

#> Balance Measures#>                Type Diff.Adj#> age         Contin.     0.01#> educ        Contin.     0.01#> race_black   Binary     0.01#> race_hispan  Binary     0.00#> race_white   Binary    -0.01#> nodegree     Binary     0.01#> married      Binary    -0.01#> re74        Contin.    -0.01#> re75        Contin.     0.01#> #> Effective sample sizes#>            Control Treated#> Unadjusted   429.      185#> Adjusted     113.9     185

We can see that all standardized mean differences are at or below .01 inabsolute value, as requested using thetols argument. Because we setmin.w = 0, some units received weights of 0, effectively dropping themfrom the sample (by default, the smallest weight allowed is$10^{-8}$).

We can useplot() to examine the dual variables for each constraint,which represent how active that constraint is at the optimal point.Highly active constraints affect the objective function value the mostwhen their tolerances are changed.

plot(ow)

We can see thatrace has the highest dual variable; relaxing theconstraint onrace would yield the biggest improvement in effectivesample size, while tightening its constraint would yield the biggestdecrease in effective sample size.

The lower-level functionoptweight.fit() operates on the covariatesand treatment variables directly.optweightMV() supports multivariate(i.e., multiple) treatments.

In addition to estimating balancing weights for estimating treatmenteffects,optweight can estimate sampling weights for generalizing anestimate to a new target population defined by covariate moments usingoptweight.svy(), which implements the methods described in Jackson,Rhodes, and Ouwens (2021)for matching-adjusted indirect comparison (MAIC).

To citeoptweight, please usecitation("optweight") to generate thecorrect reference. Be sure to include the version of the package. Pleasesubmit bug reports, questions, comments, or other issues tohttps://github.com/ngreifer/optweight/issues.