Weights not working in datasummary_balance

[kleuveld]

datasummary_balance's help file states: "If data includes columns named "blocks", "clusters", or "weights", this information will be taken into account automatically by estimatr::difference_in_means."

However, this doesn't seem to be the case:

set.seed(1)
library(randomizr)
library(estimatr)
library(dplyr)
library(modelsummary)


dat <- tibble(x = rnorm(n = 1000),
       clusters = simple_ra(N = 1000,num_arms = 50),
       Z = cluster_ra(clusters = clusters),
       y = x + Z * 5,
       weights = if_else(y > 5,0,1)) # making sure the weights actually make a difference


# reports means of 0 and 5, and a difference of 5
datasummary_balance(y ~ Z, data = dat, output = "flextable")


# hoewer, the weighted mean of the treatment group is 4.2
weighted.mean(dat$y[dat$Z == 1], dat$weights[dat$Z == 1])


# and these should be the differences
difference_in_means(y ~ Z, 
                    clusters = clusters,
                    weights = weights,
                    data = dat)


# this is what datasummary_balance reports
difference_in_means(y ~ Z, 
                    data = dat)

I don't think the clusters are applied correctly either, but I couldn't reproduce that.

Session info:
R version 4.4.1 (2024-06-14 ucrt)
Platform: x86_64-w64-mingw32/x64
Running under: Windows 11 x64 (build 22631)

Matrix products: default

locale:
[1] LC_COLLATE=English_United Kingdom.utf8 LC_CTYPE=English_United Kingdom.utf8 LC_MONETARY=English_United Kingdom.utf8
[4] LC_NUMERIC=C LC_TIME=English_United Kingdom.utf8

time zone: Europe/Amsterdam
tzcode source: internal

attached base packages:
[1] stats graphics grDevices utils datasets methods base

other attached packages:
[1] modelsummary_2.2.0.4 dplyr_1.1.4 estimatr_1.0.4 randomizr_1.0.0

loaded via a namespace (and not attached):
[1] utf8_1.2.4 generics_0.1.3 fontLiberation_0.1.0 xml2_1.3.6 httpcode_0.3.0
[6] digest_0.6.35 magrittr_2.0.3 evaluate_1.0.1 grid_4.4.1 estimability_1.5.1
[11] mvtnorm_1.3-1 flextable_0.9.6 fastmap_1.2.0 jsonlite_1.8.9 zip_2.3.1
[16] backports_1.5.0 crul_1.5.0.91 Formula_1.2-5 promises_1.3.0 fansi_1.0.6
[21] box_1.2.0 fontBitstreamVera_0.1.1 textshaping_0.4.0 cli_3.6.2 shiny_1.9.1
[26] rlang_1.1.3 fontquiver_0.2.1 crayon_1.5.3 gfonts_0.2.0 gdtools_0.3.7
[31] officer_0.6.6 tools_4.4.1 uuid_1.2-0 checkmate_2.3.2 httpuv_1.6.15
[36] curl_5.2.1 vctrs_0.6.5 R6_2.5.1 mime_0.12 lifecycle_1.0.4
[41] emmeans_1.10.5 ragg_1.3.2 insight_1.0.0.1 pkgconfig_2.0.3 pillar_1.9.0
[46] later_1.3.2 data.table_1.15.4 glue_1.8.0 Rcpp_1.0.12 systemfonts_1.1.0
[51] xfun_0.45 tibble_3.2.1 tidyselect_1.2.1 rstudioapi_0.17.1 knitr_1.48
[56] xtable_1.8-4 htmltools_0.5.8.1 tables_0.9.28 rmarkdown_2.28 compiler_4.4.1
[61] askpass_1.2.1 openssl_2.2.0

[vincentarelbundock]

Thanks a lot for the detailed report @kleuveld. I really appreciate that you took the time to craft a reproducible example.

I can replicate and confirm the problem on my computer.

Life is kind of crazy right now, but I plan to prep a new modelsummary release by the end of the month, and expect to look into a fix for this.

[kleuveld]

Hi,

Don't worry, even with a bug report that's waaaaay more elaborate than this, modelsummary would have still saved me so much time! Thanks for the good work!

Perhaps the quickest fix is simply removing mention of weights in the help file?

[vincentarelbundock]

Thanks again for the report!

This should be fixed in the development version on Github, so you can install and use it now. I plan to release a new version to CRAN before the end of January.