Incomplete data is a common challenge when analyzing real-world data. Another challenge is choosing from many available multiple imputation approaches to fill-in the missing values. Marbach (2021) suggests adopting the imputation approach that generates a density of imputed values most similar to those of the observed values for an incomplete variable after balancing all other covariates. The package facilitates the computation of discrepancy statistics summarizing differences between the density of imputed and observed values and the construction of weights to balance covariates.
Installation
remotes:::install_github("sumtxt/missDiag")To install the remotes package use install.packages("remotes").
Requirements
The package can be installed with few dependencies but to construct weights the package relies on either the ebal package (Hainmueller, 2012) or the sbw package (Zubizarreta 2015). These two packages have be installed by the users separately.
To install these packages from CRAN use:
remotes::install_cran("ebal")
remotes::install_cran("sbw")Usage
The package comes with some example data from the 2008 American National
Election Study (anes08) in which, for example, about 11% of the values
in the vote-choice variable are missing. The data have been imputed five
times using random value imputation (output data: anes08_rng) and
predictive mean matching imputation (anes08_pmm). We use the
missDiag package to corroborate that predictive mean matching produces
a density of imputed values that is more similar to the (reweighted)
density of observed values and should therefore be preferred over random
value imputation. We use entropy-balancing weights from the ebal
package as their computation is faster but stable weights from the sbw
package are preferable as their variance is smaller.
library(tidyverse)
#> ── Attaching packages ───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────── tidyverse 1.3.1 ──
#> ✔ ggplot2 3.3.5 ✔ purrr 0.3.4
#> ✔ tibble 3.1.3 ✔ dplyr 1.0.7
#> ✔ tidyr 1.1.3 ✔ stringr 1.4.0
#> ✔ readr 2.0.0 ✔ forcats 0.5.1
#> ── Conflicts ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────── tidyverse_conflicts() ──
#> ✖ dplyr::filter() masks stats::filter()
#> ✖ dplyr::lag() masks stats::lag()
library(missDiag)
diag_rng <- missDiag(
original=anes08,
imputed=anes08_rng,
adjust = 'ebal',
formula = vote ~ .)
diag_pmm <- missDiag(
original=anes08,
imputed=anes08_pmm,
adjust = 'ebal',
formula = vote ~ .)Averaging across the five multiple imputed datasets, we compare the standardized mean differences (SMD) for each of the vote choice categories and find that predictive mean matching does much better than random value imputation. The standardized mean differences are, on average, lower for predictive mean matching.
diag_rng %>% group_by(vname) %>%
summarize( smd=mean(diff_adj) )
#> # A tibble: 3 × 2
#> vname smd
#> <chr> <dbl>
#> 1 vote_McCain 0.0621
#> 2 vote_No vote/Other 0.0731
#> 3 vote_Obama 0.0577
diag_pmm %>% group_by(vname) %>%
summarize( smd=mean(diff_adj) )
#> # A tibble: 3 × 2
#> vname smd
#> <chr> <dbl>
#> 1 vote_McCain 0.0562
#> 2 vote_No vote/Other 0.0417
#> 3 vote_Obama 0.0512References
Moritz Marbach. 2021. Choosing Imputation Models, Political Analysis (forthcoming).
José R Zubizarreta. 2015. Stable Weights that Balance Covariates for Estimation with Incomplete Outcome Data, Journal of the American Statistical Association, 110(511): 910-922.
Jens Hainmueller. 2012. Entropy Balancing for Causal Effects: A Multivariate Reweighting Method to Produce Balanced Samples in Observational Studies, Political Analysis 20(1): 25–46.