README.md

In carpenitoThomas/misl: Multiple Imputation by Super Learning

misl

Multiple Imputation by Super Learning

The goal of MISL (Multiple Imputation by Super Learning) is to create multiply imputed datasets using the super learning framework. This package builds heavily off of the sl3 and mice packages.

This method has been accepted for publication in Statistical Methods in Medical Research. A link to the publication is forthcoming.

Installation

The MISL algorithm is not yet available on CRAN; instead, please use the development version available here on Github. To download the development version use:

install.packages("devtools")
devtools::install_github("carpenitoThomas/misl")

Issues

If you find any issues with this software or would like to make a request, please do so by filing an issue.

Example

Here’s an example with abalone data in which we use misl() imputation and then pool the results:

library(misl)
set.seed(123)

misl_imp <- misl(abalone, maxit = 10, m = 10, quiet = TRUE,
    con_method = c("Lrnr_glm_fast", "Lrnr_earth", "Lrnr_ranger"),
    bin_method = c("Lrnr_earth", "Lrnr_glm_fast", "Lrnr_ranger"),
    cat_method = c("Lrnr_independent_binomial", "Lrnr_ranger"))

misl_modeling <- lapply(misl_imp, function(y){
  stats::lm(Whole_Weight ~ Sex + Length + Diameter + Height + Older_12, data = y$datasets)
})

summary(mice::pool(misl_modeling), conf.int = TRUE)
#>          term      estimate   std.error    statistic        df      p.value
#> 1 (Intercept) -0.9917138520 0.017134443 -57.87838125 889.35930 0.000000e+00
#> 2        SexI -0.0426686559 0.009137655  -4.66954137 135.62160 7.174625e-06
#> 3        SexM  0.0002759165 0.007257143   0.03801998 174.92055 9.697151e-01
#> 4      Length  1.3620546956 0.206252463   6.60382269  30.35273 2.469290e-07
#> 5    Diameter  2.2108707044 0.250137186   8.83863267  32.66127 3.547722e-10
#> 6      Height  1.4354390562 0.138282429  10.38048769 101.74196 0.000000e+00
#> 7    Older_12  0.0474856116 0.008259594   5.74914621  45.01115 7.343897e-07
#>         2.5 %      97.5 %
#> 1 -1.02534251 -0.95808519
#> 2 -0.06073938 -0.02459794
#> 3 -0.01404692  0.01459875
#> 4  0.94103615  1.78307325
#> 5  1.70176219  2.71997922
#> 6  1.16114816  1.70972995
#> 7  0.03085005  0.06412117

We can also look at the traceplot of the imputations as well:

misl::trace_plot(misl_imp)

License

© 2021-2022 Thomas Carpenito, Justin Manjourides

The contents of this repository are distributed under the GPL-3 license. See file LICENSE for details.

References

Carpenito, Thomas and Manjourides, Justin. “MISL: Multiple Imputation by Super Learning.” Statistical Methods in Medical Research, In Press.

carpenitoThomas/misl documentation built on June 2, 2022, 12:30 p.m.