mantar - Missingness Alleviation for NeTwork Analysis in R

mantar provides several methods for estimating psychological networks with and without missing data. For network estimation, two main approaches are implemented:

Neighborhood selection with information-criteria-based model selection
Regularized network estimation (e.g., glasso) combined with information criteria

For missing data handling, the preferred two strategies are:

stacked multiple imputation
two-step EM algorithm

The workflow is designed to support typical use cases in psychological research, including varying sample sizes, missingness levels, and variable types.

Installation

The current stable version (0.3.1) is available on CRAN and can be installed using the usual approach:

install.packages("mantar")

You can install the development version of mantar from GitHub. To do so, you need the remotes package.

# install.packages("remotes")
remotes::install_github("kai-nehler/mantar@develop")

The suffix @develop ensures you install the latest development version with new features and updates.

Basic Example

mantar offers a few dummy data sets for demonstration and testing purposes. Here, we illustrate how to estimate a neighborhood selection network on a continuous data set with missing values using stacked multiple imputation.

library(mantar)
data(mantar_dummy_mis_cont)

Network analysis based on neighborhood selection can be performed using the neighborhood_net() function. By default, it uses the Bayesian Information Criterion (BIC) for model selection.

result <- neighborhood_net(mantar_dummy_full_cont,
                           missing_handling = "stacked-mi")
#> No missing values in data. Sample size for each variable is equal to the number of rows in the data.
summary(result)
#> The density of the estimated network is 0.321
#> 
#> Network was estimated using neighborhood selection with the information criterion: BIC
#> and the 'and' rule for the inclusion of edges based on a full data set.
#> 
#> The sample sizes used for the nodewise regressions were as follows:
#> EmoReactivity     TendWorry    StressSens     SelfAware     Moodiness 
#>           400           400           400           400           400 
#>      Cautious ThoughtFuture RespCriticism 
#>           400           400           400
result$pcor
#>               EmoReactivity TendWorry StressSens SelfAware Moodiness  Cautious
#> EmoReactivity     0.0000000 0.2055099  0.0000000 0.0000000 0.0000000 0.0000000
#> TendWorry         0.2055099 0.0000000  0.0000000 0.3396183 0.0000000 0.0000000
#> StressSens        0.0000000 0.0000000  0.0000000 0.0000000 0.1143278 0.0000000
#> SelfAware         0.0000000 0.3396183  0.0000000 0.0000000 0.0000000 0.0000000
#> Moodiness         0.0000000 0.0000000  0.1143278 0.0000000 0.0000000 0.3290979
#> Cautious          0.0000000 0.0000000  0.0000000 0.0000000 0.3290979 0.0000000
#> ThoughtFuture     0.0000000 0.1855478  0.1728659 0.0000000 0.0000000 0.0000000
#> RespCriticism     0.0000000 0.0000000  0.0000000 0.1833561 0.2670060 0.2939731
#>               ThoughtFuture RespCriticism
#> EmoReactivity     0.0000000     0.0000000
#> TendWorry         0.1855478     0.0000000
#> StressSens        0.1728659     0.0000000
#> SelfAware         0.0000000     0.1833561
#> Moodiness         0.0000000     0.2670060
#> Cautious          0.0000000     0.2939731
#> ThoughtFuture     0.0000000     0.0000000
#> RespCriticism     0.0000000     0.0000000

The summary shows the density of the estimated network and details about the estimation procedure. The partial correlation matrix can be accessed via result$pcor.

Further Documentation

The package vignette provides:

network estimation workflows
missing-data recommendations
examples for both estimation approaches