Handling missing values when using neighborhood selection for network analysis

The handling of missing values in cross-sectional network analysis has been primarily studied in conjunction with regularization techniques. However, nonregularized alternatives, such as neighborhood selection via the Bayesian information criterion (BIC) based on node-wise multiple regression, have been shown to be viable alternatives for psychological networks. Moreover, its localized approach in model selection renders neighborhood selection particularly suitable for situations in which variables show very uneven rates of missing values. In this study, we present two approaches based on multiple imputation (MI), namely stacked and grouped MI, alongside direct and two-step expectation-maximization procedures, to handle missing values. Furthermore, various approaches to calculating sample size, used for computing log-likelihood and BIC, are discussed and evaluated. A simulation study was conducted to assess the performance of these missing data handling methods and sample size definitions. Evaluation criteria included edge recovery, as well as bias in partial correlations and network statistics. The findings indicate that stacked MI performs best overall. The two-step expectation-maximization approach is the fastest and offers adequate performance when the number of observations is very large relative to the proportion of missingness and the size of the network. For sample size calculation, particularly under high levels of missingness, using the local number of observations per node yielded the least bias. The most effective methods for handling missing values in neighborhood selection via BIC are implemented in the new R package mantar. (PsycInfo Database Record (c) 2026 APA, all rights reserved).