An overview of all established frameworks that to the authors’ knowledge offer DAG-based simulation functionalities, describing for each package the main purpose, the type of data it simulates, the functional forms used, and the additional simulation utilities provided.

The bnlearn package [9] can both infer parameters and simulate data from a model, with numerical variables and functional forms restricted to full conditional probability tables and linear regression models. The pgmpy package [16] is similar to bnlearn in terms of its purpose and simulation functionalities. The package simCausal [11] is more aimed toward causal inference problems and thus focuses on simulating longitudinal data based on SEMs. The main goal of the simMixedDAG package [12] is to simulate “real life” datasets based on a learned generalised additive model or user-defined parametric linear models. The package MXM [7] simulates data from multivariate Gaussian distributions based on a user-defined or randomly generated adjacency matrix, while abn [6] simulates data from Poisson, multinomial, and Gaussian distributions based on a user-defined adjacency matrix. The packages dagitty [10], dagR [17], and lavaan [8] provide similar functionalities for simulating data based on SEMs.