Many researchers construct directed acyclic graph (DAG) models manually based on domain knowledge. Although numerous causal discovery algorithms were developed to automatically learn DAGs and other causal models from data, these remain challenging to use due to their tendency to produce results that contradict domain knowledge, among other issues. Here we propose a hybrid, iterative structure learning approach that combines domain knowledge with data-driven insights to assist researchers in constructing DAGs. Our method leverages conditional independence testing to iteratively identify variable pairs where an edge is either missing or superfluous. Based on this information, we can choose to add missing edges with appropriate orientation based on domain knowledge or remove unnecessary ones. We also give a method to rank these missing edges based on their impact on the overall model fit. In a simulation study, we find that this iterative approach to leverage domain knowledge already starts outperforming purely data-driven structure learning if the orientation of new edge is correctly determined in at least two out of three cases. We present a proof-of-concept implementation using a large language model as a domain expert and a graphical user interface designed to assist human experts with DAG construction.

Bayesian Networks (BNs) are used in various fields for modeling, prediction, and decision making. pgmpy is a python package that provides a collection of algorithms and tools to work with BNs and related models. It implements algorithms for structure learning, parameter estimation, approximate and exact inference, causal inference, and simulations. These implementations focus on modularity and easy extensibility to allow users to quickly modify/add to existing algorithms, or to implement new algorithms for different use cases. pgmpy is released under the MIT License; the source code is available at: https://github.com/pgmpy/pgmpy, and the documentation at: https://pgmpy.org.

Conditional independence (CI) tests underlie many approaches to model testing and structure learning in causal inference. Most existing CI tests for categorical and ordinal data stratify the sample by the conditioning variables, perform simple independence tests in each stratum, and combine the results. Unfortunately, the statistical power of this approach degrades rapidly as the number of conditioning variables increases. Here we propose a simple unified CI test for ordinal and categorical data that maintains reasonable calibration and power in high dimensions. We show that our test outperforms existing baselines in model testing and structure learning for dense directed graphical models while being comparable for sparse models. Our approach could be attractive for causal model testing because it is easy to implement, can be used with non-parametric or parametric probability models, has the symmetry property, and has reasonable computational requirements.

Measurement error is ubiquitous in many variables “latent-to-observed” (L2O) transformation from the MIIV approach and develop an equivalent graphical L2O transformation that allows applying existing graphical criteria to latent parameters in SEMs. We combine L2O transformation with graphical instrumental variable criteria to obtain an efficient algorithm for non-iterative parameter identification in SEMs with latent variables. We prove that this graphical L2O transformation with the instrumental set criterion is equivalent to the state-of-the-art MIIV approach for SEMs, and show that it can lead to novel identification strategies when combined with other graphical criteria.