Stop adjusting for anything: How the backdoor and adjustment criteria empower causal human-centric research

Causal inference is increasingly recognized as essential in human-centric building research, yet selecting appropriate adjustment variables remains a methodological challenge when estimating causal effects. Misusing statistical adjustments can introduce or exacerbate bias, leading to distorted effect estimates and suboptimal design or policy decisions that may fail to support—or even harm—people. This study addresses this challenge by advancing the use of causal inference tools in human-centric research. We introduce how to apply causal diagrams, fundamental causal structures, and the backdoor and adjustment criteria to guide the selection of valid adjustment sets. Using simulation-based examples grounded in human-related scenarios, we demonstrate how bias can emerge—and be mitigated—within both frequentist and Bayesian frameworks. Unlike complex statistical techniques, this causal diagram framework emphasizes visual reasoning and transparent assumptions, making it particularly accessible to building scientists who may not have extensive statistical training but still require methodological rigor. We contend that adopting this framework enhances the validity, transparency, interpretability, and reproducibility of causal research in human-centric building science.