From 'What-is' to 'What-if' in Human-Factor Analysis: A Post-Occupancy Evaluation Case

TL;DR

Addressing the gap between descriptive analyses and causal questions in human-factor settings, the paper applies causal discovery to POE data from the CBE Occupant Survey using the Greedy Equivalence Search to learn a causal skeleton. It shows that only 28 of 124 variables are causally connected, with glare/Reflections emerging as an upstream driver whose modification yields cascading improvements downstream. Through a do-operator–like estimation and open-ended response validation, the study demonstrates the practical value of prioritizing upstream interventions over downstream symptoms. The work advocates integrating causal inference into POE practice and highlights limitations like unmeasured confounders and cross-sectional data, while outlining a path toward broader application and validation.

Abstract

Human-factor analysis typically employs correlation analysis and significance testing to identify relationships between variables. However, these descriptive ('what-is') methods, while effective for identifying associations, are often insufficient for answering causal ('what-if') questions. Their application in such contexts often overlooks confounding and colliding variables, potentially leading to bias and suboptimal or incorrect decisions. We advocate for explicitly distinguishing descriptive from interventional questions in human-factor analysis, and applying causal inference frameworks specifically to these problems to prevent methodological mismatches. This approach disentangles complex variable relationships and enables counterfactual reasoning. Using post-occupancy evaluation (POE) data from the Center for the Built Environment's (CBE) Occupant Survey as a demonstration case, we show how causal discovery reveals intervention hierarchies and directional relationships that traditional associational analysis misses. The systematic distinction between causally associated and independent variables, combined with intervention prioritization capabilities, offers broad applicability to complex human-centric systems, for example, in building science or ergonomics, where understanding intervention effects is critical for optimization and decision-making.

Figures (7)

• Figure 1: Survivorship bias during World War II: Survivor aircraft with skewed bullet hole distribution caused engineers to reinforce less important parts mangel1984abraham. The misuse of 'What-is' tools to address 'What-if' questions could lead to spurious conclusions. In contrast, most of the traditional statistical analysis methods are 'What-is' tools -- credit of aircraft picture: Martin Grandjean, McGeddon, Cameron Moll, CC BY-SA 4.0.
• Figure 2: From data to a causal skeleton: A visual workflow for uncovering the hidden structure of POE data, discovered by the GES algorithm. As shown on the right, the graph is read from top to bottom, with variables higher in the hierarchy (ancestor nodes) causally influencing those below them (descendant nodes).
• Figure 3: Estimating the impact of interventions using causal effects of ancestor and descendant nodes. The left panel shows the causal hierarchy, while the right panel quantifies the average effect of a hypothetical intervention (simulating the do-operator) on other variables by comparing high and low satisfaction groups of the target node. Variable-level results and numerical outputs are provided in the code repository.
• Figure 4: Comparison of how the different 'what-is' and 'what-if' tools can lead to oppositely different suggestions for intervention.
• Figure 5: Statistical summary of the satisfaction votes in the case study.