Quantifying the Comfort Cost of Clothing Adjustment in Building Operation: A Causal Inference Study

• Quantifies the comfort cost of clothing adjustment using causal machine learning. • A 1-clo increase reduces comfort probability by about 4 percentage points. • The comfort cost is larger in naturally ventilated buildings and older groups. • Findings support behavior-aware building operation and HVAC control. Clothing adjustment is commonly treated as a low-cost adaptive mechanism in energy-efficient building operation and is frequently invoked to justify widened indoor temperature setpoints. However, whether clothing-based adaptation is effectively cost-free under real operating conditions remains unclear. Using large-scale field data from the Chinese Thermal Comfort Database across 49 cities, this study estimates the average and conditional causal effects of clothing insulation on thermal comfort acceptability using a domain-informed causal machine learning framework. A directed acyclic graph is used to represent environment–behavior–comfort interactions, and Double Machine Learning is applied to address high-dimensional confounding. The results indicate a small but statistically robust negative average effect: a 1-clo increase is associated with an approximately 4 percentage-point reduction in comfort probability. The effect is also highly context-dependent, becoming more negative in naturally ventilated buildings and among older occupants, while approaching zero near thermal-neutral conditions. These results suggest that clothing adjustment should be treated as a finite adaptive resource rather than a cost-free comfort buffer, and that explicit consideration of behavioral cost can support more robust energy-efficient building operation.