Predicting sex-based thermal sensation differences: a multilevel ordinal regression approach

Energy consumption in buildings constitutes a significant portion of global energy use, where a substantial share of this energy is used to provide thermally comfortable indoor environments. Despite these considerable energy investments, many occupants remain dissatisfied with their thermal environment [1]. This dissatisfaction could be attributed to the prevalent ‘one-size-fits-all’ approach (e.g., Fanger’s PMV model), which overlooks individual differences. Among them, sex-based differences are disputed, where well-known variations in morphology, body composition and metabolism are neglected in practice.

Understanding and integrating these differences for the provision of comfortable spaces is essential for developing inclusive standards and enhancing energy efficiency while ensuring occupant well-being. Although some statistical analyses support these differences, the thermal comfort literature does not provide clear and consistent results [2]. Conclusions often rely on classifying results as either ‘significant’ or ‘nonsignificant’ (i.e., p-values ≤ 0.05 or p-values > 0.05) to determine the presence or absence of an effect. This approach is detrimental as it disregards the variability in human response and the uncertainty associated with statistical inference.

The present study aims to predict the thermal sensation votes of males and females exposed to a wide range of steady-state thermal environments. The data was obtained from a laboratory experiment by Rohles in 1970, one of the most extensive laboratory studies on thermal sensation. A total of 1,600 college-age students, comprising 800 males and 800 females, participated in the experiment. Divided into groups of ten (five males and five females), participants were asked to describe their thermal sensations (on an ordinal scale) under one of the 160 combinations of air temperature and relative humidity examined. We applied a multilevel ordinal regression within a Bayesian framework to analyse the data. This model was selected to account for the ordinal nature of the response variable (i.e., thermal sensation) and the clustering in the data.

Although the predicted probabilities are highly variable, preliminary results suggest that, on average, females are more sensitive to cold and hot environments than males. This difference is more pronounced at lower and higher temperatures. As a result, it is essential not to overlook sex differences in indoor thermal environments. Finally, a comparison with Fanger’s PMV model reveals that taking measures to minimise the probability of ‘cold’ and ‘hot’ thermal sensations for females will likely result in a comfortable environment for all.