Data-driven research into the inaccuracy of traditional models of thermal comfort in offices

The accurate prediction of thermal sensation among office workers, at design and post-occupancy stages, is crucial for controlling indoor temperature efficiently and correcting deficiencies in workspaces, ensuring healthy and productive working conditions. Traditional analytical comfort models are still the best tool for this purpose given their potential for interpretation. However, their reliability is undermined by their poor accuracy. Based on 304 data series of point-in-time measurements of quantitative and qualitative comfort-related parameters collected in an experimental campaign in three office buildings, one air-conditioned and two in free evolution, in San Luis Potosi (Mexico), this work aims to identify the major error-causing factors of steady and adaptive comfort models. The divergences between predicted and reported thermal sensation were set as a dependant variable of two multiple regressions, one for each model. Eighteen independent demographic, environmental, contextual and subjective variables were considered. No multicollinearity problems were identified. Our findings show that contextual factors and humidity perception were relevant in the adaptive model error. Clothing insulation highly impacted the accuracy of both models while age and body mass were not statistically significant for either of them. Metabolic rate was the factor with the greatest influence in the error of the steady model. Although not covered, other influential factors played a key role in models' accuracy and further research is needed to integrate these in a new generation of more accurate and flexible analytical models.