Synthetic Data Simulation to Assess the Effects of Modelling Assumptions in Thermal Comfort Data Analysis and Inference

Designing and operating buildings that are safe, comfortable, and conducive to productivity is essential for promoting human well-being. Human-centred approaches are gaining popularity, especially in addressing factors that influence well-being and building performance (e.g., thermal comfort). Unfortunately, statistical practices for analysing thermal comfort data often rely on simplified assumptions, resulting in crucial aspects of data analysis being often absent or overlooked. A common phenomenon in thermal comfort research is ignoring the clustering of observations, and methods that do not account for clustering (e.g., single-level models) are often used. This study proposes a synthetic data simulation workflow to investigate the tangible effects of ignoring clustering in the analysis of thermal comfort data. The results show that a multilevel model is more effective for predictions and produces more reasonable inferences compared with a single-level model. Overall, the results are relevant for improving the development and interpretation of thermal comfort models. Ultimately, this research aims to bridge the gap between current practices in statistical analysis and building research to promote better inferences of complex human-building interactions.