Causal Effect Estimation in Observational Field Studies of Thermal Comfort

Causal Effect Estimation in Observational Field Studies of Thermal Comfort

The correlational relationship describes a “seeing” pattern in which a variable changes with another. In contrast, the causal relationship represents a “doing” knowledge that if interventions were made on one variable, it would lead to changes or outcomes in another. For instance, there might be a “seeing” correlation between ice cream sales and shark attacks (because they both increase during the summer). However, it is clearly erroneous to infer that eating ice cream would cause more shark attacks.

Building scientists are typically interested in whether interventions in the design, construction, and operation of buildings lead to improvements in building performance and/or occupant well-being. For example, interventions might include obtaining a green or healthy building certification, increasing outdoor ventilation rates, or providing a window view. The outcomes could include building energy consumption, occupant satisfaction, work performance, and health.

Experiments involve “doing” interventions to uncover causal relationships. However, conducting experiments is not always feasible, especially when interventions pose health risks to human subjects or are too costly to implement. In such cases, observational studies may be the only option. While these studies often have limitations in establishing causality, causal reasoning can sometimes be applied. In this paper, we introduce a causal inference framework and demonstrate its application in thermal comfort field studies in China.

A typical thermal comfort field study is an observational study (“seeing”) and generally does not enable the identification of causal knowledge (“doing”). As we showed in a previous paper that won the Building and Environment 2024 Best Paper Award, it is risky to directly interpret the observed correlational relationship between indoor temperature and thermal sensations as the causal effect of conditioning the indoor temperature to a specific value, leading directly to that particular thermal sensation. Many confounding factors in field studies can affect both indoor temperature and thermal sensations, resulting in skewed correlations between them.

We identify an opportunity to transform observational studies into natural experiments, enabling the identification and quantification of causal relationships related to thermal comfort in actual buildings. We utilize the winter district heating policy threshold in China, which is based on geographical location relative to the Qingling-Huaihe Line (referred to as the Huai River). Cities north of the line are provided with district heating, while cities south of it are not.

Causal effects can be estimated by comparing two potential outcomes for the same group of cities the Huai River threshold, where they should have similar culture, economics, climate, etc. One potential outcome is with district heating, while the other is without. This is equivalent to a randomized experiment that creates two groups that are very similar to each other, except one group would receive the intervention while another group does not.

We found causal effects of district heating on indoor operative temperature and indoor thermal sensation votes. However, we did not find causal effects of district heating on mean clothing insulation, air velocity, or metabolic rate. As shown in Fig.1, district heating led to mean indoor operative temperatures being 4.3°C higher (from 17.3°C to 21.6°C) and mean thermal sensation votes being 0.6 warmer on a seven-point scale, from -0.4 (cooler than neutral) to 0.2 (warmer than neutral).

Figure 1. Causal effects of district heating on the mean indoor operative temperature (a) and on mean thermal sensation vote (b).

We further infer that the 4.3°C increase in indoor operative temperature causes the 0.6 increase in the thermal sensation vote, assuming that the district heating policy mainly affects building occupants’ thermal sensation votes through the indoor operative temperature. We also calculated the effects of the indoor temperature changes on thermal sensation vote using PMV models. The delta PMV is 0.61, which matches our causal effect estimation.

We also use conventional correlational analysis to demonstrate that the correlation observed in these thermal comfort field studies does not accurately reflect the causal relationship. As shown in Fig.2., a variation of 4.3°C in the indoor operative temperature is associated with a 0.4 variation in the thermal sensation votes. However, this cannot be interpreted as increasing indoor temperature by 4.3°C would lead to an increase of 0.4 in thermal sensation vote. We also show that the indoor operative temperature could be either positively or negatively correlated with occupants’ thermal satisfaction. However, we cannot conclude that increasing the indoor operative temperature in these circumstances will necessarily lead to higher or lower thermal satisfaction.

Figure 2. Correlational analysis between indoor operative temperature and thermal sensation vote (a) and thermal satisfaction rate (b).

Our findings highlight the importance of the causal inference framework and method, particularly in observational field studies. Causal inference can be applied in other domains of building science where sufficient observational data is collected to enhance our understanding beyond conventional correlational methods, including indoor air quality field studies, post-occupancy evaluation, and building energy audits. This work also has practical implications for sustainable building design and operation that should be based on actionable causal knowledge.

Acknowledgments

This research has been supported by the industry consortium members of the Center for the Built Environment (CBE), University of California, Berkeley.

References

1. Sun, R., Schiavon, S., Brager, G., Yan, H., & Parkinson, T. (2025). Causal effects estimation: Using natural experiments in observational field studies in building science. Indoor Environments, 100080. https://doi.org/10.1016/j.indenv.2025.100080

2. Sun, R., Schiavon, S., Brager, G., Arens, E., Zhang, H., Parkinson, T., & Zhang, C. (2024). Causal thinking: uncovering hidden assumptions and interpretations of statistical analysis in building science. Building and Environment, 259, 111530. https://doi.org/10.1016/j.buildenv.2024.111530