Uncertainty in building airtightness tests: Comparison of regression techniques using a comprehensive dataset of 6,000 tests

Abstract

Building airtightness is important for enhancing energy efficiency and indoor air quality. Consequently, many regulations now mandate specific airtightness levels for new constructions, necessitating accurate measurements to ensure compliance. The conventional Ordinary Least Squares (OLS) regression, recommended by ISO 9972, is inaccurate, particularly under varying environmental conditions. Thus, this study evaluates different regression techniques, including Ordinary Least Squares (OLS), Weighted Least Squares (WLS), and Weighted Line of Organic Correlation (WLOC) regression, to improve uncertainty estimates in building airtightness measurements. Analyzing over 6,000 blower door tests across 127 configurations in six houses, this research compares the accuracy of these methods and introduces a second variant of Weighted Line of Organic Correlation (WLOC_2), which applies a simplified weighting procedure. Findings indicate that while all methods are similar in predicting the airflow at 50 Pa, WLS and WLOC_2 reduce prediction error by up to 6 percentage points at 4 Pa, where reference values are determined by averaging measurements taken at low wind speeds for each configuration. At 50 Pa, the OLS 95 % confidence interval covers the reference airtightness value for only 25 % of the data, compared to WLOC_2 with 42 % and WLS with 91 %. At 4 Pa, OLS includes only 21 % of measurements, while WLS overestimates uncertainty, covering all measurements, and WLOC_2 includes 82 % of the measurements within its confidence intervals. These results support incorporating weighted regression methods, which better account for variability at low pressures, into airtightness testing standards.