Regression Techniques under ISO 9972 – A Comparative Analysis

Abstract

Building airtightness is increasingly important for enhancing energy efficiency, indoor air quality, and ensuring regulatory compliance. The commonly used Ordinary Least Squares (OLS) regression approach, recommended by ISO 9972 Annex C, inadequately addresses uncertainties inherent in real-world testing conditions, particularly under the influence of wind and at low pressures. This study evaluates various regression methods—Ordinary Least Squares (OLS), Weighted Least Squares (WLS), and the Weighted Line of Organic Correlation (WLOC)—to improve uncertainty estimation in building airtightness testing. Analyzing over 6,000 blower door tests across 127 distinct building configurations, this research compares the predictive accuracy and uncertainty estimation capabilities of these regression approaches. Furthermore, the study introduces a simplified variant, WLOC_2. Findings indicate that all regression methods provide similar predictive accuracy at the standard test pressure of 50 Pa. However, at a realistic infiltration pressure of 4 Pa, WLS and WLOC_2 significantly reduce prediction error by up to 6 percentage points compared to traditional OLS. Confidence interval analysis reveals that at 50 Pa, OLS covers the reference airtightness values for only 25% of the data, while WLOC_2 achieves 42% and WLS 91%. At the lower 4 Pa pressure, OLS confidence intervals cover merely 21% of measurements, whereas WLS demonstrates overly conservative uncertainty, covering all measurements, and WLOC_2 provides a balanced performance with 82% coverage. These results strongly advocate incorporating weighted regression methods like WLS and WLOC_2 into airtightness measurement standards, as they better account for variability and uncertainty, particularly at low-pressure conditions relevant for realistic infiltration assessments.