Why models underestimate historic building performance

There are 6.5m buildings in England, Scotland and Wales that date from before 1919 – around 20–30% of Britain's building stock – and reducing their carbon emissions is critical.

Improving the energy efficiency of traditional buildings may also enhance their thermal comfort and support their continued use.

However, if professionals are to advise clients effectively to improve performance and reduce emissions, they need an accurate understanding of how these properties behave.

Although the procedures are relatively inexpensive and straightforward ways to calculate and compare the energy performance of residential properties, they have been widely criticised. Indeed, critics often argue that RdSAP disproportionately disadvantages traditional buildings, with modelled U-values for their solid walls shown in research to assume poorer performance than in-situ measurements.

While the assumed U-value for solid walls constructed before 1975 is now 1.7 W/m²K, evidence indicates that the actual value of a number of these walls will be lower.

More recently, the UK government-funded Demonstration of Energy Efficiency Potential (DEEP) research project found that pre-retrofit U-values across the studied buildings differed widely, reflecting the variability of the original fabric.

DEEP also found that the default U-values adopted in calculations were not always well aligned with in-situ values, and could be up to 15% worse than in-situ values, highlighting that outputs from models such as RdSAP may not be robust.

This has implications not only for energy performance certificates (EPCs) and related minimum energy efficiency standards (MEES), but also for predicting the potential carbon reductions, energy savings and returns on investment following energy efficiency interventions. Therefore what we advise our clients may change if the actual building performance is different from what is assumed.

The sample consisted of three solid-brick semi-detached properties and six detached properties made of solid stone, limestone or sandstone. The gross internal floor areas of each ranged from 72m² to 245m², with a median of 124m².

By way of comparison, the average dwelling size in England is 97m² and 76m² for private rental properties.

The study measured heat loss through the external walls of the dwellings only – rather than other building elements – using U-values, enabling comparison with the assumptions generally used in models such as SAP and RdSAP.

The compared U-value, as defined by ISO 7345: 2018 Thermal performance of buildings and building components – Physical quantities and definitions, represents the heat flow rate through a material per unit of area and temperature difference.

The work followed ISO 9869-1: 2014 Thermal insulation – Building elements – In-situ measurement of thermal resistance and thermal transmittance, which uses a moving average to calculate in-situ U-values.

All work was conducted during the heating season, which was from October to March, in order to ensure an acceptable temperature differential of greater than 10°C, reducing measurement uncertainty. Data analysis was conducted over a minimum period of 72 hours to ensure stable measurement and conforming to ISO 9869-1.

Air permeability increases heat loss through convection, so this was also measured using a blower door test at 50 Pascals conducted by a certified third-party provider following ATTMA Technical Standard L1.

These measured U-values were between 12.1% and 46.7% better than assumed. In-situ airtightness measurements were more variable, though, as we might expect from a small study, in particular of heritage buildings of this type. While seven of the nine properties outperformed the models, the overall range of difference extended from 51.7% better to 57.1% worse than assumed.

Together, these numbers help us to calculate whole-dwelling heat-loss figures. These also varied considerably when comparing in-situ measurements with assumed values, ranging from 12.1% to 46.7% better than assumed. This showed that the buildings in all cases lost less heat than had been modelled.

When these results are consolidated using a simple RdSAP method, we find that the measured performance can be significantly different from the modelled value. Indeed, four of the properties in the research were able to move up by one EPC rating band when measured data replaced assumed values in either the SAP or RdSAP models for the dwellings.

However, although our case study was small, it points to the fact that such models do not cater well for some of our older building stock. This has several implications.

Firstly, when we set our baseline before designing retrofits then this figure may understate actual performance and, as a result, retrofit designs and final performance figures and savings will be skewed.

Secondly, when we consider MEES the purely modelled approach using assumed values may misrepresent older buildings and suggest that it is not cost-effective or practical to treat them.

However, readers should keep in mind that RdSAP was updated in June 2025 and now enables the recording of the thickness of older buildings' stone and solid walls in calculating heat loss.

The wall thickness has been a data collection point up until now, but not previously used to calculate the U-value of the wall. This inclusion will lead to a change in the resultant EPC rating following the introduction of RdSAP 10.