3. Analysis on neighbourhoods: old and standard houses
3.2 Case-study approach on two neighbourhoods
3.4.2 Data accuracy and completeness influencing statistical analyses
The results revealed important variations in real user profiles and differences between these profiles and modelling assumptions, e.g. with regard to the heating set-point temperatures. These explain part of the variations in real energy use and part of the gap between theoretical and real energy use. However, some of these variations are not always found to be statistically significant in studies analysing larger data sets (see Chapter 2). The lack of statistical significance in those studies can also be explained in part by findings on these two case-study neighbourhoods.
Explanatory variables
Comparing the measured temperatures with the reported heating profiles showed good agreements with regard to the duration of the heating periods, except for 6 out of 59 households, all from cs1, omitting to report switching off the heating system at night. This could result from the heating periods being interpreted by the respondents as including the times when only the pilot flame of the gas furnace remains on. As reported by Weihl and Gladhart [142] and observed in this study, it is still easier for inhabitants to report their recurrent daily schedules than to make accurate estimates of the temperatures. In fact, we found large discrepancies and no correlation between the self-reported heating set-point temperatures and the measured values. This confirms findings from previous studies comparing measured with self-reported temperatures [51,52,98,140,142], showing even higher discrepancies for individual cases, exceeding 5°C. This can explain the lack of correlation found between self-reported temperatures and energy use found in statistical studies on real data ([47], see also Chapter 2 [25]), as opposed to the strong influence of set-point temperatures revealed by simulation based sensitivity analyses [97,153,154]. One exception is the statistical study by Steemers et al. [49] where a correlation was found between real energy use and set-point temperatures on a dataset of approximately 4800 houses, however the correlation was small and the data set did include actual thermostat settings, not only self-reported estimates. In other studies where significant correlations were found between energy use and self-reported indoor temperatures, the latter often referred to self-reported time-weighted set-point temperatures or to self-reported average indoor temperatures [47,155–157].
Instead of being only related to the heating set-point temperature during occupancy, the former parameter includes also an estimate of the night-time set-back temperature and duration while, in addition, the latter parameter also results from the technical properties of the building (e.g. the insulation level influencing the temperature drop during heating set-back periods). This could explain why significant correlations with energy use are more easily found for self-reported heating parameters that result not only from temperature-estimation but also from time-estimation.
Not only user related parameters reported by the inhabitants, but also technical parameters reported by professional energy performance assessors can be
inaccurate and therefore limit the power of statistical analysis on those parameters or on derived values, e.g. the theoretical energy use. The study reported in Chapter 2 [25] revealed important prediction biases associated with the use of default values instead of measured air permeability values or more accurately calculated system efficiencies. The measured air flow rates in cs2 proved that real technical properties can strongly diverge from their regulated design values. In fact, notwithstanding the ventilation system were exactly of the same type and installed by the same company, the real ventilation flow rates varied more as a result of different tuning than as a result of different user behaviour. Similar findings about the installation of ventilation systems, revealing the need for better quality control, and about the use of ventilation systems were made in other field studies in Belgium [158,159], in Finland [160]
and in the Netherlands [47,161].
Reported behavioural and technical data from surveys and EPB-assessments can thus contain considerable errors, affecting not only the accuracy of energy calculation models, but also the power of statistical analyses investigating to what extent those inaccurately reported parameters influence the real energy use.
Dependent variables
The lack of accurate values on those explanatory parameters is not the only lack of accuracy reducing the power of the statistical analysis on real energy use and prediction errors. The figures representing the real energy use, used as the dependent variable, can also lack accuracy or representativeness for the actual energy performance of a building on the long term. This was illustrated by the uncertainty on the normalized real energy use for space heating in both neighbourhoods, caused by simplified and standard assumptions in the degree day based method and possible variations over time in physical properties of the buildings (e.g. resulting from the drying of initial moisture content [119,120]) or in the use of the building. These elements can thus cause a mismatch between explanatory variables and the dependent variables, affecting the correlations between both variables. The fact that the statistical analysis reported in 3.3.4 did identify a few significant correlations in spite of the uncertainty on the consumption data reported in 3.3.2 results in part from approach of the study.
Case-studies, methodology and indirect correlations
A structured research approach based on data from uniform neighbourhoods was presented. It showed the value of combining different types of measurements and surveys (e.g. for defining heating profiles) and the need for more detailed methods for distinguishing different end-uses from aggregated consumption data and for normalizing the energy use (e.g. the energy use for space heating based on gas meter readings). The uniformity of the neighbourhoods allowed identifying variability in workmanship even for one building team (e.g. with regard to ventilation systems). Together with the detailed data collection approach, the uniformity within the neighbourhood also enabled statistical analysis to reveal the important correlation between set-point temperatures and
real energy use that is often not identified in much larger datasets. However, with regard to the selection of the case-studies, as a result of the limited number of neighbourhoods, the study cannot claim to be exhaustive with regards to socio-demographic variations, technical variations or the association between them.
The two neighbourhoods differed not only with regard to their building properties, being representative of different building periods and corresponding energy performance levels, but also with regard to their household characteristics and the ownership status. This is symptomatic of underlying socio-demographic differences, as lower insulation levels are commonly found in the houses of elderly people and low-income households [10,139,162] and as lower income levels are more highly represented in rental houses, especially in rented social houses [162]. While this makes the case-studies representative for a considerable segment of the Flemish residential building stock [162], this strong association between technical and socio-demographic parameters questions the general applicability of this study’s findings to other combinations of households and houses, e.g. to high performance social housings that will have presence profiles similar to cs1 but technical properties similar to cs2 or to the dataset from previous chapter. Furthermore, the study was limited to terraced and semi-detached single-family houses, while literature indicates that housing typologies (e.g. apartments versus detached houses) can also be a factor within user behaviour [52,163] and while many other technical variations were not included in the dataset (light weight construction types, balanced ventilation systems with heat recovery etc.). Therefore, additional complementary case-study neighbourhoods are needed to further disentangle the causal relationships between different parameters and results and to verify the applicability of the findings to other variations and combinations of types of buildings, systems and households.