Identifying determinants of energy use of schools in England for benchmarking purposes

Abstract

Carbon emissions from the school stock account for approximately 14% of emissions from the UK's public sector. Energy performance benchmarks play an important role in the built environment which encourage building operators to achieve higher energy efficiency. The robustness of benchmarks and the methodology in CIBSE TM46 which underpins the DEC scheme is vital to obtaining an accurate evaluation of operational performance. Potential ways to further improve comparability of buildings via complex statistical approaches are identified. Statistical analyses of the school stock and information on building characteristics and students have led to identification of determinants of energy use of primary and secondary schools in England. The ways in which the identified determinants could be utilised to improve the relevance of benchmarking process are explored.

Full text

CIBSE Technical Symposium, Liverpool
John Moores University, Liverpool, UK, 11-12 April 2013
1
Identifying determinants of energy use of schools in England for
benchmarking purposes
Sung-Min Honga, Sherleen Pangb, Greig Patersonb, Dejan Mumovicb, Philip
Steadmana
a UCL Energy Institute, University College London, Gower St., WC1E 6BT London,
United Kingdom
b The Bartlett School of Graduate Studies, University College London, Gower St.,
WC1E 6BT London, United
Abstract
Carbon emissions from the school stock account for approximately 14% of emissions
from the UK’s public sector. Energy performance benchmarks play an important role
in the built environment which encourage building operators to achieve higher
energy efficiency. The robustness of benchmarks and the methodology in CIBSE
TM46 which underpins the DEC scheme is vital to obtaining an accurate evaluation
of operational performance. Potential ways to further improve comparability of
buildings via complex statistical approaches are identified. Statistical analyses of the
school stock and information on building characteristics and students have led to
identification of determinants of energy use of primary and secondary schools in
England. The ways in which the identified determinants could be utilised to improve
the relevance of benchmarking process are explored.
Keywords Benchmark, Energy Performance, School, Display Energy Certificate
Number of words: 4,893

CIBSE Technical Symposium, Liverpool
John Moores University, Liverpool, UK, 11-12 April 2013
2
1.0 Introduction
There are over 25,000 schools in the UK, and in total they are responsible for
approximately 14% of the UK public sector’s total carbon emissions (1). With the UK
Government’s commitment to reduce the CO2 emissions by at least 80% by 2050,
relative to 1990 baseline levels (2), improving the energy efficiency of the non-
domestic building stock, including schools, is vital to achieving the target.
Benchmarking energy performance is a technique widely used as part of energy
management practice to understand how well a building is being operated. The
feedback from benchmarking assists and encourages owners and building operators
to set targets and take actions to achieve higher levels of energy efficiency, hence it
is a technique that is important in improving the energy efficiency of the UK’s school
stock.
In recent years, benchmarking has gained renewed interest due to the introduction of
the Display Energy Certificates (DEC) scheme under the European Energy
Performance of Buildings Directive (EPBD). The scheme aims to raise public
awareness of energy use as well as assist and motivate improvement of energy
efficiency of the existing built stock in England and Wales by displaying DECs in
public buildings with total useful floor area above 1,000m2 (3). The certificate shows
an Operational Rating (OR), a numerical indicator of a building’s performance, which
is based on a comparison of the building’s energy performance against the
performance indicators presented in CIBSE TM46 benchmarks (4).
The TM46 benchmarks present “typical” performance figures for each of the 29
benchmark categories which are medians derived from a distributions of electricity
and fossil-thermal energy consumption in the stock (Bruhns 2011). One of the key
components of benchmarking energy performance is the degree of correspondence
between a building and the benchmarks so that the evaluation of energy
performance indicates whether a building is being operated in an energy efficient
manner. What is unique about the methodology in CIBSE TM46 (6) is that it was
developed such that there is a procedure to improve the relevance of comparing
energy performance of a building to an appropriate benchmark. A key feature of the
benchmarks is that they were developed to standard conditions so that they can be
normalised to account for variations in weather and occupancy level of the buildings
that are being benchmarked. Moreover, an option to discount any separable energy
uses that do not commonly exist in an activity type further improves the comparability
of a building to the benchmarks. While this methodology is a significant improvement
on the previously developed benchmarks such as the CIBSE Guide F (7), the impact
of large variation in building characteristics, building services and occupant activity
that exist in the stock (8) on energy use of schools is yet to be assessed. There are
benchmarking methods that use complex statistical methods to assess the impact of
various building and occupant characteristics based on which primary determinants
are identified (9–12). Such approaches suggest a potential way to improve the
comparability of buildings in the UK by including additional parameters that could
explain patterns of energy use beyond weather, occupancy and separable energy
uses.
In this study, information on building characteristics and students are statistically
analysed to assess and identify the determinants of energy use of primary and

CIBSE Technical Symposium, Liverpool
John Moores University, Liverpool, UK, 11-12 April 2013
3
secondary schools in England. Ways are explored in which the identified
determinants could be utilised to improve the relevance of the benchmarking process.
2.0 Methodology
This section describes the details of the data and the statistical analyses undertaken
in this study. The statistical analyses were based on two datasets, a dataset of DEC
records and another set of various building characteristics data that were collected
through a desk‐based methodology using publicly available information sources. The
following section describes various statistical methods used to analyse the datasets.
2.1 Data collection
2.1.1 DEC data
The energy consumption data for schools in London was acquired from the database
behind the non-domestic energy performance certificate register managed by
Landmark
1
. The data used was provided by CIBSE which includes approximately
45,000 DEC records which were lodged from 2008 until February 2010.
The DEC records for schools can be found under the “Schools and seasonal public
buildings” benchmark category. The records are further sub-categorised into 26
“building types” which describe the activity in more detail. However, these building
types cover a wide range of activities including those not associated with school
activities (e.g. clubhouse, dogs racecourse) and, therefore, it was imperative to be
selective of the building types so that only the records from schools are used for
deriving benchmarks. Out of 26 building types, it was decided that records under four
building types (Primary schools, State Primary schools, Secondary school and State
Secondary schools) were the most appropriate for deriving benchmarks for primary
and secondary schools in England.
In order to prepare the raw data for analyses, a set of criteria largely based on the
criteria used by (13) was used to select records for the study:
 Total useful floor area is greater than 50m2
 Operational rating is > 5, < 1,000 and not 200 (Operational rating given to
default DECs with insufficient information)
 Fossil-thermal energy use index (EUI) greater than 0 kWh/m2
 Electricity EUI is greater than 0 kWh/m2
 Total carbon dioxide emission is less than 100,000 tonnesCO2/yr
 The DEC record with the latest assessment end date for each building
 Not a DEC produced based on a composite benchmark
2
methodology
Moreover, records from buildings which use electricity as the main heating fuel were
discounted to ensure that electrically heated buildings, which have considerably
different patterns of energy use from the majority of the stock, do not skew the
results.
Process
Primary
Secondary
Total
Full DEC dataset
-
-
44,939
1
Non-domestic energy performance certificate register: https://www.ndepcregister.com/
2
A composite benchmark is used when an assessed building comprises occupied spaces that fall into
more than one benchmark category (4)

CIBSE Technical Symposium, Liverpool
John Moores University, Liverpool, UK, 11-12 April 2013
4
Schools and seasonal public buildings
-
-
21,567
Data selection
13,649
4,756
18,405
Data cleaning
9,440
3,268
12,708
Final data
7,563
1,407
8,970
Table 2-1 Summary of changes in the number of records during the data
processing
As the DEC dataset included data collected during 2008, 2009 and 2010, heating
energy consumption figures were normalised to a standard weather condition for the
average climate of the UK and London. Based on the typical percentage of fossil fuel
used for space heating illustrated in the Energy Consumption Guide 73 (14) as well
as opinion from the experts in the industry, 80% of the fossil-thermal energy use
from DECs was assumed to be weather-dependent. Assessment end dates recorded
in the DEC database were used as a basis for correcting heating consumption. For
the stock analyses, heating consumption of DEC records covering England were
normalised to 2,021 heating degree days (15.5°C base temperature) as illustrated in
CIBSE TM46 (6). The heating consumption figures for schools in the Greater London
area were adjusted to 1,840 Heating Degree Days (15.5°C base temperature) for
London Heathrow calculated by Oxford University’s Environmental Change Institute
(15).
In order to examine the impact that pupils have on energy consumption of schools,
information on the number of pupils was acquired from the Department of
Education’s Edubase Public Portal
3
in April 2012. Once the two datasets were
merged, a manual inspection of all combinations was conducted to validate the
process.
2.1.2 Building characteristics
In order to conduct a multiple linear regression analysis, a database was developed
with much more detailed information, but smaller in sample size, that describes
various building characteristics. This section describes the data collection process
and the method used to analyse the data collected.
The sample of schools for the building characteristics database was selected from
within the Greater London area based on the availability of resources as well as the
benefit of minimising the effect of regional weather variation on fossil-thermal energy
use and any regional trends in energy use.
A list of selection criteria was developed in order to collect building characteristics
from schools that would enable analysis of determinants of energy use on a building
basis while facilitating the desk-based collection approach:
 Schools with one main building
 Uniform building characteristics (age, construction material, etc.)
 Has a valid DEC record
 Can be seen using Google street view or the Bing Bird’s eye view function
3
Edubase: http://www.education.gov.uk/edubase/home.xhtml

CIBSE Technical Symposium, Liverpool
John Moores University, Liverpool, UK, 11-12 April 2013
5
The annual energy consumption data and a number of building characteristics of
primary and secondary schools were extracted from the dataset discussed in chapter
2.1.1. These included total useful floor area, type of fuel used mainly for heating,
HVAC system type and the mode of heat distribution as well as number of pupils.
The building characteristics parameters included in this study were pre‐designed
such that the information could be gathered from publicly available online sources
using a desk-based research approach. Table 2-2 presents the initial list of
parameters.
Characteristics
Description
Construction year
Year the building was built
School type
Primary or Secondary
Site exposure
Exposed, semi-exposed or sheltered
Orientation factor
+/- 45 degrees from due North
Building perimeter (m)
Total and exposed perimeters
Façade lengths (m)
N, S, E, W
Footprint area (m2)
-
Number of storeys
Above ground
Height of building
-
Building shape
Singular, courtyard, crescent, bend or branch
Roof shape
Flat, sloped or inverted roof
Glazing type
Single or double
Glazing percentage
-
External wall material
Brick, masonry, timber etc.
External roof material
Polymer, ceramic, metal etc.
External shading
Yes/No
Evidence of atrium
Yes/No
Evidence of windcatchers
Yes/No
Compactness ratio
-
Depth ratio
-
Glazing ratio
-
Exposure ratio
-
Table 2-2 List of collected and derived building characteristics
The measurements of the geometric shape of schools were used to derive various
ratios that express the building characteristics into a single parameter that could be
used to understand the impact of the built form on energy use.
The following five ratios have been included in this study:
 Exposure ratio = volume / exposed surface area
 Depth ratio = volume / exposed wall area
 Compactness ratio = perimeter of building / perimeter of a circle with same
area
 Glazing ratio (total) = total glazed wall area on building / total floor area
 Pupil density = total floor area / number of pupils

CIBSE Technical Symposium, Liverpool
John Moores University, Liverpool, UK, 11-12 April 2013
6
2.2 Statistical analyses
2.2.1 School stock analyses
To identify factors that have impact on the energy use of schools across England the
variables in the DEC and school dataset that describe the activity, building services
and occupants were analysed using Statistical Analysis Software (SAS) 9.3
4
.
The deviation of electricity and fossil-thermal energy use variables from a normal
distribution was initially tested using the Kolmogorov-Smirnov test to identify the
adequate type of statistical test to use. Once the skewness of the distribution of the
variables was assessed, Wilcoxon-Mann-Whitney test and Kruskal-Wallis test were
conducted, rather than ANOVA, to examine whether there is a significant difference
in the pattern of energy use between the two school types and internal environment
types. A Bonferroni correction was applied when conducting these tests on large
number of categories to reduce type one error, hence tests were reported at the
0.0125 level of significance.
Scatter plots were produced to determine and compare the correlation between
energy use and floor area and number of pupils based on the R2 value of the linear
regression line. Moreover, significance of various building characteristics was
assessed using Pearson’s correlation coefficients, F-test and T-test prior to the
multiple regression analyses.
2.2.2 Multiple linear regression model
Multiple linear regression analysis was carried out to study the relative significance
of the building characteristics in determining electricity and fossil-thermal energy
consumption of primary schools in the Greater London area. The ‘goodness of fit’ or
accuracy of prediction of the regression model would be indicated by the R2 values,
which would be equal to unity in the case of a perfect fit. The statistical add‐in
package to Microsoft Excel, XLSTATS, was used to perform the analysis.
Starting with an initial regression model with the building characteristics and derived
variables, the regression model was refined through an iterative process using
sensitivity analysis and multicollinearity test.
Based on the changes in R2 value outputs, a sensitivity analysis was undertaken by
adding or subtracting parameters according to whether they had a notable impact on
the dependent EUIs. The relative importance of each input parameter within the
regression models was indicated by the absolute values of their standardised
coefficients (16).
A multicollinearity test was carried out alongside the regression analysis to identify
any strong correlations between predictor parameters within the regression model.
This was to prevent any existing linear correlation between predictor variables in the
model from not quantifying the significance of independent parameters correctly (17).
Parameters found to have high Variance Inflation Factor (VIF) values (> 5.0) were
removed from the subsequent model (18).
4
Statistical Analysis Software: http://www.sas.com/software/sas9/

CIBSE Technical Symposium, Liverpool
John Moores University, Liverpool, UK, 11-12 April 2013
7
3.0 Results
3.1 Statistical analyses of the school stock in England
3.1.1 Energy consumption by type of school
Figure 3-1 Cumulative frequency distribution of electricity and fossil-thermal
energy use by school type
The distributions of electricity and fossil-thermal energy use of primary and
secondary schools show the difference in pattern of energy use between the school
types with electricity use medians of 43kWh/m2 and 51kWh/m2, and fossil-thermal
energy use medians of 139kWh/m2 and 136kWh/m2, respectively. The electricity use
of schools was found to be significantly different depending on the type of school
(Wilcoxon-Mann-Whitney, p < .0001). This is likely to be due to the more intensive
uptake of ICT equipment such as computers in secondary schools (1). On the other
hand, there was no significant difference in the trends of fossil-thermal energy use
between primary and secondary schools (Wilcoxon-Mann-Whitney, p > .05).
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
020 40 60 80 100 120 140 160 180 200 220 240 260 280 300 320 340 360 380
Cumulative frequency (%)
Energy consumption (kWh/m2)
Primary school (Electricity) Secondary school (Electricity)
Primary school (Fossil-thermal) Secondary school (Fossil-thermal)
Sample size
Primary: 7,563
Secondary: 1,407

CIBSE Technical Symposium, Liverpool
John Moores University, Liverpool, UK, 11-12 April 2013
8
3.1.2 Energy consumption by number of pupils
Figure 3-2 Cumulative frequency distribution of energy use by number of
pupils
The cumulative frequency distribution of energy consumption per pupil (Figure 3-2)
shows significant differences in electricity consumption, median at 271kWh/pupil and
461kWh/pupil, and fossil-thermal energy use, median at 821kWh/pupil and 1540
kWh/pupil, of primary and secondary schools respectively. These trends are likely
due to the intrinsic difference in activity of schools, hence requirements for energy,
where secondary schools generally require more equipment as well as spaces for
facilities such as laboratories, workshops, and libraries (1). Wilcoxon-Mann-Whitney
tests indicated that the amount of annual electricity (p < .0001) and fossil-thermal
energy (p < .0001) used per pupil were found to be significantly affected by the type
of school.
3.1.3 Internal environment
The DEC database holds description of the primary ventilation strategies used in buildings based on a
predefined set of categories as shown in *buildings operating in mixed-mode have an option
to alternate between natural and mechanical ventilation and/or air conditioning
Table 3-1 (3). The implications of varying levels of HVAC systems in schools on
electricity and fossil-thermal energy use are assessed.
Internal environment
N
Primary
Secondary
Electricity
(kWh/m2)
Fossil-thermal
(kWh/m2)
Electricity
(kWh/m2)
Fossil-thermal
(kWh/m2)
Median
Median
Median
Median
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
2400
2600
2800
3000
3200
3400
3600
3800
4000
4200
Cumulative frequency (%)
Energy consumption by pupil (kWh/Pupil)
Primary school (Electricity) Secondary school (Electricity)
Primary school (Fossil-thermal) Secondary school (Fossil-thermal)
Sample size
Primary: 7,563
Secondary: 1,407

CIBSE Technical Symposium, Liverpool
John Moores University, Liverpool, UK, 11-12 April 2013
9
Heating and Natural
Ventilation
8,351
43
133
50
132
Mixed-mode* with Natural
Ventilation
274
44
124
52
122
Mixed-mode with
Mechanical Ventilation
76
50
133
66
91
Heating and Mechanical
Ventilation
251
48
118
64
111
Air Conditioning
18
40
113
51
91
All
8,970
*buildings operating in mixed-mode have an option to alternate between natural and
mechanical ventilation and/or air conditioning
Table 3-1 Energy consumption statistics by internal environment types
As shown in *buildings operating in mixed-mode have an option to alternate between
natural and mechanical ventilation and/or air conditioning
Table 3-1, the intensity of electricity consumption gradually increases as the
buildings become more heavily serviced with mechanical ventilation in both school
types. By contrast, it can be seen that the fossil-thermal energy use gradually
decreases which is likely due to buildings becoming less dependent on operable
windows, therefore, reduced ventilative heat loss. Moreover, heating demands may
have been offset by higher internal heat gains in some air conditioned schools.
A series of analyses were carried out to identify whether the observed differences
between the HVAC system categories were statistically significant. A comparison of
trends in energy use by schools with varying HVAC system types found significant
differences (Kruskal-Wallis, p < .0001). Wilcoxon-Mann-Whitney tests were carried
out to identify which of the HVAC systems types are significantly different.
Internal Environment
Wilcoxon-Mann-Whitney test
Electricity EUI
(kWh/m2)
Fossil-thermal EUI
(kWh/m2)
P-value
P-value
Heating and Natural Ventilation
< 0.0125
< 0.0125
Mixed-mode with Natural Ventilation
Heating and Natural Ventilation
< 0.0001
< 0.0001
Heating and Mechanical Ventilation
Mixed-mode with Natural Ventilation
< 0.0001
> 0.0125
Mixed-mode with Mechanical Ventilation
Mixed-mode with Natural Ventilation
> 0.0125
> 0.0125
Air Conditioning
Table 3-2 Summary of results from Wilcoxon-Mann-Whitney tests between
internal environment categories
As shown in Table 3-2, energy consumption of schools with “Heating and Natural
Ventilation” was significantly different from schools with “Mixed-mode with Natural
Ventilation” and “Heating and Mechanical Ventilation”. Moreover, it was found that
there was only a significant difference in electricity use between schools with “Mixed-

CIBSE Technical Symposium, Liverpool
John Moores University, Liverpool, UK, 11-12 April 2013
10
mode with Natural Ventilation” and schools with “Mixed-mode with Mechanical
Ventilation” but no significant difference with schools with “Air Conditioning”. This
suggests that categorisation of servicing strategies in CIBSE TM46 could be
simplified by grouping the categories with insignificant difference into a single
category. However, it should be noted that the majority of schools in the dataset
were naturally ventilated. Therefore, there is a necessity to acquire more data from
schools that are either air conditioned or mechanically ventilated to recognise this
trend with greater certainty.
3.1.4 Energy performance metric for schools
Figure 3-3 Scatter plot of annual electricity consumption and number of pupils

CIBSE Technical Symposium, Liverpool
John Moores University, Liverpool, UK, 11-12 April 2013
11
Figure 3-4 Scatter plot of annual electricity consumption and total floor area
A stronger correlation has been found between the annual electricity consumption of
primary and secondary schools to the floor area for, R2 values of 0.29 and 0.61, over
the number of pupils, R2 of 0.17 and 0.36, respectively, therefore, validating the
robustness of using floor area as a denominator of energy performance metric
(kWh/m2) for schools. However, the number of pupils was also found to explain a
considerable proportion of the electricity consumption of primary and secondary
schools as well as fossil-thermal energy, R2 values of 0.18 for both school types.
3.2 Multiple regression analyses
Prior to developing multiple linear regression models, a set of analyses were
conducted to develop an understanding of the continuous, binomial and categorical
variables in the building characteristics dataset for schools in Greater London.
Pearson’s correlation and linear regression analyses, and hypothesis tests were
used to identify the variables which have statistically significant effects on electricity
or fossil-thermal EUI. Subsequently, only the significant variables were used in the
multiple regression models to reduce the complexity as well as the uncertainty of the
model stemming from having a large number of dummy variables.
3.2.1 Determinants of electricity use in schools
Results from the iterative process of refining the regression model for electricity use
are summarised in Table 3-3. Among the 13 parameters which were included in the
initial model, a series of sensitivity analyses and multicollinearity tests identified 6
parameters which have little impact on electricity use, little correlation with other
parameters, or a weaker link to electricity use in schools (e.g. exposure ratio). The
final regression model for electricity use had an R2 value of 0.257, indicating
approximately 26% of the variation electricity consumption could be explained by the
parameters in the model.
Model
Number
Description
R2 value
P-Value of
model
Model 1
All variables
0.332
<0.001
Model 2
Built age, shading, exposure ratio removed
0.26
<0.001
Model 3
Glazing type, roof shape removed
0.259
<0.001
Final Model
Roof glazing removed
0.257
<0.001
Table 3-3 Development of the model for electricity use
The parameters and the standardised coefficients of Model1 and the Final Model are
shown in Figure 3-5 and Figure 3-6, respectively. The main determinants were
identified as the different ventilation modes, total floor area, and pupil density.

CIBSE Technical Symposium, Liverpool
John Moores University, Liverpool, UK, 11-12 April 2013
12
Figure 3-5 Standardised coefficients of the initial regression model for
electricity EUI (95% confidence interval)
Figure 3-6 Standardised coefficients of the final regression model for
electricity EUI (95% confidence interval)
3.2.2 Determinants of fossil-thermal energy use of schools
Results of the development in the regression model for fossil-thermal energy are
summarised in Table 3-4. From an initial model which had 12 parameters, a series of
sensitivity analyses and multicollinearity tests have allowed the model to be refined
down to 8 parameters that have relatively greater impact on fossil-thermal energy
use, hence the reduction in R2 value from 0.291 to 0.289 in the final model.
Model
Number
Description
R2 value
P-Value of
model
Model 1
All parameters
0.291
<0.001

CIBSE Technical Symposium, Liverpool
John Moores University, Liverpool, UK, 11-12 April 2013
13
Model 2
Depth ratio removed
0.29
<0.001
Model 3
External shading and ventilation mode removed
0.289
<0.001
Final Model
Compactness ratio removed
0.289
<0.001
Table 3-4 Development of the model for fossil-thermal energy use
The parameters included in the initial and final regression model for fossil-thermal
energy and their respective standardised coefficients are presented in Figure 3-7
and Figure 3-8, respectively. The differences in the coefficients shown in both figures
indicate that exposure ratio and building age were the most prominent factors
influencing fossil-thermal energy use of primary schools in London. In addition to
these parameters, existence of glazing on the roof, degree of site exposure, pupil
density and total glazing ratio in respect to the floor area were also found to have
notable influences on fossil-thermal EUI.
Figure 3-7 Standardised coefficients of the initial regression model for fossil-
thermal EUI (95% confidence interval)

CIBSE Technical Symposium, Liverpool
John Moores University, Liverpool, UK, 11-12 April 2013
14
Figure 3-8 Standardised coefficients of the final regression model for fossil-
thermal EUI (95% confidence interval)
3.2.3 Multiple Regression Equations
The final multiple regression equations for predicting heating and electric EUI can be
expressed as follows:
Heating EUI
(kwh/m2.yr)
=
197.71 + 0.62(BA) – 16.52(SE) + 5.61(RS) – 2.97(GT)
– 23.12(RG) – 43.27(ExpR) + 1.6(PD) + 90.5(TGR)
Electric EUI
(kwh/m2.yr)
=
81.57 – 4.44(SE) + 16.58(VM) –0.006(TFA) –
0.95(DepR) – 2.80(CompR) – 0.53(PD) – 24.49(TGR)
Figure 3-9 Equations from the final multiple linear regression models
Abbreviations for the building characteristic parameters are given in the table below.
Abbreviation
Parameter
Input values
BA
Built age
numerical value
ExpR
Exposure ratio
numerical value
CompR
Compactness
ratio
numerical value
DepR
Depth ratio
numerical value
GT
Glazing type
0 = single glazed; 1 = double glazed
PD
Pupil density
numerical value
RS
Roof shape
0 = pitched or sloped; 1 = flat
SE
Site exposure
0 = sheltered; 1 = semi-exposed
TFA
Total floor area
numerical value
TGR
Total glazing ratio
numerical value
VM
Ventilation mode
0 = natural vent.; 1 = mechanical vent. or mixed mode
Table 3-5 Abbreviations for Heating and Electric EUI regression equations

CIBSE Technical Symposium, Liverpool
John Moores University, Liverpool, UK, 11-12 April 2013
15
4.0 Discussion
4.1 Determinants of energy use and benchmarking
Stock wide statistical analyses of the distribution of energy consumption per unit floor
area and per pupil in primary and secondary schools have shown significant
differences in the pattern of energy use for each school type, although not in regards
to the fossil-thermal energy use per unit floor area. The significance of different
school types provides a rationale for acknowledging the intrinsic difference in the
pattern of energy use and, therefore, a need to derive separate benchmarks for each
school type so that the level of energy efficiency of schools can be evaluated with
greater accuracy.
In both the school stock analyses and the multiple linear regression analysis, HVAC
system types of schools were found to be of considerable importance, particularly in
regards to electricity consumption. However, this does not necessarily mean that
benchmarks should be provided for varying levels of HVAC system types, as heavily
serviced buildings have the potential to be as energy efficient as their less serviced
counter parts (19). Hence, there is a necessity to better understand the difference in
the pattern of energy use of schools with different servicing strategies so that
incorporating and reflecting the different HVAC system types into benchmarking
process (e.g. classification in ECON 19 (20)) can be examined further.
The number of pupils, although found to be less significant than floor area, was
found to explain a notable share of energy use in schools from the school stock
analyses. This was further recognised in the multiple linear regression analysis
where pupil density was identified as having notable influence on electricity and
fossil-thermal energy use of schools. These findings indicate the importance of
taking the impact of students, and perhaps the staff and their behaviour, into account
when assessing and evaluating the energy performance of schools, hence prompting
a need to further investigate the potential to utilise this parameter, either as a
variable in a model or adjustment procedure, so that schools with different occupant
conditions can be compared on a more equivalent basis.
The multiple linear regression analysis has identified the derived parameter
‘exposure ratio’ as one of the strongest determinants of fossil-thermal energy use in
schools. This indicates that the energy used for heating is influenced by the variation
in the geometric shape and the local environments surrounding schools which
highlights the importance of the passive design principles on energy use. Moreover,
this suggests a potential to use the derived variables that express the built form of
variety of schools, not just the exposure ratio, as part of the benchmarking process
to account for the impact of built form on energy use of schools.
4.2 Building characteristics as determinants
Despite the large number of parameters included in the multiple regression models
the correlations between building characteristics and EUIs were relatively weak. Low
R2 values in the final multiple regression models (0.29 for fossil-thermal EUI; 0.26 for
electricity EUI) suggested that only a small proportion of energy consumption of
primary schools could be explained by changes in the building parameters included
in the regression equations. Other parameters that have not been measured are
likely to be more significant determinants of energy use within the sampled buildings.
These may include insulation levels, infiltration rates, installed capacity of lighting,

CIBSE Technical Symposium, Liverpool
John Moores University, Liverpool, UK, 11-12 April 2013
16
heating system set point temperatures and heating system efficiency. This was in
line with results from a multiple regression modelling study conducted for office
buildings in Hong Kong (21), which reported that architectural parameters including
aspect ratios, floor heights and compactness ratio had minimal or no influence on
building energy consumption when compared to other factors such as fabric U-
values, cooling and heating system efficiencies and internal set point temperatures.
However, the collection of more detailed building characteristic data beyond the desk
based methodology adopted in this study for such a large sample of schools will
require significantly more time and resources.
Another reason for the low R2 values of the final models could be due to an
assumption that the relationship between the dependant variables (electricity and
fossil-thermal energy use) and the independent variables (the building characteristics
variables) is linear. A study comparing the performance of various statistical models
has found that non-linear multivariate methods such as Artificial Neural Network
(ANN) produced considerably more accurate predictions of energy performance over
a multiple linear regression model (22). Therefore, this raises the need to explore
possibilities of using a non-linear multivariate model such as ANN to identify the
determinants with greater accuracy.
5.0 Conclusion
This study focused on identifying determinants of energy use in primary and
secondary schools in England to explore ways to improve the accuracy of the
benchmarking process. Databases comprising information from DEC, student
information from EduBase and detailed building characteristics of schools in London
were developed.
The building and occupant characteristics in the databases were found to have the
following implications on the pattern of energy use of schools:
 A statistically significant difference in electricity consumption was found
between primary and secondary schools but the difference in fossil-thermal
energy use was insignificant.
 The total floor area was found to explain considerable proportion (R2 of 0.29
and 0.61) of the variation in energy consumption of primary and secondary
schools respectively.
 The number of pupils was found to be an important factor in assessing energy
performance of primary and secondary schools (R2 of 0.17 and 0.39).
 The pattern of energy use between naturally and mechanically ventilated
schools was significantly different. However, the differences between
mechanical ventilation and schools operating in ‘mixed-mode’ were
insignificant.
 The building age and exposure ratio were the most prominent determinants of
heating energy consumption.
 For electric energy consumption, ventilation mode, total floor area and pupil
densities were the main determinants.
 Coefficient of determination (R2) values for the two regression models
indicated that approximately 29% and 26% of the variation in heating and
electric EUI for the sampled schools, respectively, could be attributed to input
parameters included in the models.

CIBSE Technical Symposium, Liverpool
John Moores University, Liverpool, UK, 11-12 April 2013
17
Based on the results, ways in which the comparability in benchmarking could be
improved are suggested:
 Provision of separate benchmarks for primary and secondary schools to take
into account the intrinsic differences in requirements for energy use,
 Further investigate the implications of mechanical ventilation systems on the
demand for energy use and assess the feasibility of incorporating the
parameter into the benchmarking methodology.
 Explore ways to normalise the energy use of schools to take into account the
impact of building and occupant characteristics, such as the built form and the
pupil density, via adjustment procedure or other statistical methods.
The results from analyses have shown that such approach is an effective way to
assess and identify determinants of energy use of not only schools but also other
types of buildings. However, limitations in the building characteristics from the DEC
database and the sample size of the detailed dataset have highlighted a need for a
follow up study based on a larger sample, both in regards to sample size and
variables that describe other aspects of the building such as building services. In
addition, the possibility of using non-linear methods such as Artificial Neural Network
(ANN) is currently being explored.
References
1. Global Action Plan. UK Schools Carbon Footprint Scoping Study. Stockholm
Environment Institute; 2006. p. 1–76.
2. HM Government. Climate Change Act 2008 (c. 27). 2008.
3. DCLG. Improving the energy efficiency of our buildings. Communities and
Local Government Publications; 2008.
4. CIBSE. CIBSE TM47 Operational Ratings and Display Energy Certificates.
CIBSE Publications; 2009.
5. Bruhns H, Jones P, Bordass B. BENCHMARKING FOR DISPLAY ENERGY
CERTIFICATES. CIBSE Journal. 2011;(March).
6. CIBSE. CIBSE TM46 Energy benchmarks. Building. CIBSE Publications;
2008.
7. CIBSE. CIBSE Guide F: Energy efficiency in buildings. 2nd ed. Chartered
Institution of Building Services Engineers; 2004.
8. Carbon Trust. Building the future today - Transforming the economic and
carbon performance of the buildings we work in (CTC765). United Kingdom:
Carbon Trust; 2009.
9. Sharp T. Energy benchmarking in commercial office buildings. Proceedings of
the ACEEE 1996 Summer Study on Energy Efficiency in Buildings. 1996. p.
321–9.

CIBSE Technical Symposium, Liverpool
John Moores University, Liverpool, UK, 11-12 April 2013
18
10. Sharp T. Benchmarking Energy Use in Schools. ACEEE 1998 Summer Study
on Energy Efficiency in Buildings. 1998. p. 305–16.
11. Environmental Protection Agency. ENERGY STAR ® Performance Ratings
Technical Methodology [Internet]. Methodology. 2011 [cited 2011 Jun 8].
Available from:
http://www.energystar.gov/ia/business/evaluate_performance/General_Overvie
w_tech_methodology.pdf
12. EIA. Energy Information Administration (EIA)- Commercial Buildings Energy
Consumption Survey (CBECS) [Internet]. [cited 2012 Jun 10]. Available from:
http://www.eia.gov/consumption/commercial/
13. Bruhns H, Jones P, Cohen R. CIBSE Review of Energy Benchmarks for
Display Energy Certificates - Analysis of DEC results to date [Internet]. 2011
[cited 2011 Jun 11]. Available from:
http://www.cibse.org/content/Technical_Resources/Technical_Reports/Technic
al Report_CIBSE Report on 45000 DECs.pdf
14. BRECSU. Energy Consumption Guide 73: Saving energy in schools. Building
Research Energy Conservation Support Unit; 1996.
15. Environmental Change Institue Oxford University. Monthly Degree Days for
Energy Management [Internet]. 2012 [cited 2012 Jul 23]. Available from:
http://www.eci.ox.ac.uk/research/energy/degreedays?weekly?monthly.php
16. Gelman A. Scaling regression inputs by dividing by two standard deviations.
2008;(October 2007):2865–73.
17. Freund RJ, Wilson WJ, Sa P. Regression Analysis: Statistical Modeling of a
Response Variable. Second. London: Elsevier Inc.; 2006.
18. Kutner MH, Nachtsheim C, Neter J. Applied linear regression models. 4th ed.
McGraw-Hill/Irwin; 2004. p. 701.
19. Bordass B, Cohen R, Standeven M, Leaman A. Assessing building
performance in use 3: energy performance of the Probe buildings. Building
Research & Information. 2001 Mar;29(2):114–28.
20. Carbon Trust. Energy Consumption Guide 19. Carbon Trust; 2003.
21. Lam JC, Wan KKW, Liu D, Tsang CL. Multiple regression models for energy
use in air-conditioned office buildings in different climates. Energy Conversion
and Management. Elsevier Ltd; 2010 Dec;51(12):2692–7.
22. Yalcintas M, Ozturk UA. An energy benchmarking model based on artificial
neural network method utilizing US Commercial Buildings Energy
Consumption Survey ( CBECS ) database. 2007;(April 2005):412–21.