Energy Efficiency (2021) 14: 57

[Link]

ORIGINAL ARTICLE

Mind the Energy Performance Gap: testing the accuracy

of building Energy Performance Certificates in Ireland
Bryan Coyne · Eleanor Denny

Received: 15 June 2020 / Accepted: 4 May 2021 / Published online: 17 July 2021
© The Author(s) 2021

Abstract Ireland’s Climate Action Plan aims Future EPCs could be improved by incorporating his-
upgrade 500,000 homes to B2 Energy Performance torical household energy usage to help improve models.
Certificate (EPC) standard by 2030. Evidence of
an Energy Performance Gap, where actual energy Keywords Residential energy use · Consumer
use differs from the EPC, could undermine pro- information · Modelling error · Energy performance
gress towards such targets. This paper studies the certificate
energy performance gap for a general housing sam-
ple (n = 9923) over multiple years. It provides a JEL classification Q40 · Q48 · D83
novel comparison between whole-home energy use
(electricity and gas) that accounts for fuel switching
and removes potential rebound effects by excluding Introduction
households that may have changed their behaviour
following a retrofit. Results suggest that actual energy Residential energy policies
use is unresponsive to the EPC, with a range of 457
kWh/year observed across EPC-level averages for the Approximately 75% of buildings do not meet energy
entire sample. This difference equated to less than 5% efficient standards as defined by the EU building
of the sample average annual energy use observed. standards (European Commission 2019a). This is
The Energy Performance Gap range features an likely because 35% of the European dwelling stock
average deficit of 17% below theoretical energy use. is over 50 years old (BPIE 2011) and only 0.4–1.2%
The least energy efficient dwellings feature an aver- of the building stock is renovated annually, depend-
age difference ranging from − 15 to − 56% of the rel- ing on the member country (European Commission
evant EPC. Conversely, energy efficient houses dis- 2019a). The residential sector represented 25.4% of
play higher-than-theoretical energy use, with average final energy use in the EU in 2016, with the majority
surpluses ranging from 39 to 54% of the relevant EPC. of energy (79.2%) used for space and water heating
Results sound a note of caution for policymakers that (Eurostat, 2019a).
rely on a theoretical EPC to deliver real energy savings. The EU has set targets for renewable generation,
emissions reduction, and energy efficiency to achieve
climate neutrality by 2050 (European Commission
B. Coyne (*) · E. Denny 2019b). 2030 climate targets include (i) sourcing 32%
Department of Economics, Arts Building, Trinity College
of the energy mix from renewable sources, (ii) reduc-
Dublin, Dublin, Ireland
e-mail: bacoyne@[Link] ing GHG emissions by 40% from 1990 levels and

Vol.:(0123456789)
13
57 Page 2 of 28 Energy Efficiency (2021) 14: 57

(iii) a 32.5% improvement in energy efficiency, rela- for A/B-rated dwellings (Fuerst et al., 2016). How-
tive to a 2007 forecast (European Parliament, 2018). ever, the authors make an important observation that
Improving energy efficiency is viewed as a key way energy performance may not be the only factor driv-
to reduce emissions, representing almost 40% of the ing this price premium, as it is likely to be correlated
potential for reducing greenhouse gases for less than with other desirable factors.
€60 per metric tonne of carbon dioxide equivalent
(McKinsey, 2010). Limitations of Energy Performance Certificates
The EU Energy Performance of Buildings Direc-
tive (EPBD) is a regulation that aims to improve The mixed evidence on the value of EPCs is unsur-
building energy efficiency in member states (Euro- prising. Evidence for the Irish EPC suggests that trust
pean Commission 2019a). It emphasises the use of in the measure could be undermined due to system-
Energy Performance Certificates (EPCs) in building atic bunching2 in the distribution of EPCs with regard
sale and rental advertisements (European Commis- to property sale prices (Hyland et al., 2016). Further-
sion 2018) to improve information for buyers and more, EPCs have been shown to suffer from a lack of
sellers on the indicative energy performance of a ex-post verification between measured and theoreti-
building. EPCs also contribute towards other aspects cal energy use (Burman et al., 2014; van Dronkelaar
of the EPBD, such as providing guidance on possible et al., 2016).
energy efficiency improvements.1 In Ireland, the Cli- There is often a disparity between the engineering
mate Action Plan plans to reduce energy use in build- model-based EPC and actual energy use (Cozza et al.,
ings through a policy to upgrade 500,000 homes to an 2020; De Wilde, 2014; Gram-Hanssen & Georg,
energy efficient B2 standard (Government of Ireland, 2018; Majcen et al., 2013; Zou et al., 2018). This is
2019). This is equivalent to a quarter of the national known commonly as the Energy Performance Gap
dwelling stock (Central Statistics Office, 2017). (EPG).3 Research has found a negative relationship
between dwelling energy efficiency and the direction
The value of Energy Performance Certificates of the EPG, with a positive EPG (higher energy use)
for energy efficient dwellings and a negative EPG
Despite policymaker enthusiasm for introducing (lower energy use) for the least efficient (Cozza et al.,
EPCs, evidence on the relationship between EPCs and 2020; Majcen et al., 2013; van den Brom et al., 2018).
property prices is mixed. Although studies for the EU Studies of the EPG have identified the influential role
and Ireland found correlations between a better rat- of occupant behaviour (De Wilde, 2014; Gram-Hans-
ing and a higher sales or rental prices in EU countries sen & Georg, 2018; Zou et al., 2018).
(European Commission, 2013; Hyland et al., 2013), There has been substantial evidence on the behav-
German homeowners found it was difficult to trans- ioural factors influencing energy use when dwelling
late EPCs into the value of energy efficiency and did energy efficiency changes (‘retrofit’). Many studies
not consider it a priority in their property purchase have identified ‘rebound’ effects, where a lower effec-
decision (Amecke, 2012). Evidence from Northern tive price of heating encourages increased energy
Ireland that applies quantile regression finds evidence use (Heesen & Madlener, 2018; Sorrell et al., 2009).
of a premium attached to energy efficient dwellings at Some studies of retrofit have found a ‘prebound’
high sales prices and discounts attached to low energy effect, where the least energy efficient dwellings con-
efficiency dwellings for sale at high prices (McCord sume less heating energy than expected (per their
et al., 2020).
EPCs have also been related to other important
outcomes. Comerford et al. (2018) find that introduc- 2
Bunching is defined by Hyland et al. (2016) as ‘an excess
ing an EPC induced investment in household energy frequency of homes on the favourable side of a threshold
efficiency in the UK. Evidence from Wales suggests accompanied by a much-reduced frequency on the unfavour-
able side of that threshold.’.
a statistically significant price premium of 12.8% 3
The Energy Performance Gap can be considered within the
broader theory of the Energy Efficiency Gap (Jaffe and Stavins
1994) which considers the under-adoption of energy-efficient
1
See [Link] goods with a positive net present value at the societal level.

13
Energy Efficiency (2021) 14: 57 Page 3 of 28 57

EPC) following a retrofit (Sunikka-blank & Galvin, i) This estimate of the EPG considers a whole-home
2012). Accurate estimates of the EPG are compli- measure of energy use. This is different to other
cated by improvements in dwelling energy efficiency studies which focus exclusively on the EPG for
that may induce any behavioural change in the occu- a single fuel source for space and water heating
pant. Aydin et al. (2017) show a negative relationship (Cozza et al., 2020; van den Brom et al., 2018).
between household income and rebound in gas use.4 Monitoring both electricity and gas demand can
Research into an Irish energy retrofit also found that capture fuel switching behaviour. This is impor-
socially vulnerable occupants often under-heat their tant due to the increasing use of alternative fuels
homes and use more energy and alternative heating in households, especially for socially vulnerable
fuels following a retrofit (Coyne et al., 2018). homes in low efficiency dwellings (Coyne et al.,
The rapid nature of technological change also 2018; van den Brom et al., 2018). Accounting
poses challenges for research into the EPG. Delghust for electricity use is especially relevant as energy
et al. (2015) note how research needs to study all fuels efficient dwellings tend to have a higher share of
used in the house, including electricity, which repre- heating from electricity (Delghust et al., 2015).
sents a greater share of energy use in efficient dwell- Failure to account for fuel switching may over-
ings and is becoming more popular due to changes in state the true EPG when measured using only one
heating systems. fuel source.
ii) This research features a generally representative
Contribution sample of households over a 2-year period. Some
earlier EPG studies only feature social housing
Evidence about the EPG suggests that policies aiming tenants (Majcen et al., 2013; van den Brom et al.,
to reach a certain EPC standard may not deliver the 2018), a cohort which has been shown elsewhere
expected energy savings (Cozza et al., 2020; Gram- to have particular energy use behaviour (Coyne
Hanssen & Georg, 2018; Zou et al., 2018). Research et al., 2018; Delghust et al., 2015). For this rea-
has noted that country-level differences in the imple- son, a general sample of households may provide
mentation of the EPBD require country-specific a more general view of the EPG.
studies of the Energy Performance Gap (Andaloro iii) The estimate of the EPG does not include changes
et al., 2010; Delghust et al., 2015). This is especially in occupant behaviour that would be induced due
true for Ireland, where there are ambitious plans to to a change in dwelling energy efficiency from a
upgrade the energy efficiency of the dwelling stock retrofit. Other studies note the potential for ret-
(Government of Ireland, 2019). rofit in their sample used to estimate the EPG
This is the first paper that tests for the presence of (Cozza et al., 2020; Heesen & Madlener, 2018).
an EPG using a measure of whole-home energy use However, research has shown that a retrofit can
for a non-social housing sample of 9923 households induce a behavioural change in energy use (Sorrell
that do not receive a retrofit. The key contribution et al., 2009; Sunikka-blank & Galvin, 2012; Webber
of this paper is the combination of (i) the analysis of et al., 2015). This is the first study to estimate
whole-home energy use, (ii) for a non-social housing the EPG for a sample that do not receive a grant
sample that (iii) does not feature behavioural changes supported retrofit during the observation period,
that would be induced by retrofit. Previous studies while addressing a need for residential modelling
have considered one or two of these aspects, but this with usage data (SEAI 2018).5
is the first study to combine all three and overcomes
some of the limitations of previous studies. The fol- The paper is laid out as follows: Section Literature
lowing three paragraphs detail the three main contri- details select relevant literature and background on
butions of the study: the Irish case study. Section Methodology and Data

4
Studies of domestic energy use are complicated by such
5
‘rebound’ effects, where improvements in energy efficiency The Sustainable Energy Authority Ireland (SEAI) promotes
lower the cost of energy services, thus increasing energy use energy efficiency in Ireland and manages the Irish EPC certifi-
(Sorrell et al., 2009). cation process.

13
57 Page 4 of 28 Energy Efficiency (2021) 14: 57

details the methodology, data and variables used. monitoring and increased heating use post-retrofit.
Section Results presents results, while Section Dis- Research into an Irish energy retrofit also found that
cussion discusses the main findings. Section Conclu- socially vulnerable occupants often under-heat their
sion concludes with some policy recommendations. homes and use alternative heating fuels (Coyne et al.,
2018).
Many studies of the EPG find occupant behaviour
Literature to be an important factor (De Wilde, 2014; Gram-Hanssen
& Georg, 2018; Zou et al., 2018). In a commercial
Studies of the Energy Performance Gap context, actual energy use can be 2.5 times larger
than predicted (Menezes et al., 2012). Herrando et al.
The Energy Performance Gap (EPG) is central to this (2016) find an average EPG of 30%. Majcen et al.
study. As noted in Section Limitations of Energy Per- (2013) find that energy inefficient homes consume
formance Certificates, there is a diverse range of stud- less than predicted and energy efficient homes con-
ies of the difference between actual energy use and sume more than predicted for a sample of 200,000
the level calculated by an EPC. Differences between social housing tenants in the Netherlands. Van den
the engineering model-based EPC and actual energy Brom et al. (2018) find similar results for a larger
use often arise (Cozza et al., 2020; De Wilde, 2014; sample of Dutch social housing tenants. They also
Gram-Hanssen & Georg, 2018; Majcen et al., 2013). find that dwelling type is more responsible for the
The different implementations of the EPBD across discrepancy in actual energy use than the EPC.
member states justify the need for country-specific Most studies only consider energy used for space
research (Andaloro et al., 2010; Delghust et al., and water heating and do not account for poten-
2015). tial fuel switching, which has been shown for select
The EPG often has a distributional aspect, where cohorts (Coyne et al., 2018; Delghust et al., 2015).
households in buildings of calculated poor energy Although research has identified discrepancies
efficiency and socially vulnerable occupants demon- between the actual and theoretical level of energy use,
strate substantial under-consumption, relative to the this message has not reached policymakers (Gram-Hanssen
EPC (Cozza et al., 2020). Studies of the EPG for a & Georg, 2018). Reasons for this discrepancy include
sample of social housing tenants found that energy the limitations of building modelling, inaccurate
efficient dwellings use more energy than calculated assumptions regarding occupant behaviour, and flaws
and vice versa (Majcen et al., 2013; van den Brom during the building design phase.
et al., 2018). In summary, research has shown that modelling
Studies of the EPG are complicated by behavioural residential energy use is challenging. Part of this chal-
changes in the occupant (‘rebound’) observed due lenge arises from how occupant behaviour changes
to retrofit, where a lower effective price of heating over time through rebound effects from changes in
encourages increased energy use (Heesen & Madle- building energy efficiency. The EPG has also been
ner, 2018; Sorrell et al., 2009). In some cases, a ret- shown to be particularly sensitive to the socioeco-
rofit has been shown to lead to a fall in energy use nomic status of occupants. For these reasons, a study
(Sunikka-blank & Galvin, 2012). Aydin et al. (2017) of the EPG using a measure of whole-home energy
highlight a negative relationship between house- use for a general sample of households that did not
hold income and rebound in gas use, with the lowest receive a retrofit is highly relevant.
income quintile featuring an average rebound almost
ten percentage points higher than the average rebound Ireland as a case study
for the rest of the distribution.
Estimates of the EPG are further complicated if Ireland intended to improve energy efficiency (low-
improvements in building energy efficiency from ering energy use) by 20% before 2020, relative to
a retrofit do not deliver the expected improvement average national energy use from the period 2001 to
(Gram-Hanssen & Georg, 2018). In the UK, Dowson 2005. This equates to energy savings of 31,925 GWh
et al. (2012) note that model predicted energy sav- (DCENR 2009). As part of the EU Energy Efficiency
ings may be halved in reality due to poor installation, Directive, member states must submit a National

13
Energy Efficiency (2021) 14: 57 Page 5 of 28 57

Energy Efficiency Action Plan with specific meas- improving the dwelling stock to a certain EPC
ures designed to improve energy efficiency (Euro- threshold. In Ireland, the government aims to retrofit
pean Union, 2012). By early 2017 Ireland had only 500,000 homes to B2 EPC standard by 2030 (Gov-
achieved a 12% improvement in energy efficiency ernment of Ireland, 2019). The presence of an EPG
and is expected to miss the 2020 target by 3.77% would cause actual savings to deviate from the level
(DCCAE, 2017). Achieving compliance for the 2020 expected.
target could cost €80–140 million.6
Despite this, Ireland has made progress in improv-
ing residential energy efficiency. Energy use per Methodology and data
dwelling has fallen by 32% from 1990 to 2015 due
to technology improvement, retrofits, building regula- Methodology
tions and macroeconomic factors (SEAI 2016). This
reduction is 37% when correcting for climate during Policymakers attempting to reduce emissions by
the period. However, there is more to be done as Ire- upgrading the dwelling stock to a certain EPC stand-
land has the fourth highest level of greenhouse gas ard face a problem if an EPC is based on assumptions
emissions in the EU of 13.3 tonnes of ­CO2 equivalent regarding theoretical occupant energy use (TQi ) that
per capita in 2017 (Eurostat, 2020). does not accurately reflect actual occupant energy use
Irish homes consume the most energy on average (AQit ). Consequently, the presence of an Energy Per-
in the EU, with the second largest average occupancy formance Gap (EPG) may limit the effectiveness of
in the EU-28 of 2.7 persons per house (SEAI 2018). policies designed to lower residential energy use by
According to EU-SILC data from 2017, 8.3 percent targeting a benchmark EPC standard. This paper fea-
of the Irish population live in apartments (Eurostat, tures three distinct research questions that explore the
2019b), lower than the EU average of 41.9 percent existence of an EPG and the factors influencing actual
and almost half the second-lowest ranked country, the energy use. Each research question directly corre-
UK (14.7 percent). Electricity plays an important role sponds to a subsection of the results.
in residential energy use. In 2017, over 20 percent of The first research question (Section Annual results)
electricity used in the Irish residential sector was for tests for the presence of an Energy Performance Gap
space and water heating.7 with a null hypothesis that the EPC accurately reflects
Irish interventions to improve residential energy actual occupant usage (Eq. 1). For a given household
efficiency aim to simplify consumer decision-mak- i in year t, actual household energy use ­(AQit) is equal
ing for durable appliances (Carroll et al., 2016), to to the theoretical EPC level of energy use ­(TQi) if
improve dwelling energy efficiency through a grant- there is no Energy Performance Gap. Since the EPC
supported retrofit (Scheer et al., 2013) or to alter is an estimate that does not account for appliance use
intraday electricity (Di Cosmo et al., 2014) and gas and occupant behaviour, it will not reflect true dwell-
(Harold et al., 2018) usage patterns. Research has ing energy use (discussed in Section Dependent Vari-
established how information from an EPC on theo- able: Actual Energy Use). It is expected that this dif-
retical dwelling energy efficiency is positively asso- ference will not be equal to zero (Cozza et al., 2020;
ciated with property sale and rental prices (Hyland van den Brom et al., 2018; Zou et al., 2018). This
et al., 2013). Lastly, Hyland et al. (2016) suggest result is presented in aggregate (kWh/year) and as
there is room to improve the Irish EPC due to system- a percentage of the EPC (Cozza et al., 2020). Equa-
atic bunching in the distribution of ratings. tion 1 shows testing for the presence of an Energy
Evidence of an EPG presents an issue if poli- Performance Gap.
cymakers expect real emissions reductions from
𝐇0 ∶ 𝐀𝐐𝐢𝐭 − 𝐓𝐐𝐢 = 0 (1)

6 This question is highly relevant since the Energy

See [Link]
missi​ng-​clima​te-​and-​energy-​targe​ts-​will-​cost-​irela​nd-​milli​ons/ Performance Gap is widely accepted in the research
7
See [Link] community but often ignored in policy discourse
key-​stati​stics/​resid​entia​l/#:​~:​text=​For%​202018%​20we%​20est​ (Gram-Hanssen & Georg, 2018). This is the first
imate%​20tha​t,%​2C%​20and%​202%​25%​20for%​20coo​king.

13
57 Page 6 of 28 Energy Efficiency (2021) 14: 57

Table 1  Data sources ID Data Details Source

A Energy use Electricity and gas readings Electric Ireland

B Building Energy Dwelling features, EPC SEAI
Rating
C Weather Heating degree days (HDDs), rainfall Met Eireann
Appendix 2 details the data cleaning process and the handling of outliers and unreliable data

study that controls for whole home energy use, the Data sources
sample and the potential for retrofit-induced behav-
ioural changes. Household energy use data of electricity and natural
The second research question (Section Bimonthly gas (A) is sourced from Electric Ireland, the largest
results) aims to quantify the extent to which key residential electricity utility in Ireland. This paper
dwelling factors influence actual energy use at the studies homes with natural gas heating observable by
bimonthly level (n = 149,518 readings). It uses a lin- meter readings. This data is observed from November
ear regression at the bimonthly time frequency and 2014 to June 2017, sixteen bimonthly8 periods. This
accounts for the influential role of seasonality in is merged with dwelling information from SEAI (B)
energy use. It considers the EPC and relevant dwell- using the common meter point number. Time-varying
ing characteristics (detailed in Sections. The Irish weather (C) controls are also included (Table 1). The
EPC and Dwelling Characteristics and Weather, sample consists of 9923 homes, 19,251 customer-
respectively). It models actual energy use ­(AQit) for year observations and 149,518 bimonthly readings.
household i in period t as a function of the theoreti- Appendix 2 details the data cleaning process, includ-
cal energy efficiency of the dwelling ­(TQi), a vector ing the removal of 333 households with highly abnor-
­(Xi) of key dwelling features such as dwelling type, mal energy use. This did not affect later results. As in
size, number of stories, age and a vector ­ (Wt) of Cozza et al. (2020), such households likely represent
time-varying weather controls (Eq. 2). Results (Sec- a holiday home that is sparingly used. The sample
tion Bimonthly results) begin by regressing actual distinguishes between households that never receive
energy use on the fully interacted EPC (Model 1) a retrofit (n = 8311) and households that receive one
and then expand to include dwelling characteristics prior to the observation period (n = 1612 houses).9
(Model 2) and a time fixed effect (Model 3). Equa-
tion 2 shows linear regression of actual energy use at The Irish EPC
bimonthly frequency.
SEAI operates the Building Energy Rating (BER)
AQit = ai + β1 TQi + β2 Xi + β3 Wt + uit (2) scheme, which is the Irish EPC. A BER is required
The third research question (Section Bimonthly for every property sold, rented or in receipt of a
results: Split by EPC) explores whether the relation- grant-supported retrofit (European Union, 2002). The
ship between actual energy use and key dwelling BER denotes the theoretical energy performance of
factors persists across each of the five EPC bands j a dwelling using a 15-point scale from A1 to G in
(Eq. 3). This is motivated by the potentially different units of kilowatt-hour of primary energy per metre
influence of covariates across the EPC spectrum. This squared per annum (Table 2). It is compliant with the
relationship is explored for each specific EPC grade EU Energy Performance of Buildings Directive and
at the bimonthly frequency (Model 4) using the lin- is based on both IS EN 13,790 and the UK Standard
ear regression used in Model 3, featuring a time fixed Assessment Procedure for dwelling energy ratings
effect. Equation 3 shows linear regression of actual
energy use at bimonthly frequency — by EPC grade. 8
The term ‘bimonthly’ denotes a period of 2 months. This is
not to be confused with the ‘twice-monthly’ frequency.
AQijt = ai + β1 TQij + β2 Xi + β3 Wt + uit (3) 9
Appendix 6 performs a robustness check of annual results,
split by subsample and finds no major differences.

13
Energy Efficiency (2021) 14: 57 Page 7 of 28 57

Table 2  Building energy rating (BER) levels and simplified EPC

BER A1 A2 A3 B1 B2 B3 C1 C2 C3 D1 D2 E1 E2 F G
< 25 > 25 > 50 > 75 > 100 > 125 > 150 > 175 > 200 > 225 > 260 > 300 > 340 > 380 > 450

Simple AB C D E FG
EPC 0–150 151–224 225–299 300–379 > 380

Source: SEAI. Note: Values in kWh/m2/year of primary energy. Simplified EPC is used in later analysis

(SEAI 2012). Equation 4 shows EPC primary energy being heated to 21 °C and the rest of the house heated
components. to 18 °C for 8 h a day (SEAI 2012). Given that space

TQTotal = QSpaceHeat + QWaterHeat + QAuxEnergy + QLighting − QPV − QCoGen (4)

The BER calculates ‘the energy required for space and water heating demand is the single largest energy
heating, ventilation, water heating and lighting, less demand in the home, the a priori expectation is that
savings from energy generation technologies’ (SEAI differences between actual and theoretical energy use
2012). Equation 4 details its components, which are would be largely driven by deviations in actual heat-
similar to other dwelling asset rating models (Majcen ing behaviour, especially after accounting for appli-
et al., 2013; van den Brom et al., 2018). The BER ance usage, which is not included in the EPC.
is influenced by factors such as dwelling size, type,
insulation, ventilation and heating system (SEAI Dependent variable: actual energy use
2014). It reflects theoretical primary energy use for
space and water heating, ventilation and lighting.10 Most studies of the Energy Performance Gap draw a
It does not include energy consumed by appliances, comparison between metered energy for heating with
estimates to be roughly 20% of delivered domestic the theoretical EPC (Cozza et al., 2020; Heesen &
energy use (SEAI 2018). Madlener, 2018; Majcen et al., 2013; Scheer et al.,
There is no formal validation of the BER awarded 2013). Such studies fail to capture electricity used as
from the in-home audit using real billing information. a secondary fuel (nor do they seek to). This omission
This deficiency has also been noted in studies of the has the potential to overstate the true Energy Perfor-
UK EPC (Burman et al., 2014; van Dronkelaar et al., mance Gap and may acutely affect the most energy
2016). Collins and Curtis (2018) examine changes in efficient homes, which feature a larger share of elec-
BER pre- and post-retrofit and find discontinuities in tricity use (Delghust et al., 2015). It may also dispro-
the national distribution of post-retrofit BERs, but portionately affect homes that engage in substantial
not in the pre-retrofit BERs. They find no evidence of fuel switching, such as low income social housing
illicit behaviour by BER assessors, but a high rate of occupants (Coyne et al., 2018). For these reasons,
low energy lighting (as a retrofit measure) prevalent we include an adjustment for appliance use to allow
in the distribution. This study is the first evaluation comparability with the EPC (since the EPC does not
of the BER using actual energy use data for a sample explicitly include appliances).
without retrofit. This study leverages the rich data available to
Weather conditions are considered at a local level, develop a measure of the Energy Performance Gap
but the model assumptions regarding occupant heat- that compares whole-home energy use (natural gas
ing behaviour are more important (SEAI 2012). The plus electricity) with the theoretical Irish EPC, which
BER assumes that the heating season runs from Octo- is denoted in units of primary energy use. Table 3
ber to May inclusive, with the primary living space summarises the primary energy factors for the sample
with an average ratio of 1.24, leading to actual meter
readings being inflated for comparison with the EPC.
10
Primary energy use includes energy delivered to the home Any mention of energy from this point is referring
and an overhead for energy lost in generation and transmission.

13
57 Page 8 of 28 Energy Efficiency (2021) 14: 57

Table 3  Summary of household-level primary energy factors

N Mean Median SD Min Max Skew Kurt

Primary energy 9923 22,674 20,165.91 11,343.98 − 28,236.7 122,777.9 1.67 8.06
Delivered energy 9923 18,505 16,459.11 9431.09 − 27,848.8 73,646.2 1.53 6.89
Ratio (P/D) 9923 1.24 1.21 0.14 1 3.11 7.31 69.4

Source: Author’s calculations using SEAI data for 9923 households. Values in kWh/year. Delivered energy includes energy assumed
to be consumed in the dwelling. Primary energy includes generation and transmission losses. The ratio helps to scale metered energy
use to reflect actual energy use

Fig. 1  Household average

bimonthly actual energy use
(AARelative)

to primary energy use, unless explicitly stated oth- Results in this paper consider two versions of appli-
erwise. While it would be interesting to separate pri- ance usage (AAj). The first (AARelative) involves a rela-
mary energy consumption into electricity and gas, the tive scaling of usage to a factor of 0.8, based on SEAI
EPC is an aggregate measure which does not distin- (2018). Equation 5 presents the construction of the
guish separate consumption levels. actual energy use variable (per square metre).
In order to compare theoretical energy use from
DeliveredEnergyi PrimaryEnergyi
the EPC, the variable of actual energy use is adjusted AQit = ∗
HeatableFloorAreai DeliveredEnergyi
∗ AAj (5)
to account for the heatable floor space and the ratio of
primary to delivered energy (Eq. 5).11 Actual energy The second version of appliance adjustment
use must also be adjusted to reflect the fact that EPCs accounts for concerns about the distributional effect
do not include energy use for appliances within the of a relative appliance adjustment across the spectrum
home. SEAI (2018) estimates that appliance usage of building energy efficiency. The second appliance
comprises, on average, 20% of Irish home energy use. adjustment (AAAbsolute) involves an absolute deduc-
tion for annual appliance usage (1357 kWh/year) for a
subset of appliances assumed common to each home
11
Appendix 1 provides more detail on the construction of the
dependent variable.

13
Energy Efficiency (2021) 14: 57 Page 9 of 28 57

Table 4  Comparison of N Mean Median SD Min Max Skew Kurt

mean annual actual and
theoretical energy use Actual energy use (AQY1) 9923 197.80 180.97 124.26 9.51 1777.77 1.50 10.80
­(AARelative)
Actual energy use (AQY2) 9328 211.16 197.7 123.73 11.29 1687.24 1.35 8.55
Theoretical energy use (TQ) 9923 235.44 214.54 101.03 39.97 1240.73 1.95 11.3
Values in kWh/m2/year

Fig. 2  Distribution of theo-

retical and actual annual
energy use (AARelative)

(Owen and Foreman 2012).12 The bimonthly panel of energy efficiency. The presence of actual energy use
9,923 households (n = 149,518 readings) has a com- in the right tail of the distribution is notable, espe-
pleteness of 94.17%, with an average of 15.06 peri- cially since the EPC lacks an upper bound on the
ods present and an average gap of 1.31 periods. Fig- theoretical energy efficiency of a G-rated home. A
ure 1 shows the seasonal pattern of household average Z-test is performed for each combination of the three
energy use for the entire sample.
The sample features 9923 households with 1 full
year of actual energy use ( 𝐀𝐐𝐘1) and 9328 observa- Table 5  Summary of continuous dwelling and weather control
variables
tions with a second full year ( 𝐀𝐐𝐘2). Table 4 com-
pares mean annual actual ( 𝐀𝐐𝐘1, 𝐀𝐐𝐘2) and theo- Variable Mean SD Min Max
retical (TQ) energy use, with all variables in units of
Number of floors 1.94 0.48 0 4
annual energy use per square metre. Figure 2 com-
Year of Construction 1979 28.98 1753 2017
pares the distributions of annual measured energy
Percentage of home that is living 21.33 9.81 0 81.1
usage ( 𝐀𝐐𝐘1, 𝐀𝐐𝐘2) to theoretical annual energy area
use (TQ). The distribution shows a higher share of Bimonthly heating degree days 53.29 10.6 10 61
observations in the A- and B-rated (most efficient) Total bimonthly precipitation (in 18.78 104 5.63 55.6
part of the distribution and a lower share of obser- cm)
vations in the C- and D-rated range of theoretical
n = 9923 homes. Weather for 16 bimonthly periods and five
weather stations. Living area is defined as the main living
space in the household. In most cases, this is the public room
12 with the largest floor area (SEAI 2012)
This is detailed further in Appendix 2.

13
57 Page 10 of 28 Energy Efficiency (2021) 14: 57

Table 6  Sample v SEAI Sample (n = 9923) SEAI population Occupied

population dwelling (n = 729,599) dwelling stock
comparison (n = 1,675,795)*
Count % Count % Count %

EPC AB 1351 13.61 104,084 14

C 4,257 42.90 270,628 37
D 2486 25.05 178,172 24
E 1059 10.67 86,401 12
FG 770 7.76 90,324 12
Dwelling type Detached 1197 12.06 232,677 32 715,133 43
Apartment 873 8.80 144,289 20 204,145 12
Semi-detached 3565 35.93 193,543 27 471,948 28
Terrace 4288 43.21 159,100 22 284,569 17
Total 9923 100 729,599 100 1,675,795 100
*
2016 Census values from Irish CSO for occupied households, excluding ‘Not stated’ and
temporary accommodation

variables. It confirms that the three distributions are dwelling stock). An additional comparison by dwell-
similar, with Z𝐀𝐐𝐘1 ,𝐀𝐐𝐘2 = 0.76, Z𝐀𝐐𝐘1 ,TQ = 0.24 and ing type also includes the 2016 census national occu-
Z𝐀𝐐𝐘2 ,TQ = 0.15. pied dwelling stock, which does not feature EPC
information. SEAI data under-represents the detached
Dwelling characteristics and weather dwellings and over-represents apartments and ter-
race homes. This is because an EPC is only required
The SEAI dataset features dwelling information on when a property is sold, leased or undergoes a retro-
house type, age, height and heatable floor area. Impor- fit. Compared to the SEAI population, the sample has
tantly, it includes a variable of theoretical energy use a higher share of C houses and a similar share of AB
in units of kWh/m2/year, which informs the categori- and D houses. The sample under-represents detached
cal EPC. Table 5 summarises key continuous variables homes and apartments and over-represents semi-
for the sample used in later bimonthly analysis that detached and terraced. It is important to bear in mind
measures the factors associated with actual energy use the potential for the sample results to understate the
(Sections Bimonthly results and Bimonthly results: true situation for FG-rated homes and to over-empha-
Split by EPC). Research has identified correlations sise results for C-rated homes at a national level.
between weather and electricity (Kavousian et al.,
2013), heating (Quayle & Diaz, 1979) and appliance
use (Hart & De Dear, 2004). To account for the poten- Results
tial influence of weather in the regression of actual
energy use, households are linked at the county level As noted in Section Methodology, results are pre-
with the nearest weather station.13 A heating degree sented in order of the three research questions. Sec-
day variable is constructed which reflects the number tion Annual results quantifies the Energy Perfor-
of days in a given period where the daily mean tem- mance Gap by testing for significant differences in
perature is below 15.5 °C. This reflects days where annual values of actual and theoretical energy use.
occupants are more likely to require heating. A total Section Bimonthly results features regressions using
bimonthly rainfall variable also features. the bimonthly data account for relevant covariates.
Table 6 compares key theoretical energy efficiency Finally, Section Bimonthly results: Split by EPC
for the sample and the population of SEAI EPC studies potential heterogenous effects by EPC cat-
records (which reflects roughly half of the national egory. Section Dependent Variable: Actual Energy
Use introduced two variants of actual energy use to
13
Available on the European Climate Assessment & Dataset account for appliance use. In each subsection, the
[Link]

13
Energy Efficiency (2021) 14: 57 Page 11 of 28 57

Table 7  Difference between annual actual (AQ) and theoretical (TQ) energy use ­(AARelative)
Actual annual energy Theoretical annual energy t-test of equality of means
use (AQ) use (TQ)
n Mean Median Mean Median TQ Difference SE P value
AQ AQ TQ
Mean %

AQAll – TQ 19,251 10,869 10,167 13,148 11,402 − 2279 − 17.33 61 0***

EPC Grade
AB 2601 10,569 9661 7571 6620 2998 39.60 122 0***
C 8269 10,880 10,334 10,826 9734 54 0.50 70 0.44
D 4835 10,917 10,231 14,353 12,826 − 3436 − 23.94 104 0***
E 2051 11,026 10,421 18,133 16,300 − 7106 − 39.19 173 0***
FG 1495 10,964 9853 24,962 22,466 − 14,000 − 56.09 290 0***
Dwelling
Apartment 1674 8115 7211 11,595 10,983 − 3481 − 30.02 163 0***
Detached 2316 13,712 13,150 19,385 17,184 − 5673 − 29.27 247 0***
Semi-detached 6905 11,398 10,917 14,008 12,495 − 2610 − 18.63 99 0***
Terrace 8356 10,197 9712 11,020 9490 − 823 − 7.47 82 0***
***
P < 0.01, **P < 0.05, *P < 0.10. Note: Units in kWh/year. Sample features 9,923 observations of one year of actual energy use and
a further 9328 observations from the same sample of houses with a second year of actual energy use. Medians reported. A test of
equality of medians (using signtest STATA command (Snedecor & Cochran, 1989)) confirms the same significant differences exist as
the t-test of means (displayed above)

first set of results features the dependent variable con- stem from differences in the samples. Results denote
structed using the RELATIVE appliance adjustment the difference between mean AQ and mean TQ (‘dif-
­(AARelative). The second set features the dependent ference’) and the percentage difference as a percent-
variable constructed using an ABSOLUTE appliance age of the mean TQ (‘% difference’), which is similar
adjustment ­(AAAbsolute). to the measure used by Cozza et al. (2020).
The most striking observation is a distinct lack
Annual results of variation in average actual energy use across the
entire sample. There is only a difference of 457 kWh/
The sample features 19,251 annual observations of year between the lowest and highest average. This
energy use (AQ), representing 9923 observations of suggests that the demand for energy is unresponsive
1 full year (AQY1) and a further 9328 observations to the energy efficiency of the dwelling. A simi-
featuring a second full year of energy use (AQY2).14 lar relationship has been observed in a study of UK
Energy use variables are in units of kilowatt-hours office buildings (Better Buildings Partnership 2019).
per year (kWh/year). Table 7 performs a test of On a comparative basis, there are significant differ-
paired differences showing that average annual actual ences between average actual and theoretical energy
energy use is significantly lower than the theoretical use. The most efficient homes (AB) feature an aver-
level from the EPC, suggesting the existence of an age positive difference of 2998 kWh per year, 39%
Energy Performance Gap (EPG). The average deficit greater than theoretical. Conversely, less efficient
in annual consumption is 2279 kWh, roughly 15% of homes (D, E, FG) exhibit actual energy use lower
the average value of 15,000 kWh/year considered by than theoretical, with an average difference rang-
the Irish utilities regulator (CRU 2017).15 Differences ing from 24% for D-rated homes to 56% for F- and
between the sample and the regulator reference likely G-rated homes. There are also significant differences
by dwelling type. Apartments and detached dwellings
14
Appendix 5 features a robustness check that splits data for feature a deficit in the region of 30%. Semi-detached
each full year of actual energy use (Y1, Y2).
15
homes semi-detached (19%) and terrace houses (7%)
The Irish regulator reference values for average energy use
are discussed in Appendix 4.
feature a smaller deficit.

13
57 Page 12 of 28 Energy Efficiency (2021) 14: 57

Table 8  Difference between annual actual (AQ) and theoretical (TQ) energy use ­(AAAbsolute)
Actual annual energy Theoretical annual energy t-test of equality of means
use (AQ) use (TQ)
n Mean Median Mean Median TQ Difference SE P value
AQ AQ TQ
Mean %

AQAll – TQ 19,251 12,044 11,158 13,148 11,402 − 1105 − 8.40 68 0***

EPC Grade
AB 2601 11,630 10,499 7571 6620 4059 53.61 147 0***
C 8269 12,039 11,356 10,826 9734 1213 11.20 83 0***
D 4835 12,119 11,207 14,353 12,826 − 2234 − 15.57 120 0***
E 2051 12,307 11,490 18,133 16,300 − 5826 − 32.13 195 0***
FG 1495 12,187 10,773 24,962 22,466 − 12,800 − 51.28 308 0***
Dwelling
Apartment 1674 8617 7507 11,595 10,983 − 2978 − 25.68 184 0***
Detached 2316 15,620 15,003 19,385 17,184 − 3765 − 19.42 267 0***
Semi-detached 6905 12,702 12,063 14,008 12,495 − 1306 − 9.32 111 0***
Terrace 8356 11,195 10,584 11,020 9490 175 1.59 93 0.06*
***
P < 0.01, **P < 0.05, *P < 0.10. Note: Units in kWh/year. Sample features 9923 observations of 1 year of actual energy use and
a further 9328 observations from the same sample of houses with a second year of actual energy use. Medians reported. A test of
equality of medians (using signtest STATA command (Snedecor & Cochran, 1989)) confirms the same significant differences exist as
the t-test of means (displayed above)

A test of equality of medians (Snedecor & amount (11.20%). FG-rated homes consume less than
Cochran, 1989) reports the same level of significance theoretical (− 51.28%), but the magnitude of this dif-
and direction of results as the t-test of means. Table 7 ference is smaller. A test of equality of medians is
shows a greater-than-theoretical energy use for effi- showing similar significance of results.
cient homes and lower-than-theoretical energy use Figure 3 illustrates the extent of the EPG across
for less efficient homes, a finding which is consistent the spectrum of dwelling energy efficiency. The left
with other estimates of the EPG (Cozza et al., 2020; panel features the EPG as a percentage of the theo-
Majcen et al., 2013; van den Brom et al., 2018). How- retical EPC. The right panel features the EPG in abso-
ever, this result has not previously been shown using lute terms. It emphasises the substantial differences in
a measure of whole-home energy use. In particular, the EPG, with a positive EPG for the most efficient
Cozza et al. (2020) find a median EPG of − 11% and dwellings and a negative EPG for the least efficient
mean EPG of − 6% for a sample of Swiss dwellings. dwellings.
In this study, the median EPG is similar (10.8%), but
the mean difference is far greater (− 17%). This con- Bimonthly results
firms the presence of an EPG in the Irish context but
suggests that the EPG may be larger. Section Annual results provided evidence of an EPG
Results using a measure of the dependent variable across the entire EPC spectrum on an annual basis.
that features an absolute appliance deduction still Sections Bimonthly results and Bimonthly results:
suggests a minor difference 677 kWh/year between Split by EPC investigate the factors associated with
the highest and lowest actual energy use averages actual energy use at the bimonthly frequency to bet-
(Table 8). Within EPC bands, results suggest that an ter understand seasonal differences (Tables 9, 10,
EPG exists. The overall average difference is smaller 11 and 12), using a linear regression (Eqs. 2 and 3).
(1105 kWh, − 8.40%), yet larger positive differences There are 149,518 data points in these regressions,
are observed for AB-rated homes (53.61%). C-rated indicated in the fourth row from the bottom in each
homes also consume more than the theoretical table. In addition to the EPC and dwelling type, each

13
Energy Efficiency (2021) 14: 57 Page 13 of 28 57

Fig. 3  Comparison of Energy Performance Gap. Note: the left ers denoting the adjacent value. Sample includes 19,251 obser-
panel presents the EPG as the difference between actual and vations with 9923 observations of 1 year of actual energy use
theoretical energy use as a percentage of theoretical energy and a further 9328 observations featuring a second year of
use. The right panel presents the EPG in absolute differences. observed actual energy use. Figure reflects the relative appli-
Each box reflects the interquartile range of EPG, with whisk- ance adjustment

regression controls for the following dwelling char- spring/summer and higher energy use in autumn/win-
acteristics that are related to dwelling energy use, ter. The coefficients for the EPC and dwelling features
obtained from the EPC: number of floors, year of con- are largely unchanged. Models 1–3 also interact the
struction, percentage of home classed as living area continuous EPC with its categorical form. Signifi-
(from Section Dwelling Characteristics and Weather). cant interactions in Model 1 suggest a heterogenous
Regressions also control for weather using a measure relationship between the continuous EPC and actual
of heating degree days and bimonthly rainfall. energy use, depending on the theoretical level of
Actual bimonthly energy use (AQt ) is modelled energy efficiency. Following AIC and BIC criteria,
as a function of the bimonthly theoretical energy use we consider Model 3 the most appropriate.16
(TQt = TQ∕6), a full interaction with the categorical Results in Table 10 are obtained using the variant
EPC, specific dwelling characteristics and weather of actual energy use that features an absolute reduc-
controls (Table 9). Standard errors are clustered at tion in appliance use (­AAAbsolute) are in line with
the household level. Table 9 shows a less than 1:1 those in Table 9. The exception is that every signifi-
relationship between changes in actual and theoreti- cant coefficient is larger in magnitude than before,
cal energy use. Model 1 suggests that, on average, a 1 reflecting a stronger association between theoretical
kWh/bimonth increase in theoretical energy use leads energy use, dwelling characteristics and actual energy
to a 0.47 kWh increase in actual bimonthly energy use.
use. In Model 2 the theoretical EPC coefficient falls to
0.198 and features significant effects for dwelling size Bimonthly results: split by EPC
and weather covariates. A one square metre increase
in dwelling size is associated with 8.69 kWh higher This section answers the third research question,
actual energy use, on average. Relative to a one floor which investigates whether the effects from Model 3
dwelling, houses with a second floor, on average, use are heterogeneous across the levels of the EPC. These
548kWh more each period. Results suggest that larger results are presented in Model 4, which splits the
homes consume more energy and that actual energy sample according to EPC (Table 11). Results show
use rises during colder or periods with more rain.
Model 3 replaces the weather variables with
a categorical time variable. Relative to Novem- 16
Appendix 7 confirms the relationship persists using a cat-
ber–December 2014, we observe lower actual use in egorical version of theoretical energy use.

13
57 Page 14 of 28 Energy Efficiency (2021) 14: 57

Table 9  Bimonthly results — continuous EPC ­(AARelative)

Dep Var: AQ actual energy Model 1 Model 2 Model 3
use (kWh/bimonth)
Coef SE Coef SE Coef SE

TQ (kWh/bimonth) 0.466*** (0.042) 0.202*** (0.063) 0.222*** (0.061)

EPC = AB (REF)
C − 33.080 (67.636) 63.796 (69.185) 67.532 (66.217)
D − 50.590 (72.906) 132.875* (76.916) 131.218* (73.388)
E − 204.105** (94.404) 14.525 (99.869) 32.140 (95.389)
FG 35.592 (117.020) 109.705 (113.436) 71.718 (111.735)
TQ#AB (REF)
TQ#C − 0.100** (0.048) − 0.027 (0.049) − 0.041 (0.047)
TQ#D − 0.183*** (0.047) − 0.071 (0.051) − 0.094* (0.049)
TQ#E − 0.185*** (0.050) − 0.043 (0.057) − 0.076 (0.055)
TQ#FG − 0.321*** (0.049) − 0.102* (0.059) − 0.124** (0.057)
Detached (REF)
Apartment 5.487 (41.743) − 56.685 (39.239)
Semi-detached − 17.547 (32.802) − 47.427 (31.008)
Terrace − 18.061 (35.134) − 53.857 (33.098)
1 Floor (REF)
2 Floors 518.289*** (33.150) 492.860*** (31.563)
3 Floors 829.710*** (46.477) 784.223*** (44.264)
4 Floors 552.210*** (206.635) 483.133** (193.952)
Floor area ­(m2) 8.480*** (1.297) 7.806*** (1.235)
Year of construction − 0.040 (0.366) − 0.052 (0.344)
Living area percent − 2.032* (1.103) − 3.294*** (1.055)
Heating degree days 27.651*** (0.313)
Total precipitation (cm) 23.491*** (0.674)
Bimonthly Time Dummy Yes Yes
Constant 1024.657*** (53.899) − 1425.67* (751.461) 1857.671*** (707.585)
N 149,518 149,518 149,518
r2 0.028 0.112 0.255
AIC 2,634,556 2,621,070 2,594,909
BIC 2,634,655 2,621,278 2,595,246

Asterisks note significance at the 10 percent (*), 5 percent (**), or 1 percent (***) level. Standard errors in brackets. Model 1 fea-
tures a significant interaction for the continuous and categorical EPC independent variables

decreasing explanatory power for the least energy for dwelling size throughout and a significant floor
efficient dwellings. This finding is similar to van den area effect in all except E-rated homes, with larger
Brom et al. (2018), who find their EPC to be more average effects (8.15–9.78) than in the pooled Model
reliable for efficient households. 3 (8.13).
Differences are observed in the magnitude of the Using the alternative version of our dependent vari-
coefficient for the continuous measure of theoretical able (absolute appliance adjustment), we see similar
energy use (EPC). Average effects range from 0.16 results (Table 12). Average effects range from 0.20 to
to 0.45, which differs from the same coefficient in 0.56, suggesting that changes in theoretical energy use
Model 3 (0.22). When split by EPC, dwelling type is are more closely related to changes in actual energy
only significantly lower for C-rated apartments and use for the most energy efficient dwellings. Although
semi-detached houses. There are significant effects results at the bimonthly level do not prove the existence

13
Energy Efficiency (2021) 14: 57 Page 15 of 28 57

Table 10  Bimonthly results — continuous EPC ­(AAAbsolute)

Dep Var: AQ actual energy Model ­1Abs Model 2 Abs Model 3 Abs
use (kWh/bimonth)
Coef SE Coef SE Coef SE

TQ (kWh/bimonth) 0.586*** (0.052) 0.257*** (0.077) 0.282*** (0.074)

EPC = AB (REF)
C − 36.686 (83.897) 80.930 (85.970) 85.474 (82.242)
D − 50.335 (90.273) 172.676* (95.467) 170.472* (91.035)
E − 234.916** (117.127) 27.987 (124.325) 49.941 (118.660)
FG 69.487 (141.041) 154.345 (137.354) 106.771 (134.915)
TQ#AB (REF)
TQ#C − 0.126** (0.060) − 0.035 (0.060) − 0.051 (0.058)
TQ#D − 0.232*** (0.058) − 0.091 (0.063) − 0.120** (0.061)
TQ#E − 0.234*** (0.062) − 0.055 (0.070) − 0.096 (0.067)
TQ#FG − 0.407*** (0.061) − 0.134* (0.073) − 0.162** (0.070)
Detached (REF)
Apartment 1.912 (51.753) − 75.859 (48.617)
Semi-detached − 26.190 (40.772) − 63.580* (38.530)
Terrace − 28.773 (43.670) − 73.615* (41.127)
1 Floor (REF)
2 Floors 640.371*** (40.979) 608.591*** (38.963)
3 Floors 1031.274*** (57.534) 974.431*** (54.734)
4 Floors 677.805*** (255.956) 591.653** (240.308)
Floor area ­(m2) 10.473*** (1.584) 9.632*** (1.504)
Year of construction − 0.180 (0.454) − 0.195 (0.426)
Living area percent − 2.413* (1.368) − 3.990*** (1.307)
Heating degree days 34.441*** (0.390)
Total precipitation (cm) 29.411*** (0.840)
Bimonthly time Yes Yes
Constant 1009.958*** (66.653) − 1770.902* (933.531) 2329.161*** (878.001)
N 149,518 149,518 149,518
r2 0.029 0.113 0.257
AIC 2,700,145 2,686,646 2,660,127
BIC 2,700,244 2,686,855 2,660,464

Asterisks note significance at the 10 percent (*), 5 percent (**), or 1 percent (***) level. Standard errors in brackets. Model 1 fea-
tures a significant interaction for the continuous and categorical EPC independent variables

of an EPG (Section Annual results), they highlight occupant demand for energy may not be as respon-
the statistically significant role of the EPC, dwelling sive to dwelling energy efficiency, which has been
characteristics and seasonality when modelling actual observed in the energy use of commercial buildings
energy use. (Better Buildings Partnership 2019). This study
also finds evidence of an Energy Performance Gap
(EPG) for the Irish EPC, with significant differences
Discussion between actual and theoretical energy use. Annual
actual energy use is below the theoretical level,
The key insight from this study is the striking lack with a mean deficit of 2279 kWh/year (− 17%)
of variation in average actual energy use across and a median deficit of 1235 kWh/year (10.83%).
the sample (457 kWh/year). This suggests that By comparison, Cozza et al. (2020) find a median

13
57 Page 16 of 28 Energy Efficiency (2021) 14: 57

Table 11  Bimonthly Dep Var: AQ actual Model 4 — EPC label

results — continuous energy use (kWh/
EPC, by EPC category bimonth) AB C D E FG
­(AARelative)
TQ (kWh/bimonth) 0.21** 0.16** 0.10 0.44*** 0.09*
(0.09) (0.06) (0.08) (0.13) (0.04)
Detached (REF)
Apartment 31.98 − 82.10 − 0.66 − 57.27 − 179.64
(101.10) (62.65) (83.19) (136.72) (166.66)
Semi-detached − 97.60 − 110.07** 27.47 − 48.44 117.17
(79.59) (46.16) (61.41) (111.91) (116.53)
Terrace − 33.00 − 76.90 − 16.62 − 53.52 − 57.04
(81.51) (50.51) (66.16) (115.57) (119.99)
1 Floor (REF)
2 Floor 357.15*** 500.60*** 598.68*** 426.30*** 441.32***
(114.05) (51.35) (54.03) (87.08) (100.78)
3 Floor 673.08*** 819.08*** 725.50*** 771.94*** 708.54***
(133.47) (67.14) (91.89) (146.46) (189.89)
4 Floors 1110.28*** 357.92 551.62*
(149.40) (246.11) (329.90)
Floor area ­(m2) 7.45*** 8.41*** 9.84*** − 9.41 8.32**
(2.05) (1.98) (3.65) (7.38) (3.87)
Year built − 1.38 − 0.58 0.01 2.81*** − 1.94*
(0.95) (0.59) (0.66) (0.84) (1.11)
Living area percent − 12.06*** − 3.86** 0.51 − 3.93 − 0.26
(2.98) (1.69) (1.92) (2.86) (3.97)
Bimonthly Time Yes Yes Yes Yes Yes
Constant 3333.49* 2764.72** 1605.77 -3851.8** 5548.46***
(2011.66) (1184.99) (1334.01) (1658.07) (2126.24)
N 20,116 64,456 37,506 15,897 11,543
r2 0.273 0.260 0.255 0.265 0.219
AIC 346,797 1,113,643 651,925 277,468 203,840
BIC 346,995 1,113,879 652,147 277,659 204,023

Asterisks note significance at the 10 percent (*), 5 percent (**), or 1 percent (***) level. Standard
errors in brackets

EPG of − 11% and mean EPG of − 6% for a sample These results have significant policy implications,
of Swiss dwellings. In this study, the median EPG as a nationwide upgrade of dwelling energy efficiency
is similar (10.8%), but the mean difference is far may be ineffective or counter-productive, with under-
greater (− 17%). This confirms the presence of an heating in the least efficient homes potentially lead-
EPG in the Irish context but suggests that the EPG ing to over-heating post-upgrade. This is an important
may be larger. consideration, as it may run counter to policy targets
The size of the EPG varies by the level of the of reducing energy use (while accepting it would
EPC. For the most efficient homes, actual energy use likely improve occupant comfort and welfare). This
exceeds theoretical, with an average difference in the is especially relevant considering the fact that the
range of 39.6% for AB-rated homes. For the least effi- Energy Performance Gap is widely accepted in the
cient homes, actual energy use is below theoretical, research community, but often ignored in policy dis-
with an average deficit of 24% for D-rated homes, course (Gram-Hanssen & Georg, 2018).
39% for E-rated homes and 56% for FG-rated homes. Findings are in line with similar studies of the EPG
(Cozza et al. 2020; Majcen et al. 2013; van den Brom

13
Energy Efficiency (2021) 14: 57 Page 17 of 28 57

Table 12  Bimonthly Dep Var: AQ actual Model ­4Abs — EPC Label

results — continuous energy use (kWh/
EPC, by EPC category bimonth) AB C D E FG
­(AAAbsolute)
EPC (kWh/bimonth) 0.28*** 0.20** 0.13 0.56*** 0.09*
(0.10) (0.08) (0.10) (0.16) (0.05)
Detached (REF)
Apartment 46.28 − 107.23 − 9.15 − 81.08 − 238.90
(125.04) (78.08) (103.02) (170.65) (199.12)
Semi-detached − 120.69 − 140.87** 30.48 − 69.20 131.12
(98.38) (57.53) (76.44) (139.03) (142.58)
Terrace − 38.88 − 102.54 − 25.43 − 80.06 − 84.14
(101.16) (62.95) (82.25) (143.65) (146.65)
1 Floor (REF)
2 Floor 447.60*** 617.86*** 735.36*** 528.56*** 550.79***
(139.06) (63.87) (66.94) (107.96) (121.96)
3 Floor 840.97*** 1016.06*** 903.99*** 963.06*** 889.66***
(162.93) (83.37) (114.67) (181.54) (230.83)
4 Floors 1383.02*** 440.42 664.84
(183.26) (305.80) (409.67)
Floor area ­(m2) 8.76*** 10.44*** 12.12*** − 12.54 11.88**
(2.48) (2.47) (4.54) (9.17) (4.65)
Year built − 1.74 − 0.86 − 0.14 3.31*** − 2.47*
(1.18) (0.73) (0.82) (1.04) (1.35)
Living area percent − 14.99*** − 4.75** 0.65 − 4.52 0.35
(3.66) (2.11) (2.38) (3.54) (4.85)
Bimonthly Time Yes Yes Yes Yes Yes
Constant 5796.75** 3779.70** 2025.20 − 4260.4** 6792.98**
(2473.12) (1479.84) (1658.99) (2108.26) (2681.35)
N 20,116 64,456 37,506 15,897 11,543
r2 0.276 0.262 0.257 0.267 0.223
AIC 355,463 1,141,976 668,373 284,384 208,743
BIC 355,660 1,142,212 668,595 284,575 208,927
Asterisks note significance at the 10 percent (*), 5 percent (**), or 1 percent (***) level. Standard
errors in brackets

et al. 2018). Similar to the observation of Delghust under-heating observed in energy inefficient homes
et al. (2015), this study emphasises the importance of could be due to other barriers to energy use such as
accounting for electricity use, instead of limiting the fuel poverty, which have been established in research
focus strictly space and water heating (Scheer et al. elsewhere (Coyne et al. 2018).
2013). This is especially important as homes become Results at the bimonthly frequency indicate that a
increasingly energy efficient and electricity reliant. 1 kWh increase in bimonthly theoretical energy use
The lack of variation in energy use across dwell- is associated with a 0.222 kWh increase in actual
ing of very different theoretical efficiency presents energy usage, on average. Other results suggest the
opportunities for research across households of dif- EPC broadly works as intended, with a less efficient
fering dwelling energy efficiency and socioeconomic EPC being associated with greater actual energy
status. For example, it could be the case that occu- use, when also controlling for key dwelling charac-
pants of energy efficient homes have paid a premium teristics and seasonality. The coefficient values for
for a home that can be heated at a lower effective dwelling floor size (7.81 kWh/bimonth) indicate that
per-unit cost. Similarly, possible explanations for the larger homes tend to consume more energy. When

13
57 Page 18 of 28 Energy Efficiency (2021) 14: 57

split by EPC category, greater explanatory power for Results show there is very little difference in actual
more efficient homes is observed. The model better average consumption for households across the EPC
explains variation for more efficient homes, suggest- spectrum. There is a less than five percent discrep-
ing there is greater uncertainty when modelling less ancy (457 kWh/year) between the highest and low-
efficient homes, a finding that is consistent with van est average value. This is a surprising observation
den Brom et al. (2018). which warrants further investigation to understand
Future work might seek to address some of the the factors underlying this result. Analysis within
limitations of this study. One concern is that the actual EPC bands shows evidence of an Energy Perfor-
energy data observed is understated if a home uses mance Gap (EPG), with lower-than-expected energy
another fuel source, e.g. open wood burning stove. This use for houses with low energy efficiency and higher-
is true of many studies that focus on one space heat- than-expected energy use for energy efficient houses.
ing fuel. In this study, this risk is minimized by focus- For more efficient homes (AB, C) the average differ-
ing homes with either natural gas or electricity as their ence ranges from + 39% to − 56% of the relevant EPC
heating fuel. To address sample attrition that may arise value. Less efficient homes (E, FG) feature actual
from customers switching energy provider,17 criteria energy use lower than predicted, with an average
based on the number of readings, the level of missing- difference ranging from − 23 to − 56% below the rel-
ness and for unrealistically low metered energy use is evant EPC. Results using a measure of actual energy
applied to ensure sufficient energy use is observed (see use with an absolute deduction for appliance usage
Appendix 2). Customer switching could be addressed (instead of relative) display similar results.
with access to data from more utilities. The addition These findings are in line with similar studies of
of household socioeconomic information could help to the EPG that focused exclusively on social housing
explain the main result of a lack of variation in actual tenants (Majcen et al., 2013; van den Brom et al.,
energy use across the sample. Finally, future research 2018). Additional results show a heterogenous rela-
could seek to quantify changes in whole-home energy tionship between theoretical energy efficiency and
use before and following a home energy retrofit. This actual energy use across EPC levels. This is consistent
would require a dataset with metered energy use and with prior work that found the EPC has less explana-
EPCs pre- and post-retrofit. tory power for the least efficient homes (Cozza et al.,
2020; Sunikka-blank & Galvin, 2012; van den Brom
et al., 2018) and the ‘prebound’ effect (Cozza et al.,
Conclusion 2020; Sunikka-blank & Galvin, 2012).
Policymakers could seek to improve the EPC by
This paper investigates the difference between theo- including historical energy use information. This could
retical energy use denoted by a residential Energy be facilitated by the upcoming rollout of residential
Performance Certificate (EPC) with actual energy smart meters as part of the Climate Action Plan (Gov-
use for a sample of 9923 households in Ireland from ernment of Ireland, 2019). Since the Irish EPC is con-
late 2014 to mid-2017. It is the first paper to test for sistent with EU guidance, it is likely that the issues
the presence of an Energy Performance Gap using a identified in this paper could be present in other con-
measure of whole-home energy use for a non-social texts, especially in the UK, as the Irish EPC is based
housing sample of dwellings that do not receive a on the UK Standard Assessment Procedure for dwelling
retrofit. It focuses on homes heated by natural gas energy ratings (SEAI 2012) and a similar lack of vari-
and electricity to profile whole-home energy use and ation in actual energy use has been observed in com-
capture fuel switching. Households that underwent a mercial buildings (Better Buildings Partnership 2019).
retrofit during the observed period are excluded from Future work could include additional utilities, soci-
the sample in order to isolate the difference in actual oeconomic information and track occupants over time
energy use and the theoretical level created by the (as is done in a retrofit study by Aydin et al. (2017))
engineering-based model that informs the EPC. to minimise customer attrition and include additional
relevant covariates. This would enable researchers to
17
There were 26,154 electricity customer switches on average understand the factors causing differences in the most
each month in 2017 (CRU, 2018). and least efficient homes.

13
Energy Efficiency (2021) 14: 57 Page 19 of 28 57

Appendix 1: Constructing a comparable measure

of actual energy use Fig. 4
The EPC is based on heatable floor area (in units of
This paper tests for differences in actual energy use kWh/m2/year). Heatable and total floor area variables are
(AQ) with the theoretical level from the Irish residential present in the SEAI data. One source of sample attrition
EPC (TQ). Whenever actualenergy use is mentioned is that approximately 11% of dwellings in the SEAI data-
(Eq. 6) is applies tothe variable created that is compara- base report a heatable floor area of zero, the majority being
ble to theoretical energy use (Eq. 7). Actual meter read- mid-floor apartments where heat loss is through the exter-
ings are aggregated, weighted by the heatable floor area nal wall (Table 13). These are excluded as they prevent the
(per EPC) and account for the difference between ‘pri- actual energy use variable from being constructed.
mary’ and ‘delivered’ energy use (per EPC). The actual Figure 5 compares the distribution of the con-
energy use variable must be adjusted to reflect ‘pri- tinuous theoretical energy use (TQ) with the actual
mary’ energy consumed. Per SEAI, ‘primary’ energy energy use (AQ) for the first and second full years
use includes the energy consumed in the house plus an observed (Y1, Y2). The distributions are within a
overhead for energy used in its generation and trans- similar range, which suggests that the removal of
mission. ‘Delivered’ energy is only what is consumed extreme values was successful.
within the home. In the SEAI database of 729,609 Fig. 5
homes, ‘primary’ is 39% larger than ‘delivered’ on
average. In the sample of 9923 homes, ‘primary’ is 22%
larger than ‘delivered’ on average (Fig. 4). We then
adjust it to only reflect energy for space and water heat- Appendix 2: Data cleaning process
ing, lighting and ventilation.
This section details the data cleaning process in this paper.
PrimaryEneryi
𝐄𝐏𝐂 ∶ TQi = (6) We describe the linking process, discuss how energy data are
HeatableFloorAreai merged and list the data cleaning and exclusion criteria used.
∑
Meter readingsit TotalPrimaryEnergyi
𝐀𝐜𝐭𝐮𝐚𝐥𝐄𝐧𝐞𝐫𝐠𝐲𝐔𝐬𝐞 ∶ AQit = ∗ ∗ AAj (7)
HeatableFloorAreai TotalDeliveredEnergyi

Fig. 4  Primary v delivered

energy

13
57 Page 20 of 28 Energy Efficiency (2021) 14: 57

Table 13  Homes with zero heatable floor area

Apartment Basement Detached Ground apartment Maisonette Mid Top Total
floor apartment floor apartment

Houses 2,359 2 1 815 2357 42,276 33,406 81,216

Source: SEAI BER database (n = 729,609)

1. Linking customers Readings are aggregated bimonthly and adjusted

to reflect the period of use, e.g. A reading in March
a. Gas fuelled homes are identified by linking 2015 reflects usage in January 2015. Additional
Electric Ireland gas customer accounts with observations are dropped for the following reasons:
the corresponding Electric Ireland electric-
ity account. • Total household metered energy usage is zero.
b. Electricity accounts are anonymously • House is not heated by gas (per SEAI).
merged with the SEAI dwelling data using • Multiple meters for a house (per SEAI).
the electric meter number (MPRN), which • A house received a grant-supported retrofit during
was unobservable to the research team. the observed period (per SEAI).
c. There are 286,523 unique customer matches • Drop electricity readings before the start of the
between the original Electric Ireland meas- gas sample (November 2014) to focus on the com-
ured energy use data and the SEAI dwelling mon period of electricity and gas use.
data. Of this, 21,198 are unique matches for • A ratio of the number of missing periods to the
a gas customer account that is linked to an number of periods present is created. This ratio
electricity customer account. is equal to 0.75 if a household is present for 16
periods but if missing for any four periods. Any
2. Energy data merge and sample restrictions
household with a ratio less than 0.5 is dropped,
which does not discriminate against homes that
The original energy dataset features 30,045,696
enter the data later.
daily energy readings (28,563,625 electricity,
• Drop any household with a gap between observa-
1,482,071 gas) beginning November 2011. We drop
tions of at least 6 months. Although the customer
households with no match in the SEAI dwelling data.
might be present during the entire sample period,

Fig. 5  Distributions of con-

tinuous theoretical v actual
energy use (Y1, Y2)

13
Energy Efficiency (2021) 14: 57 Page 21 of 28 57

Fig. 6  Bimonthly heating

degree days and rainfall
(by station). Source: Met
Eireann. Note: A Heating
Degree Day occurs when
the mean temperature falls
below 15.5 °C

such a large missing period makes it unsuitable for in Appendix 1, the SEAI dataset includes two vari-
analysis, especially for annual values. ables of calculated annual use, one reflecting con-
• Drop any house with fewer than six observations sumed energy (delivered energy) and the other
(a full year of readings). including an overhead for energy generation (pri-
• Drop any house with an annual energy usage value mary energy).
(Y1, Y2) reading in the top or bottom one percent
of the distribution to observe households with Appendix 3: Weather and energy price controls
realistic energy use.
• Drop homes with an SEAI heatable floor area of Fig. 6
less than ­10m2. Mostly apartments. Fig. 7
• Remove households with a delivered energy value
in the top or bottom 1% of the distribution. As noted

Fig. 7  EU 2016 H2

electricity and gas price.
Source: Eurostat H2 2016
Energy Prices

13
57 Page 22 of 28 Energy Efficiency (2021) 14: 57

Table 14  Assumptions regarding appliance-specific electricity energy use (for constructing ­AAAbsolute)
Fridge-freezer Oven Microwave Electric kettle Toaster Washing ­machine1 Vacuum LCD TV Total

Annual consumption 427 290 56 167 22 178 18 199 1357

(kWh)

Source: Owen & Foreman (2012). [Link] assume a multiple person household. The annual average lighting energy consumption of
548 kWh is disregarded in our analysis as the EPC accounts for lighting

Appendix 4: The Irish regulator annual energy use by tariff, with urban and rural 24 h tariffs
use benchmark below the national average. However, urban and
rural day/night tariffs show an average above the
The Irish Commission for Regulation of Utili- national mean. The CRU gas data shows that the
ties (CRU) provides reference values for annual national average of 11,000 is higher than what
domestic energy use to be used by price com- would be expected for an apartment. Unfortu-
parison websites and energy providers. 2017 nately, we cannot control for tariff type in our
annual averages were set at 4200 kWh and 11,000 data. This underscores the need to consider appro-
kWh for electricity and natural gas, respectively. priate reference points for annual national aver-
Tables 14 and 15 shows the CRU annual means ages when considering our constructed variable of
split by electricity meter type and dwelling type actual energy use and how it might vary by tariff
for gas. The CRU observes average electricity and property type.

Table 15  CRU average Annual average electric- Annual average gas

energy consumption ity consumption (kWh) consumption (kWh)

CRU mean (2017) 4200 11,000

Electricity tariff type Gas dwelling type
Urban 24 h 3600 (− 14%) Apartment (1–3 bed) 7000 (− 46%)
Urban day/night tariff 6200 (+ 48%) House (1–3 bed) 10,500 (− 5%)
Rural 24 h 3900 (− 7%) Large house (4–6 bed) 13,000 (+ 18%)
Rural day/night tariff 12,000 (+ 286%) Standalone residential 15,000 (+ 36%)

Source: CRU decision paper CER/17042

Table 16  Year 1: n Mean ­AQY1 Mean EPC Difference % Difference SE P value

difference between annual
actual (AQ) and theoretical AQAll,Y1 – TQ 9923 10,532 13,152 − 2620 − 19.92% 86 0***
(TQ) energy use ­(AARelative)
EPC grade
AB 1351 10,392 7509 2882 38.38% 171 0***
C 4257 10,521 10,842 − 322 − 2.97% 100 0.002***
D 2486 10,546 14,373 − 3827 − 26.62% 147 0***
E 1059 10,745 18,159 − 7414 − 40.83% 246 0***
FG 770 10,499 24,990 − 14,500 − 58.02% 400 0***
Dwelling type
Apartment 873 7,914 11,625 − 3711 − 31.92% 227 0***
Detached 1197 13,261 19,323 − 6061 − 31.37% 346 0***
Semi-detached 3565 11,026 13,996 − 2970 − 21.22% 140 0***
Terrace 4288 9,892 11,039 − 1146 − 10.38% 116 0***
***
P < 0.01, **P < 0.05, *P < 0.10. Sample of 9923 homes with 9923 observations of 1 year of
actual energy use and a further 9328 observations for the same houses with a second year of
observed actual energy use

13
Energy Efficiency (2021) 14: 57 Page 23 of 28 57

Table 17  Year 2: n Mean ­AQY2 Mean EPC Difference % Difference SE P value

difference between annual
actual (AQ) and theoretical AQAll,Y2 – TQ 9328 11,229 13,144 − 1916 − 14.57% 87 0***
(TQ) energy use ­(AARelative)
EPC grade
AB 1250 10,761 7637 3124 40.90% 175 0***
C 4012 11,262 10,808 453 4.20% 99 0***
D 2349 11,309 14,332 − 3022 − 21.09% 147 0.6
E 992 11,326 18,105 − 6778 − 37.44% 242 0***
FG 725 11,458 24,932 − 13,500 − 54.15% 421 0***
Dwelling type
Apartment 801 8333 11,563 − 3230 − 27.93% 233 0.002***
Detached 1119 14,194 19,451 − 5258 − 27.03% 354 0.181
Semi-detached 3340 11,796 14,021 − 2226 − 15.87% 139 0***
Terrace 4068 10,518 11,000 − 483 − 4.39% 117 0***
***
P < 0.01, **P < 0.05, *P < 0.10. Sample of 9923 homes with 9923 observations of 1 year of
actual energy use and a further 9328 observations for the same houses with a second year of
observed actual energy use

Appendix 5: Robustness check 1 — annual results deficit between actual and theoretical energy use,
(split by year of energy use) with smaller deficits observed across every level of
the EPC and property type. The only exception is the
This section provides a robustness check on the ear- most energy efficient homes, with a slightly larger
lier test of significant differences between annual val- deficit for AB- and C-rated homes.
ues of theoretical and actual energy use. The body of
the paper aggregates 19,251 observations of actual
annual energy use. Here we split the sample into Appendix 6: Robustness check 2 — annual results
9923 household-level observations for 1 full year of (split by subsample)
energy use and a further 9328 observations of a sec-
ond full year of energy use. Comparing Tables 16 and As noted in the body of the paper, the sample of 9923
17 we observe a similar trend in differences between houses (149,518 readings) is divided across 8311
actual and theoretical energy use. Results for the houses (124,763 readings) that never receive a retro-
second year of energy use feature a slightly smaller fit and a further 1612 houses (24,755) that completed

Table 18  Never retrofit — n Mean AQ Mean EPC Difference % Difference SE P value

difference between annual
actual (AQ) and theoretical AQControl – TQ 16,091 10,657 12,985 − 2327 − 17.92% 68 0***
(TQ) energy use ­(AARelative)
EPC grade
AB 2317 10,205 7087 3118 43.99% 129 0***
C 6590 10,604 10,346 258 2.49% 77 0.001***
D 3861 10,711 14,045 − 3335 − 23.74% 114 0***
E 1858 11,081 17,995 − 6914 − 38.42% 179 0***
FG 1465 10,937 25,035 − 14,100 − 56.32% 294 0***
Dwelling type
Apartment 1622 8088 11,501 − 3413 − 29.68% 165 0***
Detached 1723 13,391 19,390 − 5998 − 30.94% 301 0***
Semi-detached 5485 11,207 14,041 − 2834 − 20.18% 113 0***
Terrace 7261 10,167 10,998 − 831 − 7.56% 90 0***
***
P < 0.01, **P < 0.05, *P < 0.10. Sample of 8311 homes that never avail of a grant-supported
retrofit. 8311 observations of 1 year of actual energy use and a further 7780 observations for the
same houses with a second year of observed actual energy use

13
57 Page 24 of 28 Energy Efficiency (2021) 14: 57

Table 19  Already retrofit n Mean AQ Mean EPC Difference % Difference SE P value

— difference between
annual actual (AQ) and AQTreated – TQ 3160 11,949 13,980 − 2031 − 14.52% 138 0***
theoretical (TQ) energy use
EPC Grade
­(AARelative)
AB 284 13,540 11,517 2023 17.57% 375 0***
C 1679 11,966 12,711 − 745 − 5.86% 171 0***
D 974 11,735 15,573 − 3838 − 24.65% 250 0***
E 193 10,495 19,456 − 8961 − 46.06% 626 0***
FG 30 12,293 21,387 − 9094 − 42.52% 1318 0***
Dwelling Type
Apartment 52 8954 14,544 − 5590 − 38.44% 1043 0***
Detached 593 14,644 19,371 − 4728 − 24.41% 408 0***
Semi-detached 1420 12,135 13,879 − 1744 − 12.57% 194 0***
Terrace 1095 10,392 11,165 − 772 − 6.92% 196 0***
***
P < 0.01, **P < 0.05, *P < 0.10. Sample of 1612 homes that avail of a grant-supported retrofit
before the period of energy use. 1612 observations of 1 year of actual energy use and a further
1548 observations for the same houses with a second year of observed actual energy use

Table 20  Bimonthly OLS results — categorical EPC ­(AARelative)

Dep Var: Bimonthly measured Model 5 Model 6 Model 7
energy use (kWh)
Coef SE Coef SE Coef SE

EPC label = AB REF REF REF

C 37.644 (26.467) 152.072*** (25.291) 143.463*** (23.823)
D 36.961 (28.886) 224.829*** (29.235) 189.794*** (27.623)
E 54.256 (35.535) 277.004*** (37.420) 231.953*** (35.467)
FG 49.832 (41.052) 310.585*** (44.758) 241.323*** (42.668)
Detached (REF)
Apartment − 9.963 (41.907) − 72.374* (39.406)
Semi-detached − 19.210 (32.817) − 49.929 (31.025)
Terrace − 26.649 (35.015) − 63.286* (32.978)
1 Floor (REF)
2 Floors 491.419*** (32.992) 467.667*** (31.387)
3 Floors 794.670*** (46.279) 750.215*** (44.048)
4 Floors 508.383** (201.707) 443.997** (190.195)
Floor area ­(m2) 13.923*** (0.566) 13.342*** (0.540)
Year built − 0.374 (0.358) − 0.408 (0.336)
Living area percent − 2.558** (1.091) − 3.708*** (1.044)
Total Heating Degree Days 27.656*** (0.313)
Total rainfall (cm) 23.470*** (0.676)
Bimonthly Time Yes Yes
Constant 1613.435*** (23.329) − 805.289 (731.714) 2537.047*** (688.257)
N 149,518 149,518 149,518
r2 0.000 0.111 0.254
AIC 2,638,800 2,621,245 2,595,109
BIC 2,638,849 2,621,404 2,595,396

Asterisks note significance at the 10 percent (*), 5 percent (**), or 1 percent (***) level. Standard errors in brackets

13
Energy Efficiency (2021) 14: 57 Page 25 of 28 57

Table 21  Difference between annual actual (AQ) and theoretical (TQ) gas use ­(AARelative)
n Mean AQ Mean TQ Difference % Difference SE P value

AQAll – TQ 19,251 9452.773 13,148.1 − 3695.331 − 28% 64.591 0***

EPC grade
AB 2601 8979.353 7570.791 1408.562 19% 134.252 0***
C 8269 9279.777 10,825.76 − 1545.985 − 14% 77.938 0***
D 4835 9628.3 14,353.03 − 4724.732 − 33% 114.68 0***
E 2051 9990.076 18,132.59 − 8142.514 − 45% 189.71 0***
FG 1495 9928.487 24,961.51 − 15,000 − 60% 303.971 0***
Dwelling type
Apartment 1674 6945.365 11,595.13 − 4649.767 − 40% 177.632 0***
Detached 2316 12,413.54 19,384.96 − 6971.412 − 36% 259.594 0***
Semi-detached 6905 9990.711 14,008.02 − 4017.31 − 29% 105.639 0***
Terrace 8356 8689.943 11,019.98 − 2330.038 − 21% 86.446 0***
***
P < 0.01, **P < 0.05, *P < 0.10. units in kWh/year. Sample features 9923 observations of 1 year of actual GAS use and a further
9328 observations from the same sample of houses with a second year of actual GAS use. A test of equality of medians (using
signtest STATA command (Snedecor & Cochran, 1989), confirms the same differences exist

a retrofit prior to the start of our observed period of Appendix 7: Robustness check 3 — bimonthly
energy use. Results in the body of the paper report results using categorical EPC
values for the entire sample, being explicit in how
the sample excludes houses that change their dwell- In addition to the continuous version of the EPC, we
ing energy efficiency over time. This appendix repli- model the categorical version of the EPC to observe
cates those results, split by subsample, as a robustness associations (Table 20). This is performed in order
check, to show no major discrepancy exists between to consider the average effect of a one unit increase
houses in the sample (Tables 18 and 19). in actual energy use for a change in the EPC. This is

Table 22  Difference between annual actual (AQ) and theoretical (TQ) electricity use ­(AAAbsolute)
n Mean AQ Mean EPC Difference % Difference SE P value

AQAll – TQ 19,251 4134 13,148 − 9014 − 69% 53 0***

EPC grade
AB 2601 4232 7571 − 3339 − 44% 81 0***
C 8269 4321 10,826 − 6505 − 60% 51 0***
D 4835 4018 14,353 − 10,300 − 72% 81 0***
E 2051 3793 18,133 − 14,300 − 79% 147 0***
FG 1495 3777 24,962 − 21,200 − 85% 272 0***
Dwelling type
Apartment 1674 3198 11,595 − 8397 − 72% 126 0***
Detached 2316 4726 19,385 − 14,700 − 76% 222 0***
Semi-detached 6905 4257 14,008 − 9751 − 70% 82 0***
Terrace 8356 4056 11,020 − 6964 − 63% 67 0***
***
P < 0.01, **P < 0.05, *P < 0.10. Sample of 9923 homes with 9923 observations of 1 year of actual electricity use and a further
9328 observations for the same houses with a second year of observed actual electricity use. A test of equality of medians (using
signtest STATA command (Snedecor & Cochran, 1989), confirms the same differences exist

13
57 Page 26 of 28 Energy Efficiency (2021) 14: 57

appealing if we believe that occupants are aware of at the Irish Economic Association (IEA) 2019 conference for
their EPC but are inattentive to the specific unit value. their valuable feedback.
These results are consistent with those in the body of
Author contribution Dr. Bryan Coyne: Methodology, soft-
the paper using a continuous EPC and confirm that ware, formal analysis, data curation, writing — original draft,
less efficient EPCs are associated with increasing lev- visualization.
els of actual energy use. Prof. Eleanor Denny: Conceptualization, supervision, writ-
ing — review and editing, project administration, funding
acquisition.

Appendix: Robustness check 4 — fuel‑specific Funding Open Access funding provided by the IReL Con-
results sortium This work has emerged from research conducted with
the financial support of Science Foundation Ireland under the
Results in the body of this paper consider a measure SFI Strategic Partnership Programme Grant Number SFI/15/
SPP/E3125. The opinions, findings and conclusions or recom-
of actual energy use that includes electricity and gas, mendations expressed in this material are those of the author(s)
while deflating to account for appliance use. This is and do not necessarily reflect the views of the Science Founda-
designed to be comparable to the EPC. Many other tion Ireland.
studies of the EPG focus on the comparison between
gas energy use and the EPC (Aydin et al. 2017; Cozza Data availability I wish to highlight that the household-
level consumption data used for this paper was provided under
et al. 2020; Sunikka-blank & Galvin, 2012). This sec- a non-disclosure agreement with an energy utility and unfor-
tion presents annual results that perform a t-test of tunately cannot provided for submission in a repository or as
means for gas. If this shows similar results to Sec- an electronic resource. The dataset of household-level dwell-
tion Annual results, this suggests that the evidence on ing characteristics and expected energy efficiency is publicly
available from the Sustainable Energy Association of Ireland
the EPG observed is all due to a difference in gas con- (SEAI) online ([Link]
sumption from the theoretical EPC level. ter/​Regis​ter.​aspx).
Table 21 highlights that the average EPG is 28%
when only studying gas energy use. This is larger Declarations
than the average deficit in the main results for total
energy use (− 17%). Interestingly, a similar negative Conflict of interest The authors declare no competing inter-
ests.
relationship exists between the size of the deficit and
energy efficiency (+ 19% for AB, − 68% for FG). This Open Access This article is licensed under a Creative Com-
result suggests that estimates of the EPG that do not mons Attribution 4.0 International License, which permits
account for electricity use in the home may be over- use, sharing, adaptation, distribution and reproduction in any
stating the true EPG as they do not account for fuel medium or format, as long as you give appropriate credit to the
original author(s) and the source, provide a link to the Crea-
switching. This is especially the case for homes with tive Commons licence, and indicate if changes were made. The
the lowest energy efficiency. The sample average defi- images or other third party material in this article are included
cit of 28% is in line with estimates of rebound effects in the article’s Creative Commons licence, unless indicated
observed in other contexts (Sorrell et al. 2009). otherwise in a credit line to the material. If material is not
included in the article’s Creative Commons licence and your
Although Table 21 made a suitable comparison intended use is not permitted by statutory regulation or exceeds
between gas energy use and the EPC, Table 22 is diffi- the permitted use, you will need to obtain permission directly
cult to interpret because electricity use in the home was from the copyright holder. To view a copy of this licence, visit
never intended to be the sole determinant of the EPC [Link]
(which reflects space and water heating). As such, the
average deficits are large negative values. This makes
sense, as electricity is more commonly used as a second-
ary fuel in this sample, in association with gas heating. References

Acknowledgements The authors acknowledge the support Amecke, H. (2012). The impact of energy performance certifi-
of the Sustainable Energy Authority of Ireland and Electric cates : A survey of German home owners. Energy Policy,
Ireland in providing access to the data for this work. We also 46, 4–14. [Link]
thank Prof. John FitzGerald, Dr. James Carroll and participants Andaloro, A. P. F., Salomone, R., Ioppolo, G., & Andaloro, L.
(2010). Energy certification of buildings: A comparative

13
Energy Efficiency (2021) 14: 57 Page 27 of 28 57

analysis of progress towards implementation in European Delghust, M., Roelens, W., Tanghe, T., De Weerdt, Y., & Jans-
countries. Energy Policy, 38(10), 5840–5866. [Link] sens, A. (2015). Regulatory energy calculations versus
org/​10.​1016/j.​enpol.​2010.​05.​039 real energy use in high-performance houses. Building
Aydin, E., Kok, N., & Brounen, D. (2017). Energy Efficiency Research and Information, 43(6), 675–690. [Link]
and household behavior: The rebound effect in the resi- org/​10.​1080/​09613​218.​2015.​10338​74
dential sector. RAND Journal of Economics, 48(3), Di Cosmo, V., Lyons, S., & Nolan, A. (2014). Estimating the
749–782. impact of time-of-use pricing on irish electricity demand.
Better Buildings Partnership. (2019). Design for performance. The Energy Journal, 35(2), 117–136. [Link]
[Link] 5547/​01956​574.​35.2.6
BPIE. (2011). Europe’s buildings under the microscope. Dowson, M., Poole, A., Harrison, D., & Susman, G. (2012).
[Link] 9789491143014 Domestic UK retrofit challenge: Barriers, incentives
Burman, E., Mumovic, D., & Kimpian, J. (2014). Towards and current performance leading into the Green Deal.
measurement and verification of energy performance Energy Policy, 50, 294–305. [Link]
under the framework of the European directive for energy enpol.​2012.​07.​019
performance of buildings. Energy, 77, 153–163. [Link] European Commission. (2013). Energy performance certifi-
doi.​org/​10.​1016/j.​energy.​2014.​05.​102 cates in buildings and their impact on transaction prices
Carroll, J., Denny, E., & Lyons, S. (2016). The effects of and rents in selected EU countries. Brussels. [Link]
energy cost labelling on appliance purchasing decisions: ec.​e uropa.​e u/​e nergy/​s ites/​e ner/​f iles/​d ocum​e nts/​2 0130​
Trial results from Ireland. Journal of Consumer Policy, 619en​ergy_​perfo​r mance_​certi​ficat​es_​in_​build​ings.​pdf.
39(1), 23–40. [Link] Accessed 1 Feb 2019.
Central Statistics Office. (2017). Housing stock - CSO - Cen- European Commission. (2018). Directive 2018/844 on the
tral Statistics Office. [Link] energy performance of buildings. Official Journal of the
icati​ons/​ep/p-​cp1hii/​cp1hii/​hs/. Accessed 12 September European Union (Vol. 2018). [Link]
2017 legal-​c onte​n t/​E N/​T XT/?​u ri=​u rise​r v%​3 AOJ.​L _.​2 018.​
Collins, M., & Curtis, J. (2018). Bunching of residential build- 156.​01.​0075.​01.​ENG. Accessed 20 June 2020.
ing energy performance certificates at threshold values. European Commission. (2019a). In Focus: Energy perfor-
Applied Energy, 211(September 2017), 662–676. [Link] mance of buildings. [Link]
doi.​org/​10.​1016/j.​apene​rgy.​2017.​11.​077 focus-​energy-​effic​iency-​build​ings-​2020-​feb-​17_​en#:​~:​
Comerford, D. A., Lange, I., & Moro, M. (2018). Proof of con- text=​Today%​2C roughly 75%25 of the materials when
cept that requiring energy labels for dwellings can induce constructing new houses. Accessed 24 April 2019
retrofitting. Energy Economics (Vol. 69). Elsevier B.V. European Commission. (2019b). Clean energy for all Euro-
[Link] peans. Luxembourg. [Link]
013 European Parliament. (2018). Directive 2018/2002/EU
Coyne, B., Lyons, S., & McCoy, D. (2018). The effects of amending Directive 2012/27/EU on Energy Efficiency.
home energy efficiency upgrades on social housing ten- Official Journal of the European Union, 328(Novem-
ants: Evidence from Ireland. Energy Efficiency, 11(8), ber), 210–230.
2077–2100. [Link] European Union. (2002). Directive 2002/91/EC energy per-
Cozza, S., Chambers, J., & Patel, M. K. (2020). Measuring the formance of buildings. Official Journal of the European
thermal Energy Performance Gap of labelled residential Union, L1(65), 65–71. [Link]
buildings in Switzerland. Energy Policy, 137. [Link] 21001​96.
org/​10.​1016/j.​enpol.​2019.​111085 European Union. (2012). Directive 2012/27/EU - energy
CRU. (2017). Review of typical domestic consumption values efficiency directive. Official Journal of the European
for electricity and gas customers. [Link] Union, L.315/1. [Link]
conte​nt/​uploa​ds/​2017/​07/​CER17​042-​Review-​of-​Typic​al-​ EN/​T XT/​P DF/?​u ri=​C ELEX:​3 2012​L 0027​&​f rom=​E N.
Consu​mption-​Figur​es-​Decis​ion-​Paper-1.​pdf. Accessed 28 Accessed 12 September 2017
Oct 2020. Eurostat. (2019a). Energy consumption in households - Sta-
CRU. (2018). Customer switching report for the electricity and tistics Explained. Eurostat Statistics Explained. [Link]
gas retail markets. [Link] ec.​e uropa.​e u/​e uros​t at/​s tati​s tics-​expla​i ned/​i ndex.​p hp/​
ds/​2018/​08/​CRU18​177-​Change-​of-​Suppl​ier-​Update-​June-​ Energy_​consu​mption_​in_​house​holds. Accessed 28
2018.​pdf October 2020
DCCAE. (2017). Ireland’s fourth national energy efficiency Eurostat. (2019b). Housing statistics. Eurostat Statistics
action plan 2017–2020. Department of Communications, Explained. [Link]
Climate Action and the Environment. ined/​index.​php/​Housi​ng_​stati​stics#​Type_​of_​dwell​ing.
DCENR. (2009). Maximising Ireland’s energy efficiency action Accessed 28 October 2020
plan 2009 - 2020. [Link] Eurostat. (2020). Greenhouse gas emissions per capita. [Link]
0b3f-​48d3-​bd6e-​bb21f​e4c54​75.​pdf. Accessed 01 Nov ec. ​ e uropa. ​ e u/ ​ e uros​ t at/ ​ d atab ​ rowser/ ​ v iew/ ​ t 2020_ ​ rd300/​
2020. defau​lt/​table?​lang=​en. Accessed 28 October 2020
De Wilde, P. (2014). The gap between predicted and measured Fuerst, F., McAllister, P., Nanda, A., & Wyatt, P. (2016).
energy performance of buildings: A framework for inves- Energy performance ratings and house prices in Wales:
tigation. Automation in Construction, 41, 40–49. [Link] An empirical study. Energy Policy, 92, 20–33. [Link]
doi.​org/​10.​1016/j.​autcon.​2014.​02.​009 org/​10.​1016/j.​enpol.​2016.​01.​024

13
57 Page 28 of 28 Energy Efficiency (2021) 14: 57

Government of Ireland. (2019). Climate Action Plan to reduce the performance gap. Applied Energy, 97, 355–
2019. Government Buildings,Dublin. [Link] 364. [Link]
ie/​en/​publi​cation/​ccb2e0-​the-​clima​te-​action-​plan-​2019/. Owen, P., & Foreman, R. (2012). Powering the nation: House-
Accessed 17 June 2019. hold electricity-using habits revealed. Energy Saving
Gram-Hanssen, K., & Georg, S. (2018). Energy performance Trust / DECC / DEFRA. London. [Link]
gaps: promises, people, practices. Building Research and ngtru​st.​org.​uk/​policy-​resea​rch/​power​ingna​tion-​house​hold-​
Information, 46(1), 1–9. [Link] elect​ricity-​using-​habits-​revea​led. Accessed 15 July 2019.
218.​2017.​13561​27 Quayle, R. G., & Diaz, H. F. (1979). Heating degree day data
Harold, J., Lyons, S., & Cullinan, J. (2018). Heterogeneity and applied to residential heating energy consumption. Journal
persistence in the effect of demand side management stim- of Applied Meteorology, 19(3), 241–246. [Link]
uli on residential gas consumption. Energy Economics, 1175/​1520-​0450(1980)​019%​3c0241:​HDDDAT%​3e2.0.​CO;2
73, 135–145. [Link] Scheer, J., Clancy, M., & Hógáin, S. N. (2013). Quantifica-
Hart, M., & De Dear, R. (2004). Weather sensitivity in household tion of energy savings from Ireland’s home energy saving
appliance energy end-use. Energy and Buildings, 36(2), scheme: An ex post billing analysis. Energy Efficiency,
161–174. [Link] 6(1), 35–48. [Link]
Heesen, F., & Madlener, R. (2018). Consumer behavior in SEAI. (2012). Dwelling energy assessment procedure ( DEAP)
energy-efficient homes: The limited merits of energy perfor- Manual Version 3.2.1. [Link]
mance ratings as benchmarks. Energy and Buildings, 172, DEAP_​Manual.​pdf. Accessed 05 Jan 2019.
405–413. [Link] SEAI. (2014). A guide to building energy rating for home-
Herrando, M., Cambra, D., Navarro, M., De, L., Millán, G., & owners. Dublin. [Link]
Zabalza, I. (2016). Energy performance certification of Your_​Guide_​to_​Build​ing_​Energy_​Rating.​pdf. Accessed
faculty buildings in Spain: The gap between estimated and 1 Sept 2019.
real energy consumption. Energy Conversion and Man- SEAI. (2016). Energy in Ireland 1990 – 2015
agement Journal, 125, 141–153. [Link] SEAI. (2018). Energy in the residential sector: 2018
encon​man.​2016.​04.​037 Report. Dublin. [Link]
Hyland, M., Alberini, A., & Lyons, S. (2016). The effect of Publi​catio​ns/​EPSSU_​Publi​catio​ns/. Accessed 31 Oct 2019.
energy efficiency labeling: Bunching and prices in the Irish Snedecor, G. W., & Cochran, W. G. (1989). Statistical methods
residential property market. Trinity Economics Papers (8th ed.). Iowa State Univ Press.
Working Paper 0516, (0516). [Link] Sorrell, S., Dimitropoulos, J., & Sommerville, M. (2009).
mics/​TEP/​2016/​tep05​16.​pdf. Accessed 18 Jan 2019. Empirical estimates of the direct rebound effect: A review.
Hyland, M., Lyons, R. C., & Lyons, S. (2013). The value of Energy Policy, 37(4), 1356–1371. [Link]
domestic building energy efficiency — evidence from Ire- 1016/j.​enpol.​2008.​11.​026
land. Energy Economics, 40, 943–952. [Link] Sunikka-blank, M., & Galvin, R. (2012). Introducing the
1016/j.​eneco.​2013.​07.​020 prebound effect- the gap between performance and actual
Jaffe, A. B., & Stavins, R. N. (1994). The energy-efficiency energy consumption. Building Research and Information,
gap: What does it mean? Energy Policy, 22(10), 804–810. 40(3), 260–273.
[Link] van den Brom, P., Meijer, A., & Visscher, H. (2018). Perfor-
Kavousian, A., Rajagopal, R., & Fischer, M. (2013). Determi- mance gaps in energy consumption: household groups
nants of residential electricity consumption: Using smart and building characteristics. Building Research & Infor-
meter data to examine the effect of climate, building mation, 46(1), 54–70. [Link]
characteristics, appliance stock, and occupants’ behavior. 2017.​13128​97
Energy, 55, 184–194. [Link] van Dronkelaar, C., Dowson, M., Spataru, C., & Mumovic, D.
2013.​03.​086 (2016). A review of the regulatory energy performance
Majcen, D., Itard, L. C. M., & Visscher, H. (2013). Theoreti- gap and its underlying causes in non-domestic buildings.
cal vs. actual energy consumption of labelled dwellings in Frontiers in Mechanical Engineering, 1(January), 1–14.
the Netherlands : Discrepancies and policy implications. [Link]
Energy Policy, 54, 125–136. [Link] Webber, P., Gouldson, A., & Kerr, N. (2015). The impacts
enpol.​2012.​11.​008 of household retrofit and domestic energy efficiency
McCord, M., Haran, M., Davis, P., McCord, J. (2020). Energy schemes: A large scale, ex post evaluation. Energy Policy,
performance certificates and house prices: A quantile 84, 35–43. [Link]
regression approach. Journal of European Real Estate Zou, P. X. W., Xu, X., Sanjayan, J., & Wang, J. (2018). Review
Research. [Link] of 10 years research on building energy performance
McKinsey. (2010). Energy efficiency : A compelling global gap: Life-cycle and stakeholder perspectives. Energy and
resource. McKinsey Sustainability & Resource Produc- Buildings, 178, 165–181. [Link]
tivity. [Link] ld.​2018.​08.​040
dotcom/​client_​servi​ce/​susta​inabi​lity/​pdfs/a_​compe​lling_​
global_​resou​rce.​ashx. Accessed 20 Dec 2019. Publisher’s note Springer Nature remains neutral with regard
Menezes, A. C., Cripps, A., Bouchlaghem, D., & Buswell, R. to jurisdictional claims in published maps and institutional
(2012). Predicted vs. actual energy performance of non- affiliations.
domestic buildings: Using post-occupancy evaluation data

13