Household Energy Consumption
classification of Greater London
David Goulvent
Submitted on the 15th of September 2012
Number of words: 9820
MSc Birkbeck University 2012
1
Abstract
Reducing energy consumption is the next frontier in the combat against climate
change. The UK, as a member of the European Union has a non-binding target of
reducing energy use by 20% by 2020, with energy efficiency at the cornerstone of
the government’s strategy.
But with increased pressure from the European Commission to move to mandatory
targets and current energy policies not delivering to their full potential, improved
targeting of energy efficiency measures at household level is key to successful
policy.
London alone, home to some 8 million people, is by far the highest number of hardto-treat properties, as far as energy efficiency improvements are concerned. The
capital has nonetheless lost out under previous carbon schemes due to the higher
cost of increasing building efficiency. Its intrinsic urban and cosmopolitan
characteristics have been diluted in national energy related classifications.
This study creates a London classification for household energy consumption build
on geodemographic principles. This classification can enable energy behaviour
profiling at a regional (Greater London) level to serve as a basis for improved
targeting. Seven distinct clusters emerged as a result of this analysis ranging from
the energy-intensive wealthy owners to the fuel-poor social renters.
The study has been carried out within the open-source philosophy. The analytical
process was based on public data and we have made the first step towards making
the outcome of this study publically downloadable on the web. Our final offline
visualisation of the classification is presented on a map following the Charles Booth
tradition, infamously tied to the very roots of geodemographic-based
classifications.
2
Acknowledgements
3
Table of contents
Abstract………………………………………………………………………………………………………………2
Acknowledgements……………………………………………………………………………………………..3
Table of contents…………………………………………………………………………………………………4
Tables, Figures and maps…………………………………………………………………………………….7
Declaration………………………………………………………………………………………………………….8
Introduction………………………………………………………………………………………………………..9
I. Overview………………………………………………………………………………………………………..15
1.1 Why Classify?................................................................................................15
1.2 What are Area Classifications and Geodemographics?...................................15
1.3 Beyond definitions……………………………………………………………………………………….17
1.4 Cluster analysis…………………………………………………………………………………………….18
1.5 Energy consumption and geodemographics…………………………………………………18
1.6 Energy efficiency and policy…………………………………………………………………………19
II. Method and data…………………………………………………………………………………………..20
2.1 Cluster analysis……………………………………………………………………………………….……20
2.1.1 Step 1: Data selection process…………………………..………………………………….…..21
2.1.1.1 Variables selection………………………………………………………………………………...21
2.1.1.2 Scale……………………………………………………………………………………………………...22
2.1.1.3 Demographic variables…………………………………………………………………………..23
2.1.1.4 House composition variables………………………………………………………………….23
2.1.1.5 Housing variables………………………………………………………………………………..…24
2.1.1.6 Socio-economic and employment variables……………………………………………25
2.1.1.7 Energy consumption variables………………………………………………………………..26
2.1.2 Step 2: Normalisation………………………………………………….……………………………28
4
2.1.2.1 Logarithmic transformation……………………………………………………………………28
2.1.2.2 Standardisation………………………………………………………………………………………28
2.1.3 Steps 3: Clustering…………………………………………………………………………………….29
III. Application, results and discussion…………………………………………..…………………..29
3.1 Application…………………………………………………………………………………………………..29
3.1.1 Running the K-means………………………………………………………………………………..29
3.1.2 Selecting the appropriate number of clusters…………………………………………...30
3.2 Analysing and mapping the new geodemographic clusters…………………………..32
3.2.1 Step 4: The clusters and their descriptions………………………………………………..32
3.2.1.1 Cluster 1: Electricity-intensive city renters………………………………………………33
3.2.1.2 Cluster 2: Fuel poor social renters…………………………………………………………..35
3.2.1.3 Cluster 3: Energy-intensive wealthy greys………………………………………………37
3.2.1.4 Cluster 4: Average consuming London renters……………..………………………..39
3.2.1.5 Cluster 5: Wealthy energy intensive owners……………………..…………………..41
3.2.1.6 Cluster 6: Average use suburban working families………………………………..43
3.2.1.7 Cluster 7: Low consuming strained renters……………………………………………44
3.3 Visualisation………………………………………………………………………………………………..46
3.4 Limits and extensions……………………………….………………………………………………….50
3.4.1 Ecological Fallacy………………………………………………………………………………………50
3.4.2 The Modifiable Areal Unit Problem………….……………………………………………….50
3.4.3 Dating data……………………………………………………………………………………………….51
3.4.4 Missing data…………………………………………..……….………………………………………..51
5
3.4.5 The Labelling of Clusters…………………………..……….………………………………………52
3.4.6 Fuzzy classification?....................................................................................52
3.4.7 Hierarchical classification………………….………………………………………………………53
3.4.8 Classification validation…………………….………………………………………………………54
Conclusions………………………………………….……………………………….……………………..……54
References………………………………..………………………………………………………………………56
Data sources……………………………………………..………………………………………………………60
Annexes…………………………………………………………………………………………………………….61
6
Tables, Figures and maps
Table 1: List of the 31 variables selected for input to the classification
Table 2: Homogeneity of the cluster membership size
Table 3: Number of LSOAs in each cluster
Figures
Figure 1: Average distance from the cluster centre by number of clusters
Figure 2: Summary of cluster 1
Figure 3: Summary of cluster 2
Figure 4: Summary of cluster 3
Figure 5: Summary of cluster 4
Figure 6: Summary of cluster 5
Figure 7: Summary of cluster 6
Figure 8: Summary of cluster 7
Figure 9: Visualisation of the building layer in Google Earth
Maps
Map 1: Study area localisation
Map 2: Localisation of LSOAs in cluster 1
Map 3: Localisation of LSOAs in cluster 2
Map 4: Localisation of LSOAs in cluster 3
Map 5: Localisation of LSOAs in cluster 4
Map 6: Localisation of LSOAs in cluster 5
Map 7: Localisation of LSOAs in cluster 6
Map 8: Localisation of LSOAs in cluster 7
7
Map 9: Visualisation of household energy consumption classification in
choropleth map
Map 10: Visualisation of household energy consumption classification in Charles
Booth’s style
8
Declaration
I have read and understood the section of the handbook that explains plagiarism,
including that related to group work. I testify that, unless otherwise acknowledged,
the work submitted herein is entirely my own.
This dissertation is my own unaided work and has not been submitted for a further
degree at any other Higher Education Institution. It does not exceed the word limit
of 10,000 words.
9
Household Energy Consumption classification of
Greater London
Introduction
According to the Department of Energy and Climate Change (DECC, 2011),
energy consumption from the domestic sector in 2011 was 38, 842 thousand
tonnes of oil equivalent, about 26% of total UK final consumption of energy
products. A quarter of the UK’s carbon emissions come from the energy used in
homes (The Energy Trust 2012). While most of the UK and the EU’s binding targets
around the reduction of carbon emissions apply to businesses and industries, there
is increasing pressure to cast the target net wider and implement change at the
household level through a cut in residential energy use. The EU has a non-binding
target to cut energy consumption by 20% by 2020 with energy efficiency the main
driver of that reduction (European Commission, 2008).
Although this target is not yet binding, there is growing consensus within the
European Commission that the only way to see Member States meet the target is to
make it binding. The UK’s Energy Act 2011 already includes energy efficiency
policies – with the forthcoming Green Deal set to soft-launch later this year. The
challenge remains their reach and implementation within the different types of
households, whether that is geodemographics or building types for example.
While current UK energy efficiency policies are more actionable at a new building
level and some progress has been made already around insulation with DECC
reporting an increase of 6% of cavity wall insulation, 9% of loft insulation and 22%
of solid wall insulation, sizable gains are to be made at the existing household level
(DECC, a, 2011).
10
With the threat of binding EU targets looming, combined with rising wholesale
energy prices, energy efficiency or what we will define as using less energy to
provide the same level of performance, comfort, and convenience, is firmly back on
the agenda. But while there are tangible technological solutions available (smart
metering, insulation, voltage optimisation, green grants) the uptake by different
types of households remains a challenge, confined in some cases to a certain
profile.
Visualising relatively homogenous groups of energy efficiency household types at a
London Lower Super Output Area (LSOA1) level through a classification method
could enable a better understanding of household behaviour and energy
characteristics, which in turn could lead to improved policy targeting.
“Area classifications provide a unique way of bringing together a real pattern for a
range of variables” (Vickers and Rees, 2007; Webber and Craig, 1978).
Geodemographic classification system has been widely used now by geographers,
policy makers and market researchers, to underline the neighbourhood effect and
help to classify geographic areas according to the characteristics of people living
there, based on the principle that people living close to each other tend to be more
similar than to people living further away.
Such a classification would also help energy suppliers – increasing at the forefront
of energy efficiency measures due to government policies – to guide their customer
base into reducing their energy use and subsequently carbon emissions.
“Geodemographic profiling also presents the opportunity to achieve savings by
targeting communication programmes at populations to whom their messages are
most appropriate.” (Longley, 2005) and is therefore a key method for government
to encourage uptake of new energy saving policies.
A large amount of studies have been published in recent years on energy
consumption, carbon emissions and the implication on climate change. However,
there still a lack of analysis linking these issues with geodemographic factors, as we
will explain in our literary review.
1
The average population of an LSOA in London in 2010 was 1,642 (http://data.london.gov.uk/)
11
This study will be limited to Greater London. The capital has by far the highest
number of needy properties of any region in the UK. According to the London
Greater Authority, London has lost out under previous carbon reducing schemes
due to higher costs of increasing building energy efficiency. According to a recent
briefing by the Mayor of London on the upcoming Green Deal: “This is because
energy companies have previously fulfilled their obligations wherever it most costeffective to do so, without regard to the potential to reduce comparatively high
levels of fuel poverty in London and carbon emissions from some hard-to-reach
housing stock” (Mayor of London, July 20122).
There is already a fear that the forthcoming national Green Deal, including the new
Energy Company Obligation (ECO), which will put the burden on energy companies,
expected to provide £1.3 billion a year towards the Green Deal for low income and
hard to insulate homes, will not deliver the required results in London. The danger
being that as under previous schemes energy companies focus on treating
properties and areas that are cheaper and easier to retrofit (Mayor of London, July
2012). Around 22% of hard-to-treat properties in England are in London.
The London Mayor briefing states that Londoners in flats and mid-terraces could be
excluded from the Green Deal, as housing stock will be unable to access a high
enough subsidy from the ECO to make the scheme viable. There is a call for area
allocation for London under ECO to take into account its unique social, architectural
and economic traits. A London energy classification could enable this.
2
The London Mayor briefing in annex
12
Map 1: Study area localisation
Hypothesis, aims and objectives
Our main hypothesis will be to assess the possibility of energy behaviour profiling at
an LSOA level for London through the creation of a household energy consumption
classification for Greater London. Will the groups created from our new
classification be distinct enough to enable improved targeting and uptake of energy
13
efficiency measures? To what extent can an open source classification of
households from a socio-economic and energy consumption perspective accelerate
decision making and actionable targets around energy efficiency?
The null hypothesis would be that such a classification does not bring anything to
the current academic energy efficiency and consumption debate.
The aim of this study is not however to match existing energy reducing policies with
different London profiles. Certain energy efficiency measures will be suggested
following the description of the clusters to show how this type of classification can
enable better targeting. An analysis of existing measures and their uptake as a
result of this classification would be the subject of another study.
As far as energy efficiency is concerned, it is not financially viable to replace older,
low energy efficient households and replace with new, more energy efficiency
buildings. So policy makers are also looking at improving older houses. Knowing
where the least energy efficient houses are and what type of people live in them
can feed the decision making over how to tackle the issue and determine who
would be responsible for improvements
Enabling and visualising of socio-economic profiling of the London population at a
LSOA level depending on a defined set of energy household characteristics can help
deliver the right campaigns and methods to see energy efficiency actions take off.
Finally through the presentation of the final maps of this research project, we will
insist on the advantages of exploratory visualization methods to help households,
policy makers and energy organisation to actually see energy efficiency in London in
a more intuitive way.
After presenting the academic foundations on which this study will be built and
outlining its relevance within the current debate, we will define our method and
data selection and analysis. The next step will contain the description of our
clusters, followed by insight into the limits of this study. Finally we will conclude
and establish whether the objectives of this study were met.
14
I. Overview
This section will define the principle concepts and theories on which this study is
based, notably classification and geodemographics. It will also situate this study
within the context of existing research, as well as touching on its relevance.
1.1 Why Classify?
According to Everitt, “a classification scheme may simply represent a convenient
method for organizing a large data set so that it can be more easily understood and
information retrieved more efficiently” (Everitt et al., 2010). Classification enables a
greater understanding of socio-economic characteristics and behaviors as it allows
for the grouping or clustering of elements (people, buildings, areas etc.) with similar
traits. One of the most well-known classification undertakings in the UK is the
Output Area classification (OAC) created in collaboration between the Office for
National Statistics (ONS) and the University of Leeds. According to the OAC user
group: “the OAC distils key results from the Census for the whole of the UK to
indicate the character of local areas” (The OAC User Group3).
Vickers, who was tasked with the creation of the OAC in 2005, alongside Rees,
created a hierarchy of seven, 21 or 52 classes from 41 Census 2001 variables in
order to portray the main characteristic of the household in UK. Vickers explains:
“[…] Clear distinctions can be made between neighbourhoods, for example on the
basis of affluence, rurality or multiculturalism. The classification can answer many
questions about the residential patterns of the UK at the start of the 21st century”
(Vickers’ at the University of Leeds; School of Geography4).
1.2 What are Area Classifications and Geodemographics?
Area classification is the classifying of areas into groups of similarity, based on the
characteristics of selected features within them (Everitt et al., 2001). In our study,
3
http://areaclassification.org.uk/
4
http://www.sheffield.ac.uk/geography/staff/vickers_dan/index
15
area classification will refer to the segmentation of LSOA in London, based on sociodemographic and energy consumption data.
We now have to define geodemographics. In the 1980s, geodemographic analysis
through “neighbourhood classification became linked to the marketing roles of
leading commercial organizations, particularly retailers” (Longley, 2005) while
neighbourhood classification was obsolete in academic and public sectors.
Geodemographics is not just a set of off- the-shelf consumer targeting products it
is: “the analysis of people by where they live” (Sleight 2004; Vickers and Rees,
2007). To Sleight’s rather concise and stripped-down definition, Birkin and Clarke
1998 add: “demography is the study of population types and their dynamics
therefore geodemographics may be labelled as the study of population types and
their dynamics as they vary by geographical area”. (Birkin and Clarke, 1998; Vickers
and Rees, 2006) Geodemographic is based on the relation between the people and
the place they live. Knowing about where somebody lives can reveal a lot of
information about that person (Vickers et al., 2005).
The concept of geodemographics is based on the Tobler’s First Law of Geography:
“Everything is related to everything else, but near things are more related than
those far apart” (Tobler 1970). Vickers et al. (2005) introduces another dimension
to this law, which will be essential within this study of an urban area such as
London. While those living closer to each other tend to be more similar, in urban
areas for example, some neighbourhoods, which are at opposite sides of a large
city, will still be have similar characteristics due to their positioning with regards to
the urban centre. In London for example, two suburban neighbourhoods at
opposite locations from the centre, could still present the same characteristics, as
some of our clusters will show.
We will retain the use of the term neighbourhood to define the different areas of
our study. The term neighbourhood enables conceptualisation of the area in which
you live (Vickers and Rees, 2007). Singleton adds that the definition of
“neighbourhood in a geodemographic sense is determined by the size of the
16
geographic areas used to output the classification” (Alex Singleton, 2007). The term
neighbourhood in our study will be the LSOA.
1.3 Beyond definitions…
Having given the relevant definitions around the notion of classification, let us step
back to pin point the origin of the movement and reiterate some of its first usages.
There is a consensus that Charles Booth and his studies of deprivation and poverty
of London between 1889 and 1899 are at the origin of the movement. Charles and
his researchers went all around London visiting all households to interview the
inhabitants. They then characterised them into seven categories, such as “Well-todo” or more derogatory…“Vicious, semi-criminal”.
Since the publication of 1981 UK Census of Population results, geodemographics
classifications have occupied a large position in the private sector, and “by the
middle of the decade four main systems were competing for dominance, ACORN,
PiN, Mosaic and Super Profiles” (Vickers et al., 2005).
Even though these commercial companies always used Census data, they added
“other type of data from Country court judgements, credit reference agencies,
vehicle registrations, and lifestyle surveys” (Harris et al., 2005). The main issue with
the commercial geodemographic classifications is they are created as ‘black -box’
systems (Longley and Singleton, 2009), with little transparency around the raw data
and methods used to elaborate their classifications.
Geodemographic classifications are still heavily used in the marketing industry.
(Vickers et al, 2005). They allow for example to determine the customer archetype
and their location for new products or estimate opinion polls. As Gale et al. (2012)
note, “the use of geodemographic classifications has become popular in different
areas with applications in health (Farr and Evans, 2005; Shelton et al., 2006),
policing (Ashby and Longley 2005), education (Singleton, 2010) and local
government (Longley and Singleton 2009)”. The use of geodemographics in a public
health setting can help for service delivery planning or targeting campaign.
17
1.4 Cluster analysis
We will base our geodemographic classification for this study on cluster analysis.
“Cluster analysis is a generic term for a wide range of numerical methods for
examining multivariate data with a view to uncovering or discovering groups or
clusters of homogeneous observations” (Everitt et al., 2001). It is a human
automatism to group similar objects in into categories. Objects in the groups or
clusters are supposed to have similar characteristics within the same cluster but are
dissimilar to the objects in other clusters. A detailed description of the steps for a
cluster analysis will be presented later, in the method section of this study.
1.5 Energy consumption and geodemographics
While the creation of the open source geodemographics system OAC by the ONS
has been an important step towards better understanding of socio-economic traits
based on location, the question over whether it was pertinent for a large city like
London was raised ahead of the latest Census. A new classification for London was
introduced for the latest 2011 Census to better capture the specificities of the UK’s
largest city.
Already in 2007, Petersen et al. created a new regional geodemographics for
London for health applications. At the time, they said “the National OAC did not
represent the variation measured as market penetration potential across the 41
Census variables in Greater London very well”. Gale and Longley assessed the OAC
2001 in order to improve the new coming OAC 2011, they underlined the fact
“London's economic, political, cultural and infrastructural characteristics set it apart
from the rest of the United Kingdom and to a large extent the rest of the world”
(Petersen et al., 2010; Gale and Longley, 2012). They insisted that the national
classification such as OAC 2001 wasn’t defining properly the diversity of London.
DECC analysis shows London as having one of the regions with the highest domestic
consumptions of gas and electricity in the UK. However, and yet again highlighting
the benefit of using a regional classification for London, it is important to notice
here that it doesn’t mean households in London are consuming more than the rest
18
of the county. In fact, energy consumption per capita in London is relatively low. To
understand and consider the specificity of regional areas within a national territory,
Peterson et al. (2010) propose to create a regional classification to co-exist
alongside a national classification. Considering the scale of this work and the
particularities of London, this study will only focus on Greater London.
A large amount of studies have been published in recent years on energy
consumption, carbon emissions and the implication on climate change. However,
there still a lack of analysis linking these issues with geodemographic factors. There
are studies showing a relation between energy consumption and income, such as
the one undertaken by Druckman and Jackson (2008) demonstrating a link between
energy consumption, carbon footprint and environmental awareness and socioeconomics elements. They are limited in their approach as they focus mainly on
correlations. In Drukman’s paper for example, he uses the two extreme opposite
groups from the OAC, which we know is not sufficient enough for a pertinent
London analysis. In this study will go beyond correlations to create London’s own
energy classification.
An outcome similar to the one we will attempt to produce here in terms of
classification was reached by Experian in 2008 when they created a carbon
emissions classification called GreenAware: “to identify priority areas and
households and to map regional variations in behaviour and attitudes toward
carbon reduction” (Experian 2008). However, as is the case with some other of their
classifications on geodemographic, the full range of data used and the method to
create this classification are obscure and not entirely presented… the infamous
black-box system.
There is currently a PHD being undertaken by Sarah Goodwin (UCL) with a similar
objective to the one presented here but it has yet to be published.
1.6 Energy efficiency and policy
The UK’s residential energy efficiency policy is changing. From October 2012, the
19
government’s efforts to improve energy efficiency will be delivered through the
Green Deal and the new Energy Company Obligation (ECO). Soft-launching next
month with the bulk of residential policies kick starting in January 2013, the Green
Deal and ECO will jump start a number of new range of measures centered around
low upfront cost for efficiency household improvements.
As member of the EU, the UK has a non-binding target to reduce energy
consumption by 20% by 2020 and although it is appears to be quietly shying away
from Brussels wish to make this target mandatory the UK has nonetheless made
progress over recent years in sectors such as transport, industry and buildings.
London alone has a large role to play in meeting targets, housing around 8 million
people and home to 22% of known, heard-to-treat properties. The Mayor has set a
target to cut carbon emission from the capital by 60% by 2025 with 35% of London
emissions coming from the households (Greater London Authority, 2012). While
national policies have a role to play in London’s aim to meet its carbon and energy
efficiency targets, the characteristics of such a dense urban area like London has led
to the creation of London only measures or more recently, calls to tweak existing or
upcoming measures to ensure London and its specific traits does not miss out on
crucial funding.
The uptake of energy efficiency measures is the main challenge the government
and local authorities face. The barriers that can prevent uptake are divers:
behaviour and motivation, financial, misaligned incentives and hidden costs. A
London classification could help mitigate some of these challenges through a more
thorough understanding of target areas and households.
II. Method and Data
2.1 Cluster analysis
Geodemographic classifications are performed following various steps: The
following outline shows the succession of the steps and elements to keep in mind
when undertaking geodemographic classification.
20
The first step is to select the data sources and variables. Then comes the
normalisation of the input data to enable comparison between different data types.
There are various methods used to normalise data and we will explain later which
one we used and the reasons behind our choice. The third step is clustering. As with
the normalisation of data, there are various methods used to undertake cluster
analysis, each with their advantages and limits. The last step, which will be
presented in the result part of our study, consists of interpreting the results by
naming the clusters or pen portrait[s] and describing their main characteristics.
2.1.1 Step 1: Data selection process
2.1.1.1 Variables selection
Data selection has been at the core of the debate surrounding the geodemographic
system. While most of the geodemographic classifications are based on the Census
data, others also combine data from commercial surveys such as Experian or other
public data. “Some of the geodemographic companies may have added non-Census
variables (credit ratings, county court judgements) but the impact of such additions
upon the classification systems is unclear” (James Debenham, 2002). For this study,
we will use Census data 2001 and energy consumption figures from DECC (2009).
There are several reasons to use only Census data as well as free publically available
data. The main one is to ensure that the final classification is open-source. Gale et
al. (2012) explain the need of open, transparent and flexible geodemographic
classifications and we will keep within this philosophy in this current analysis.
The pursuit of open-source output is increasingly facilitated by the current
expansion of open data initiatives that have resulted in an ever-increasing amount
of data sources becoming available to the public (Gale et al., 2012).
The main advantage of making it open-source, is the possibility of being able to
freely publish the results and maps on the web so that third parties, either
21
governmental organisation, private company but also households can use it. Later,
we will determine whether this aim has been reached.
Beyond the wish to conduct a fully open-source and freely available study, the use
of publically available data stems from the issues around accessibility of other types
of data. Including data from lifestyles databases derived from consumers’ survey
can be costly and legally difficult to publish in a raw state once used. Data collected
from the private sector also has the inconvenience of not being representative
either geographically or demographically. As Vickers et al. in 2003 explained, these
data can be transferred in other spatial scale but none of the method available
today has been proven as strong enough to secure a satisfactory level of accuracy.
With OAC, we noted a lack of variable on financial incomes. Vickers and Rees
assume that census variables such as car ownership and type of housing which
provide a good proxy for income (Vickers and Rees, 2006). We did not select the car
owner variable here but we will in the description of the variable selected explain
how we aim to overcome this missing information.
We first had to consider all the data available and find a way to condense them to a
level that would give us enough information on our subject without clouding the
analysis with unnecessary details. As Openshaw and Blake underline, the choice of
variables and their specification has to reflect the explicit purpose (Blake &
Openshaw, 1995) and must be able to inform us on the household general
characteristics, as well as explain their energy consumption. The selection of data
has been influenced by the creation of the OAC as a benchmark of geodemographic
classification. As well as personal judgement, in some cases we looked at whether
certain variables were highly correlated. However, this does not mean we excluded
all correlated variables.
2.1.1.2 Scale
Our choice to use LSOA as the primary scale for this study has been driven by data
availability. While the Census 2001 data were available at Output Area (OA) level -
22
the smallest geographical unit for Census data, energy consumption data were only
available at LSOA. For this study, we chose the following types of data:
demographic structure, household composition, housing, socio-economic and
energy consumption.
2.1.1.3 Demographic variables
As for the OAC, gender was rejected as most of the areas have similar gender
proportion. With regards to age, we took the same age range as for the OAC. They
grouped Census categories together to avoid data variable redundancy. We merged
the age variables to have only 4 variables representing 5 to 14 year olds, the 25 to
44, the 45 to 64 and the 65 and over. We did not include the 15 to 24 year olds as
this group tended to be prone to changes in residential circumstance. We did
however include the student variable, which was correlated to this age group. In
term of demographic data, we also included the number of persons per hectare, a
measure of population density, as it seems as a good indicator on the type of urban
area and puts high-energy consumption areas into perspective.
2.1.1.4 House composition variables
“Household composition is, as expected, another significant factor in domestic
energy use and associated carbon emissions,” (Druckman and Jackson, 2008). We
have selected the following variables: two adults with children and two adults
without children. They both represent the aggregation of cohabiting and married
couples. Households with non-dependent children were rejected, as they made up
only a small proportion of households. We included one person in the household,
one-person pensioner and lone-parent households, as there is an increasing
tendency towards the one-person household. The ONS noted that the proportion of
lone-parent families, and subsequent smaller family sizes had been increasing since
1991 (ONS, a 2011). We chose not to include students as a variable at this stage,
23
under house composition, as we have selected a student related variable under our
socio-economic data.
2.1.1.5 Housing variables
As well as being related to energy consumption, housing characteristics are also a
good indicator of the household wealth. For this study, we included owned, private
and socially rented households. This type of knowledge would enable better
targeting of energy efficiency measures. As highlighted by Druckman and Jackson,
the issue of uptake of energy efficiency measures in rented accommodation
remains current (Druckman and Jackson, 2008). They underline the high proportion
of dwellings with less loft insulation in private rented sector, as it is difficult to
convince landlords to make any improvement to their houses, as they have to pay
for it without directly benefit from the energy efficiency and financial gains. As we
mention later in this study, a new policy ECO, under the forthcoming Green Deal, is
looking to address how to increase energy efficiency in rented housing stock.
Inefficient households will no longer be fit to rent come 2018.
‘House type’ refers to whether dwellings are semi‐detached, terraced houses,
detached houses or flats. “This is significant in energy terms because heating energy
is related to external wall area and window area. Flats tend to have less external
wall area compared to their floor area (so have less heat loss in winter), while
detached houses typically have more external wall and more windows than
equivalent homes of other types,” (DECC, b, 2011). The average heat losses of
different types of dwellings range from 365W/°C for a detached house down to 182
W/°C for a flat (Druckman and Jackson, 2008). We will also include data showing
the average number of rooms per household and average number of people per
room. These two variables are strong indications of the level of wealth, as well as
having an impact on energy consumption.
24
2.1.1.6 Socio-economic and employment variables
Selecting variables to best understand social and economic status of households
can prove to be a more subtle exercise. Based on previous geodemographic
classification, we considered variables on the occupation group, economic activity
or The National Statistics Socio-economic Classification (NS-SEC). The number of
variables on the type of occupation was relatively high. We could have aggregated
certain occupations as they were highly correlated but we would still have had at
least 5 groups of occupation that could create confusion in the interpretation. As
for the economic activity variables, they lacked insight around the full time
employee. People under the variables descriptive full time employee can have
radically different statuses and therefore incomes. “The NS-SEC is the primary social
classification in the United Kingdom” (Wikipedia 2012). The full version of NS-SEC
has 17 main categories and is collapsible down to three categories. The advantage
on using the NS-SEC is that “it has been constructed to measure the employment
relations and conditions of occupations” (ONS, b, 2012). It groups jobs of similar
social and economic status in classes using the occupation and employment type
data and therefore can be seen as a condensed result of the association of these
two set of data.
There are 3 versions of NS-SEC: eight classes; five classes or three classes. For our
study we will use the three-class version, which “is the only representation […] that
might be assumed to involve some kind of hierarchy” (Rose et al 1998). We add an
additional class Never-worked and long-term unemployed. We also include the
variable full time student and retired. We choose at this stage to include the
variable retired despite having also already chosen one-pensioner household under
house composition, which in this case and as we will see in the cluster composition,
more characteristic of financially strained as opposed to retired , which we will see
appeared with other more affluent characteristics. Not including the variable
retired at this stage would have omitted all households characterised by retired
couples. The industry sector variables, including agriculture, fishing, mining were
not representative enough of London characteristics as some sectors were quasi
inexistent in London.
25
We also included the percentage of not qualified people and the percentage of
educated people to degree level or more. Even though not qualified people are
correlated to other variables such as social rented housing, education
characteristics allow understanding and insight around how to transmit new
policies. Education is also correlated with green awareness and therefore an
important element to consider when deciding on what approach to take for
different groups.
2.1.1.7 Energy consumption variables
DECC has collated and analysed property level electricity and gas consumption data
since 2004. “These administrative datasets provide total and average consumption
of domestic ordinary electricity, and gas at LSOA level. The data cover annual
consumption for 2009. The data cover all metered domestic gas and electricity
consumption,” (DECC5). To produce these electricity consumption estimates,
annualised consumption data is provided to DECC at Meter Point Administration
Number (MPAN) level by the data aggregators (DAs). DAs are agents of the
electricity suppliers who collate/aggregate electricity consumption levels for each
electricity meter in Great Britain” (DECC). A similar process is used to compile gas
data. Gas transporters supply DECC with the Annualised Quantity (AQ) for each
Meter Point Reference Number (MPRN) or gas meter as well as address point data,”
(DECC).
We also decided that is was necessary for the relevance of this study to include the
fuel poverty variable. “Low‐income households, who spend proportionately more
of their incomes on energy, are hit much harder by energy cost rises. Their demand
for energy tends to be more elastic than wealthier households, meaning that they
tend to use less if prices rise” (DECC, b, 2011). As we will discover in what we will
define as Fuel Poor Households, consumption of energy is low and this is not due to
green awareness but to financial barriers. Any measures looking at increasing
5
From Energy consumption 2009 Metadata
26
energy prices as a way to curb energy demand must look hard at the impact
(health, well-being) on the fuel-poor.
Variables
Definition
Demographic
v1
Age 5–14 : percentage of resident population aged 5–14 years
v2
Age 25–44: percentage of resident population aged 25–44 years
v3
Age 45–64: percentage of resident population aged 45–64 years
v4
Age 65+: percentage of resident population aged 65 or more years
v5
Population density : population density (the number of people per hectare)
Household composition
v6
Single pensioner household: percentage of households which are single-pensioner households
v7
Single person household (not pensioner): percentage of households
with one person who is not a pensioner
v8
Two adults with children: percentage of households which are
cohabiting or married couple households with children
v9
Two adults no children: percentage of households which are
cohabiting or married couple households with no children
v10
Lone parent household: percentage of households which are
lone parent households with dependent children
Housing
v11
Owned: percentage of households that are owned
v12
Social rented: percentage of households that are public sector rented accommodation
v13
Private rented: percentage of households that are private or other rented accommodation
v14
Detached Housing: percentage of all household spaces which are detached
v15
Semi-detached housing: percentage of all household spaces which are semi-detached
v16
Terraced housing: percentage of all household spaces which are terraced
v17
All flats: percentage of households which are flats
v18
People per room : average number of people per room
v19
Average house size : average house size (rooms per household)
v20
without central heating; with sole use of bath/shower and toilet
Socio-economic and employment
v21
Managerial and professional occupations
v22
Intermediate occupations
v23
Routine and manual occupations
v24
Never worked and long-term unemployed
v25
Full time student (economically active)
v26
Retired
v27
No qualifications
v28
Level_4_5
27
Energy consumption
v29
Consumption of Ordinary Domestic Electricity 2009
v30
Consumption of Domestic Gas 2009
v31
Fuel Poverty 2009
Table 1: List of the 31 variables selected for input to the classification
2.1.2 Step 2: Normalisation
2.1.2.1 Logarithmic transformation
Before standardisation and cluster of the data, we transformed the data to a log
scale. "Using logs is one of several ways in which the effect of outliers can be
reduced" (Harris et al. 2005). The second aim to log scale the data is it “reduces the
likelihood of a highly skewed distribution within a variable which would create
uneven cluster sizes” (Vickers & Rees, 2005). To undertake the log scale, we used
the software SPSS 17. Log transform can only be applied to numbers above 0 so we
added the value 1 to all the data in the dataset.
2.1.2.2 Standardisation
After transforming the data to a log scale, we standardised the input data. This step
is essential in transforming the data into rate or measure to the same scale,
enabling comparisons.
Each variable has been standardised following the Range method. “In this method
[Range] the minimum and the maximum of the data values are computed (thus the
range – max-min) and each value has the min. subtracted from it and is then
divided by the range” (De Smith et al., 2007). To not standardise the data would
give too much weight to variables with larger value or important variations. This
method was also used in the ONS 1991 classification of local authorities and for the
UK national Statistic 2001 AOC.
28
2.1.3 Steps 3: Clustering
Different methods that can be used to create such a classification have been widely
discussed. There is an abundance of different clustering algorithms available:
hierarchical clustering, K-means, fuzzy K-means to cite only a few. For this study we
will use “the k-means classification as one of the most commonly used methods in
the geodemographics industry” (Harris et al., 2005). This method also suits large
data sets.
For the k-means analysis, once you’ve chosen the number of clusters, “the
algorithm iteratively estimates the cluster means and assigns each case to the
cluster for which its distance to the cluster mean is the smallest” (SPSS 17). “The
data is randomly split into K clusters and the distances between the cluster centres
and the observation values in m-dimensional space are measured using the
Pythagorean equation for Euclidean distances” (Debenham et al., 2001). It then
recalculates the centre means from each object in the clusters. These new centre
means are used to classify again the objects. This step needs to be repeated until
their values are close or similar to the previous ones at which point we define the
situation as stable.
III. Application, results and discussion
3.1 Application
3.1.1 Running the K-means
There are several statistical packages that enable clustering analysis (R, SPSS, and
Microsoft Excel). For this work, we used SPSS 17 as Birkbeck College benefits from
free access for students and a user-friendly interface. R could have been an
interesting free alternative for this work.
29
It is advised to run the algorithm until the results are stable. This part was time
consuming as we ran the cluster analysis to the point where the means of the
cluster are identical for two successive runs.
We encountered a problem with one LSOA, which formed a group on its own. Once
identified, we removed it from the dataset before running the cluster analysis
again. The LSOA was added later to the appropriate cluster. To decide which cluster
this LSOA should belong to, we compared the average of the difference between
this LSOA variables values and the mean centre of each clusters.
The number of clusters (K) has to be specified before running the analysis.
Callingham (2003) suggested that “the most useful number of clusters in the first
level would be around 6” (Vickers et al., 2005) so we repeated the clustering
process with different values of K from 3 to 8 to find the best results.
3.1.2 Selecting the appropriate number of clusters
There isn’t one way to choose the number of clusters and it mainly depends on the
data selected, as well as personal interpretation of how many typologies the
classification should create (Debenham et al., 2001; Vickers and Rees, 2007). To
help our choice we conducted two recommend basics analysis: observing the
average distance of each object from the mean centres for each cluster and looking
at the number of cases in the clusters.
Figure 1 shows the average distance of each case from its cluster centre. It is
normal to see the values decreasing as the number of cluster increase. It is hard to
see an obvious solution; however we can see a higher intensity in the increase of
the line from 6 to 5, 5 to 4 and 4 to 3 clusters. For this reason we can consider to be
reasonable to keep 6, 7 and 8 as possible number of clusters.
30
Figure 1: Average distance from the cluster centre by number of clusters
The second element weighing on the choice of the number of clusters is the
number of members in each cluster. It is best to have the most evenly distributed
members in the groups. Although we cannot expect exactly the same number of
members, it is imperative to avoid having the majority of cases in one or two
clusters and then a number of sparsely populated groups (Debenham et al., 2001).
To assess it for each K possibilities, we calculated “the average difference between
the number of members in each cluster from the mean -the mean is the optimal
solution as all clusters will have the same number of members” (Vickers and Rees,
2007). Results can be seen in the table 2. The values 7 and 8 stood out from this
method.
Number of members
in each cluster
Average distance from the
mean
3
clusters
1375
1586
1803
-
4
clusters
1304
1102
855
1503
-
5
clusters
938
1438
731
723
934
-
6
clusters
718
559
727
649
1057
1054
-
7
clusters
693
648
446
997
704
660
616
-
8
clusters
496
693
389
677
557
802
548
602
143
213
194
174
101
98
Table 2: Homogeneity of the cluster membership size
31
Considering the results from the two methods, it appears the optimal number of
clusters should be 7 or 8. After looking at the cluster characteristics for 7 and 8
clusters, we decided to take 7 clusters, as in the 8 clusters solution there was a
redundancy of type of areas.
LSOAs
% of LSOAs
1
2
3
4
5
6
7
Range
693
648
446
997
704
660
616
551
14.55
13.6
9.36
20.93
14.78
13.85
12.93
-
Table 3: Number of LSOAs in each cluster
3.2 Analysing and mapping the new geodemographic clusters
3.2.1 Step 4: The clusters and their descriptions
Pen portraits are “small descriptive analyses of the clusters that draw upon their
main identifiable characteristics” (Debenham et al., 2001).
The aim to defining clusters is to find a name, which can quickly convey the main
type of group without offending or introducing negative connotations. For example,
Vickers and Rees reject the word “elderly” as it could portray old age in a negative
sense (Vickers & Rees, 2007). Often names can be too specific and give a stereotype
of a cluster that although may be an accurate representation of the mean values or
the cluster centre it does not represent any of the diversity within the cluster
(Vickers, 2006). Vickers has criticised commercial geodemographic labelling as some
cluster names emphasize certain groups or have negative undertones (Vickers,
2006). They also insist on not repeating any group name which could have been
already use by other classifications.
After characterising the main elements of each cluster, we will briefly put forward
policies from the set-to-launch Green Deal as well as other existing measures in
place to curb energy use. However, as stated in the introduction, the aim of this
study is not to match up existing and forthcoming legislation and policy around
32
energy efficiency and the different clusters but to create a classification that would
enable a more precise targeting of energy efficiency policies.
For the presentation of the cluster order, we had the choice between using the
energy consumption variables to create an ordinal order or just use the random
order generated from the cluster analysis. We preferred to describe the clusters in
the same order as they appeared after the cluster analysis, to avoid any subjective
idea of superiority or inferiority between them. For each cluster portrait, we will
add the radial plot showing the values for each variable. The values are the
difference from the mean for that variable. The mean, which is therefore 1, it also
represented in each radial plot. We have also added a separate basic map for each
cluster representing the distribution of the LSOAs, which allows to geographically
localise each group.
3.2.1.1 Cluster 1: Electricity-intensive city renters
Distinctive Variables:
V30
V29
V28
V27
V26
V25
V24
V310.25
V1
High:
V2
0.2
0.15
0.1
0.05
0
-0.05
-0.1
-0.15
-0.2
V3
V4
V5
V6
V7
V8
V9
V23
V10
V22
V11
V21
V12
V20
V13
V19
V18 V17
Age 25-44
Single Person Household
Two adults with children
Two adults no children
Private rented
All flats
Managerial and professional occupations
Educated
V16 V15
V14
Cluster 1
Mean
Low:
Age 5-14
Semi-detached
People per room
Average house size
Routine and manual occupations
Retired
No qualification
Fuel poverty
Figure 2: Summary of cluster 1
33
Map 2: Localisation of LSOAs in cluster 1
This group, predominantly situated in or close to the city centre, is characterised by
a high proportion of managerial and professional occupations within the 25-44
years age bracket. They are mainly geographically concentrated in the inner city
and live in flats rather than houses. They are educated people in the prime of the
active life. The house composition is essentially represented by single, couples and
couples with kids younger than 5 years old. The low proportion of not educated
people, fuel poverty and routine and manual occupation within this cluster
underlines the fact this group represents young wealthy professional still renting,
yet to get on to the property ladder.
Well-equipped and usually living in flats, their electric consumption is the highest of
all the groups, while their gas consumption is close to the average. This group has
the advantage of being aware of climate change issues in relation to their energy
consumption but the disadvantaged of being trapped in the landlord-renter
34
dilemma, as they can’t improve the energy efficiency of their home without
landlord buy-in.
3.2.1.2 Cluster 2: Fuel poor social renters
Distinctive Variables:
V30
V29
V28
V27
V26
V25
V24
V31 0.3
V1
V2
0.25
0.2
0.15
0.1
0.05
0
-0.05
-0.1
-0.15
-0.2
-0.25
-0.3
-0.35
-0.4
High:
V3
V4
V5
V6
V7
V8
V9
V23
V10
V22
Population density
Lone parent household
Social rented
All flats
Never worked and long-term unemployed
No qualifications
Fuel Poverty
Low:
Owned
Detached
V12
Semi-detached
V13
Terrace
V14
People per room
V16 V15
Average house size
Cluster 2
Managerial and professional occupations
Mean
V11
V21
V20
V19
V18 V17
Figure 3: Summary of cluster 2
35
Map 3: Localisation of LSOAs in cluster 2
The proportion of social renters, people who has never worked or in long-term
unemployment is the highest of all clusters. These areas are also characterised by a
large amount of non-qualified people living in small flats. This group probably
represents the most financially strained type of household with the highest rate of
social renters and households fuel-poverty. A household is defined as in fuel poverty
when it spends more than 10% of its revenue on heating. John Hills in his recent
report suggests another definition which he takes from the Warm Homes and
Energy Conservation Act 2000, which introduces the notion, if not directly
mentioned, of inefficient households: those living on a lower income in a home that
cannot be kept warm and a reasonable cost (Hills, J, 2012).
This group has the highest rate of lone parent households. Detached, semi-detached
or terraced houses are rare. The consumption of gas or electricity is the lowest of all
clusters. For this group energy consumption is dictated by their budget rather than
36
their aspiration to environmental benefits. There is clear overlap between low
income and the inefficiency of the homes people live in and successful policies for
this group would benefit from not only addressing the environmental issues
associated with energy inefficient homes but also from reducing low income
household spend on energy. This group is roughly situated in what we will call the
inner city areas in often difficult to treat housing stock.
3.2.1.3 Cluster 3: Energy-intensive wealthy greys
Distinctive variables:
High:
V30
V29
V28
V27
V26
V25
V24
V31 0.3
V1
V2
0.25
0.2
0.15
0.1
0.05
0
-0.05
-0.1
-0.15
-0.2
-0.25
-0.3
V3
V4
V5
V6
V7
V8
V9
V23
V10
V22
V11
V21
V12
V20
V13
V19
V18 V17
Age 45-64
Age 65+
Owned
Detached
Semi-detached
Average house size
Retired
V16 V15
V14
Cluster 3
Mean
Low:
Population density
One person household
Two adults with children
Two adults no children
Lone parent household
Social rented
Private rented
Terrace
All flats
Without central heating
Never worked and long-term unemployed
Figure 4: Summary of cluster 3
37
Map 4: Localisation of LSOAs in cluster 3
This cluster represents the older generation with a high proportion of people aged
between 45 to over 65, generally towards the end of their professionally active life
or already retired. The low proportion of couples with children confirms the
population’s later age. They are mainly working or worked in intermediate or
managerial and professional occupations. They are the second highest gas
consumer and their electricity consumption is also important. They own large
detached or semi-detached houses in well-established residential suburbs. It
contains the least number of cases compare to the other clusters with 446 LSOAs.
This group is likely to be able to afford to maintain their current energy
consumption. However, going forward and with changes to pension policies, energy
efficiency is likely to become more of an attractive option to the members of this
group. Further educating this group over the benefits of energy efficiency could
increase uptake in policies. A recent survey from non-profit energy supplier Ebico6
6
https://www.ebico.org.uk/blog/2012/04/30/the-green-deal-an-opportunity-for-the-newly-retired/
38
on how the forthcoming Green Deal could benefit newly or soon to retire still
highlighted a suspicion and misunderstanding over potential benefits for this group
with only around a third of respondents saying they would take advantage of the
new measures.
3.2.1.4 Cluster 4: Average consuming London renters
V30
V29
V28
V310.15
V1
V2
Distinctive variables:
V3
V4
0.1
High:
V5
0.05
V27
V6
0
V26
V7
-0.05
V25
V8
-0.1
V24
V9
V23
Lone parent households
Social rented
Private rented
Terrace
Never Worked and long-term unemployed
Low:
V10
V22
V11
V21
V12
V20
Age 65+
Retired
Consumption of Electricity
V13
V19
V18 V17
V16 V15
V14
Cluster 4
Mean
Figure 5: Summary of cluster 4
39
Map 5: Localisation of LSOAs in cluster 4
As for cluster 2, this group includes a strong proportion of social renters and of
people who have never worked or are in long -term unemployment. However by
comparing the mean centres for these 2 variables to the cluster 2, it is apparent
that this group is not as financially deprived as the fuel-poor social renters. This
group also includes a large proportion of privately rented and the primary type of
household is terraced rather than flats, as is the case for cluster 2. There is also a
non-negligible rate of the population working in routine or manual jobs, implying
that although not in the best-paid lines of work, they are still professionally active
and are not only dependent of governmental aids. The mean centres of the other
variables including the two energy consumption variables are close to the average.
This cluster also has the highest number of LSOAs. This group could represent in
some ways the average household in London.
40
Being the largest group with some inevitable nuances, policies would have to
include measures for landlords as well as awareness campaigns around renters’
rights with regards to energy efficiency.
3.2.1.5 Cluster 5: Wealthy energy intensive owners
V30
V29
V28
V310.15
V1
V2
Distinctive variables:
V3
V4
0.1
0
V27
High:
V5
0.05
V6
-0.05
V26
V7
-0.1
V25
V8
-0.15
-0.2
V24
V9
V23
V10
V22
V11
V21
V12
V20
V13
V19
V18 V17
V16 V15
V14
Cluster 5
Mean
Owned
Privated rented
Detached
Semi-detached
Managerial and professional occupations
Educated
Low:
Lone parent households
Social rented
Routine and manual occupations
Never worked and long-term unemployed
No qualifications
Fuel poverty
Figure 6: Summary of cluster 5
41
Map 6: Localisation of LSOAs in cluster 5
This group comprises people with high levels of education, who occupy highresponsibility positions. They are mainly owners of large detached or semi-detached
house but there is still a fair proportion of privately rented accommodation. This
cluster is the main consumer in gas, as well as being the second biggest consumer
of electricity. In this cluster there is a strong correlation between high-energy
consumption and wealth.
In terms of energy efficiency measures, the fact that a high proportion of this group
is highly educated can help awareness and buy-in around energy efficiency
measures. However money not really an issue for this group so policies would
potentially have to strike a chord with regards to climate change.
42
3.2.1.6 Cluster 6: Average use suburban working families
Distinctive variables
V30
V29
V31 0.2
V1
High:
V2
0.15
V3
V4
0.1
V28
V5
0.05
V27
V6
0
-0.05
V26
V7
-0.1
V25
V8
-0.15
-0.2
V24
V9
V23
V10
V22
Owned
Semi-detached
Terrace
People per room
Average house size
Intermediate occuaptions
Routine and manual occupations
Retired
No qualifications
Low:
V11
V21
Single person household (not pensioner)
Two adults with children
Two adults no children
Social rented
V12
V20
V13
V19
V18 V17
V16 V15
V14
Cluster 6
Mean
Private rented
All flats
Figure 7: Summary of cluster 6
43
Map 7: Localisation of LSOAs in cluster 6
A majority of the housing stock in this group is semi-detached and terraced. The
average is 45 to 64 years old as well as 5 to 14 years. Adding the fact the number of
people per room is high; we can conclude this cluster represents older families
possibly with teenagers or grown-up kids. They are mainly working in intermediate,
routine and manual occupations and the unemployment rate is low. Even though
they own their house, the low level of qualification, type of job and the choice of
area where houses are more affordable than central London suggest, “money is still
a concern”. As opposed to the previous group, the suburban working family is less
concerned per se about climate change issues but more about saving money
through energy efficiency.
3.2.1.7 Cluster 7: Low consuming strained renters
Distinctive variables:
V30
V29
V28
V27
V26
V25
V24
V31 0.2
V1
V2
0.15
0.1
0.05
0
-0.05
-0.1
-0.15
-0.2
-0.25
-0.3
High:
V3
V4
V5
V6
V7
V8
V9
V23
V10
V22
Age 5-14
Age 65+
One person pensioner
Lone parent households
Social rented
Semi-detached
Routine and manual occupations
No qualifications
Fuel poverty
V11
V21
V12
V20
V13
V19
V18 V17
V16 V15
V14
Cluster 7
Mean
Low:
Age 25-44
Private rented
Managerial and professional occupations
Educated
Figure 8: Summary of cluster 7
44
Map 8: Localisation of LSOAs in cluster 7
This cluster is characterised by a much higher proportion of the population without
qualifications and working in routine and manual occupations when comparing to
the London average and to the other clusters. The proportion of people in the 5-14
age group and of lone parent households with dependent children, as well as
retired people, suggest a more mixed type of household with the common
characteristic to be in some proportion deprived. This idea is supplemented by the
large proportion of social renters. This group also has the second highest rate of
fuel poverty. Both their electricity and gas consumption are below average and only
cluster 2 has both rates lower than cluster 7. This group lives predominantly on the
outskirts of London.
45
3.3 Visualisation
Our first map (map 9) representing cluster spatial distribution follows a
conventional approach, by using choropleth maps. However, we found the visual
impact to be greater when assigning the cluster to the buildings included in the
LSOA instead of the full area. An alternative view of this geodemographic
classification can be seen in map 10. It follows the ideas of Mark O’Brien, himself
inspired by Charles Booth’s poverty map. This visualisation method assigns cluster
colours to buildings included in the LSOA instead of the full area. Only the buildings,
roads, parks and water-ways can be seen. This type of map not only helps the user
to situate the groups but it also adds visual realism.
46
Map 9: Visualisation of household energy consumption classification in
choropleth map
47
Map 10: Visualisation of household energy consumption classification in Charles
Booth’s style
48
We added a link7 to download the choropleth map in KML (Keyhole Markup
Language) as well as the building layer from map 10. KML is a “XML-based file
format used to display geographic data in an Earth browser such as Google Earth,
Google Map or ESRI ArcGIS Explorer” (Google Developers, ESRI). Visualising our
maps allows the user to pan or zoom in, out or around the map. However there is a
specific limitation to the size and complexity of loaded KML files (Google
Developers), which means we could not export the entire map10 but only the layer
with the buildings. Hosting the map on this type of platform can also be considered
as a way of sharing data. Anyone with the link to download the file will be able to
freely visualise our maps in Google Earth. We could argue that there are easier
ways to share data. In this case, a third party would have to download the file, as
opposed to if we would have hosted it on a web-based application which is as easy
to access as any other website. It is worth noting also that even though visualising
our maps in Google Earth gives enables user interactivity,” it is an incomplete
exploratory spatial data analysis (ESDA) tool because it lacks the functionality of
brushing” (Gibin et al., 2008).
Figure 9: Visualisation of the building layer in Google Earth
7
https://sites.google.com/site/nrgclassification/kml-links
49
3.4 Limits and extensions:
Geodemogropahic classification, although still being used and also evolving (new
UK OAC will come from the Census 2011) has its limits. Some limits can be
described as inherent to most classifications, while others can be specifically
applied to the commercial classification (i.e. lack of methods details). Two main
issues have to be raised when working on area-based data: ecological fallacy and
Modifiable Areal Unit Problem (MAUP).
3.4.1 Ecological Fallacy
An ecological fallacy occurs when it is inferred that results based on aggregate zonal
(or grouped) data can be applied to the individuals who form the zones or groups
being studied (Openshaw, 1984). For example, it is right to say the average
consumption of ordinary domestic electricity for an area is 3000 Megawatt Hours in
2009 but strongly wrong to assume that households living in that area are
consuming 3000 Megawatt Hours in 2009. The ecological fallacy occurs when we
interpret that the association observed at this area level reflects the same
association at the individual level. When looking at the results, it is important that
people are aware of this to avoid wrong assumptions.
3.4.2 The Modifiable Areal Unit Problem
Openshaw describes two related elements of MAUP: The ‘scale problem’ is “the
variation in results that can be obtained when data for one set of areal units are
progressively aggregated into fewer and larger units for analysis” (Openshaw 1984).
The “aggregation Problem” is when individual data are regrouped in areas; the
statistical interest will have different values depending on the area used. In our
study, we focussed on the LSOA level. By using Output Area or wards level for
example, it is likely that characteristics and the spatial repartition of the clusters
50
would have been different. Openshaw advises to minimise this problem by using
the smallest spatial unit. This study would have no doubt gained in pertinence by
using Output area, which is the finer Census area possible. But as Singleton notes,
“the scale at
which geodemographic
classifications
may plausibly
be
created depends crucially on the resolution of the input data” (Singleton, 2007). We
used LSOA level to match the essential energy consumption data available.
3.4.3 Dating data
A constant criticism of any geodemographic is the question around the time
relevancy of the data used. We have here used the Census 2001 and therefore can
expect changes from 2001 to today. In London, population size has increased by 9.1
per cent from 2001 to 2010 (ONS, c, 2011). At the time of writing, a new Census
(2011) has already been conducted but the outcomes are yet to be made available.
“Some (Sleight 2004) have suggested this change does not matter as certain areas
will always be dominated by certain types of people and as people move out similar
people move in” (Longley et al. 2011; Gale and Longley, 2012). This study would
benefit from being updated with the new Census 2011 data and more recent data
on the energy consumption in order to reflect a more real-time profile of London
household energy characteristics. It would also allow us to seeing the evolution in
attitudes towards energy consumption as well as impacts and effects of house
improvement, new energy efficient building developments, uptake of energy
efficiency policies etc.
3.4.4 Missing data
This section will question some of limitations of this study with regards to the data
selection. An interesting data set to include, if it would have been publically
available, is the age of the house. “The highest consuming properties are those built
between 1919 and 1945, 12 per cent higher than the average and 7 per cent higher
than the oldest property group pre- 1919” (DECC, c, 2011). Another variable or
51
aspect that could be associated to our research would have been household data
on building efficiency. Energy performance certificate could have made it possible
to see the difference between high-energy consumption due to house structure,
lack of isolation, or due to behaviours. This data is now compulsory on sale of a
house but is not yet available for all households and is not publically stored or
gathered.
We are studying the average energy consumption over a year, but it still not
implicitly showing the individual habits around energy use. We can only suggest
their consumption habits depending on the type of house or other socio-economic
characteristics. For the electricity-intensive city renters, for example, we assume
that one of the reasons behind their high power use is their high use of electrical
appliances.
3.4.5 The Labelling of Clusters
Finding adequate names for each cluster to give the reader a clear, first idea of the
group’s characteristics was challenging. Even though the name will not change the
content of the cluster, Vickers insists many users will not look past the names to
provide them with an impression of what an area is like (Vickers, 2006). For this
study we wanted to incorporate energy consumption characteristics as well as one
or two of the main distinctive characteristics drawn from the chosen variables.
When labelling clusters there is always the risk of misinterpretation through the
very nature of generalising groups through naming.
3.4.6 Fuzzy classification?
Another typical criticism of classification is the crisp nature of classification (Vickers,
2006), which implies one area can only be in one cluster. The fact of assigning an
area in a specific group while this same area can be closed to another group can
therefore be seen as a limit for this sort of classification. A solution to this problem
52
could be the use of a fuzzy classification. The idea of fuzzy geodemographics is that
“areas are not seen as a member of one type but as partial members of all types
dependent on values” (Vickers, 2006). However, Vickers claims that “classic”
geodemographics are appreciated for their simplicity.
3.4.7 Hierarchical classification
While other geodemographic classification such as OAC or MOSAIC have several
levels of hierarchy, with super groups which are themselves separated in to smaller
groups, our classification comport only one level. Based on Callingham’s work
(2003), Vickers and Rees (2007) explained that each level of hierarchy, three in this
case, represents a different contribution. While the super groups allow a good
visualisation and cluster labelling, the second level of aggregation has its main
contribution for conceptual customer profiling (Vickers and Rees, 2007). The next
level of aggregation can be used for market propensity measures from the larger
commercial surveys such as TGI [target group index] and the readership surveys
(Vickers and Rees, 2007). There are various reasons why we only created one level
of aggregation here. We were only working on Greater London, as oppose as the
whole of the UK for the OAC, so we created a regional geodemographic
classification instead of a national one. Therefore the number of possible levels of
aggregation was reduced. This choice has also been directed by the fact we were
using data at the LSOA level and in consequence we had less data available in term
of number of objects or areas. We didn’t want to create clusters with too few areas,
which could have result with the creation of groups representing only outliers and
could have been misinterpreted later on. A consequent extension of our study in
view of this issue would therefore be the possibility to create this classification at
OAs level and include a hierarchy.
53
3.4.8 Classification validation
We have intentionally omitted to process a validation of our classification because
of the limited scale of this dissertation. The classification validation could be
considered as a study in itself. Vickers and Rees came back on the validation of the
OAC in 2001, “considering internal validation of data inputs, peer reviews of
methods and external validation against other variables” (Vickers and Rees, 2011).
They insisted on the specificity of ground-truthing, its difficulties and benefits.
Conclusions
The aim of this study was to create a household energy consumption classification
for Greater London in line with academicals values. A classification that would
enable energy behaviour profiling at a small enough level to serve as a basis for
improved policy targeting. The result produced through the methods presented in
the early stages of this study, and despite the limitations around the data, is 7
clusters, distinctive enough in their main characteristics to use a basis for public or
private energy efficiency initiatives.
Area classifications and geodemographics have since Charles Booth’s study of
London continued to be developed and used (Vickers, 2006). While they have for a
period of time been essentially used by commercial companies, they are now
widely spread in the academic world with the best example being the UK OAC. The
academic world has insisted in creating open-source classifications as opposed to
commercial geodemographics developed behind closed doors. The classification
carried out in this study fits those of applied academic geodemographics.
The end result is an open-source classification, which could be used as a tool for
local authorities but also energy suppliers and households to find the adequate
solutions to reduce energy consumption. While the aim of this study was not to
proceed with a detailed policy match with the different profile groups once defined,
we did judge it necessary to make some suggestions to support the overall
objective of this report.
54
An interesting element in this study was the scale of our target area. Because of the
special status of Greater London in its geodemographic structure and energy
consumption determinants, it needs specific area classifications to be developed
around it to avoid its intrinsic characteristics being diluted in a UK-wide
classification. Our study follows this trend of regional geodemographics to support
a more representative view on energy consumption and London. An interesting
development of our study would be to consider if the method used here could be
applicable at a national and/or regional scale. Gale et al. (2012) insist that the
known limitations of the 2001 Geodemographic OAC for London will feed into the
elaboration of the 2011 OAC. One of the main aims being to develop a clear release
strategy as well as “a flexible open reproducible methodology for the 2011 UK OAC
in order to allow the creation of regional classifications” (Gale et al., 2012).
Another extension of the result of this analysis would be the exploration of
visualisation methods to help households, policy makers and energy organisation to
actually see energy efficiency in London in a more intuitive way. The creation of a
mash-up with Google Map could for example allow the results of this study to reach
a wider audience. Similar applications include “Londonprofiler” created by Maurizio
Gibin and the “Geodemographics of Housing in Great Britain: A new visualisation in
the style of Charles Booth's map” by Mark O’Brien. Due to the nature and scale of
this particular study, we instead took first step towards making the outcome of this
study publically available and interactive through a downloadable KLM map.
55
References:
ArcGIS Resource Center, About KML, Copyright © 1995-2012 Esri. Accessed 9th
September 2012:
http://help.arcgis.com/en/arcgisonline/help/index.html#//010q00000066000000
Briefing from the Mayor of London, 2nd Of July 2012, The Green Deal Statutory
Instrument – Second Delegate Legislation Committee. Accessed 05th September
from: http://www.energyforlondon.org/
European Commission (2008) Climate and Energy Package. Accessed 11th
September 2012:
http://ec.europa.eu/clima/policies/package/index_en.htm
Birkin, M. and Clarke, G. (1998) GIS, Geodemographics, and Spatial Modeling in the
U.K. Financial Service Industry. Journal of Housing Research. Vol 9, Issue1
Blake, M & Openshaw, S. (1995) Selecting Variables for Small Area Classifications of
1991 UK Census Data. Working Paper 95/2, School of Geography, University of
Leeds.
Debenham, J., Clarke, G. and Stillwell, J. (2001) Deriving supply-side variables to
extend geodemographic classification.
Debenham, J. (2002) Understanding Geodemographic Classification: Creating The
Building Blocks For An Extension. Working Paper. School of Geography , University
of Leeds.
DECC (2011) Energy consumption in the United Kingdom: 2012. Accessed 11th of
September 2012:
http://www.decc.gov.uk/assets/decc/11/stats/publications/energyconsumption/2323-domestic-energy-consumption-factsheet.pdf
DECC, a (2011) Estimates of Home Insulation Levels in Great Britain: January 2012.
Accessed 10th September 2012
56
http://www.decc.gov.uk/assets/decc/11/stats/energy/energy-efficiency/4537statistical-release-estimates-of-home-insulation-.pdf
DECC, b (2011) Great Britain’s housing energy fact life. Accessed 11th of September
2012:
http://www.decc.gov.uk/assets/decc/11/stats/climate-change/3224-great-britainshousing-energy-fact-file-2011.pdf
DECC, c (2011) National energy efficiency data-framework. Accessed 11th of
September 2012:
http://www.decc.gov.uk/assets/decc/11/stats/energy/energy-efficiency/2078need-data-framework-report.pdf
De Smith, M. J., Goodchild, M. F., Longley, P. A. (2007) Geospatial Analysis: a
Comprehensive Guide to Principle Techniques and Software Tools (Leicester, UK,
Troubador)
Druckman, A. & Jackson, T. (2008) Household Energy Consumption in the UK: A
Highly Geographically and Socio- economically Disaggregated Model. Energy Policy
, vol. 36, pp. 3177- 3192.
Everitt, B. S., Landau, S. and Leese, M. (2001) Cluster Analysis. 4th edn. London:
Arnold
Experian (2008) GreenAware: A Segmentation of Environmentally- Relevant
Behaviours, Attitudes and Carbon Footprint. Retrieved April, 19, 2012, from
http://www.experian.co.uk/assets/business
strategies/brochures/GreenAware_factsheet[1].pdf
Gale, C.G., Adnan, M., Longley, P.A. (2012) Open Geodemographics: Open Tools and
the 2011 OAC. Not published yet, accessed through University College of London.
Gale, C.G., Longley, P.A. (2012) Geodemographic Output Area Classifications for
London, 2001-2011. Not published yet, accessed through University College of
London.
57
Gibin, M., Singleton, A.D., Mateos, P., Longley, P.A. (2008) Exploratory Cartographic
Visualisation of London using the Google Maps API. Applied Spatial Analysis and
Policy, 1(2), 85-97.
Google Developers (2012) KML Tutorial. Accessed 9th September 2012:
https://developers.google.com/kml/documentation/kml_tut
Greater London Authority (2012) Low carbon London. Accessed the 12th
September 2012:
http://www.london.gov.uk/priorities/business-economy/low-carbon-economy
Harris, R, Sleight, P. & Webber, R. (2005) Geodemographics: GIS and
Neighbourhood Targeting. Wiley-Blackwell.
Hills, J. (2012) Getting the Measure of Fuel Poverty 2012. Accessed 12 th of
September 2012:
http://sticerd.lse.ac.uk/dps/case/cr/CASEreport72.pdf
Longley, P. A. (2005) Geographical Information Systems: a renaissance of
geodemographics for public service delivery, Progress in Human Geography 29, 1
(2005) pp. 57–63
Longley, P.A., Cheshire, J.A. and Mateos, P. (2011) Creating a regional geography of
Britain through the spatial analysis of surnames. Geoforum 42, 506-516.
ONS, a (2011) Household and families. Social Trends 41. Accessed 11th of
September 2012:
http://www.pdfdownload.org/pdf2html/pdf2html.php?url=http%3A%2F%2Fwww.o
ns.gov.uk%2Fons%2Frel%2Fsocial-trends-rd%2Fsocial-trends%2Fsocial-trends41%2Fsocial-trends-41---household-and-families.pdf&images=yes
ONS, b (2012) The National Statistics Socio-economic Classification (NS-SEC rebased
on the SOC2010). Accessed in 28th August 2012
58
http://www.ons.gov.uk/ons/guide-method/classifications/current-standardclassifications/soc2010/soc2010-volume-3-ns-sec--rebased-on-soc2010--usermanual/index.html
ONS, c (2011) Methodology Note on production of Super Output Area Population
Estimates. Accessed 10th September 2012
http://www.ons.gov.uk/ons/guide-method/method-quality/specific/populationand-migration/pop-ests/methodology-note-on-production-of-super-output-areapopulation-estimates.pdf
Openshaw, S. (1984) The modifiable areal unit problem. Concepts and Techniques
in Modern Geography 38: 41.
Petersen, J., Gibin, M., Longley, P., Mateos, P., Atkinson, P. and Ashby, D. (2010)
Geodemographics as a tool for targeting neighbourhoods in public health
campaigns. J Geogr Syst 13(2), 173-192.
Petersen, J and Ashby, D and Atkinson, P. (2007) Health applications for open
geodemographics. In: Winstanley, AC, (ed.) (Proceedings) Geographical Information
Science Research UK. pp. 160 - 166. National university of Ireland: Maynooth, IE.
Petersen, J., Gibin, M., Longley, P., Mateos, P., Atkinson, P. and Ashby, D. (2010)
Geodemographics as a tool for targeting neighbourhoods in public health
campaigns. J Geogr Syst 13(2), 173-192.
Räsänen, T., Ruuskanen, J. and Kolehmainen, M. (2008) Reducing energy
consumption by using self-organizing maps to create more personalized electricity
use information. Applied Energy 85, 830-840.
Rose, D. and O'Reilly, K. (1998) The ESRC Review of Government Social
Classifications: Final Report
Singleton, A.D. (2007) Comparing Classifications: Some Preliminary Speculations on
an
Appropriate
Scale
for
Neighbourhood
Analysis
with
Reference
to
Geodemographic Information Systems. UCL, Working Papers Series, Paper 127.
59
Singleton, A.D., Longley, P.A. (2009) Creating Open Source Geodemographic
Classifications for Higher Education Applications. UCL, Working Papers Series, Paper
134.
Sleight, P. (2004) Targeting Customers: How to Use Geodemographic and Lifestyle
Data in Your Business. Henley-on-Thames: World Advertising Research Centre.
SPSS Inc. (2011) K-Means Cluster Analysis, Chapter 26. pp. 184-187 in IBM SPSS
StatisticBase 20, User’s Guide. Chicago, SPSS Inc.
The Energy Trust. (2012) http://www.energysavingtrust.org.uk/
Vickers, D.W., Rees, P.H. and Birkin, M. (2005). Creating the National Classification
of Census Output Areas: Data, Methods and Results. Working Paper 05/2, School of
Geography, University of Leeds, Leeds
Vickers, D.W. and Rees, P.H. (2006). Introducing the National Classification of
Census Output Areas, Population Trends, 125.
Vickers, D.W. (2006) PhD Thesis: Multi-Level Integrated Classifications Based on the
2001 Census. University of Leeds
Vickers, D.W. and Rees, P.H. (2007). Creating the National Statistics 2001 Output
Area Classification. Journal of the Royal Statistical Society, Series A
Vickers, D. W., Rees, P.H. (2011) Ground-truthing Geodemogrpahics. Applied Spatial
Analysis, 4:3-21
Webber, R. and Craig, J. (1978) Socio-economic Classifications of Local Authority
Areas. London: Office of Population Censuses and Surveys.
Data sources
UK
Census
data
2001.
Retrieved
from
the
CASWEB
website:
http://census.ac.uk/casweb/
60
DECC (2009) Domestic Energy Consumption. Retrieved from Department of Energy
and Climate Change in January 2012:
http://www.decc.gov.uk/en/content/cms/statistics/energy_stats/energy_stats.aspx
Boundary data from Ordnance Survey data © Crown copyright and database right
2011. Retrieved from:
http://edina.ac.uk/
Annex 1: Accessed 05th September from: http://www.energyforlondon.org/
The Green Deal Statutory Instruments - Second Delegated Legislation Committee
Monday 2 July
Briefing from the Mayor of London
Key Messages
● The Mayor of London is supportive of the Government’s Green Deal agenda, and
is clear that its ambitious aims can only be achieved nationally if they are achieved
in London. The capital has by far the highest number of needy properties of any
region, but there is a real danger that these properties could be sidelined by Green
Deal providers as a result of the current framework being proposed by
Government.
● To ensure that the Green Deal succeeds in the UK, and that the most needy
properties in London are treated, we have therefore suggested some small but
extremely important refinements to DECC’s current plans. These refinements will
ensure that London’s unique situation is taken into account and will help prioritise
the capital’s fuel-poor as well as helping to significantly decrease carbon emissions.
● Most crucially, there is a pressing need for an area allocation for the Energy
Company Obligation (ECO). Without such a target there is a real danger that London
61
will miss out on the attention it needs, as energy companies and Green Deal
providers focus on treating areas that are cheaper and easier to retrofit. As 22% of
hard-to-treat properties in England are in London, this has the potential to
significantly undermine the success of the Green Deal as a whole.
● Furthermore, without an area allocation to ensure that flats and mid-terraces key markets in London for the Green Deal - are able to access ECO subsidies, there
is a serious threat that these homes will miss out on the benefits of the scheme. If
these properties are side-lined, Londoners could end up paying an additional
£390m on their energy bills to fund the Green Deal nationally, while the capital
receives investment of only £156m in return – an unacceptable possibility given
London’s specific needs.
Background
● London has lost out under previous, similar schemes such as the Carbon
Emissions Reduction Target (CERT), under which the city only received 4.7% of
installations despite having 12% of housing. We estimate that London has missed
out on £480m of CERT funding since 2005.
● This is because energy companies have previously fulfilled their obligations
wherever it most cost-effective to do so, without regard to the potential to reduce
the comparatively high levels of fuel poverty in London and carbon emissions from
some hard-to-reach housing stock. The comparatively high cost of treating London
homes, compared to those outside the capital, has meant that energy companies
have focused their attentions elsewhere.
The need for area allocations
● The Government’s current suggested plans for ECO could exclude Londoners in
flats and mid-terraces from the Green Deal, as such housing stock will be unable to
access a high enough subsidy from the ECO to make the scheme viable. DECC’s
Green Deal impact assessment confirms this possibility, though as these housing
62
types make up nearly two thirds of solid wall properties in the capital it is vital for
the success of the Green Deal that they are not excluded in this way.
● If London continues to receive a lower than equitable share of national energy
efficiency funding this will mean ECO will fail to adequately tackle fuel poverty in
the capital. This, in turn, will affect the success of the Green Deal as a whole.
● Area allocations should therefore be put into effect, based on the relative share
of solid wall properties for the carbon target and the relative share of fuel poverty
for the affordable warmth target. This will ensure that key properties in London,
including flats and mid-terraces, are prioritised and in turn help to ensure the
success of the Green Deal programme.
● Such area allocations would not only ensure fuel poverty in the capital is tackled,
but also create certainty as to the size of the market and encourage more Londonbased suppliers to enter the market. Only 1/3 of the suppliers for London’s unique
RE:NEW scheme, which provides energy efficiency measures to needy London
homes, have expressed interest in becoming Green Deal providers so far, and it is
extremely important for the success of the scheme that more providers are
encouraged to get involved in the capital.
● Given that the Green Deal is a new programme, area allocations should be
indicative, and then monitored on a quarterly basis so we have a clear
understanding of where the ECO is being delivered and can check that delivery is
equitable across the country. This will help ensure that the Green Deal is a success,
including in areas that are more expensive to treat. Provision for area allocations
could be delivered through secondary legislation or within the ECO brokerage
document, which will set out how Green Deal providers can access ECO funding.
For further information please contact Greg Taylor, Senior Government Relations
Officer, on 020 7983 4498 or at greg.taylor@london.gov.uk
63
Annex 2:
Suggested policies:
This table provides a brief overview of possible matching of existing or forthcoming
energy efficiency policies with the different groups.
For more information on the schemes mentioned please refer to the Energy saving
Trust http://www.energysavingtrust.org.uk/.
Cluster 1:
Electricity-intensive,
-
Energy
Saving This group is unlikely to
Trust Recommend qualify for means tested
city
Label
grants.
-
Renewable tariffs
majority of renters, this
-
Smart Metering
group might benefit from
-
Energy
renters
Company ECO
Obligation (ECO)
-
Comprising
Climate
through
their
landlord. Climate change
change campaigns
campaigns
a
linked
to
reduced energy use could
strike a cord with this
group, which has a high
electricity use.
Cluster 2:
-
Carbon
savings This group would qualify
community
for a range of energy
obligation
efficient schemes at no
-
Smart metering
cost.
-
Free
Fuel-poor, social renters
grants
But
having
the
insulation measures in place is one
thing, getting households
to
apply
is
another.
Efforts will have to be
made to raise awareness
around the benefits of
reducing
household
64
inefficiency.
Cluster 3:
-
Energy-intensive, wealthy
greys
-
Micro-generation
The older part of this
grants
group would benefit from
Free
insulation existing and forthcoming
grants
policies although there is
-
Smart Metering
still wariness among this
-
Renewable
heat age group around the
incentives
benefits versus the effort
(applying, works etc.) The
recently retired or soon to
be retired could buy-in to
micro-generation
(favorable house types)
with the right, targeted
messages.
Cluster 4:
-
Savings The largest group of all
Obligation
our clusters, the average
-
ECO
consuming London renter
-
Affordable
would benefit from any
Average-consuming
London renter
Carbon
Warmth Obligation landlord led initiatives,
-
Smart Metering
which will come into play
-
Green Deal
with the Green Deal. The
more financially strained
could still qualify for some
of
the
means-tested
schemes. However being
the most heterogeneous
groups, a one-size fit all
approach
will
necessary work.
65
not
Cluster 5:
-
Green Deal
With money not really an
Wealthy energy intensive
-
Renewable tariffs
issue
owners
-
Energy
-
-
for
Saving successful
this
group,
uptake
of
Trust
energy
Recommended
measures will have to be
Label
driven by a convincing
Climate
change story.
efficiency
A
majority
of
campaigns
owners, this group will
Smart Metering
have some options under
the Green Deal.
Cluster 6:
Average
use
suburban
working families
-
Green Deal
Family
orientated,
-
Insulation grants
group would benefit from
-
Smart Metering
targeting campaigns to
-
ECO
promote
the
this
benefits
mainly financial of taking
up some of the energy
efficiency measures on
offer to this type of
households and income.
Cluster 7:
-
Insulation grants
This group has the second
Low consuming strained
-
Smart Metering
largest proportion of fuel
renters
-
Carbon
-
Savings poor and will therefore
community
benefit from a range of
obligation
measures for financially
Affordable
strained
households,
Warmth Obligation whether on benefits or in
low paid jobs. But as with
cluster 2, educating the
households in this group
66
around what options are
available to them at no
financial
cost
is
biggest hurdle.
67
the