research and analysis the impact of social factors and …re.indiaenvironmentportal.org.in/files/the...

R E S E A R C H A N D A N A LYS I S

The Impact of Social Factorsand Consumer Behavior onCarbon Dioxide Emissions inthe United KingdomA Regression Based on Input−Outputand Geodemographic ConsumerSegmentation Data

Giovanni Baiocchi, Jan Minx, and Klaus Hubacek

Keywords:

consumer behavioreconometricsenvironmental economicsindustrial ecologyinput−output analysis (IOA)sustainable consumption and

production (SCP)

Address correspondence to:Giovanni BaiocchiDurham Business SchoolDurham UniversityDurham DH1 3HYUnited [email protected]

c© 2010 by Yale UniversityDOI: 10.1111/j.1530-9290.2009.00216.x

Volume 14, Number 1

Summary

In this article we apply geodemographic consumer segmen-tation data in an input−output framework to understand thedirect and indirect carbon dioxide (CO2) emissions associatedwith consumer behavior of different lifestyles in the UnitedKingdom. In a subsequent regression analysis, we utilize thelifestyle segments contained in the dataset to control for as-pects of behavioral differences related to lifestyles in an analy-sis of the impact of various socioeconomic variables on CO2

emissions, such as individual aspirations and people’s attitudestoward the environment, as well as the physical context inwhich people act.

This approach enables us to (1) test for the significanceof lifestyles in determining CO2 emissions, (2) quantify theimportance of a variety of individual socioeconomic determi-nants, and (3) provide a visual representation of “where” thevarious factors exert the greatest impact, by exploiting thespatial information contained in the lifestyle data.

Our results indicate the importance of consumer behaviorand lifestyles in understanding CO2 emissions in the UnitedKingdom. Across lifestyle groups, CO2 emissions can vary bya factor of between 2 and 3. Our regression results providesupport for the idea that sociodemographic variables are im-portant in explaining emissions. For instance, controlling forlifestyles and other determinants, we find that emissions areincreasing with income and decreasing with education. Us-ing the spatial information, we illustrate how the lifestyle mixof households in the United Kingdom affects the geographicdistribution of environmental impacts.

50 Journal of Industrial Ecology www.blackwellpublishing.com/jie


Introduction and LiteratureReview

Evidence increasingly suggests that thehuman-induced release of greenhouse gasesinto the atmosphere might cause serious, po-tentially irreversible changes to the globalclimate within the next decades (IPCC 2007).Although further warming is inevitable, the chal-lenge has become checking human-induced in-creases in global mean temperature enough toavoid the most catastrophic consequences. Thisrequires deep-rooted cuts in the global releaseof carbon dioxide (CO2) emissions—particularlyin industrialized countries, such as theUnited Kingdom.

In its climate change bill (OPSI 2008), theUnited Kingdom was the first country to com-mit to a set of legally binding reduction tar-gets, including greenhouse gas emission reduc-tions, through action in the United Kingdomand abroad, of at least 80% by 2050. In addi-tion, the bill aims for CO2 emission reductionsof at least 26% by 2020, against a 1990 base-line. There is agreement in the sustainable con-sumption literature (Lintott 1998; Charkiewiczet al. 2001; Princen et al. 2002) and in the policycommunity (HM Government 2005, 2006; Sus-tainable Consumption Roundtable 2006) thatthis reduction will require substantial increasesin the carbon efficiency of production pro-cesses and changes in the way people live andconsume.

“Lifestyle analysis” provides a systemic viewof the entire sociotechnological system that linksconsumption and induced production activitiestogether in one analytical framework. Lifestylesthemselves are usually seen to be reflected in theconsumption patterns of societal groups with dif-ferent socioeconomic characteristics, such as so-cial identity, education, employment, or familystatus (Hertwich and Katzmayr 2004). Lifestyle-related emissions therefore include the carbonemissions consumers release directly from heat-ing their homes and driving their cars but alsoall the indirect emissions released throughoutthe global supply chain from the production ofthe final goods and services purchased. Becauselifestyle analysis accounts for both kinds of emis-sions, it can generate a comprehensive carbon

output measure for a set of consumer behaviorsattached to a particular lifestyle.

Environmental input−output models havebeen used most frequently for the assessment ofthe direct and indirect carbon emissions associ-ated with different lifestyles (e.g., Symons et al.1994; Duchin 1998; Labandeira and Labeaga1999; Weber and Perrels 2000; Wier et al.2001; Pachauri and Spreng 2002; Lenzenet al. 2004b; Bin and Dowlatabadi 2005; Cohenet al. 2005; Hertwich 2005; Hubacek and Sun2005; Tukker et al. 2005; Tukker and Jansen2006; Lenzen et al. 2008; Ornetzeder et al. 2008;Hubacek et al. 2009). This standard environmen-tal input−output-based lifestyle approach canalso be challenged on various grounds, however.Four lines of criticism are briefly summarized here.

The first line of criticism is directed towardthe treatment of import-related CO2 emissions inlifestyle-related studies. Because lifestyle analysisis interested in all CO2 emissions associated withthe consumption patterns of a particular groupof households (e.g., within a country), studiesconventionally compile emission inventories onthe basis of the principle of consumer responsi-bility. These inventories include import-relatedand exclude export-related CO2 emissions (seeMunksgaard and Pedersen 2001; Munksgaardet al. 2009; Lenzen et al. 2007; Peters 2008).Due to limited data availability and the com-plexity of the task, however, most studies haveassumed that the structure of the economy andthe sectoral CO2 intensities are the same abroadas at home (Weber and Perrels 2000; Lenzenet al. 2004a; Bin and Dowlatabadi 2005; Cohenet al. 2005; Druckman and Jackson 2008). Lenzenand colleagues (2004a) have shown that such a“single-region assumption” can lead to a sig-nificant estimation error (Munksgaard et al.2005; Wiedmann et al. 2007; Wiedmann 2009).A proposed way forward is to base estima-tions on a multiregional input−output model(Lenzen et al. 2004a; Munksgaard et al. 2009;Peters and Hertwich 2008; Weber and Matthews2008; Minx et al. 2009; Wiedmann et al.2010).

The second line of criticism is directed to-ward the purely expenditure-based characteriza-tion of a lifestyle. Schipper and colleagues (1989)highlight the need for a redefinition of lifestyle in

Baiocchi et al., The Impact of Social Factors and Consumer Behavior on CO2 Emissions in the UK 51


terms of human activity patterns, with a focus onwhat people do rather than on what they spend.In this context, time-use data have been pro-posed as a complement for operationalizing such ahuman-activity-based approach to lifestyle anal-ysis (Minx and Baiocchi 2009). These consid-erations have been recently incorporated in theinput−output literature (Brodersen 1990; Jalas2002, 2005; Stahmer 2004; Schaffer and Stah-mer 2005; Kondo and Takase 2007; Minx andBaiocchi 2009).

The third line of criticism addresses the waylifestyle groups are conventionally classified ininput−output studies. Most frequently, lifestylegroups are identified on the basis of a small setof preselected variables. These variables are typi-cally closely related to the occupation and associ-ated socioeconomic status of individuals (Duchin1998). The most important problem of such atop-down classification is that it fails to recognizespatial aspects associated with lifestyles and thatit often includes only a very limited number ofsocioeconomic background variables for furtheranalysis.

Marketing practitioners have long recognizedthese problems and have adopted a geodemo-graphic approach in the classification of lifestylesto better understand consumer behaviors. Geode-mographics might be best defined as the “analy-sis of people by where they live” (Harris et al.2005, 2). The rationale behind geodemographicsis that places and people are inextricably linked.Knowledge about the whereabouts of people re-veals information about them. Such an approachhas been shown to work well, because people withsimilar lifestyles tend to cluster—a longstandingtheoretical and empirical finding in the socio-logical literature (Schelling 1969; Harris et al.2005; Pancs and Vriend 2007; Vickers and Rees2007).

Geodemographic lifestyle classifications arebuilt in a bottom-up procedure based on a largeset of spatially specific variables that cover char-acteristics of both people and places. Lifestylegroup names, such as “villages with wealthy com-muters,” “affluent urban professionals, large flats,”or “single elderly people, high-rise flats,” are a re-flection of this. In the context of environmen-tal research, such classifications seem superiorto conventional “occupation-based” systems, as

they account for important aspects of the imme-diate physical environment in which people oper-ate, which have considerable impact on people’semission patterns. For example, people in ruralareas can afford bigger houses—which often havegreater heating requirements (Boardmann 2007).Access to public transport and to other privateand public services, together with the distanceto shops, the commuting requirement, or the ageand the condition of the housing stock, are otherimportant neighborhood-specific determinants ofcarbon emissions associated with a lifestyle. Re-gardless of the appeal, however, so far, only veryfew studies have started to explore the poten-tial of geodemographic data for lifestyle analysis(e.g., Duchin 1998; Duchin and Hubacek 2003;Druckman and Jackson 2008; Druckman et al.2008; Minx et al. 2009).

A final line of criticism can be directed towardthe fact that most input−output-based lifestylestudies remain purely descriptive in their analysisof the results. Simply estimating the level of emis-sions associated with different consumption pat-terns across societal groups might provide impor-tant insights, but unless links can be establishedbetween emissions and different socioeconomicfactors, such as education, income, or gender, itwill be difficult to improve our understanding ofthe relationship between lifestyles and emissions.Even though researchers have made more andmore attempts to further investigate this issue onthe basis of univariate (e.g., Lenzen 1994; Vringerand Blok 1995; Cohen et al. 2005; Hertwichand Peters 2009) and multivariate regressionmodels (e.g., Weber and Perrels 2000; Ferrer-iCarbonell and van den Bergh 2004; Lenzen et al.2006; Weber and Matthews 2008), no study sofar has established the importance of lifestyles inexplaining emissions and accounted for them inits empirical investigations of emissions and theirdeterminants.

Although we have addressed the first two setsof criticisms elsewhere (e.g., Minx and Baioc-chi 2009; Minx et al. 2009), this study addsto the last two strands of literature and at-tempts to make the following contributions: Weuse geodemographic consumer segmentation dataprovided by the ACORN database for our analy-sis of the CO2 emissions associated with differentlifestyles in the United Kingdom. This adds to the

52 Journal of Industrial Ecology


literature a very detailed carbon account for 56lifestyle types and allows a general investigationof the environmental significance of lifestyles (seealso Duchin 1998; Duchin and Hubacek 2003;Druckman and Jackson 2008; Druckman et al.2008).

By applying a regression approach, we are ableto exploit the unique geodemographic nature ofthe data to identify the impact of lifestyles on therelationship between emissions and a set of im-portant socioeconomic determinants. By allow-ing for “lifestyle effects,” we are able to controlnot only aspects of the immediate physical en-vironment in which people operate that haveconsiderable effect on their environmental im-pact but also for aspects of economic and socialbehavioral differences related to lifestyles, suchas individual aspirations and people’s attitudestoward the environment (e.g., toward consump-tion, education, housing, transport, and leisureactivities). We finally provide a visual represen-tation of where the various socioeconomic factorsexert the greatest impact and discuss the applica-tions of our approach to sustainable consumptionpolicy.

In the next section, we describe the un-derlying input−output model and data, fur-ther explain the nature of the geodemographiclifestyle data used in this study, and discussthe results on emissions associated with differ-ent lifestyles. In the subsequent section, we de-scribe and apply a panel data approach to de-termine the impact of socioeconomic factors onemissions. Conclusions are provided in a finalsection.

Carbon Emissions of LifestyleTypes in the United Kingdom

Input−Output Model

The total CO2 emissions, phh,tot, from house-hold consumption of s different lifestyle groupscan be expressed most generally as the sumof their direct (phh,dir) and indirect emissions(phh,ind),

phh,tot = phh,dir + phh,ind (1)

The direct CO2 emissions, phh,dir, are associ-ated with domestic energy consumption and pri-

vate transport. We obtain lifestyle-group-specificestimates by assigning direct emissions of allhouseholds across each of the s lifestyle groupsproportionally to their energy and transport ex-penditures.

We can calculate the indirect emissions bymultiplying a vector of total CO2 intensities,ε i nd , of n different production sectors with Yhh

= [ykl], a matrix of detailed household con-sumption expenditures of the s different lifestylegroups in m functional spending categories—thatis,

phh,ind = (ε ind)′Yhh = (ε ind)′AhhYhh,soc (2)

Ahh = [aik] = yik∑nj =1 yjk

is an n × m matrix of di-

rect coefficients indicating the proportion of finalhousehold demand for products provided by then different sectors across the m different func-tional spending categories. Yhh,soc is a matrix ofhousehold consumption expenditures of the s dif-ferent socioeconomic groups in the m spendingcategories.

The vector of indirect CO2 intensities, ε ind,from the n different sectors is derived from a sup-ply and use model (Miller and Blair 2009).1 Thisvector can be estimated as follows:

ε ind = r′(I − DB)−1D (3)

where r is a vector of sectoral direct CO2 inten-sities indicating the amount of CO2 emitted perunit of final output of the n different sectors, In isan identity matrix of order n, D = [dji] = vji

qiis a

coefficient matrix of size n × n based on an indus-try technology assumption indicating the supplyvji of commodity i to industry j per unit of totalsupply qi of commodity i, and B = [bij] = uij

x jis a

technical coefficient matrix of size n × n provid-ing information about the use uij of commodityi by industry j per unit of output xj of industryj. We used this method to estimate the directand indirect CO2 emissions from the differentlifestyle groups as presented in the ACORN con-sumer segmentation database. The approach isdescribed in detail by Wiedmann and colleagues(2006).



Input−Output and GeodemographicData

Input−Output DataFor the input−output estimations, we used

supply and use tables provided by the Office forNational Statistics for 2000 (ONS 2003b), incombination with sectoral CO2 data from the UKEnvironmental Accounts for 2000 (ONS 2005a).All calculations were carried out at the 76-sectoraggregation level.

To use the published supply and use tablesfor a modeling application, we had to undertakea variety of steps, including price conversions,updating, and rebalancing procedures. The stepsare described in detail by Wiedmann and col-leagues (2006).2 Furthermore, we used officiallypublished statistics to break down the householdconsumption expenditure vector into 39 func-tional spending categories, following the Classi-fication of Individual Consumption by Purpose(COICOP; ONS 2003b).

Geodemographic DataLifestyles are distinguished according to

CACI’s ACORN classification.3 ACORN standsfor “a classification of residential neighborhoods.”The classification includes every street in thecountry and groups spatial areas according to so-cioeconomic profiles of the residents. The mostrecent ACORN classification distinguishes 17household groups and 56 household types (seetable 1).4

ACORN data are designed for companies tounderstand consumers and their wants. Morethan 400 variables were used to build ACORN ina cluster analysis procedure and help to describethe different ACORN types. More than 30%of the data used for this purpose were sourcedfrom the 2001 census. These data are geograph-ically coded down to the postal code level. Theremainder were derived from CACI’s consumerlifestyle databases, which cover all of the UnitedKingdom’s 46 million adults in 23 million house-holds and are built from other private and publicsurvey data (CACI 2004; ONS 2005c).

Central to the estimations here is a func-tional consumer expenditure table, which pro-vides weekly per-household spending estimatesacross 34 COICOP consumption categories for

ACORN categories, groups, and types. This ta-ble has been built from the United Kingdom’sFamily Expenditure Survey for 2004. One canderive total expenditure figures by multiplyingthis functional spending table by the numberof households in each ACORN category, group,and type. This table can then be linked with theinput−output data.

To reconcile the ACORN functional spend-ing with the input−output household final de-mand matrix, we needed to use imputation to fillthe data gaps in the CACI data for COICOP cate-gories “purchase of vehicles” (7.2), “accommoda-tion services” (11.2), “social protection” (12.4),and “insurance” (12.5). In the absence of betterinformation, we roughly grouped the 56 ACORNtypes into income deciles and used respective esti-mates from the United Kingdom’s Family Expen-diture Survey for gap-filling (ONS 2005b). This islikely to lead to some error, as the ACORN typesare less homogenous in their income profile. Allspending estimates were calibrated to the spend-ing levels of 2000 in each COICOP category.Hence, the estimates reflect the spending levelsof 2000 and the composition of consumption bas-kets of 2004. Finally, to derive a spatial pictureof CO2 emissions associated with lifestyles in dif-ferent areas of the United Kingdom, we used thesmall-area household and population estimatesfrom CACI’s ACORN data set.5

Data LimitationsThe major sources of uncertainty in the

input−output data are associated with the re-quired price conversions and updating proce-dures. A recent Monte Carlo experiment byWiedmann and colleagues (2008) has shown thatthis is unlikely to heavily affect estimates at highaggregation levels, as errors tend to cancel out,even though estimates at the detailed sectorallevel can show considerable uncertainties undercertain circumstances.6 Therefore, we do not per-ceive these sources of uncertainty to be a ma-jor obstacle for the current analysis, as all theinput−output results reported in this study arehighly aggregated.

Further uncertainties are introduced by thesingle-region technology assumption applied inthe estimation of the CO2 emissions embod-ied in imported goods and services. We adjusted



Tabl

e1

AC

OR

Nco

nsum

ercl

assifi

catio

n

AC

OR

Ngr

oup

AC

OR

Nty

peA

CO

RN

grou

pA

CO

RN

type

1–

Wea

lthy

exec

utiv

es1

–W

ealt

hym

atur

epr

ofes

sion

als,

larg

eho

uses

2–

Wea

lthy

wor

king

fam

ilies

wit

hm

ortg

ages

3–

Vill

ages

wit

hw

ealt

hyco

mm

uter

s4

–W

ell-

offm

anag

ers,

larg

eho

uses

9–

Sett

led

subu

rbia

32–

Ret

ired

hom

eow

ners

33–

Mid

dle

inco

me,

olde

rcou

ples

34–

Low

erin

com

e,ol

derc

oupl

es,s

emis

2–

Affl

uent

gray

s5

–O

lder

afflu

entp

rofe

ssio

nals

6–

Farm

ing

com

mun

itie

s7

–O

ldpe

ople

,det

ache

dho

uses

8–

Mat

ure

coup

les,

smal

lerd

etac

hed

hom

es

10–

Prud

entp

ensi

oner

s35

–El

derl

ysi

ngle

s,pu

rpos

e-bu

iltfla

ts36

–O

lder

peop

le,fl

ats

3–

Flou

rish

ing

fam

ilies

9–

Old

erfa

mili

es,p

rosp

erou

ssub

urbs

10–

Wel

l-of

fwor

king

fam

ilies

wit

hm

ortg

ages

11–

Wel

l-of

fman

gers

,det

ache

dho

uses

12–

Larg

efa

mili

esan

dho

uses

inru

rala

reas

11–

Asi

anco

mm

unit

iesa

37–

Cro

wde

dA

sian

terr

aces

38–

Low

-inc

ome

Asi

anfa

mili

es

4–

Pros

pero

uspr

ofes

sion

als

13–

Wel

l-of

fpro

fess

iona

ls,l

arge

rhou

ses

14–

Old

erpr

ofes

sion

alsi

nsu

burb

anho

uses

12–

Post

indu

stri

alfa

mili

es39

–Sk

illed

olde

rfam

ilies

,ter

race

s40

–Y

oung

wor

king

fam

ilies

5–

Educ

ated

urba

nite

s15

–A

fflue

ntur

ban

prof

essi

onal

s,fla

ts16

–Pr

ospe

rous

youn

gpr

ofes

sion

als,

flats

17–

You

nged

ucat

edw

orke

rs,fl

ats

18–

Mul

tiet

hnic

youn

g,co

nver

ted

flats

19–

Subu

rban

priv

atel

yre

ntin

gpr

ofes

sion

als

13–

Blu

e-co

llarr

oots

41–

Skill

edw

orke

rs,s

emis

and

terr

aces

42–

Hom

e-ow

ning

fam

ilies

,ter

race

s43

–O

lder

peop

le,r

ente

dte

rrac

es

6–

Asp

irin

gsi

ngle

s20

–St

uden

tflat

sand

cosm

opol

itan

shar

ers

21–

Sing

lesa

ndsh

arer

s,m

ulti

ethn

icar

eas

22–

Low

-inc

ome

sing

les,

smal

lren

ted

flats

23–

Stud

entt

erra

ces

14–

Stru

gglin

gfa

mili

es44

–Lo

win

com

ela

rger

fam

ilies

,sem

is45

–Lo

win

com

eol

derp

eopl

e,sm

alle

rsem

is46

–Lo

win

com

e,ro

utin

ejo

bs,t

erra

cesa

ndfla

ts47

–Lo

win

com

efa

mili

es,t

erra

ced

esta

tes

48–

Fam

ilies

and

sing

lepa

rent

s,se

mis

and

terr

aces

49–

Larg

efa

mili

esan

dsi

ngle

pare

nts,

man

ych

ildre

n

7–

Star

ting

out

24–

You

ngco

uple

s,fla

tsan

dte

rrac

es25

–W

hite

-col

lars

ingl

esan

dsh

arer

s,te

rrac

es15

–B

urde

ned

sing

les

50–

Sing

leel

derl

ype

ople

,cou

ncil

flats

51–

Sing

lepa

rent

sand

pens

ione

rs,c

ounc

ilte

rrac

es52

–Fa

mili

esan

dsi

ngle

pare

nts,

coun

cilfl

ats C

onti

nued

.



Tabl

e1

Con

tinue

d

AC

OR

Ngr

oup

AC

OR

Nty

peA

CO

RN

grou

pA

CO

RN

type

8–

Secu

refa

mili

es26

–Y

oung

erw

hite

-col

larc

oupl

esw

ith

mor

tgag

e27

–M

iddl

e-in

com

e,ho

me-

owni

ngar

eas

28–

Wor

king

fam

ilies

wit

hm

ortg

ages

29–

Mat

ure

fam

ilies

insu

burb

anse

mis

30–

Esta

blis

hed

hom

e-ow

ning

wor

kers

31–

Hom

e-ow

ning

Asi

anfa

mili

esa

16–

Hig

h-ri

seha

rdsh

ip53

–O

ldpe

ople

,man

yhi

gh-r

ise

flats

54–

Sing

lesa

ndsi

ngle

pare

nts,

high

-ris

ees

tate

s

17–

Inne

rcit

yad

vers

ity

55–

Mul

tiet

hnic

purp

ose-

built

esta

tes

56–

Mul

tiet

hnic

,cro

wde

dfla

ts

Not

e:Fl

ats

are

apar

tmen

ts,s

emis

are

sem

idet

atch

edho

uses

,and

terr

aces

are

row

hous

ing.

a Alt

houg

hA

CO

RN

clas

sific

atio

nus

esth

eet

hnic

labe

lAsia

nto

help

desc

ribe

Gro

up11

and

apo

rtio

nof

Gro

up8,

we

wan

tto

note

that

the

term

isve

rybr

oad

and

does

not

defin

ea

pers

on’s

lifes

tyle

,nor

does

itpr

edic

ton

e’s

lifes

tyle

-rel

ated

CO

2em

issi

ons.

Alt

houg

hth

eA

CO

RN

clas

sific

atio

nin

dica

tes

that

indi

vidu

als

who

fall

into

the

note

dgr

oupi

ngs

tend

tobe

Asi

an,o

ne’s

bein

gA

sian

does

notn

eces

sari

lym

ean

incl

usio

nin

thes

epa

rtic

ular

nam

edA

CO

RN

grou

ping

s.

sectoral direct CO2 intensities to reflect globalaverages for imported goods and services, how-ever, as explained by Wiedmann and colleagues(2006). The resulting estimates are very close tothe ones derived in a recent UK analysis that useda multiregional input−output model (see Wied-mann et al. 2010). Other sources of uncertaintiesthat are typical for this kind of input−outputestimation are discussed in work by Suh and col-leagues (2004).

Uncertainties in the ACORN data are associ-ated with the imputation of missing information.However, imputation was only necessary in 4 ofthe 39 functional spending categories (mainlyservices and transport equipment), and is not ex-pected to affect the results in a way that wouldalter the main conclusions of the article. Othersources of uncertainties are difficult to quantify,as this type of commercial information is not aswell documented as official government statis-tics. This is certainly an important limitation ofcommercial geodemographic data (Harris et al.2005). Still, expenditure estimates should be ofreasonable quality, as the Office for NationalStatistics appended ACORN codes to each in-dividual observation of the UK Family Expen-diture Survey, making use of the best availableinformation.7

For spatial emission estimates, we need to as-sume that there is no regional or local variationin the expenditure profiles. Households belong-ing to the same type are presumed to have thesame spending patterns no matter where they arelocated in the territory. This is bound to be acrude but, we hope, useful approximation of real-ity. Minx and colleagues (2009) explain how thisassumption can be overcome through the inclu-sion of regional and local data.

Results

Table 2 shows that in 2000, all final demandactivities in the United Kingdom were responsi-ble for 681 million tonnes of CO2 emissions glob-ally.8 This is very close to the 682 million tonnesreported by Wiedmann and colleagues (2010)and is approximately 11% higher than territorialemissions in the United Kingdom, as reportedin the Environmental Accounts (ONS 2005a).Seventy-five percent, or 505 of the 681 million



tonnes of CO2 emissions from UK consumption,is directly or indirectly related to private house-holds. This explains much of the attention this is-sue has received in the literature. Of this amount,70%, or 358 million tonnes, occurs in the globalsupply chains of products consumed in the UnitedKingdom, whereas the remaining 30%, or 147million tonnes, is directly emitted by households:86 million tonnes of CO2 through domestic en-ergy use, and 61 million tonnes for fueling privatevehicles.

In table 3, we provide a more detailedoverview of how carbon emissions from house-hold consumption of 17 ACORN groups breakdown into basic emission components. We dis-tinguish direct, indirect, and total CO2 emis-sions and express these emission figures in ab-solute, per household, and per capita terms. Thesociodemographic characteristics are taken fromthe ACORN user guide (CACI 2004). Each char-acteristic is expressed as an index, where 100 rep-resents the UK average.9 As one moves from theleft to the right in table 3, ACORN groups tendto become less wealthy and live in more urbanareas.

In absolute terms, the most CO2 is producedby Group 8, “secure families.” This is also thelargest ACORN group, however; it comprises3.64 million households. When we express re-sults from the input−output model in per house-hold terms, Group 11, “Asian communities,”is identified as living the most CO2-intensive

Table 2 Total consumer emissions in million tonnes(megatons, or Mt) of carbon dioxide (CO2) by finaldemand (FD)

EmissionsFD category (in Mt)

Domestic energy consumption 86.2(household direct)

Private transport 61.3(household direct)

Household consumption 357.9(household indirect)

Government 61.9Capital investment 109.2Other final demand 4.7

Total 681.3

lifestyle. This is somewhat surprising, as thisACORN group comprises households with com-paratively moderate means (see the income in-dex in table 3), but the finding is explained bythe group’s higher transport emission component,which stems from a high demand for transport.10

Other high-impact households tend to come fromwealthier backgrounds and live in bigger housesin rural areas, such as ACORN Group 1, “wealthyexecutives,” which is also the richest group inthe sample, or ACORN Group 3, “flourishingfamilies.”

Once we express the total CO2 emissionsof each lifestyle group in per capita terms, thepicture changes again, as the number of house-hold members tends to be higher in rural areasthan in urban areas. The large household sizealso partially explains the high per householdemissions of Asian communities, which have thelargest households in the sample, with 3.27 mem-bers on average. Even though their per capitaCO2 impact still remains high, some wealthiergroups have higher impacts, such as Group 4,“prosperous professionals,” or Group 5, “educatedurbanites.”

The variation in CO2 emissions from con-sumption across ACORN groups is substantial. Inper capita terms, the difference between the high-est (Group 5, “educated urbanites”) and lowest(Group 14, “struggling families”) emitting groupis almost a factor of 2, and on a per household ba-sis it is even a factor of 3 (Group 11, Asian com-munities, and Group 16, “high-rise hardship”).11

Among the 12 functional spending categories,we find that transportation and housing havethe largest impacts. This is in agreement withevidence from other input−output studies (e.g.,Tukker et al. 2005; Tukker and Jansen 2006).

Across lifestyle groups, the share of transport-related emissions seems to decrease when onemoves from the wealthier, rural lifestyles on theleft to the poorer, urban lifestyles on the rightin table 3, with the exception of Group 11,“Asian communities.” On the one hand, peo-ple in urban environments have less need forlong-distance travel.12 On the other hand, poorerlifestyle groups simply might not be able to af-ford a car or long recreational trips by plane.The opposite trend is found for housing-relatedemissions: Housing-related CO2 emissions are



Tabl

e3

Sum

mar

yst

atist

ics

for

AC

OR

Ngr

oups

AC

OR

Ngr

oup

12

34

56

78

910

1112

1314

1516

17

Emis

sion

Indi

rect

a36

.528

.031

.99.

922

.814

.59.

251

.819

9.6

6.9

14.5

23.7

37.3

12.6

4.9

7.5

com

pone

ntD

omes

tic

ener

gya

8.2

7.6

7.7

2.2

5.1

4.1

2.3

12.4

5.1

2.6

2.2

3.7

6.7

9.5

3.2

1.2

2.3

Priv

ate

tran

spor

t6.

9a5.

96.

21.

73.

22.

21.

710

.03.

91.

80.

72.

74.

46.

52.

00.

70.

8T

otal

CO

2a51

.641

.445

.913

.731

.120

.813

.274

.228

.014

.09.

920

.934

.953

.417

.86.

810

.6

CO

ICO

Pgr

oup

1−

Food

and

non-

alco

holic

drin

ks4.

24.

44.

54.

03.

54.

14.

34.

74.

94.

43.

24.

94.

95.

45.

35.

04.

3(%

ofto

tal)

2−

Alc

ohol

icdr

inks

,tob

acco

and

narc

otic

s0.

80.

80.

90.

80.

80.

90.

90.

90.

90.

90.

40.

90.

91.

01.

11.

10.

73

−C

loth

ing

and

foot

wea

r1.

41.

41.

51.

41.

41.

41.

51.

61.

51.

41.

31.

61.

61.

71.

61.

61.

64

−H

ousi

ng(n

et),

fuel

and

pow

er32

.136

.634

.133

.334

.440

.735

.834

.337

.137

.644

.236

.439

.839

.340

.440

.646

.95

−H

ouse

hold

good

sand

serv

ices

5.4

5.7

5.7

5.1

5.1

5.3

5.8

6.0

6.3

5.8

3.6

6.4

6.5

7.1

7.3

7.3

5.0

6−

Hea

lth

0.7

0.8

0.7

0.8

0.6

0.7

0.7

0.7

0.7

0.7

0.4

0.7

0.6

0.6

0.6

0.5

0.5

7−

Tra

nspo

rt32

30.2

30.7

31.8

30.7

27.2

30.1

30.2

28.3

29.1

34.9

28.3

26.1

25.4

2525

.622

.88

−C

omm

unic

atio

n0.

70.

80.

80.

80.

80.

90.

90.

90.

90.

90.

70.

90.

91.

01.

01.

01.

09

−R

ecre

atio

nan

dcu

ltur

e6.

66.

66.

85.

95.

15.

96.

47.

17.

16.

23.

97.

16.

87.

37.

16.

65.

110

−Ed

ucat

ion

1.0

0.6

0.7

0.9

1.0

0.9

0.6

0.7

0.5

0.6

0.4

0.6

0.5

0.4

0.4

0.4

0.7

11−

Res

taur

ants

&ho

tels

5.6

4.7

5.4

5.8

6.1

4.9

5.2

5.3

5.1

5.0

3.2

5.3

5.1

5.1

5.0

4.9

4.8

12−

Mis

cella

neou

s9.

57.

38.

29.

410

.47.

17.

87.

76.

87.

63.

86.

96.

45.

75.

25.

46.

7

Hou

seho

ldsb

1,94

51,

978

1,98

656

61,

471

1,03

372

63,

644

1,63

780

029

41,

086

2,12

93,

383

1,35

861

556

6to

fCO

2pe

r26

.521

.023

.124

.321

.220

.218

.220

.417

.117

.533

.519

.316

.415

.813

.111

.118

.7ho

useh

old

Popu

lati

onb

5,17

04,

574

5,27

11,

339

2,80

22,

363

1,48

09,

219

3,56

91,

553

963

2,84

64,

713

8,32

22,

691

949

1,29

1C

O2

perc

apit

ac10

.09.

18.

710

.311

.18.

88.

98.

07.

99.

010

.27.

47.

46.

46.

67.

28.

2In

com

ed16

511

213

315

514

497

107

108

8598

7392

7461

5349

82Ed

ucat

iond

166

116

116

225

241

116

138

8969

119

7165

5836

3848

114

Une

mpl

oym

entd

4757

5764

117

140

7970

6582

192

9511

615

818

123

622

6

Con

tinu

ed.



Tabl

e3

Con

tinue

d

AC

OR

Ngr

oup

12

34

56

78

910

1112

1314

1516

17

Larg

edw

ellin

gd30

417

819

218

353

6153

9675

8277

6650

3722

1219

Smal

ldw

ellin

gd18

2821

9039

423

112

733

3716

613

133

7263

167

369

403

Soci

alho

usin

gd11

2118

2383

9932

3342

7410

473

7723

930

736

334

1Sh

arin

gd61

6264

153

297

270

154

7551

8020

098

8573

6265

192

Nat

iona

lTru

std

187

184

141

170

9061

9510

611

415

530

8365

4944

4538

Not

e:Fo

rAC

OR

Ngr

oups

,see

tabl

e1.

a 106

tonn

es(t

)C

O2.

b Tho

usan

ds.

c tonn

es(t

)C

O2/

capi

ta.

d UK

aver

age

=10

0.

more prominent in the consumption patterns ofpoorer groups. This is closely related to the issueof fuel poverty, which has been high on the polit-ical agenda in the United Kingdom (Dresner andEkins 2005).

The descriptive analysis as presented abovehas clear limits in revealing the influence of in-dividual factors underlying people’s lifestyles un-ambiguously. In the next section, we thereforefurther analyze the results of our lifestyle analysisin a regression approach.

Lifestyles and the Impact ofSocioeconomic Factors onEmissions

Regression Approach

In this section, we outline the methodologyused to estimate the relationship between house-hold CO2 emissions by ACORN type, obtainedin the previous section, and several socioeco-nomic factors that characterize these households.In the Income Elasticity and Spatial Data subsec-tion, we will illustrate how the estimated rela-tionship, jointly with the additional geographi-cal information available about the location ofhousehold types, can be used to provide a vi-sually clear representation of where the varioussocioeconomic factors exert the greatest impacton the map. This analysis hinges on the spe-cific nature of the geodemographic lifestyle dataused, which have been designed to enable firmsto understand people’s preferences and consump-tion activities in the context of their everydaylife.13 The ACORN data are arranged in a cross-sectional data set in which each household typeobservation belongs to a well-defined group (re-fer back to the Input−Output and GeodemographicData subsection for details). Due to the specificway the data are constructed, we can interpretthe different ACORN groups, such as wealthyexecutives or educated urbanites (see table 1), aslifestyles. Although in principle we can analyzethese data using ordinary regression, we want toemphasize the data’s special features for makinguse of the additional information content that isspecific to the geodemographic lifestyle classifi-cation. We can treat the ACORN data, for thepurpose of analysis, as a panel or longitudinal data



set14 (for simplicity, referred to as panel here), inthat lifestyle groups are observed repeatedly. Wecan exploit the variation of types within lifestylegroups to improve on the standard regression ap-proach. The general form of the panel data modelthat incorporates heterogeneity among house-hold types is given by

ln Eit = αi +p∑

j =1

β j (ln X0it )j

+k−p∑

j =1

β j +p ln X jit + εit

(4)

with i = 1,..., N and t = 1,..., Ti represent-ing groups and types, respectively. Here, Eit isa measure of household CO2 emissions, X0it isincome, and X jit , for j = 1,..., k − p, denote theother sociodemographic control variables fromthe ACORN database that correspond to addi-tional determinants of emissions. p is the order ofthe polynomial in the log of income, k is the totalnumber of regressors, and εit is the classical errorterm.

From a statistical viewpoint, incorporatingheterogeneity can mitigate the problem of omit-ted variable bias, which occurs when relevant re-gressors are excluded from the model (see, e.g.,Verbeek 2008, 358–359). Allowing for group- orlifestyle-specific parameters αi provides a way to“control” for heterogeneity among households,including economic and social behavioral differ-ences related to lifestyles, such as individual as-pirations and people’s attitudes toward the envi-ronment. Also, if all other things are equal, paneldata methods yield more efficient estimators (see,e.g., Diggle et al. 2002, 24–26). In particular, thehigher the positive correlation between obser-vations is, the more efficient the estimator is.Because geodemographic classification is basedon the principle of (positive) spatial autocorrela-tion (i.e., residents of the same neighborhood aretaken to share several common socioeconomicand behavioral characteristics), the gains in ef-ficiency from the use of a panel method can besubstantial.

The functional specification is inspired by theliterature on the environmental Kuznets curve(EKC; Grossman and Krueger 1993). The EKChypothesis envisages an inverted-U-shaped re-

lationship between income and environmentaldegradation. EKC models have been estimatedin either logarithmic, quadratic, or cubic polyno-mial functional form specifications between pol-lutant concentration and per capita income (fora survey, see, e.g., Panayotou 2000; Stern 2004).Although the method is usually applied in cross-country studies, it is used here for the provisionof an intrasocietal perspective on the relation-ship between CO2 emissions and income. By us-ing this approach, we acknowledge the particularimportance of income in the determination ofCO2 from households (e.g., Ferrer-i Carbonelland van den Bergh 2004; Lenzen et al. 2006)in the model specification while also control-ling for a wide range of other sociodemographicdeterminants.

Regressions that incorporate heterogeneityare typically estimated with fixed-effects (FE) andrandom-effects (RE) models. The FE model treatsthe αi as regression parameters, whereas the REmodel considers them components of random er-ror. To decide between fixed and random ap-proaches, one can use the Hausman statistic,which tests the hypothesis that the regressorsare correlated with the individual effect. If theindividual effects, αi, are correlated with the re-gressors, the RE estimator is inconsistent (for fur-ther details on panel methods, see, e.g., Verbeek2008).

Emission Determinants

Socioeconomic variables that can affect CO2

emissions in the United Kingdom were obtainedfrom the ACORN data set. The variables in-cluded in the final regression were selected froma larger set of possible determinants belonging tothe wider categories of housing (one to two rooms,seven or more rooms, etc.), families (couplewith children, single parent, etc.), education (de-gree or equivalent, no qualifications, etc.), work(self-employed, looking for work, etc.), finance(average family income, have mortgage, etc.),and so on,15 through the widely used backwardelimination statistical procedure, as describedin standard applied regression references, suchas the work of Draper and Smith (1998, 339–340) or Weisberg (2005, 222), and theoreticalconsiderations.16



Table 4 Description and summary statistics of regression variables

Standard UKVariable name Label Unit Mean Deviation Minimum Maximum average

Total CO2 per household CO t/hh 20.30 5.81 10.58 44.58 19.36t/hha

No. individuals HHSIZE Count 2.37 0.46 1.40 3.93 2.36Average family income INC Indexb 102.11 37.9 47 188 £447c

Degree or equivalent EDU Indexb 108.50 66.4 27 302 20%f

Large houses (7+ rooms) BHO Indexb 138.20 111.9 16 451 20%f

Families with children FWC Indexb 98.12 44.46 13 225 21%f

Pensioners PENS Indexb 94.59 45.20 28 225 23%f

Single nonpensioner SING Indexb 109.66 62.6 37 320 16%f

Use Internet for e-mail INTER Indexb 106.16 39.53 32 199 55%df

Social housing SOCH Indexb 106.86 114.55 5 370 20%f

National Trust member NTRUST Indexb 97.70 55.20 21 240 0.013%ef

aFrom table 3.bUK average = 100.cAverage weekly disposable income from UK Expenditure and Food Survey (ONS 2005b).dFrom UK households with access to the Internet (ONS 2008).eDerived from membership of selected environmental organizations (ONS 2003a).f Percentage of UK households represented by each category.

The most important criticism of any selectionstrategy of this kind, which one should keep inmind when interpreting the regression findings,is that it can overstate the significance of the re-sults, as multiple testing is performed with thesame body of data. This problem is mostly re-ferred to as “data mining” in statistical literature(see, e.g., Lovell 1983). Wilkinson and Dallal(1981) showed with Monte Carlo that final re-gressions obtained by stepwise selection, said tobe significant at the 0.1% level, were in fact onlysignificant at the 5% level.

There is no generally accepted simple so-lution to this problem. The recommended ap-proach of splitting the available data into a subsetfor exploratory analysis and another for test-ing may only increase the chance of selectinga wrong model if too few observations are avail-able (Lovell 1983). We address this concern bytesting the resulting model for several specifica-tion problems. The importance of specificationtesting to recover the correct model has beenhighlighted in a Monte Carlo study by Hooverand Perez (1999). Also, our final model includesvariables that are significant at the 0.1% level.

The variables used in this study, together withdescriptive statistics for the sample, are reported

in table 4. The final sample consists of 56 obser-vations, determined by the maximum number ofhousehold types distinguished in the data set; theobservations fall (unevenly) into the 17 house-hold groups (see table 1) and are restricted to theyear 2000.17

A correlation matrix between determinantsof emissions is presented as a scatterplot ma-trix in figure 1. The figure clearly illustrates theimportance of using a regression approach. Wefind that emissions are positively correlated withhousehold size, having children, Internet use, in-come, and education and negatively correlatedwith being a pensioner and being single. It is im-portant to stress, however, that these are onlymarginal relationships—that is, they do not takeinto account the influence of other variables. Forinstance, education and the use of the Internetare highly correlated with each other and withincome; therefore, we need to “control” for in-come to determine the variables’ direct impacton emissions to avoid biased estimates.

Regression Estimation Results

Table 5 presents the results of our estima-tion of equation (4) with “pooled” ordinary least



Carbon Emissions (CO)

With children (FWC)

HH size (HHSIZE)

National Trust(NTRUST)

Education (EDU)

Family income (INC)

Use internet(INTER)

Single (SING)

Social housing (SOCH)

Pensioners (PENS)

Large houses (BHO)

Ca

rbo

n E

mis

sio

ns (

CO

)

With

ch

ildre

n (

FW

C)

HH

siz

e (

HH

SIZ

E)

Na

tio

na

l Tru

st(

NT

RU

ST

)

Ed

uca

tio

n (

ED

U)

Fa

mily

in

co

me

(IN

C)

Use

in

tern

et(

INT

ER

)

Sin

gle

(S

ING

)

So

cia

l h

ou

sin

g (

SO

CH

)

Pe

nsio

ne

rs (

PE

NS

)

La

rge

ho

use

s (

BH

O)

Figure 1 Pairwise correlation between regression variables (correlation times 100). Variables aretransformed in natural logs. Ellipses are added to the scatterplot matrix to visually display both the sign andthe magnitude of the correlation. The online version of this article uses blue for positive values and pink fornegative values; either way, intensity increases uniformly as the correlation value moves away from 0.

squares (OLS) and two-panel estimation meth-ods. Figures in parentheses below coefficients aret statistics.18

The observed F statistic for the joint signif-icance of the 17 household group effects is sig-nificant at any conventional level. We can con-clude that lifestyles are considerably importantand that panel methods are necessary to deter-mine the impact of socioeconomic factors on theemissions. The statistically significant Hausmanstatistic implies that only the FE model can beusefully interpreted.

Interpreting individual coefficients reveals in-teresting insights into the links between socioe-conomic factors and CO2 emissions in the UKeconomy and might provide important insightsfor the formation of demand-side policies for fac-

ing the climate change challenge. Due to thereasons outlined above, we limit the discussionto the results of the FE model. To conserve space,we limit the more detailed analysis to a fewfindings.19

Results for IncomeThe results show that the income variables

(LINC, LINC2, and LINC3) are all highly sig-nificant. We are unable to find an environ-mental Kuznets curve, as emissions monotoni-cally increase, with decreasing rates up to thesample mean income and increasing rates after-ward. Our findings seem to agree with a grow-ing body of theoretical and empirical evidencethat has cast doubt on the previous conjec-ture that the demand for environmental quality



Table 5 Results from panel regressions

Model OLS FE RE

Constant −50.28∗∗∗∗ (−3.66) −60.17∗∗∗∗ (−7.22)LINC 33.45∗∗∗∗ (3.68) 37.00∗∗∗∗ (6.35) 39.54∗∗∗∗ (7.22)LINC2 −7.23∗∗∗∗ (−3.62) −7.86∗∗∗∗ (−6.13) −8.45∗∗∗∗ (−6.28)LINC3 0·53∗∗∗∗ (3.60) 0.56∗∗∗∗ (6.05) 0.61∗∗∗∗ (6.20)LEDU −0.05 (−0.64) −0.16∗∗∗ (−3.47) −0.12∗∗ (−2.52)LFWC −0.38∗∗∗∗ (−6.41) −0.31∗∗∗∗ (−8·47) −0.31∗∗∗∗ (−8.14)LHHSIZE 2.45∗∗∗ (11.86) 1.96∗∗∗∗ (16·62) 2.06∗∗∗∗ (16.40)LSING 0.30∗∗∗ (4.90) 0.20∗∗∗∗ (6·15) 0.23∗∗∗∗ (6.39)LPENS 0.08∗ (1.85) 0.06∗∗ (2.62) 0.06∗∗ (2.56)LBHO 0.02∗ (1.98) 0.04∗∗∗∗ (4.64) 0.03∗∗∗∗ (4.65)LSOCH −0.07∗∗∗∗ (−4.46) −0.03∗∗∗ (−3.17) −0.04∗∗∗ (−3.63)LINTER −0.18∗∗ (−2.13) −0.17∗∗∗ (−3·60) −0.16∗∗∗ (−3.31)LNTRUST 0.08∗∗ (2.12) 0.10∗∗∗∗ (5.81) 0.09∗∗∗∗ (5.05)

Note: Robust t statistics are reported in parentheses. The symbols ∗, ∗∗, ∗∗∗, and ∗∗∗∗, denote significance levels, α,of 10%, 5%, 1%, and 0.1% respectively. Figures in parentheses beside coefficients are t statistics. The F statistic forthe joint significance of the 17 household group effects is 14.3∗∗∗∗. The Hausman statistic is 21.7∗∗. OLS = ordinaryleast squares; FE = panel fixed effect estimator; RE = panel random effects estimator; LINC = natural-log-transformedaverage family income; LEDU = natural-log-transformed education; LFWC = natural-log-transformed families withchildren; LHHSIZE = natural-log-transformed household size; LSING = natural-log-transformed single nonpensioners;LPENS = natural-log-transformed pensioners; LBHO = natural-log-transformed large houses; LSOCH = natural-log-transformed social housing; LINTER = natural-log-transformed use of Internet for e-mail; LNTRUST = natural-log-transformed National Trust member.

increases more than proportionally with income,so that emissions may rise at first with economicgrowth but eventually fall as income continuesto rise (see, e.g., Flores and Carson 1997; Mc-Connell 1997; Hokby and Soderqvist 2003). Alook at the description of ACORN’s wealthi-est households, such as wealthy executives, sug-gests that perhaps their demand for improvedenvironmental quality is expressed through de-mand for luxury homes in high-status subur-ban or rural neighborhoods, regular holidays inrelatively pollution-free locations, and engage-ment in golf and gardening—that is, these house-holds adopt a carbon-intensive lifestyle. Furtheranalysis is required to verify this hypothesis,however.

Coefficients from a regression model in whichthe dependent and independent variable of in-terest are in natural log form and linearly relatedto each other can be conveniently interpreted asthe average percentage change in the dependentvariable, corresponding to a percentage changein the independent variable (see, e.g., Verbeek2008, 55). In economics, such coefficients areknown as elasticities. In general, for nonlinearlyrelated log-transformed variables, one can obtain

the “impact” in terms of elasticity of a variableof interest on the dependent variable by takingthe differential of the regression equation, hold-ing all independent variables constant except therelevant regressor. It can be shown that the emis-sion elasticity of income, the average percentagechange in emissions resulting from a percentagechange in income, ceteris paribus, is a U-shapedfunction of income.20 At the national averageincome level of GBP £447 per week, a 10% in-crease in income determines a rise of about 6%in CO2 emissions. For the same 10% increase inincome, at the highest household-type incomelevel (about twice the national average, for thetype “wealthy executives”), the increase in CO2

emissions is about 12%; at the lowest household-type income, (about half the national average, forthe “old people, many high-rise flats” type), it isabout 16%.

Results for Other DeterminantsKeeping in mind that a regression approach

on its own is not sufficient to establish a causalrelationship among variables, we find that house-hold types with large homes (LBHO) or largefamily sizes (LHHSIZE) and those that consist



of single nonpensioners (LSING), of pensioners(LPENS), or of members of the National Trust(LNTRUST) are associated with higher emis-sions. The factors of education (LEDU), socialhousing (LSOCH), Internet usage (LINTER),and families with children (LFWC) seem to re-duce emissions.

Some of the variables can be interpreted inlight of general sociodemographic trends in theUnited Kingdom. For instance, households com-posed of pensioners tend to increase emissionsdue to older adults’ greater requirements forwarmth and the fact that they tend to spend amuch larger portion of the day at home. As theUK population ages, this could become an issueof concern. Haq and colleagues (2007) provideda detailed analysis of what determines the carbonemissions of older people’s lifestyles, how this re-lates to the United Kingdom’s demographic pat-terns, and what the implications for environmen-tal policies might be.

The result for education could be seen tosupport the green consumerism argument (e.g.,Pettit and Sheppard 1992) and could be usedin justification of information campaigns andenvironmental education of the general public,as, for example, undertaken in the UK ClimateChange Communication Programme (DEFRA2007). Our findings are also consistent with theidea that Internet use substitutes other, moreCO2-intensive activities, as suggested, for ex-ample, by Wilsdon (2002). Further research isneeded to verify these interpretations, however.

Other variables are more difficult to explain.For instance, household types with NationalTrust memberships tend to have higher CO2

emissions. This might be counterintuitive at first,as we interpret membership in this environmen-tal organization as an indicator of a positive at-titude toward the environment. National Trustmembership is culturally very segregated, how-ever, in that predominantly middle-income andhigh-income households are enrolled, so the vari-able seems to act as a proxy for wealth, anotherimportant determinant of consumption. This isalso shown by the high correlation with incomein figure 1.21 The same argument can be used toexplain the negative correlation for social hous-ing. Social housing has, in fact, the highest neg-ative correlation with income.

Income Elasticity and Spatial Data

Using the panel-fixed-effect estimation re-sults, we can illustrate how the spatial informa-tion content of our geodemographic data set canbe used to link lifestyle-related CO2 emissions tospecific locations within the United Kingdom.

For simplicity of notation, we assume that allvariables are in logs and that Ti = T, with i = 1,. . ., N. We can obtain predicted emissions forhousehold types by computing

E = [Zα, X] (α, β)′, (5)

where E is the NT×1 vector of emissions,X is the NT×k matrix of data, that is, X =[X0, X2

0, X30, X1, . . . , Xp−k ], α′ = (α1, . . . , αN),

and β ′ = (β1, . . . , βk). Zα = IN ⊗ ιT , where IN

is an identity matrix of order N, ιT is a vector of1s of dimension T, and ⊗ denotes the Kroneckermatrix product. Note that Zα is the matrix ofgroup dummies included in the regression to es-timate the αi when they are assumed to be fixed.

ACORN data provide detailed spatial in-formation on households classified by lifestylegroups. In particular, the data provide the num-ber of households of each ACORN lifestyle typeliving in every local authority area in Englandand Wales.22 For details on the methods involvedin producing this kind of data, see the work ofWebber (1998, 2007) and Harris and colleagues(2005). We arrange the data on the number of56 different ACORN household types living ineach of the 410 local authority areas in Englandand Wales in a matrix T of size 410 × 56. Hence,each of the 410 rows of T contains the numberof households belonging to each one of the 56household types for one specific local authorityarea. T can be used to map CO2 emissions byhousehold types, E, to map emissions by UK lo-cal authority areas, EL A.

The emissions for 410 local authority areas inEngland and Wales can be computed as

EL A = T exp(E) (6)

where exp denotes the vector-valued exponentialfunction.

Equations (5) and (6), estimated for the avail-able sample, can serve as a benchmark againstwhich to measure the effect of counterfactualscenarios. Using these equations, we can evaluate



Figure 2 Maps showing the impact of incomeincreases on direct and indirect carbon dioxide(CO2) emissions attributable to consumers’ lifestyletypes, by local authority area in England. Classintervals are based on sample quantile breaks. Notethat “(a,b)” denotes the set of values x, such asa ≤ x < b. (a) Map of absolute changes from a 10%increase in household income, with all othervariables fixed at their sample values, in direct andindirect CO2 emissions from the 56 lifestyle types ofconsumers living in 390 unitary authority and localauthority areas in England, EL A. One kiloton(kt) = 103 tonnes (t) ≈ 1.102 × 103 short tons.(b) Map of relative changes from a 10% increase inhousehold income, with all other variables fixed at

the impact of a suitable change �Xj on emissions,�EL A.

Using equation (6), we can map changes inemissions for household types into changes for lo-cal authority areas. We find that a 10% increasein income determines a rise of 7% in total CO2

emissions, if we keep all other variables constantat their sample values. The impact for specific lo-cal authority areas depends on the lifestyle mix ofhouseholds living in those areas. Figure 2 presentsthe distribution of CO2 emission changes over390 local authority areas for England.23

It is not surprising that if more households ofany lifestyle inhabit a particular area, the envi-ronmental consequences are greater. The map infigure 2(a) shows that the local areas with largerabsolute impact of income on CO2 emissions cor-respond closely to densely populated areas in Eng-land, including the urban West Midlands, north-ern cities, and Greater London.

Figure 2(b) presents a map of the relativechanges in CO2 emissions for England deter-mined by income increases, ceteris paribus. Largevisual impacts of relative changes reflect the im-portance of households’ lifestyle for the environ-ment. The highest percentage changes occur inlocal areas that have the highest concentrationof households with highly polluting lifestyles,such as those from the wealthiest types (ACORNGroup 1) and from the poorest types (ACORNTypes 51 through 54). Figure 3 shows maps ofthe number of the richer and poorer households,with large estimated emission elasticities of in-come. With the aid of these maps, we can seethat in figure 2(b), the large visual impacts for thenorthern areas are mostly due to the large num-ber of poorer families, such as single parents andpensioners, council flats (ACORN Type 51); oldpeople, many high-rise flats (Type 53);24 and “sin-gles and single parents, high-rise estates” (Type54; see figure 3b). Similarly, the large visual im-pact for the south can be explained by the large

←−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−their sample values, in direct and indirect CO2

emissions from the 56 lifestyle types of consumersliving in 390 unitary and local authority areas inEngland, �EL A

EL A.



Figure 3 Maps of the number of household typeswith large emission elasticity of income, by localauthority area in England. (a) Map of the number ofpoorest households, belonging to Groups 15 and 16(see tables 1 and 3 for reference) with largeemission elasticity of income, by 390 local authorityareas in England. The types include “single parentsand pensioners, council terraces” (ACORN LifestyleType 51), “old people, many high-rise flats” (Type53), and “singles and single parents, high-rise estates”(Type 54). In some of the northern cities, such asManchester, Middlesborough, Newcastle upon Tyne,and Sunderland, more than 20% of householdsbelong to these types. (b) Map of the number ofrichest households, belonging to Group 1 (see

concentration of the wealthiest, more pollutinglifestyles, including “wealthy mature profession-als, large houses” (ACORN Type 1), “wealthyworking families with mortgages” (Type 2), “vil-lages with wealthy commuters” (Type 3), and“well-off managers, large houses” (Type 4; seefigure 3b).

Conclusion

Climate change is a key issue on the UK policyagenda. It has been acknowledged that changesin both production technology and lifestyle arerequired to achieve the government’s ambitiousCO2 emission reduction targets. In this article, wehave explored the potential of geo-demographicmarketing data for analyzing CO2 emissions as-sociated with lifestyles in the United Kingdom.

In a first step, we have applied geodemo-graphic segmentation data in an input−outputmodel to assess the direct and indirect CO2 emis-sions associated with the consumption patternsof different lifestyle groups in the United King-dom. In a second step, we have used a regressionapproach to estimate the relationship betweenhousehold CO2 emissions and several socioeco-nomic factors characterizing these households.We have also illustrated how the estimated rela-tionship, together with the additional geograph-ical information available about the location ofhousehold types, can be used to provide a vi-sual representation of where the various socioe-conomic factors cause the greatest emissions onthe map.

Our results demonstrate the value of ap-plying rich geodemographic data sets, with

←−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−tables 1 and 3 for reference), with large emissionelasticity of income by 390 local authority areas inEngland. The types include “wealthy matureprofessionals, large houses” (ACORN Lifestyle Type1), “wealthy working families with mortgages” (Type2), “villages with wealthy commuters” (Type 3), and“well-off managers, large houses” (Type 4). Some ofthe rural, semirural, and suburban southern areas,including Chiltern, Mole Valley, Rushcliffe, andWaverley, have more than 40% households of thesetypes.



information about people’s attitudes, their behav-iors, and the neighborhoods they live in, to theanalysis of CO2 emissions associated with dif-ferent lifestyles. We find that lifestyles are im-portant for determining CO2 emissions associ-ated with UK consumption. Seventy-five percentof the United Kingdom’s consumer CO2 emis-sions are associated with people’s consumptionchoices. Differences in consumption choices leadto variations in CO2 across lifestyle groups by afactor of 2 to 3 (on a per household or per capitabasis), with transport- and housing-related ex-penditures accounting for the largest shares.

Our results provide support for the idea thatsociodemographic variables are important in ex-plaining emissions. For instance, emissions areincreasing with income, but at increasing ratesfor richer households. This finding contributes toa growing body of literature that has cast doubtson the previously widely held conjecture that thedemand for environmental quality increases withincome, so that emissions may rise at first witheconomic growth but eventually fall as incomecontinues to rise. This view was often used to sup-port the environmental Kuznets curve hypothe-sis, which posits an inverted-U-shaped relation-ship between environmental degradation andincome. Having found a similar relationship us-ing a completely different data set (see Minxet al. 2009), we are reasonably confident aboutthis general finding.

Apart from the fact that transport emissionsare particularly high for wealthier households,however, our analysis does not provide deeperinsights into why this might be the case. Onepossible explanation, suggested by the descriptorsof lifestyles, is that the demand for improved en-vironmental quality from the wealthiest house-holds is expressed through a surrogate demandfor luxury homes in high-status suburban and ru-ral neighborhoods, regular holidays in relativelypollution-free locations, membership in countryclubs, and so on.

As another example, our analysis finds thathigher levels of education are associated with re-duced emissions. This could be seen as evidencein favor of the “green consumerism” argument.Although further evidence is required to confirmthis finding, it could give support to the UnitedKingdom’s government’s climate change commu-

nication program, which focuses on providing in-formation about potential ways to reduce carbonemissions to particularly ill-informed groups ofsociety (DEFRA 2007).

Finally, we also demonstrate how geodemo-graphic data can help in estimating lifestyle-related emissions at smaller spatial scales andtherefore in determining where households withcarbon-intensive lifestyles can seem to cluster.Such information has direct application to pol-icy, as government might want to work with high-emission households in a targeted approach to re-duce the carbon emissions associated with theirlifestyle. Equally, our findings could provide im-portant information in explaining emission pat-terns attached to a particular lifestyle. The phys-ical infrastructure of a particular neighborhoodcould be one key determinant of lifestyle-relatedemission that could also act as a barrier to lifestylechange. Such potential infrastructure bottlenecksto emission reductions are still relatively littleunderstood and are one important avenue of re-search where geodemographic data could play animportant role in the future. Recent activitiesof the IPCC as well as the fast-growing city-levelclimate change mitigation research underline theurgency of this issue (VandeWehge and Kennedy2007; IPCC 2009; Kennedy et al. 2010). Wehope that our initial findings will stimulate fur-ther much-needed research in geodemographic-based lifestyle analysis, making use of better dataand improved methods.

Acknowledgements

We thank the editors and three anonymousreferees for helpful comments and suggestions.

Notes

1. Note that energy use at work is part of the indirectemission component. The energy use from work-related activities is embedded in the goods and ser-vices households buy and is assigned to them onthe basis of the type and quantity of products theychoose.

2. The price conversion was undertaken by CambridgeEconometrics.

3. CACI is a commercial marketing data firm head-quartered in London (see www.caci.co.uk). CACIand its ACORN data are approved data suppliers


http://www.caci.co.uk


of the Office for National Statistics in the UnitedKingdom and meet “the agreed standards of dataanalysis and dissemination” (ONS, 2005c, 1).

4. A complete listing of all household types can befound at www.caci.co.uk/acorn.

5. We are prepared to supply specific data on request,except when the data are only commercially avail-able.

6. In fact, uncertainties are greatest for sectors withvery small outputs and sectors that are heavily in-volved in international trade.

7. The main information deficit concerns the assump-tion made during the construction of the ACORNclassifications.

8. This includes imported goods and services con-sumed in the United Kingdom and excludes allexports from the United Kingdom. One milliontonnes = 1 megaton (Mt) = 106 tonnes (t) = oneteragram (Tg, SI) ≈ 1.102 × 106 short tons.

9. These variables were derived by CACI, mainlyfrom census data (see www.neighbourhood.statistics.gov.uk).

10. The “Asian communities” group seemed to be anoutlier. Discussion with the ACORN data provider,CACI, revealed that these higher travel demandsare a particular feature of this group.

11. Keep in mind that we are talking about averages ofrather large segments of society, and substantial in-group variations still exist (see Weber and Matthews2008).

12. We have reconfirmed this finding in a detailed spa-tial analysis elsewhere (SEI 2008).

13. As highlighted previously, this also includes the im-mediate physical environment in which people live.

14. The term panel data is more in use among economists(see, e.g., Wooldridge 2002), whereas other socialscientists and statisticians prefer the the expressionlongitudinal data (see, e.g., Diggle et al. 2002).

15. For details, see the ACORN user guide (CACI2004).

16. This model selection procedure is known as the“general-to-specific approach” in econometrics. Theprocedure followed in this study starts with a largenumber of variables and sequentially reduces themby removing the least significant variable, one at thetime, if its p value is above a chosen threshold, rees-timating the model each time with the remainingvariables. The initial selection criterion used wasp value greater than .2 to remove. Once the pro-cedure stops, because all variables are significant atthe .2 level, the selection criterion is gradually mademore stringent until the predetermined final levelof significance is obtained for the remaining sub-set of variables. In this first stage, only linear terms

are considered. Nonlinear terms (cross-productsand squares) were tried and tested in a secondstage.

17. To be precise, we note that the ACORN data arefrom 2004 and the input−output and environmen-tal account data are from 2000, as outlined in theInput−Output and Geodemographic Data subsectionof this article. Further note that a suitable time seriesof data was not available due to a reclassification ofthe ACORN data between 2003 and 2004 as wellas their limited availability for previous years.

18. We calculated the t statistics using a co-variance matrix estimator, which is robust toheteroskedasticity and serial correlation of un-known form, which is known to have good proper-ties in “short” panels (see, e.g., Wooldridge 2002).As emissions are a function of consumption, it isimportant to allow for heteroskedastic and auto-correlated errors for valid statistical inference. Infact, because consumption above subsistence levelsis more likely to be discretionary, we would expecta greater variability in consumption and, thereforeemissions as income increases.

19. We can provide additional results and data on re-quest.

20. If we take the differential of equation (4) as a func-tion of income, X0, in its cubic form (p = 3), af-ter rearranging we get εE,X0 = β1 + 2β2 ln(X0) +3β3(ln X0)2, where εE,X0 = d E

d X0

X0E ≈ %�E

%�X0is the

elasticity of emissions with respect to income. Asβ3 > 0, the elasticity is a U-shaped curve.

21. Economic theory posits (“permanent-income hy-pothesis” and “life-cycle hypothesis”) that consump-tion decisions are based not only on the level ofcurrently disposable income but also on lifetime in-come and wealth.

22. For information about local authority areas in theUnited Kingdom, see the work of the Local Gov-ernment Association (2005).

23. These maps are based on vector-based spatial data(administrative boundary) provided through ED-INA Digimap, an online service to Ordnance Sur-vey digital map data (http://edina.ac.uk/digimap),with the support of the ESRC and JISC, and usesboundary material that is copyrighted by the Crown.We used R release 2.8.1, the standard Win32 releaseavailable at the time we wrote this article, togetherwith the routines for manipulating and reading ge-ographic data provided by the maptools R package,version 0.7–23, developed by Nicholas J. Lewin-Kohand Roger Bivand.

24. The type “families and single parents, council flats”(ACORN Type 52) is found in significant numbersonly in Scotland.


http://www.caci.co.uk/acorn

http://www.neighbourhood.statistics.gov.uk

http://www.neighbourhood.statistics.gov.uk

http://edina.ac.uk/digimap


References

Bin, S. and Dowlatabadi, H. 2005. Consumer lifestyleapproach to US energy use and related CO2 emis-sions. Energy Policy 33(2): 197–208.

Boardman, B. 2007. Home truth: A low carbon strategyto reduce UK housing emissions by 50%. Technicalreport. Oxford, UK: Friends of Earth.

Brodersen, S. 1990. Time and consumption. Copen-hagen, Denmark: Statistics Denmark.

CACI. 2004. The ACORN user guide. Technical report.London: CACI.

Charkiewicz, E., S. V. Bennekom, and A. Young. 2001.Transitions to sustainable production and consump-tion: Concepts, policies and actions. Maastricht, theNetherlands: Shaker.

Cohen, C., M. Lenzen, and R. Schaeffer. 2005. Energyrequirements of households in Brazil. Energy Policy33(4): 555–562.

DEFRA (UK Department for Environment, Food andRural Affairs). 2007. Climate change communica-tion: Tomorrow’s climate, today’s challenge. Lon-don: DEFRA.

Diggle, P., P. Heagerty, K. Y. Liang, and S. Zeger. 2002.Analysis of longitudinal data. Oxford, UK: OxfordUniversity Press.

Draper, N. R. and H. Smith. 1998. Applied regressionanalysis. New York: Wiley-Interscience.

Dresner, S. and P. Ekins. 2005. Climate change and fuelpoverty. London: Policy Studies Institute.

Druckman, A. and T. Jackson. 2008. Household energyconsumption in the UK: A highly geographicallyand socio-economically disaggregated model. En-ergy Policy 36(8): 3177–3192.

Druckman, A., P. Sinclair, and T. Jackson. 2008. Ageographically and socio-economically disaggre-gated local household consumption model for theUK. Journal of Cleaner Production 16(7): 870–880.

Duchin, F. 1998. Structural economics: Measuring changein technology, lifestyle and the environment. Wash-ington, DC: Island Press.

Duchin, F. and K. Hubacek. 2003. Linking social ex-penditures to household lifestyles. Futures 35(1):61–74.

Ferrer-i Carbonell, A. and J. van den Bergh. 2004.A micro-econometric analysis of determinants ofunsustainable consumption in the Netherlands.Environmental and Resource Economics 27(4): 367–389.

Flores, N. E. and R. T. Carson. 1997. The relation-ship between the income elasticities of demandand willingness to pay. Journal of EnvironmentalEconomics and Management 33(3): 287–295.

Grossman, G. and A. Krueger. 1993. Environmental

impacts of a North American Free Trade Agree-ment. In US–Mexico free trade agreement, editedby P. Gaber. Cambridge, MA: MIT Press.

Haq, G., J. Minx, J. Whitelegg, and A. Owen. 2007.Greening the greys: Climate change and the over 50s.Technical report. York, UK: Stockholm Environ-ment Institute.

Harris, R., P. Sleight, and R. Webber. 2005. Geodemo-graphics, GIS and neighbourhood targeting. Chich-ester, UK: Wiley.

Hertwich, E. G. 2005. Life cycle approaches to sustain-able consumption: A critical review. Environmen-tal Science and Technology 39(13): 4673–4684.

Hertwich, E. and M. Katzmayr. 2004. Examples ofsustainable consumption: Review, classification andanalysis. Report 5. Trondheim, Norway: Norwe-gian University of Science and Technology In-dustrial Ecology Programme.

Hertwich, E. and G. Peters. 2009. The carbon footprintof nations. Environmental Science and Technology43(16): 6414–6420.

HM Government. 2005. Securing the future: DeliveringUK sustainable development strategy. Technical re-port. London: Department for Environment, Foodand Rural Affairs.

HM Government. 2006. Climate change: The UK gov-ernment 2006 programme. Technical report. Lon-don: Department for Environment, Food and Ru-ral Affairs.

Hokby, S. and T. Soderqvist. 2003. Elasticities of de-mand and willingness to pay for environmentalservices in Sweden. Environmental and ResourceEconomics 26(3): 361–383.

Hoover, K. D. and S. J. Perez. 1999. Data miningreconsidered: Encompassing and the general-to-specific approach to specification search. Econo-metrics Journal 2(2): 167–191.

Hubacek, K. and L. Sun. 2005. Changes in China’seconomy and society and their effects on wateruse: A scenario analysis. Journal of Industrial Ecol-ogy 9(1–2): 187–200.

Hubacek, K., D. Guan, J. Barrett, and T. Wiedmann.2009. Environmental implications of urbaniza-tion and lifestyle change in China: Ecological andwater footprints. Journal of Cleaner Production 17:1241–1248.

IPCC. 2007. Climate change 2007: Contributions ofWorking Groups i, ii and iii to the fourth assessmentreport of the Intergovernmental Panel on ClimateChange. Technical report. Geneva, Switzerland:IPCC.

IPCC. 2009. Concept note for an expert meeting on humansettlement, water, energy and infrastructure. Tech-nical report. Geneva, Switzerland: IPCC.



Jalas, M. 2002. A time use perspective on the ma-terial intensity of consumption. Ecological Eco-nomics 41(1): 109–123.

Jalas, M. 2005. The everyday context of increasing en-ergy demands: Time use survey data in a decompo-sition analysis. Journal of Industrial Ecology 9(1–2):129–145.

Kennedy, C., J. Steinberger, B. Gasson, Y. Hansen,T. Hillman, M. Havranek, D. Pataki, A. Ph-dungsilp, A. Ramaswami, and G. V. Mendez.2010. Methodology for inventorying greenhousegas emissions from global cities. Energy Policy.Forthcoming.

Kondo, Y. and K. Takase. 2007. Advances in life cycleengineering for sustainable manufacturing businesses.London: Springer.

Labandeira, X. and J. M. Labeaga. 1999. Combininginput-output analysis and micro-simulation to as-sess the effects of carbon taxation on Spanishhouseholds. Fiscal Studies 20(3): 305–320.

Lenzen, M. 1994. Energy and greenhouse gas cost ofliving for Australia during 1993/94. Energy 23(6):497–516.

Lenzen, M., L. Pade, and J. Munksgaard. 2004a.CO2 multipliers in multi-region input-outputmodels. Economic Systems Research 16(4): 391–412.

Lenzen, M., C. Dey, and B. Foran. 2004b. Energy re-quirements of Sydney households. Ecological Eco-nomics 49: 375–399.

Lenzen, M., M. Wier, C. Cohen, H. Hayami, S.Pachauri, and R. Schaeffer. 2006. A comparativemultivariate analysis of household energy require-ments in Australia, Brazil, Denmark, India andJapan. Energy 31: 181–207.

Lenzen, M., J. Murray, F. Sack, and T. Wiedmann.2007. Shared producer and consumer responsi-bility: Theory and practice. Ecological Economics61(1): 27–42.

Lenzen, M., R. Wood, and B. Foran. 2008. Direct ver-sus embodied energy: The need for urban lifestyletransitions. In Urban energy transition, edited byP. Droege. Amsterdam: Elsevier.

Lintott, J. 1998. Beyond the economics of more: Theplace of consumption in ecological economics.Ecological Economics 25: 239–248.

Local Government Association. 2005. Local govern-ment matters: Facts and figures about local coun-cils 2005–2006. Technical report. London: UKDepartment of Environment, Food and Rural Af-fairs.

Lovell, M. C. 1983. Data mining. Review of Economicsand Statistics 65(1): 1–12.

McConnell, K. E. 1997. Income and the demand for

environmental quality. Environment and Develop-ment Economics 2(04): 383–399.

Miller, R. and P. Blair. 2009. Input-output analysis:Foundations and extensions. Second edition. Cam-bridge: Cambridge University Press.

Minx, J. C. and G. Baiocchi. 2009. Time-use and sus-tainability. In Handbook of input-output economicsin industrial ecology, edited by S. Suh. Dordrecht,the Netherlands: Springer.

Minx, J., T. Wiedmann, R. Wood, G. Peters, M.Lenzen, A. Owen, K. Scott, et al. 2009. Multi-regional input-output analysis and carbon foot-printing: An overview of UK applications.. Eco-nomic Systems Research 21(3): 187–216.

Munksgaard, J. and K. A. Pedersen. 2001. CO2 ac-counts for open economies: Producer or consumerresponsibility? Energy Policy 29: 327–334.

Munksgaard, J., J. C. Minx, L. Christoffersen, andL.-L. Pade. 2009. Models for national CO2 ac-counting. In Handbook of input-output economicsin industrial ecology. Eco-efficiency in industry andscience, Vol. 23, edited by S. Suh. Dordrecht:Springer.

Munksgaard, J., L.-L. Paade, J. Minx, and M. Lenzen.2005. Influence of trade on national CO2 emis-sions. International Journal of Global Energy Issues23(4): 324–336.

ONS (UK Office for National Statistics). 2003a.Membership of selected environmental organisations,1971–2002: Social trends 33. London: ONS.

ONS. 2003b. United Kingdom input-output analyses.London: ONS. Download supply and use tablesat www.statistics.gov.uk/input-output.

ONS. 2005a. Environmental accounts 2005. London:ONS.

ONS. 2005b. Family spending: A report of the 2004–05expenditure and food survey. London: ONS.

ONS. 2005c. Official suppliers: Census 2001. London:ONS.

ONS. 2008. Internet access 2008 households and individ-uals. London: ONS.

OPSI (Office of Public Sector Information).2008. The Climate Change Act 2008, Lon-don. www.opsi.gov.uk/acts/acts2008/pdf/ukpga_20080027_en.pdf. Accessed 17 January 2010.

Ornetzeder, M., E. G. Hertwich, K. Hubacek, K. Kory-tarova, and W. Haas. 2008. The environmentaleffect of car-free housing: A case in Vienna. Eco-logical Economics 65(3): 516–530.

Pachauri, S. and D. Spreng. 2002. Direct and indirectenergy requirements of households in India. En-ergy Policy 30: 511–523.

Panayotou, T. 2000. Economic growth, environment,Kuznets curve. Technical report, CID Working


http://www.statistics.gov.uk/input-output

http://www.opsi.gov.uk/acts/acts2008/pdf/ukpga_20080027_en.pdf

http://www.opsi.gov.uk/acts/acts2008/pdf/ukpga_20080027_en.pdf


Paper No. 56, Center for International Develop-ment at Harvard University.

Pancs, R. and N. J. Vriend. 2007. Schelling’s spatialproximity model of segregation revisited. Journalof Public Economics 91(1–2): 1–24.

Peters, G. P. 2008. From production-based toconsumption-based national emission invento-ries. Ecological Economics 65(1): 13–23.

Peters, G. P. and E. G. Hertwich. 2008. The importanceof imports for household environmental impacts.Journal for Industrial Ecology 10(3): 89–109.

Pettit, D. and J. Sheppard. 1992. It’s not easy beinggreen: The limits of green consumerism in lightof the logic of collective action. Queens Quarterly99(3): 328–350.

Princen, T., M. Maniates, and K. Conca. 2002. Con-fronting consumption. London: MIT Press.

Schaffer, A. and C. Stahmer. 2005. The part time so-ciety: A concept for more sustainable patterns ofproduction and consumption. GAIA 14(3): 229–239.

Schelling, T. C. 1969. Models of segregation. AmericanEconomic Review 59(2): 488–493.

Schipper, L., S. Bartlett, D. Hawk, and E. Vine.1989. Linking lifestyles and energy use: A mat-ter of time? Annual Review of Energy 14: 273–320.

SEI (Stockholm Environment Institute). 2008. Theright climate for change: Using the carbon foot-print to reduce CO2 emissions—a guide for localauthorities. Technical report. Washington, DC:WWF.

Stahmer, C. 2004. Social accounting matrices and ex-tended input-output tables. In Measuring sustain-able development: Integrated economic, environmen-tal and social frameworks, edited by OECD. Paris:OECD.

Stern, D. I. 2004. The rise and fall of the environmentalKuznets curve. World Development 32(8): 1419–1439.

Suh, S., M. Lenzen, G. J. Treloar, H. Hondo, A. Hor-vath, G. Huppes, O. Jolliet, et al. 2004. Systemboundary selection in life-cycle inventories us-ing hybrid approaches. Environmental Science andTechnology 38(3): 657–664.

Sustainable Consumption Roundtable. 2006. I will ifyou will. Technical report. London: SustainableDevelopment Commission and National Con-sumer Council.

Symons, E., J. Proops, and P. Gay. 1994. Carbon taxes,consumer demand and carbon dioxide emissions:A simulation analysis for the UK. Fiscal Studies15(2): 19–43.

Tukker, A. and B. Jansen. 2006. Environment impacts

of products: A detailed review of studies. Journalof Industrial Ecology 10(3): 159–182.

Tukker, A., G. Huppes, J. Guinee, R. Heijungs, A. deKoning, L. van Oers, S. Suh, et al. 2005. Environ-mental impact of products (EIPRO): Analysis of thelife cycle environmental impacts related to the totalfinal consumption of the EU25. Technical report.Institute for Prospective TechnologicalStudies.

VandeWehge, J. and C. Kennedy. 2007. A spatial anal-ysis of residential greenhouse gas emissions in theToronto census metropolitan area. Journal of In-dustrial Ecology 11(2): 133–144.

Verbeek, M. 2008. A guide to modern econometrics. Thirded. Chichester, UK: Wiley.

Vickers, D. and P. Rees. 2007. Creating the UK na-tional statistics 2001 output area classification.Journal of the Royal Statistical Society 170(2): 379–403.

Vringer, K. and K. Blok. 1995. The direct and indirectenergy requirements of households in the Nether-lands. Energy Policy 23(10): 893–910.

Webber, R. 1998. Developments in cross border stan-dards for geodemographic segmentation. In Euro-pean geographic information infrastructures: Oppor-tunities and pitfalls, edited by P. Burrough and I.Masser. London: Taylor and Francis.

Webber, R. 2007. The metropolitan habitus: Its man-ifestations, locations, and consumption profiles.Environment and Planning A 39(1): 182–207.

Weber, C. and A. Perrels. 2000. Modelling lifestyleeffects on energy demand and related emissions.Energy Policy 28: 549–566.

Weber, C. L. and H. S. Matthews. 2008. Quantifyingthe global and distributional aspects of Americanhousehold carbon footprint. Ecological Economics66(2–3): 379–391.

Weisberg, S. 2005. Applied regression analysis. Seconded. Hoboken, NJ: Wiley.

Wiedmann, T. 2009. A review of recent multi-regioninput-output models used for consumption-basedemission and resource accounting. Ecological Eco-nomics 69(2): 211–222.

Wiedmann, T., M. Lenzen, K. Turner, and J. Barrett.2007. Examining the global environmental im-pact of regional consumption activities: part 2.Review of input-output models for the assessmentof environmental impacts embodied in trade. Eco-logical Economics 61(1): 15–26.

Wiedmann, T., M. Lenzen, and R. Wood. 2008. Uncer-tainty analysis of the UK-MRIO model: Results froma Monte-Carlo analysis of the UK multi-region input-output model. Technical report. London: Depart-ment for Environment, Food and Rural Affairs.



Wiedmann, T., J. Minx, J. Barrett, and M. Wacker-nagel. 2006. Allocating ecological footprints tofinal consumption categories with input-outputanalysis. Ecological Economics 56: 28–48.

Wiedmann, T., R. Wood, J. Minx, M. Lenzen, D. Guan,and R. Harris. 2010. A carbon footprint time se-ries of the UK: Results from a multi-region input-output model. Economic Systems Research. Forth-coming.

Wier, M., M. Lenzen, J. Munksgaard, and S. Smed.2001. Effects of household consumption patternson CO2 requirements. Economic Systems Research13(3): 259–274.

Wilkinson, L. and G. Dallal. 1981. Tests of significancein forward selection regression with an F-to-enterstopping rule. Technometrics 23: 377–380.

Wilsdon, D. 2002. Digital futures: Living in a dot.comworld. London: Earthscan.

Wooldridge, J. M. 2002. Econometric analysis of crosssection and panel data. Cambridge, MA: MIT Press.

About the Authors

Giovanni Baiocchi is a lecturer in economicsat Durham Business School at Durham Univer-sity in Durham, UK. Jan Minx is a senior researchfellow at the Stockholm Environment Institute,working at the Institute for the Economics of Cli-mate Change at Technische Universitat Berlin inBerlin, Germany. Klaus Hubacek is a reader withthe School of Earth and Environment, Univer-sity of Leeds, in Leeds, UK, and is also an adjunctprofessor at the Department of Geography, Uni-versity of Maryland, in College Park, Maryland,USA.


research and analysis the impact of social factors and …re.indiaenvironmentportal.org.in/files/the...

Documents