05.12.2016 Views

The Direct and Indirect Rebound Effects for Residential Heating in Switzerland

n?u=RePEc:irn:wpaper:16-11&r=reg

n?u=RePEc:irn:wpaper:16-11&r=reg

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

University of Neuchatel<br />

Institute of Economic Research<br />

IRENE, Work<strong>in</strong>g paper 16-11<br />

<strong>The</strong> <strong>Direct</strong> <strong>and</strong> <strong>Indirect</strong> <strong>Rebound</strong> <strong>Effects</strong> <strong>for</strong><br />

<strong>Residential</strong> <strong>Heat<strong>in</strong>g</strong> <strong>in</strong> Switzerl<strong>and</strong><br />

Cécile Hediger* Mehdi Farsi** Sylva<strong>in</strong> Weber***<br />

* University of Neuchâtel, Institute of Economic Research, Rue Abram-Louis<br />

Breguet 2, 2000 Neuchtel, Switzerl<strong>and</strong>. Contact: cecile.hediger@un<strong>in</strong>e.ch.<br />

* University of Neuchâtel, Institute of Economic Research, Rue Abram-Louis<br />

Breguet 2, 2000 Neuchtel, Switzerl<strong>and</strong>. Contact: mehdi.farsi@un<strong>in</strong>e.ch<br />

* University of Neuchâtel, Institute of Economic Research, Rue Abram-Louis<br />

Breguet 2, 2000 Neuchtel, Switzerl<strong>and</strong>. Contact: sylva<strong>in</strong>.weber@un<strong>in</strong>e.ch.


<strong>The</strong> <strong>Direct</strong> <strong>and</strong> <strong>Indirect</strong> <strong>Rebound</strong> <strong>Effects</strong> <strong>for</strong><br />

<strong>Residential</strong> <strong>Heat<strong>in</strong>g</strong> <strong>in</strong> Switzerl<strong>and</strong> ∗<br />

Work<strong>in</strong>g paper<br />

Cécile Hediger, Mehdi Farsi, Sylva<strong>in</strong> Weber †<br />

This version: October 7, 2016<br />

Abstract<br />

Improvements <strong>in</strong> energy efficiency are seen as a key tool to decrease<br />

energy consumption, yet less is known about the reaction of household<br />

to such improvements. <strong>The</strong>y could adapt (<strong>in</strong>crease) their dem<strong>and</strong> because<br />

the relative price of the energy service has dim<strong>in</strong>ished thanks<br />

to the efficiency ga<strong>in</strong>s. Focus<strong>in</strong>g on the residential heat<strong>in</strong>g sector, we<br />

study how the households react to a ga<strong>in</strong> <strong>in</strong> efficiency of their heat<strong>in</strong>g<br />

system. An <strong>in</strong>creased dem<strong>and</strong> <strong>for</strong> the heat<strong>in</strong>g service is the direct rebound<br />

effect, while the <strong>in</strong>direct rebound effect is the <strong>in</strong>come effect on<br />

the consumption of all other goods. Both effects take back some of the<br />

energy sav<strong>in</strong>gs <strong>in</strong>itially expected. To estimate these rebound effects,<br />

we use a specific survey conta<strong>in</strong><strong>in</strong>g <strong>in</strong>novative choice experiments. We<br />

f<strong>in</strong>d that 10 to 15% of the expected energy sav<strong>in</strong>gs are not realised <strong>in</strong><br />

the space heat<strong>in</strong>g sector due to the direct rebound effect. Includ<strong>in</strong>g<br />

the <strong>in</strong>direct rebound effects, it is one third of the potential sav<strong>in</strong>gs<br />

which are lost. We also highlight the heterogeneity among people:<br />

some do not rebound at all, while others display a large rebound. We<br />

analyse this heterogeneity us<strong>in</strong>g socio-economic characteristics, f<strong>in</strong>d<strong>in</strong>g<br />

that people with a low <strong>in</strong>come, less educated, less environmentally<br />

friendly, <strong>and</strong> less satisfied of their heat<strong>in</strong>g com<strong>for</strong>t have the largest<br />

direct rebound effect.<br />

JEL Classification: D12, Q41, Q47, R22.<br />

Keywords: <strong>Rebound</strong> <strong>Effects</strong>, Energy Efficiency, <strong>Residential</strong> <strong>Heat<strong>in</strong>g</strong><br />

∗ This research was f<strong>in</strong>ancially supported by the Swiss National Science Foundation<br />

(SNSF), grant No 100018-144310.<br />

† University of Neuchâtel. Correspond<strong>in</strong>g author: cecile.hediger@un<strong>in</strong>e.ch.<br />

1


1 Introduction<br />

<strong>The</strong> easiest way to save energy is not to waste it. Improv<strong>in</strong>g energy efficiency<br />

is thus considered a key tool to reduce energy consumption while<br />

keep<strong>in</strong>g the same level of com<strong>for</strong>t. Energy efficiency is <strong>in</strong>deed seen as the<br />

“Fifth Fuel” or “Invisible Fuel” (<strong>The</strong> Economist, 2015) of the energy transition.<br />

In Switzerl<strong>and</strong>, like <strong>in</strong> many other <strong>in</strong>dustrialized countries, room <strong>for</strong><br />

improvements <strong>in</strong> energy efficiency is still considerable, especially <strong>in</strong> the area<br />

of build<strong>in</strong>gs <strong>and</strong> heat<strong>in</strong>g systems. Accord<strong>in</strong>g to the Swiss Federal Office of<br />

Energy, 30% of Switzerl<strong>and</strong>’s primary energy consumption is attributable to<br />

heat<strong>in</strong>g, air-condition<strong>in</strong>g <strong>and</strong> hot water <strong>in</strong> the build<strong>in</strong>gs, with much of the<br />

energy be<strong>in</strong>g wasted. Promot<strong>in</strong>g energy efficiency is the first pillar of the<br />

energy strategy 2050 aim<strong>in</strong>g to phase out nuclear power, com<strong>in</strong>g be<strong>for</strong>e the<br />

promotion of renewable. In terms of efficiency, four sectors are targeted, with<br />

build<strong>in</strong>gs <strong>in</strong> the first position, <strong>and</strong> then mobility, <strong>in</strong>dustry, <strong>and</strong> appliances.<br />

Yet, sett<strong>in</strong>g ambitious efficiency st<strong>and</strong>ards might not be sufficient to achieve<br />

the new targets of energy consumption, because people will react <strong>and</strong> adjust<br />

to such policies. <strong>The</strong>se reactions must be acknowledged <strong>in</strong> order to design<br />

effective policies. Follow<strong>in</strong>g efficiency improvements, consumers could react<br />

by consum<strong>in</strong>g more goods <strong>and</strong> services, with the risk of offsett<strong>in</strong>g completely<br />

the <strong>in</strong>itial ga<strong>in</strong>s <strong>in</strong> energy consumption. For example, suppose the heat<strong>in</strong>g<br />

system of your house has just been renovated <strong>and</strong> is now more efficient. It<br />

means that it is now cheaper to heat your house. Because of that, you could<br />

decide to modify your heat<strong>in</strong>g habits, <strong>for</strong> <strong>in</strong>stance set the temperature by 1<br />

or 2 degrees higher, or start heat<strong>in</strong>g earlier <strong>in</strong> the season, air more often,<br />

etc. This mechanism is known as the “rebound effect”, or more precisely the<br />

direct rebound effect. In the literature, the direct rebound effect is commonly<br />

def<strong>in</strong>ed as “an <strong>in</strong>creased consumption of energy services [e.g., heat<strong>in</strong>g] follow<strong>in</strong>g<br />

an improvement <strong>in</strong> the technical efficiency of deliver<strong>in</strong>g those services”<br />

(Sorrell <strong>and</strong> Dimitropoulos, 2008).<br />

An <strong>in</strong>direct rebound effect might also arise because products <strong>and</strong>/or services<br />

will be purchased with the sav<strong>in</strong>gs made on heat<strong>in</strong>g. <strong>The</strong> use of the term<br />

2


<strong>in</strong>direct rebound is <strong>in</strong>consistent <strong>in</strong> the literature, but it most commonly describes<br />

the <strong>in</strong>come effect on the consumption of all other goods (Chan <strong>and</strong><br />

Gill<strong>in</strong>gham, 2015). As any product <strong>and</strong> service requires energy to be created<br />

<strong>and</strong> used, overall energy consumption will decrease by less than the amount<br />

predicted after the renovation of an heat<strong>in</strong>g system, even <strong>in</strong> the absence<br />

of any direct rebound effect. In the worst case, if the products or services<br />

purchased with the sav<strong>in</strong>gs made on heat<strong>in</strong>g are very energy <strong>in</strong>tensive (<strong>for</strong> <strong>in</strong>stance<br />

travell<strong>in</strong>g by plane), overall energy consumption could even <strong>in</strong>crease.<br />

<strong>The</strong> latter situation is known as backfire.<br />

In this paper, we <strong>in</strong>vestigate the direct <strong>and</strong> <strong>in</strong>direct rebound effects <strong>for</strong> residential<br />

heat<strong>in</strong>g <strong>in</strong> Switzerl<strong>and</strong>. As efficiency is seen as a key tool to decrease<br />

energy consumption <strong>and</strong> <strong>in</strong> turn CO 2 emissions, a reliable estimation of<br />

these rebound effects is highly valuable <strong>for</strong> policy makers <strong>in</strong> charge of the<br />

energy transition. Besides, we also analyse <strong>in</strong>dividual heterogeneity <strong>and</strong> determ<strong>in</strong>e<br />

who rebounds the most. In the literature, hypotheses suggest that<br />

people less satisfied with their heat<strong>in</strong>g com<strong>for</strong>t will rebound more. Yet, such<br />

hypotheses are never tested due to the lack of data, except <strong>for</strong> the impact<br />

of <strong>in</strong>come (Chitnis et al., 2014) <strong>and</strong> ownership (Madlener <strong>and</strong> Hauertmann,<br />

2011).<br />

Our data come from a survey that we created <strong>for</strong> this purpose. <strong>The</strong> 3,500<br />

respondents answered a choice experiment <strong>in</strong> which improvements <strong>in</strong> the efficiency<br />

of their heat<strong>in</strong>g system were simulated. To our knowledge, this k<strong>in</strong>d of<br />

experiment has never been conducted to estimate both direct <strong>and</strong> <strong>in</strong>direct rebound<br />

effects. Furthermore, measur<strong>in</strong>g rebound effects <strong>for</strong> residential heat<strong>in</strong>g<br />

has received few attention, with only a h<strong>and</strong>ful studies <strong>in</strong> Europe (Haas <strong>and</strong><br />

Biermayr, 2000; Madlener <strong>and</strong> Hauertmann, 2011; Druckman et al., 2011;<br />

Chitnis et al., 2013). However, results show that the direct rebound effect<br />

<strong>in</strong> this doma<strong>in</strong> is not negligible, with a “best guess” (Sorrell et al., 2009)<br />

between 10 <strong>and</strong> 30% .<br />

Our results <strong>in</strong>dicate an average direct rebound effect between 10 <strong>and</strong> 15%.<br />

Add<strong>in</strong>g the <strong>in</strong>direct rebound effect, we f<strong>in</strong>d an average total household rebound<br />

effect of 36%. It means that policy makers should not overestimate<br />

3


the power of efficiency to dim<strong>in</strong>ish energy consumption <strong>in</strong> the residential<br />

heat<strong>in</strong>g sector, because one third of the benefits of efficiency is lost through<br />

direct <strong>and</strong> <strong>in</strong>direct rebound effects. We also <strong>in</strong>vestigate who rebounds the<br />

most, <strong>and</strong> obta<strong>in</strong> among other that less educated people <strong>and</strong> people with a<br />

lower <strong>in</strong>come have a larger direct rebound effect.<br />

<strong>The</strong> rema<strong>in</strong>der of this article is structured as follows. In section 2, we provide<br />

an overview of how the rebound effect has been def<strong>in</strong>ed <strong>and</strong> measured <strong>in</strong> the<br />

literature, section 3 presents the data we use, <strong>and</strong> section 4 describes the<br />

models we estimate <strong>in</strong> section 5. Conclusions <strong>and</strong> policy implications are<br />

discussed <strong>in</strong> section 6.<br />

2 <strong>Rebound</strong> <strong>Effects</strong> <strong>in</strong> <strong>Residential</strong> <strong>Heat<strong>in</strong>g</strong><br />

Sorrell <strong>and</strong> Dimitropoulos (2008) provide an extensive discussion of the def<strong>in</strong>ition<br />

of the rebound effects (RE). <strong>The</strong> direct RE is basically def<strong>in</strong>ed as an<br />

<strong>in</strong>crease <strong>in</strong> the consumption of an energy service follow<strong>in</strong>g a decrease <strong>in</strong> the<br />

effective price of that service due to an improvement <strong>in</strong> the energy efficiency<br />

of deliver<strong>in</strong>g that service. Energy efficiency may be def<strong>in</strong>ed as ε = S/E, where<br />

E represents energy <strong>in</strong>put <strong>and</strong> S service dem<strong>and</strong>. In our study, S represents<br />

the services provided by heat<strong>in</strong>g. It is not only an <strong>in</strong>ternal temperature, but<br />

heat<strong>in</strong>g com<strong>for</strong>t <strong>in</strong> general. <strong>The</strong> rationale is that heat<strong>in</strong>g dem<strong>and</strong> may <strong>in</strong>crease<br />

because heat<strong>in</strong>g is turned on earlier <strong>in</strong> the season, or because of more<br />

frequent air<strong>in</strong>g, <strong>and</strong> not only because <strong>in</strong>ternal temperature is <strong>in</strong>creased.<br />

Us<strong>in</strong>g elasticities, the direct RE is def<strong>in</strong>ed as the elasticity of the service<br />

dem<strong>and</strong> (S) with respect to efficiency (ε):<br />

η ε (S) = ∂S<br />

∂ε · ε<br />

S<br />

(1)<br />

Because of the lack of data, this def<strong>in</strong>ition is seldom used <strong>in</strong> empirical studies.<br />

Authors usually rely on alternative def<strong>in</strong>itions such as the elasticity of the<br />

service dem<strong>and</strong> with respect to energy price. In the case of heat<strong>in</strong>g, an<br />

4


estimation of the price elasticity <strong>for</strong> heat<strong>in</strong>g fuel is often used <strong>in</strong> the literature<br />

to approximate the direct RE (Madlener <strong>and</strong> Hauertmann, 2011; Haas <strong>and</strong><br />

Biermayr, 2000). Yet, strong assumptions are needed to do so: people have<br />

to react symmetrically to a change <strong>in</strong> price <strong>and</strong> to a change <strong>in</strong> efficiency,<br />

an hypothesis rejected by Greene (2012) <strong>in</strong> the context of private mobility.<br />

Chan <strong>and</strong> Gill<strong>in</strong>gham (2015) criticize strongly this approach, demonstrat<strong>in</strong>g<br />

that it leads to biased estimations of rebound effects.<br />

In the present paper, we do not need to make such restrictive assumptions.<br />

We rely on a second def<strong>in</strong>ition of the RE, which is given by the proportion<br />

of the energy sav<strong>in</strong>gs which are not realised (as Chitnis et al., 2013, do), <strong>in</strong><br />

other words:<br />

<strong>Direct</strong> RE = 1 −<br />

Actual energy sav<strong>in</strong>gs<br />

Potential energy sav<strong>in</strong>gs<br />

(2)<br />

Note that def<strong>in</strong>itions (1) <strong>and</strong> (2) are equivalent s<strong>in</strong>ce Actual energy sav<strong>in</strong>gs =<br />

1 − η ε (S).<br />

<strong>The</strong>re is no consensus <strong>in</strong> the literature about the magnitude of the direct <strong>and</strong><br />

<strong>in</strong>direct rebound effects. For residential space heat<strong>in</strong>g, Sorrell et al. (2009)<br />

review the literature <strong>and</strong> collect estimates of the direct RE rang<strong>in</strong>g between<br />

10 <strong>and</strong> 58% <strong>in</strong> the short run <strong>and</strong> between 1.4 <strong>and</strong> 60% <strong>in</strong> the long run. <strong>The</strong>y<br />

suggest a mean value of 20% <strong>for</strong> the direct rebound effect <strong>for</strong> household<br />

heat<strong>in</strong>g. Nadel (2012), also <strong>in</strong> a review of the literature, advocates a likely<br />

range would be between 1 <strong>and</strong> 12%. He questions studies claim<strong>in</strong>g higher<br />

direct RE, because they are mostly based on price elasticity. In a recent<br />

article (Nadel, 2016), he summarizes the f<strong>in</strong>d<strong>in</strong>gs of studies look<strong>in</strong>g at both<br />

direct <strong>and</strong> <strong>in</strong>direct RE. For residential space heat<strong>in</strong>g, he f<strong>in</strong>ds that direct RE<br />

is generally about 10%, <strong>and</strong> <strong>in</strong>direct RE <strong>in</strong> the range of 10-20%, lead<strong>in</strong>g to<br />

a total rebound of 20 to 30%.<br />

We are not aware of any study that focuses on rebound effects <strong>in</strong> space<br />

heat<strong>in</strong>g <strong>in</strong> Switzerl<strong>and</strong>. In their study about the effects of global warm<strong>in</strong>g on<br />

energy use <strong>in</strong> Switzerl<strong>and</strong>, Gonseth et al. (2015) f<strong>in</strong>d a direct RE of 35% <strong>in</strong><br />

the long-run (up to 2060) us<strong>in</strong>g a CGE model. Madlener <strong>and</strong> Hauertmann<br />

5


(2011) <strong>and</strong> Haas <strong>and</strong> Biermayr (2000) <strong>in</strong>vestigate a similar topic <strong>for</strong> two<br />

neighbour<strong>in</strong>g countries, namely Germany <strong>and</strong> Austria. For Germany, a direct<br />

RE between 12 to 49% is found, with tenants hav<strong>in</strong>g a higher rebound than<br />

owners, <strong>and</strong> between 20 to 32% <strong>for</strong> Austria. <strong>The</strong>y both use price-elasticity.<br />

Some studies rely on eng<strong>in</strong>eer<strong>in</strong>g calculations, compar<strong>in</strong>g predicted energy<br />

use with observed energy use. For <strong>in</strong>stance Ayd<strong>in</strong> et al. (2014) study 500,000<br />

households <strong>in</strong> the Netherl<strong>and</strong>s, compar<strong>in</strong>g the energy label of dwell<strong>in</strong>gs with<br />

their energy consumption. <strong>The</strong>y f<strong>in</strong>d a direct RE of 28% <strong>for</strong> owners <strong>and</strong><br />

42% <strong>for</strong> tenants. This technique rely<strong>in</strong>g on eng<strong>in</strong>eer<strong>in</strong>g predictions has also<br />

been criticised, mostly because these predictions over-estimate the potential<br />

energy sav<strong>in</strong>gs of the efficiency improvements. Fowlie et al. (2015) study<br />

30,000 households participat<strong>in</strong>g <strong>in</strong> an energy efficiency program <strong>in</strong> the USA.<br />

<strong>The</strong>y f<strong>in</strong>d that the projected sav<strong>in</strong>gs by eng<strong>in</strong>eers are roughly 2.5 times the<br />

actual sav<strong>in</strong>gs. A part of this discrepancy may be due to direct RE. As they<br />

def<strong>in</strong>e it (an <strong>in</strong>crease <strong>in</strong> the <strong>in</strong>door temperature), they f<strong>in</strong>d no evidence <strong>for</strong><br />

it, attribut<strong>in</strong>g all the discrepancy to over-estimated eng<strong>in</strong>eer calculations.<br />

Although we argue <strong>in</strong> this article that direct RE is not only due to higher<br />

temperatures, a large part of the discrepancy rema<strong>in</strong>s expla<strong>in</strong>ed by overestimations<br />

of the eng<strong>in</strong>eers’ projections.<br />

<strong>The</strong> way we measure the direct RE, by us<strong>in</strong>g scenarios <strong>in</strong> a survey <strong>and</strong><br />

ask<strong>in</strong>g respondents about their reaction, is <strong>in</strong>novative <strong>in</strong> the field. It is the<br />

first time that this k<strong>in</strong>d of experiment is used to assess the rebound effect<br />

<strong>for</strong> residential heat<strong>in</strong>g. Only two authors previously used a survey <strong>in</strong> the<br />

rebound effect literature; Schleich et al. (2014) <strong>in</strong> Germany <strong>for</strong> lightn<strong>in</strong>g,<br />

<strong>and</strong> Yu et al. (2013) <strong>in</strong> Japan <strong>for</strong> vehicles.<br />

Most of the literature focuses solely on the direct RE, yet the <strong>in</strong>direct is<br />

also important <strong>for</strong> energy policies. <strong>The</strong> <strong>in</strong>direct rebound effect is l<strong>in</strong>ked<br />

to <strong>in</strong>come <strong>and</strong> substitution effects, although the latter is often put aside<br />

<strong>in</strong> the literature because it is considered as small (Thomas <strong>and</strong> Azevedo,<br />

2013a). We also consider only the <strong>in</strong>come effect, i.e., how the <strong>in</strong>come freed<br />

up by the efficiency improvement is spent. By spend<strong>in</strong>g it, overall energy<br />

consumption will decrease by less than the amount predicted by the efficiency<br />

6


improvement, as any product <strong>and</strong> service requires energy to be created <strong>and</strong><br />

used. To estimate this variation <strong>in</strong> energy consumption we take <strong>in</strong>to account<br />

not only the direct energy but also the embodied (grey) energy.<br />

Most of the studies on the <strong>in</strong>direct RE are based upon <strong>in</strong>come elasticities <strong>for</strong><br />

diverse categories of goods <strong>and</strong> services (Chitnis et al., 2013, <strong>for</strong> <strong>in</strong>stance) or<br />

<strong>in</strong>put-output tables (Thomas <strong>and</strong> Azevedo, 2013a; Lenzen <strong>and</strong> Dey, 2002).<br />

Data on the energy <strong>in</strong>tensity of the categories of goods <strong>and</strong> services are<br />

added to estimate the variation <strong>in</strong> energy consumption. Our estimation of<br />

the <strong>in</strong>direct RE is based on self-reported respend<strong>in</strong>g patterns <strong>and</strong> data on the<br />

embodied energy <strong>in</strong> kWh <strong>for</strong> categories of goods <strong>and</strong> services. We estimate<br />

at both the direct <strong>and</strong> the <strong>in</strong>direct RE, which very few studies have achieved<br />

be<strong>for</strong>e: Druckman et al. (2011) <strong>and</strong> Chitnis et al. (2013) conduct such analyses<br />

<strong>in</strong> terms of GHG emissions, f<strong>in</strong>d<strong>in</strong>g respectively a total rebound effect<br />

between 12 <strong>and</strong> 34% <strong>and</strong> between 5 <strong>and</strong> 15%.<br />

3 Data<br />

Our data come from an orig<strong>in</strong>al onl<strong>in</strong>e survey carried out <strong>in</strong> September 2015<br />

with 3,555 respondents representative of the Swiss population (exclud<strong>in</strong>g the<br />

Italian-speak<strong>in</strong>g region). Quotas match<strong>in</strong>g data from the Swiss Federal Office<br />

of Statistics were imposed <strong>for</strong> age, gender <strong>and</strong> region 1 . For <strong>in</strong>come levels we<br />

checked afterwards whether our sample was consistent with national data,<br />

<strong>and</strong> it differs only slightly (see figure A.4 <strong>in</strong> the appendix). <strong>The</strong> lowest <strong>and</strong><br />

the highest categories are under-represented <strong>in</strong> our sample. Yet we do not<br />

consider this as a problem <strong>in</strong> the estimation of the direct RE, s<strong>in</strong>ce we show<br />

later that both categories have an opposite effect on the magnitude of the<br />

direct RE. Thus it does not lead to a systematic upward or downward bias.<br />

<strong>The</strong> key questions of the survey regard<strong>in</strong>g the direct RE were asked us<strong>in</strong>g<br />

scenarios, which simulated an improvement <strong>in</strong> the efficiency of the heat<strong>in</strong>g<br />

1 Four regions represent the country: Rom<strong>and</strong>ie, Alps <strong>and</strong> Pre-Alps, Plateau West <strong>and</strong><br />

Plateau East.<br />

7


system. Each respondent was faced with three different scenarios: 1) a small<br />

improvement <strong>in</strong> efficiency (10, 15 or 20% r<strong>and</strong>omly chosen); 2) a large improvement<br />

(40, 50 or 60%); 3) a medium improvement (25 or 30%) aris<strong>in</strong>g<br />

from a CO 2 neutral technology. Under each scenario, we asked respondents<br />

whether they would modify their heat<strong>in</strong>g habits, <strong>and</strong> if yes by how much 2 .<br />

We used both qualitative <strong>and</strong> quantitative questions to ensure respondents<br />

understood the potential actions offered to them follow<strong>in</strong>g an efficiency improvement<br />

<strong>and</strong> to check they answered consistently.<br />

We should also acknowledge that our survey rules out superconservation (i.e.,<br />

negative rebound) <strong>and</strong> backfire (i.e., rebound larger than 100%) by design.<br />

Indeed, <strong>in</strong> the quantitative question, respondents were constra<strong>in</strong>ed to choose<br />

a number between 0 <strong>and</strong> the efficiency improvement <strong>in</strong> %. For <strong>in</strong>stance, if<br />

the scenario was based on an efficiency improvement of 20%, the scale of the<br />

slider along which respondents could make a choice went from 0 to 20 (like<br />

<strong>in</strong> figure A.2). We did so deliberately, <strong>in</strong> order to avoid potential confusion<br />

<strong>in</strong> the respondents, who would probably have wondered how such cases are<br />

possible. In light of the estimates of the direct rebound effect available <strong>in</strong><br />

the literature, most of which be<strong>in</strong>g around 20% <strong>and</strong> none be<strong>in</strong>g outside the<br />

range from 0 to 100%, we do not consider these constra<strong>in</strong>ts as a flaw of our<br />

design.<br />

To estimate the direct RE <strong>and</strong> to analyse who rebounds the most, we use<br />

a set of controls, namely education, <strong>in</strong>come, owner or tenant status, satisfaction<br />

regard<strong>in</strong>g heat<strong>in</strong>g com<strong>for</strong>t, environmental attitudes, <strong>and</strong> <strong>in</strong>door<br />

2 First, <strong>in</strong> a qualitative question, we provided possible reactions, such as sett<strong>in</strong>g the<br />

thermostat higher, air<strong>in</strong>g more often, or heat<strong>in</strong>g earlier/later <strong>in</strong> the season. This qualitative<br />

question was <strong>in</strong>tended to help the respondents th<strong>in</strong>k about the potential actions<br />

that have an impact on their fuel consumption related to heat<strong>in</strong>g <strong>and</strong> to frame them <strong>for</strong><br />

the next quantitative question. If at least one of the possible reactions was chosen, we<br />

requested the respondent to state how much of the sav<strong>in</strong>gs would be used to <strong>in</strong>crease heat<strong>in</strong>g<br />

services. Based on the def<strong>in</strong>ition of efficiency ε = S/E, we observe that when efficiency<br />

improves by ∆ε%, services will <strong>in</strong>crease by ∆ε% if energy consumption is kept constant,<br />

or energy consumption will decrease by ∆ε% if services are kept constant, <strong>and</strong> any other<br />

comb<strong>in</strong>ation with ∆S% − ∆E% = ∆ε% is possible. This is the range of possibilities we<br />

offer the respondents <strong>in</strong> the qualitative question. <strong>The</strong> questions relative to the direct RE<br />

are available <strong>in</strong> the appendix figures A.1 <strong>and</strong> A.2.<br />

8


temperature. Table 1 shows descriptive statistics <strong>for</strong> these variables.<br />

Table 1: Summary Statistics<br />

Variable Mean Std. dev. M<strong>in</strong>. Max. N<br />

Age 47.78 16.27 19 89 3,555<br />

Education 7.37 1.92 2 9 3,554<br />

Income level 3.97 1.41 1 6 2,924<br />

Female 0.51 0.50 0 1 3,555<br />

Tenants 0.58 0.49 0 1 3,555<br />

Environmental attitudes 22.88 6.41 0 36 3,543<br />

<strong>Heat<strong>in</strong>g</strong> satisfaction 10.40 3.22 0 15 3,477<br />

Indoor temperature 21.20 1.60 9 30 3,444<br />

Education goes from 1 (compulsory school not f<strong>in</strong>ished) to 9 (University <strong>and</strong><br />

higher vocational tra<strong>in</strong><strong>in</strong>g). Figure A.3 shows the frequencies. We group<br />

some levels together <strong>for</strong> the analysis, follow<strong>in</strong>g groups from the Federal Office<br />

of Statistics (FOS). <strong>The</strong> four groups are described <strong>in</strong> table A.1, as well as the<br />

percentage from the FOS. Compared to the whole population, people with a<br />

low education level (only compulsory school f<strong>in</strong>ished) are under-represented<br />

<strong>in</strong> our sample, <strong>and</strong> people with a high degree of education over-represented.<br />

This may slightly under-estimate the direct rebound effects, as we show <strong>in</strong><br />

the analysis that people less educated have a higher direct RE.<br />

<strong>Heat<strong>in</strong>g</strong> satisfaction is constructed as an <strong>in</strong>dex us<strong>in</strong>g the answers to five questions<br />

about the satisfaction on <strong>in</strong>ternal temperature <strong>in</strong> w<strong>in</strong>ter, uni<strong>for</strong>mity<br />

of temperature, quality of the ventilation, ease to modify the temperature<br />

<strong>and</strong> the number of days with the heat<strong>in</strong>g on. Respondents could be totally<br />

unsatisfied, rather unsatisfied, rather satisfied <strong>and</strong> totally satisfied of each<br />

item. To create the <strong>in</strong>dex, we attributed po<strong>in</strong>ts, respectively from 0 (totally<br />

unsatisfied) to 3 (totally satisfied). Consequently the respondent totally unsatisfied<br />

<strong>for</strong> the 5 items gets 0 po<strong>in</strong>t, whereas the respondent totally satisfied<br />

<strong>for</strong> the 5 items gets 15 po<strong>in</strong>ts. We proceeded <strong>in</strong> a similar way to build an<br />

<strong>in</strong>dex of environmental attitudes. Such an <strong>in</strong>dex has been used previously <strong>in</strong><br />

the literature, by Best <strong>and</strong> Mayerl (2013). N<strong>in</strong>e items were presented to the<br />

respondents who have to disagree-agree on a five-po<strong>in</strong>t scale. If the respon-<br />

9


dent completely disagrees with one item, he gets 0 po<strong>in</strong>t, if he completely<br />

agrees 4 po<strong>in</strong>ts. Hence the m<strong>in</strong>imal score is 0 (<strong>for</strong> those who care very little<br />

about environment) <strong>and</strong> the maximum is 36. <strong>The</strong> (stated) <strong>in</strong>door temperature<br />

goes from 9 to 30 degrees, but 96% of the answers lie between 18 <strong>and</strong> 25<br />

degrees. To dim<strong>in</strong>ish the impact of the outliers <strong>and</strong> because the answers are<br />

not cont<strong>in</strong>uous , we create three categories:


83 studies conta<strong>in</strong><strong>in</strong>g 616 stated-revealed preferences comparisons <strong>for</strong> quasipublic<br />

goods f<strong>in</strong>d that stated preferences estimates are smaller, but only<br />

slightly. However the downward bias is not systematic, <strong>and</strong> they conclude<br />

that there is no general need to correct stated preferences data. More recently,<br />

Grijalva et al. (2002) studied visitors’ behaviour <strong>in</strong> a climb<strong>in</strong>g area.<br />

<strong>The</strong>y f<strong>in</strong>d that <strong>in</strong>tended behaviour matches aggregate actual data. Loomis<br />

<strong>and</strong> Richardson (2006) exam<strong>in</strong>e the number of visits to a national park. <strong>The</strong>y<br />

conclude that the estimates from the stated <strong>and</strong> revealed preference methods<br />

are not statistically different from one another.<br />

To our knowledge, hypothetical choices have never been used to estimate direct<br />

rebound effect <strong>in</strong> the heat<strong>in</strong>g sector. In view of the literature compar<strong>in</strong>g<br />

stated <strong>and</strong> revealed preferences, we acknowledge that there may be a bias,<br />

but limited. To obta<strong>in</strong> results as robust as possible, we used two complementary<br />

strategies. First, we collected <strong>in</strong><strong>for</strong>mation on the rebound effect <strong>in</strong><br />

two different flavors, once through a qualitative question <strong>and</strong> once through<br />

a quantitative one. Based on these questions, we can dist<strong>in</strong>guish consistent<br />

<strong>and</strong> <strong>in</strong>consistent respondents, with the possibility to keep only the consistent<br />

ones <strong>for</strong> the analysis. Second, every respondent was faced with three different<br />

scenarios, under which he had to provide answers to the same questions.<br />

<strong>The</strong>se repeated questions allow us to build a panel. When controll<strong>in</strong>g <strong>for</strong><br />

<strong>in</strong>dividual heterogeneity <strong>in</strong> this way, our estimates will be based on changes<br />

with<strong>in</strong> <strong>in</strong>dividuals from one scenario to the other.<br />

Moreover there are many advantages <strong>in</strong> us<strong>in</strong>g stated preferences, Whitehead<br />

et al. (2008) summarize them. Stated preferences data are useful <strong>for</strong> the<br />

analysis of new policies, <strong>for</strong> which revealed preferences data do not exist.<br />

It is particularly relevant <strong>in</strong> the case of new policies lead<strong>in</strong>g to a change <strong>in</strong><br />

behaviour, <strong>in</strong> this case hypothetical choices may be the only way to gather<br />

<strong>in</strong><strong>for</strong>mation. Additionally, it <strong>in</strong>creases the sample size <strong>in</strong> a cross-section,<br />

as usually the respondents answer several times the hypothetical questions.<br />

<strong>The</strong>ir choices can be treated as panel data, provid<strong>in</strong>g more <strong>in</strong><strong>for</strong>mation about<br />

the preferences of each respondent.<br />

11


4 Empirical strategy<br />

We estimate the direct rebound effect us<strong>in</strong>g <strong>for</strong>mula (2). In the choice<br />

experiment of the survey, potential sav<strong>in</strong>gs are given. Respondents can subsequently<br />

choose the variation of service level, i.e., the part of the potential<br />

energy sav<strong>in</strong>gs which is taken back because of people’s reaction. Answers<br />

collected through this choice experiment thus provide all necessary data to<br />

apply <strong>for</strong>mula (2) directly: potential energy sav<strong>in</strong>gs are given by the efficiency<br />

improvement, while actual energy sav<strong>in</strong>gs are deduced from respondents’ answers.<br />

This strategy provides a different rebound effect <strong>for</strong> every <strong>in</strong>dividual<br />

<strong>and</strong> each scenario We use it to test <strong>for</strong> the impact of <strong>in</strong>dividual characteristics<br />

on its magnitude. <strong>The</strong> estimated equation is:<br />

<strong>Direct</strong> RE i,t = α + β 1 Educ i,t + β 2 Income i,t + β 3 F emale i,t<br />

+ β 4 Owner i,t + β 5 Environ i,t + β 6 Satisf i,t (3)<br />

+ β 7 T emperature i,t + β 8 Scenario i,t + u i<br />

Where i denotes the <strong>in</strong>dividual, t (= 1, 2, 3) denotes the scenario, <strong>and</strong> u is a<br />

r<strong>and</strong>om <strong>in</strong>tercept .<br />

Satisf is the score on the <strong>in</strong>dex of heat<strong>in</strong>g satisfaction <strong>and</strong> Environ the score<br />

on the environmental attitudes <strong>in</strong>dex. Income is divided <strong>in</strong> three classes<br />

(CHF): 9000. <strong>The</strong> variable Owner <strong>in</strong>dicates whether<br />

the <strong>in</strong>dividual owns or rents is residence <strong>and</strong> whether the heat<strong>in</strong>g bill is<br />

calculated on the <strong>in</strong>dividual consumption or shared among all the <strong>in</strong>habitants<br />

(calculated as a proportion of the m 2 of the home). Four categories exist <strong>for</strong><br />

this variable: owners with <strong>in</strong>dividual costs, owners with shared costs, tenants<br />

with <strong>in</strong>dividual costs, <strong>and</strong> tenants with shared costs. <strong>The</strong> rationale <strong>for</strong> that<br />

dist<strong>in</strong>ction is that tenants with <strong>in</strong>dividual costs should behave as owners.<br />

This hypothesis has never been tested be<strong>for</strong>e. Scenario is the small, middle<br />

(with the CO 2 neutral technology) or large efficiency improvement.<br />

We use a two-part model to estimate equation (3). As described by Belotti<br />

et al. (2015), a two-part model is appropriate when the outcome (y i ) has<br />

12


two features: 1) y i ≥ 0, <strong>and</strong> 2) y i = 0 is observed often enough. In our<br />

dataset, the direct RE is either zero or a positive value, with about half<br />

of the observations be<strong>in</strong>g zero. <strong>The</strong> two-part model accounts <strong>for</strong> the mass<br />

of zeroes by first fitt<strong>in</strong>g a b<strong>in</strong>ary choice model, <strong>and</strong> then, conditional on a<br />

positive outcome, fitt<strong>in</strong>g a l<strong>in</strong>ear regression model <strong>for</strong> the positive outcomes.<br />

Equation (3) is the second part of the two-part model. <strong>The</strong> first part is the<br />

b<strong>in</strong>ary choice model, <strong>and</strong> we use a probit model:<br />

P r(Y i,t = 1|X i,t ) = φ(X ′ i,tβ) (4)<br />

<strong>The</strong> vector X i ′ <strong>in</strong>cludes the same variables as <strong>in</strong> equation (3). In the two<br />

equations of the model we use panel data regression method. It is more efficient<br />

(lower st<strong>and</strong>ard errors) than pooled models, because the pooled model<br />

assumes <strong>in</strong>dependence over i <strong>and</strong> t. In our case, because all the <strong>in</strong>dependent<br />

variables are time-<strong>in</strong>variant, r<strong>and</strong>om effects are chosen. <strong>The</strong> r<strong>and</strong>om<br />

<strong>and</strong> <strong>in</strong>dividual <strong>in</strong>tercept u i accounts <strong>for</strong> autocorrelation due to the omitted<br />

time-<strong>in</strong>variant variables <strong>for</strong> a respondent.<br />

We expect a negative effect on the direct RE of education, <strong>in</strong>come, heat<strong>in</strong>g<br />

satisfaction <strong>and</strong> environmental attitude. It is obvious that people already<br />

fully satisfied with their heat<strong>in</strong>g com<strong>for</strong>t will be less prone to a direct RE,<br />

as well as people pay<strong>in</strong>g more attention to the environment. <strong>The</strong> effect<br />

of <strong>in</strong>come has been analysed <strong>in</strong> two previous studies (Chitnis et al., 2014;<br />

Madlener <strong>and</strong> Hauertmann, 2011), both f<strong>in</strong>d<strong>in</strong>g that richer people rebound<br />

less. An explanation is that they were not restrict<strong>in</strong>g their usage of heat<strong>in</strong>g<br />

be<strong>for</strong>e the efficiency improvement, <strong>and</strong> hence were already at their maximum<br />

level of com<strong>for</strong>t. Some evidence of the impact of <strong>in</strong>come on the rebound effect<br />

can also be found <strong>in</strong> macro-level studies <strong>and</strong> <strong>in</strong> various fields. For <strong>in</strong>stance,<br />

Small <strong>and</strong> Van Dender (2007) f<strong>in</strong>d that the rebound effect <strong>for</strong> motor vehicles<br />

decl<strong>in</strong>ed over 1966-2001 <strong>in</strong> the US, which they expla<strong>in</strong> by the rise <strong>in</strong><br />

<strong>in</strong>comes. More <strong>in</strong> general, rebound effect estimates are larger <strong>for</strong> develop<strong>in</strong>g<br />

countries than <strong>for</strong> developed ones, which is <strong>in</strong>terpreted as the impact of<br />

<strong>in</strong>come constra<strong>in</strong>ts <strong>in</strong> the <strong>for</strong>mer (e.g., Azevedo, 2014). <strong>The</strong>re<strong>for</strong>e, an <strong>in</strong>ter-<br />

13


est<strong>in</strong>g contribution of our paper is that we are able to determ<strong>in</strong>e the effect<br />

of <strong>in</strong>come on the rebound effect at the <strong>in</strong>dividual level. For education it is<br />

less obvious why less educated people would rebound more. A possibility is<br />

that they are less aware of the impact of their actions on the climate, there<strong>for</strong>e<br />

they will not restrict their heat<strong>in</strong>g usage <strong>for</strong> environmental concerns.<br />

F<strong>in</strong>ally we expect that the owners with shared or <strong>in</strong>dividual costs <strong>and</strong> the<br />

tenants with <strong>in</strong>dividual costs rebound more than the tenants with shared<br />

costs. <strong>The</strong> latter have fewer <strong>in</strong>centives to limit their heat<strong>in</strong>g usage <strong>and</strong> then<br />

could already have reached a maximum heat<strong>in</strong>g com<strong>for</strong>t prior to the efficiency<br />

improvement. Madlener <strong>and</strong> Hauertmann (2011) f<strong>in</strong>d that tenants rebound<br />

more than owners, but they do not exam<strong>in</strong>e the features of the heat<strong>in</strong>g bill.<br />

For temperature we expect a negative coefficient, as people gets closer to<br />

their satiety po<strong>in</strong>t.<br />

With the scenario dummies, we test whether the direct RE is l<strong>in</strong>ear with the<br />

efficiency improvement, i.e., whether a larger improvement would <strong>in</strong>duce a<br />

larger direct RE. <strong>The</strong> green technology <strong>in</strong> the second scenario is used to see<br />

whether people react more (have a larger direct RE) if on top of the efficiency<br />

improvement they benefit from a green technology. <strong>The</strong> hypothesis is that<br />

people would react more because they get rid of a sort of guilt of consum<strong>in</strong>g<br />

more of a pollut<strong>in</strong>g service. We also tested <strong>for</strong> the impact of other variables:<br />

age, the region, the hous<strong>in</strong>g type, the size of the house or flat, the size of the<br />

household, the number of children, rural or urban household. None of them<br />

was significant.<br />

5 Results<br />

5.1 Estimations <strong>and</strong> Analysis of the <strong>Direct</strong><br />

<strong>Rebound</strong> Effect<br />

<strong>The</strong> implementation of <strong>for</strong>mula (2) gives an <strong>in</strong>dividual RE <strong>for</strong> each respondent<br />

<strong>in</strong> each scenario, i.e., three estimates <strong>for</strong> every respondent. <strong>The</strong> overall<br />

14


esults are very similar <strong>for</strong> the three scenarios, with an average direct RE estimated<br />

between 10 <strong>and</strong> 15%. When the three scenarios are pooled together,<br />

the average direct RE is 12.2%. Table 2 reports the values obta<strong>in</strong>ed <strong>and</strong><br />

figure A.5 displays the distribution of the strictly positive values. A direct<br />

RE of 12.2% is rather <strong>in</strong> the low-end of the estimations obta<strong>in</strong>ed so far <strong>in</strong><br />

the literature, but it is consistent with the reviews by Sorrell et al. (2009)<br />

(10-30%) <strong>and</strong> Nadel (2016) (1-12%).<br />

Table 2:<br />

Estimation of the <strong>Direct</strong> <strong>Rebound</strong> Effect<br />

Mean Std. dev. M<strong>in</strong>. Max. N<br />

<strong>Direct</strong> RE (Small eff. improvement) 0.147 0.226 0 1 3,555<br />

<strong>Direct</strong> RE (Large eff. improvement) 0.106 0.187 0 1 3,555<br />

<strong>Direct</strong> RE (Middle eff. imp. + CO 2 neutral) 0.113 0.193 0 1 3,555<br />

<strong>Direct</strong> RE: pooled 0.122 0.203 0 1 10,665<br />

Note: <strong>The</strong> means are calculated us<strong>in</strong>g <strong>for</strong>mula (2) <strong>and</strong> can be multiplied by 100<br />

to obta<strong>in</strong> to direct RE <strong>in</strong> %. About half of the respondents have a zero direct RE.<br />

In the literature, the only possibility usually considered <strong>for</strong> a direct RE <strong>in</strong><br />

heat<strong>in</strong>g is an <strong>in</strong>crease <strong>in</strong> the temperature. Yet, people can react to an efficiency<br />

improvement <strong>in</strong> various ways, <strong>for</strong> <strong>in</strong>stance by air<strong>in</strong>g more often their<br />

house, or start<strong>in</strong>g to heat earlier <strong>in</strong> the season, etc. <strong>The</strong> purpose of the qualitative<br />

question <strong>in</strong> the survey is precisely to <strong>in</strong>vestigate such possibilities. <strong>The</strong><br />

question is made of 5 items about the hypothetical change <strong>in</strong> behaviour, with<br />

three possible choices: no, maybe, yes. Figure A.6 <strong>in</strong> the appendix shows<br />

the answers. Air<strong>in</strong>g longer <strong>and</strong> more frequently is <strong>in</strong>deed what respondents<br />

chose the more often, be<strong>for</strong>e pay<strong>in</strong>g less attention to heat<strong>in</strong>g <strong>in</strong> general. Increas<strong>in</strong>g<br />

temperature only comes as the third most popular choice. It shows<br />

that by consider<strong>in</strong>g only the <strong>in</strong>crease <strong>in</strong> temperature, the direct RE will be<br />

underestimated (see also Voll<strong>and</strong>, 2016).<br />

15


One concern with data com<strong>in</strong>g from a choice experiment is to make sure<br />

that respondents understood correctly the questions. In our sett<strong>in</strong>g, we can<br />

control <strong>for</strong> that by compar<strong>in</strong>g<br />

answers to the qualitative question on the<br />

rebound effect (an easy question) <strong>and</strong> to the quantitative question (more<br />

difficult). We attributed zero, one <strong>and</strong> two po<strong>in</strong>ts to the qualitative question<br />

(no, maybe, yes) to obta<strong>in</strong> an overall view of the changes <strong>in</strong> the heat<strong>in</strong>g<br />

habits. Hence, more po<strong>in</strong>ts <strong>in</strong>dicate more substantial modifications of habits<br />

<strong>and</strong> a larger direct rebound effect. Someone who would change noth<strong>in</strong>g gets<br />

0 po<strong>in</strong>t. Table 3 highlights that respondents answered consistently to both<br />

questions. <strong>The</strong> more they decided to change <strong>in</strong> the qualitative question, the<br />

largest changes they reported <strong>in</strong> the quantitative question.<br />

Table 3: Consistency <strong>in</strong> the stated preferences<br />

Qualitative RE 0 1 2 3 4 5 6<br />

Quantitative RE 0% 17% 21% 26% 29.8% 33.9% 32.8%<br />

Respondents with 0 po<strong>in</strong>t <strong>in</strong> the qualitative question get mechanically a direct RE<br />

of zero as the quantitative question did not appear <strong>for</strong> them. <strong>The</strong> qualitative RE is<br />

an <strong>in</strong>dex, while the quantitative rebound is estimated with equation (2).<br />

By compar<strong>in</strong>g the answers of each respondent to the qualitative <strong>and</strong> quantitative<br />

questions, we can def<strong>in</strong>e who is <strong>in</strong>consistent. An <strong>in</strong>consistent respondent<br />

is someone who said <strong>in</strong> the qualitative question that he would change his heat<strong>in</strong>g<br />

behaviour along at least one dimension, but did not report any change<br />

<strong>in</strong> the correspond<strong>in</strong>g quantitative question. Hence <strong>for</strong> him the quantitative<br />

direct rebound we calculate is zero, while his qualitative rebound is non-zero.<br />

It is likely that he did not underst<strong>and</strong> well the quantitative question, or he<br />

was not able to answer it. Inconsistent respondents are less educated <strong>and</strong><br />

have a lower <strong>in</strong>come than the average. As a robustness check, we per<strong>for</strong>m<br />

our analysis by exclud<strong>in</strong>g them. It does not change significantly our results.<br />

Once the direct RE estimated <strong>for</strong> each <strong>in</strong>dividual, we can assess which characteristics<br />

<strong>in</strong>fluence the magnitude of the direct RE, or who rebounds the<br />

most. This is <strong>in</strong>vestigated us<strong>in</strong>g equation (3) with a two-part model, whose<br />

results are reported <strong>in</strong> table 4). For the first part of the model (a probit es-<br />

16


timation), we report the marg<strong>in</strong>al effects at the means. <strong>The</strong> marg<strong>in</strong>al effects<br />

<strong>in</strong>dicate the <strong>in</strong>crease <strong>in</strong> the probability of hav<strong>in</strong>g a positive direct rebound<br />

(versus a zero direct rebound) associated with a one-unit <strong>in</strong>crease of the correspond<strong>in</strong>g<br />

variables. Only respondents with a positive direct RE are kept <strong>for</strong><br />

the second part, <strong>and</strong> the coefficients <strong>in</strong> this estimation are to be <strong>in</strong>terpreted<br />

as the <strong>in</strong>crease <strong>in</strong> the magnitude of the direct rebound follow<strong>in</strong>g a one-unit<br />

<strong>in</strong>crease of the correspond<strong>in</strong>g variable.<br />

In both estimations, education <strong>and</strong> <strong>in</strong>come have the expected negative sign.<br />

For <strong>in</strong>stance, people with the highest level of education have a 20% lower<br />

probability to have a positive direct rebound, compared to people who stopped<br />

at compulsory school. However the effect of education is not significant <strong>in</strong><br />

the l<strong>in</strong>ear regression. Be<strong>in</strong>g a woman decreases both the probability <strong>and</strong><br />

the magnitude of the direct RE. Environmental attitudes <strong>and</strong> heat<strong>in</strong>g satisfaction<br />

also have the expected negative sign: an <strong>in</strong>crease <strong>in</strong> one st<strong>and</strong>ard<br />

deviation <strong>for</strong> environmental attitudes leads to a decrease by 4% <strong>in</strong> the probability<br />

of direct rebound, <strong>for</strong> heat<strong>in</strong>g satisfaction it is by 8%. People more<br />

environmentally conscious have also a lower direct RE.<br />

<strong>The</strong> coefficients related to <strong>in</strong>door temperature are unexpectedly positive. It<br />

is robust even if we change the choice of the degrees <strong>in</strong> the categories, <strong>and</strong><br />

also if we drop the outliers (less than 17 degrees or more than 25). However<br />

if the reference category is (19.5-22.5), the last category is not significantly<br />

different. Our <strong>in</strong>terpretation is that a part of the people with a low <strong>in</strong>door<br />

temperature have a preference <strong>for</strong> this <strong>and</strong> will not change even if the service<br />

is cheaper. A question <strong>in</strong> the survey is “Would you be will<strong>in</strong>g to dim<strong>in</strong>ish<br />

the temperature <strong>in</strong> w<strong>in</strong>ter?” <strong>The</strong>y are 56% answer<strong>in</strong>g that they already<br />

do it, versus 38 <strong>and</strong> 20% respectively <strong>for</strong> the middle <strong>and</strong> high temperature<br />

categories.<br />

<strong>The</strong> negative signs on the magnitude of the efficiency improvement shows that<br />

people rebound, but up to a certa<strong>in</strong> po<strong>in</strong>t <strong>and</strong> not <strong>in</strong>def<strong>in</strong>itely. Imag<strong>in</strong>e you<br />

would feel more com<strong>for</strong>table if you heat your home slightly more, <strong>for</strong> <strong>in</strong>stance<br />

us<strong>in</strong>g 15% more of heat<strong>in</strong>g energy. <strong>The</strong>n if the efficiency improvement is 15%<br />

you have a direct RE of 100%, but if the efficiency improvement is 50% your<br />

17


Table 4:<br />

Two-part model <strong>for</strong> the <strong>Direct</strong> RE<br />

Probit model, MEs<br />

L<strong>in</strong>ear regression<br />

Apprenticeship −0.151 ** −0.037<br />

(0.076) (0.042)<br />

Baccalaureate −0.186 ** −0.036<br />

(0.079) (0.044)<br />

University <strong>and</strong> higher vocational<br />

tra<strong>in</strong><strong>in</strong>g<br />

−0.200 *** −0.067<br />

(0.076) (0.042)<br />

Income 4500-9000 −0.029 −0.025 *<br />

(0.023) (0.014)<br />

Income >9000 −0.074 *** −0.023<br />

(0.025) (0.015)<br />

Female −0.035 ** −0.035 ***<br />

(0.017) (0.010)<br />

Owners shared costs 0.032 0.044 **<br />

(0.035) (0.021)<br />

Tenants shared costs −0.029 −0.015<br />

(0.023) (0.014)<br />

Tenants <strong>in</strong>dividual costs −0.002 0.004<br />

(0.021) (0.013)<br />

Environmental attitudes −0.006 *** −0.002 ***<br />

(0.001) (0.001)<br />

<strong>Heat<strong>in</strong>g</strong> satisfaction −0.024 *** −0.002<br />

(0.003) (0.002)<br />

19.5-22.5 degrees 0.066 ** −0.005<br />

(0.029) (0.019)<br />

>22.5 degrees 0.089 *** 0.004<br />

(0.034) (0.021)<br />

Large eff. improv. −0.020 ** −0.079 ***<br />

(0.008) (0.005)<br />

Middle eff. improv. + CO2 neutral −0.033 *** −0.057 ***<br />

(0.008) (0.005)<br />

Constant − 0.468 ***<br />

#Obs. 8, 391 3, 812<br />

Note: St<strong>and</strong>ard errors <strong>in</strong> parentheses. ∗ p < .1, ∗∗ p < .05, ∗∗∗<br />

p < .01<br />

<strong>The</strong> dependent variable is the direct RE as estimated with <strong>for</strong>mula<br />

(2).<br />

18<br />

(0.054)


direct RE is only 30% (= 15/50). Here the magnitude of the improvement<br />

decreases both the probability to rebound <strong>and</strong> the magnitude of the direct<br />

rebound. In the probit model, the scenario with the middle improvement<br />

<strong>and</strong> the CO 2 neutral technology leads to a higher decrease <strong>in</strong> probability<br />

of rebound than the large efficiency improvement (the two coefficients are<br />

statistically different at the 10% level). Contrary to what was expected, the<br />

green technology does not seem to lead to a higher direct rebound, but rather<br />

to a decrease <strong>in</strong> the probability of rebound. On the magnitude of the rebound,<br />

the effect is not clear. Some people rebound only <strong>in</strong> this third scenario,<br />

but they are only 116 respondents (3.3% of the sample). Interest<strong>in</strong>gly, a<br />

probit model shows that environmental attitudes as well as a low <strong>in</strong>door<br />

temperature <strong>in</strong>crease the likelihood of belong<strong>in</strong>g to this group of respondents.<br />

It thus seems that a “green rebound” does exist, but only a marg<strong>in</strong>al one.<br />

Hypotheses on who will rebound the most have often been made <strong>in</strong> previous<br />

research, but very rarely tested due to the lack of data. We here show that<br />

education, <strong>in</strong>come, environmental attitudes <strong>and</strong> heat<strong>in</strong>g staisfaction decrease<br />

the probability of the direct rebound effect. Our f<strong>in</strong>d<strong>in</strong>gs highlight important<br />

facts: First, the magnitude of the direct RE is very heterogeneous among<br />

people. Second, the direct RE depends on the level of the efficiency improvement,<br />

large improvements do not mechanically imply a higher rebound than<br />

smaller improvements. Third, measures target<strong>in</strong>g low <strong>in</strong>come households (<strong>for</strong><br />

<strong>in</strong>stance subsidies conditional on <strong>in</strong>come) will be less effective <strong>in</strong> terms of energy<br />

saved, because they are associated with larger rebound effects. Chitnis<br />

et al. (2014), by us<strong>in</strong>g expenditure elasticities <strong>for</strong> different socio-economic<br />

groups, have reached similar conclusions concern<strong>in</strong>g <strong>in</strong>come.<br />

5.2 Estimation <strong>and</strong> Analysis of the <strong>Indirect</strong><br />

<strong>Rebound</strong> Effect<br />

To estimate the <strong>in</strong>direct RE, we use the answers to the re-spend<strong>in</strong>g question<br />

(i.e., how<br />

respondents would allocate CHF 1,000 saved on heat<strong>in</strong>g among<br />

19


n<strong>in</strong>e categories of goods <strong>and</strong> services) <strong>and</strong> the embodied energy <strong>for</strong> each<br />

category. Potential sav<strong>in</strong>gs are calculated as the amount of embodied energy<br />

saved when spend<strong>in</strong>g CHF 1,000 less on heat<strong>in</strong>g: 1,000 [CHF] × 10.64<br />

[kWh/CHF] = 10,640 [kWh] (see figure A.7 <strong>in</strong> appendix). We then compute<br />

the emodied energy amount consumed when re-spend<strong>in</strong>g the CHF 1,000 on<br />

various categories of goods. We thus have potential <strong>and</strong> actual sav<strong>in</strong>gs <strong>for</strong><br />

each household, which allows us to apply <strong>for</strong>mula 2 <strong>and</strong> retrieve an estimation<br />

of the <strong>in</strong>direct rebound effect.<br />

We obta<strong>in</strong> an average <strong>in</strong>direct RE of 24.7%. This result means that on<br />

top of the direct RE, about a quarter of the potential energy sav<strong>in</strong>gs are<br />

lost due to the re-spend<strong>in</strong>g of the sav<strong>in</strong>gs <strong>in</strong>itially made on heat<strong>in</strong>g. <strong>The</strong><br />

distribution of the <strong>in</strong>dividual <strong>in</strong>direct RE is shown <strong>in</strong> appendix figure A.8.<br />

<strong>The</strong> median rebound effect is at 14.5% <strong>and</strong> the largest values are above<br />

100%, which corresponds to a backfire situation because the <strong>in</strong>dividual ends<br />

up consum<strong>in</strong>g more energy than be<strong>for</strong>e the efficiency improvement. Such a<br />

situation is possible if a large part of the sav<strong>in</strong>gs made on heat<strong>in</strong>g is used<br />

<strong>for</strong> additional air travel.<br />

As <strong>for</strong> the direct RE, we can assess which personal characteristics have an<br />

impact on the magnitude of the <strong>in</strong>direct RE. However, there is an important<br />

difference with the direct RE, <strong>in</strong> the sense that the <strong>in</strong>direct RE is less salient<br />

<strong>and</strong> tangible <strong>for</strong> people because we take <strong>in</strong>to account the grey (embodied)<br />

energy. <strong>The</strong>y may know that travell<strong>in</strong>g by plane or driv<strong>in</strong>g their car is energy<br />

<strong>in</strong>tensive, but they are much less likely to know whether go<strong>in</strong>g to a restaurant<br />

is more or less energy <strong>in</strong>tensive than buy<strong>in</strong>g clothes.<br />

In our model, we <strong>in</strong>clude similar determ<strong>in</strong>ants as <strong>in</strong> equation (3): socioeconomic<br />

characteristics, environmental attitudes, <strong>and</strong> tenant/owner status.<br />

Furthermore, <strong>in</strong> order to <strong>in</strong>vestigate potential l<strong>in</strong>ks between the direct <strong>and</strong><br />

<strong>in</strong>direct rebound effect, we <strong>in</strong>clude the direct rebound effect estimated earlier<br />

<strong>in</strong> the determ<strong>in</strong>ants of the <strong>in</strong>direct rebound effect. In theory, there is an<br />

automatic negative l<strong>in</strong>k between the direct <strong>and</strong> <strong>in</strong>direct RE, because the<br />

larger the direct RE, the lowest the rema<strong>in</strong><strong>in</strong>g sav<strong>in</strong>gs to spend on other goods<br />

(Thomas <strong>and</strong> Azevedo, 2013b). However there could be other l<strong>in</strong>ks between<br />

20


the two RE, <strong>for</strong> <strong>in</strong>stance if people car<strong>in</strong>g very few about the environment<br />

would display both a large direct RE <strong>and</strong> a large <strong>in</strong>direct RE. Thanks to the<br />

design of our survey, this automatic l<strong>in</strong>k is absent, <strong>and</strong> we are able to study<br />

an empirical <strong>and</strong> unconstra<strong>in</strong>ed relation that could potentilly exists between<br />

the direct <strong>and</strong> the <strong>in</strong>direct RE. We use the natural logarithm of the <strong>in</strong>direct<br />

RE to dim<strong>in</strong>ish the impact of the long right tail <strong>in</strong> the distribution. <strong>The</strong><br />

model is:<br />

ln (<strong>Indirect</strong> RE i ) = α + β 1 <strong>Direct</strong> RE i + β 2 Educ i<br />

+ β 3 Income i + β 4 Environ i + β 5 Owner i + ε i (5)<br />

We expect a negative coefficient <strong>for</strong> environmental attitudes, as people with<br />

green values pay probably more attention to the CO 2 footpr<strong>in</strong>t of their consumption.<br />

For the other explanatory variables, we do not make any hypothesis.<br />

<strong>The</strong> results are presented <strong>in</strong> table 5.<br />

<strong>The</strong> environmental attitudes have the expected negative sign. By compar<strong>in</strong>g<br />

the re-spend<strong>in</strong>g patterns of people <strong>in</strong> the first quartile <strong>and</strong> the last quartile<br />

of the environmental attitudes, we observe that people <strong>in</strong> the first quartile<br />

re-spend more on air travel, car fuel <strong>and</strong> electrical appliances, <strong>and</strong> less on<br />

public transport <strong>and</strong> leisure. <strong>The</strong> differences were tested with t-tests on the<br />

means. Tenants have a higher <strong>in</strong>direct RE than owners, <strong>and</strong> now the type<br />

of the heat<strong>in</strong>g bill does not matter any more. <strong>The</strong> ma<strong>in</strong> difference <strong>in</strong> the<br />

re-spend<strong>in</strong>g patterns of tenants is that they choose to allocate less to sav<strong>in</strong>gs<br />

than owners, <strong>and</strong> as sav<strong>in</strong>gs have a low embodied energy it expla<strong>in</strong>s why they<br />

have a larger <strong>in</strong>direct RE.<br />

<strong>The</strong> positive coefficient of the direct RE shows that the larger the direct RE,<br />

the larger the <strong>in</strong>direct one. <strong>The</strong> ma<strong>in</strong> difference <strong>in</strong> terms of re-spend<strong>in</strong>g<br />

patterns between people who never rebound <strong>and</strong> those who always rebound,<br />

is that the latter save less money. <strong>The</strong>y also re-spend a larger part on air<br />

travel <strong>and</strong> car fuel.<br />

21


Table 5: <strong>Indirect</strong> <strong>Rebound</strong> Effect<br />

<strong>Direct</strong> RE 0.212 ***<br />

(0.057)<br />

Apprenticeship 0.188<br />

(0.117)<br />

Baccalaureate 0.247 **<br />

(0.122)<br />

University <strong>and</strong> higher 0.188<br />

(0.117)<br />

Income 4500-9000 −0.037<br />

(0.037)<br />

Income >9000 −0.066 *<br />

(0.039)<br />

Environmental attitudes −0.008 ***<br />

(0.002)<br />

Owners shared costs 0.060<br />

(0.055)<br />

Tenants shared costs 0.123 ***<br />

(0.034)<br />

Tenants <strong>in</strong>dividual costs 0.124 ***<br />

(0.031)<br />

Constant −1.774 ***<br />

#Obs. 2, 822<br />

L<strong>in</strong>ear regression<br />

(0.129)<br />

Note: St<strong>and</strong>ard errors <strong>in</strong> parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01.<br />

<strong>The</strong> dependent variable is the natural logarithm of the <strong>in</strong>direct RE.<br />

22


5.3 Estimation of the Total Microeconomic<br />

Household <strong>Rebound</strong> Effect<br />

To compute the total household rebound effect we add up the direct <strong>and</strong><br />

<strong>in</strong>direct rebound effects. We obta<strong>in</strong> an average total household RE of 36.7%<br />

with a median at 23.2%. Figure A.9 <strong>in</strong> appendix shows the distribution of<br />

this total RE.<br />

Our ma<strong>in</strong> f<strong>in</strong>d<strong>in</strong>g is that, <strong>in</strong> the doma<strong>in</strong> of residential heat<strong>in</strong>g, about one<br />

third of the energy sav<strong>in</strong>gs (<strong>in</strong> terms of kWh) <strong>in</strong>itially expected after an<br />

efficiency improvement is lost due to the direct <strong>and</strong> <strong>in</strong>direct rebound effects.<br />

However, heterogeneity is large, <strong>and</strong> this total household RE is an average.<br />

Very few studies use the same dataset to estimate direct <strong>and</strong> <strong>in</strong>direct RE<br />

together. One comparable result is from Chitnis et al. (2014) <strong>for</strong> the UK.<br />

Us<strong>in</strong>g expenditure elasticities, they f<strong>in</strong>d a rebound effect (direct <strong>and</strong> <strong>in</strong>direct)<br />

between 0 <strong>and</strong> 32% but <strong>in</strong> terms of greenhouse gas.<br />

6 Conclusion<br />

In this paper we estimate the rebound effect <strong>in</strong> residential heat<strong>in</strong>g <strong>in</strong> Switzerl<strong>and</strong><br />

at the household level, <strong>in</strong>clud<strong>in</strong>g both direct <strong>and</strong> <strong>in</strong>direct rebound effects<br />

(RE). <strong>Direct</strong> RE arises when households adapt to an improvement <strong>in</strong><br />

efficiency of their heat<strong>in</strong>g system by <strong>in</strong>creas<strong>in</strong>g their dem<strong>and</strong> <strong>for</strong> heat<strong>in</strong>g services.<br />

<strong>The</strong> rationale is as follows: improved efficiency means a cheaper service,<br />

then why not consume more of this service? Us<strong>in</strong>g an orig<strong>in</strong>al dataset<br />

created <strong>for</strong> this purpose, we estimate an average direct RE between 10 <strong>and</strong><br />

15%. It means that only 85-90% of the energy sav<strong>in</strong>gs expected after an<br />

efficiency improvement from an eng<strong>in</strong>eer<strong>in</strong>g perspective are <strong>in</strong> fact realised.<br />

This result is not too worry<strong>in</strong>g <strong>for</strong> policy makers who can still count on<br />

efficiency to play a major role <strong>in</strong> the energy transition. However to see the<br />

whole picture, one should also take the <strong>in</strong>direct rebound effect <strong>in</strong>to account.<br />

<strong>Indirect</strong> RE stems from the sav<strong>in</strong>gs made on heat<strong>in</strong>g after an improvement <strong>in</strong><br />

23


efficiency, <strong>and</strong> which are re-spent <strong>in</strong> various activities. Purchas<strong>in</strong>g any good<br />

or service implies the consumption of energy, because every good <strong>and</strong> service<br />

consumes energy <strong>in</strong> order to be produced <strong>and</strong> used. Even bank sav<strong>in</strong>gs<br />

use energy as they are lent <strong>for</strong> some other purposes. As the majority of<br />

goods <strong>and</strong> services are less energy <strong>in</strong>tensive than heat<strong>in</strong>g, overall energy<br />

consumption will generally decrease after an improvement <strong>in</strong> efficiency of<br />

the heat<strong>in</strong>g system, except if the money saved is used on a very energy<br />

<strong>in</strong>tensive good (such as travell<strong>in</strong>g by plane). In such cases, overall energy<br />

consumption may even <strong>in</strong>crease, which is called a backfire effect or <strong>in</strong> other<br />

words a rebound effect of more than 100%.<br />

Our analysis shows that backfire is extremely rare <strong>and</strong> that the average total<br />

household rebound effect (the direct + <strong>in</strong>direct RE) is around 36%. This<br />

amount is not negligible, <strong>and</strong> means that about one third of the energy<br />

sav<strong>in</strong>gs <strong>in</strong>itially expected is lost due to the direct <strong>and</strong> <strong>in</strong>direct rebound effects.<br />

For policy makers <strong>in</strong> charge of the energy transition it means that the role<br />

of efficiency improvement is currently probably overestimated.<br />

Our study also highlights the large heterogeneity of the magnitude of the<br />

direct <strong>and</strong> <strong>in</strong>direct RE among <strong>in</strong>dividuals. Those who are affluent, educated,<br />

satisfied with their heat<strong>in</strong>g com<strong>for</strong>t, <strong>and</strong> environmentally conscious are less<br />

prone to have a direct rebound. We show <strong>for</strong> <strong>in</strong>stance that people with a<br />

lower <strong>in</strong>come have a larger direct rebound effect. As a consequence, measures<br />

target<strong>in</strong>g those people will be less effective <strong>in</strong> terms of energy saved. F<strong>in</strong>ally<br />

we also show that the level of efficiency improvement impacts negatively the<br />

direct rebound, i.e., it is not because you face a large efficiency improvement<br />

that you will rebound more.<br />

24


Appendix A<br />

Appendix<br />

Figure A.1: <strong>The</strong> qualitative question about the direct RE<br />

Note: <strong>The</strong> scenario is presented be<strong>for</strong>e, with the efficiency improvement <strong>and</strong> the<br />

costs saved given <strong>in</strong> % of their current heat<strong>in</strong>g costs.<br />

Figure A.2: <strong>The</strong> quantitative question about the direct RE<br />

Note: <strong>The</strong> slider is constra<strong>in</strong>ed to be at the maximum equal to the efficiency<br />

improvement <strong>in</strong> %.<br />

25


Figure A.3: Level of education<br />

Compulsory school<br />

School of commerce/household school<br />

Elementary vocational tra<strong>in</strong><strong>in</strong>g<br />

Apprenticeship<br />

Diploma<br />

Full−time vocational school<br />

Baccalaureate<br />

University <strong>and</strong> higher vocational tra<strong>in</strong><strong>in</strong>g<br />

0 200 400 600 800 1,000 1,200 1,400 1,600 1,800<br />

Frequency<br />

Note: Highest degree of education achieved by the respondent.<br />

Table A.1: Level of education <strong>in</strong> four categories<br />

Education Freq. Percent Percent <strong>in</strong> Switzerl<strong>and</strong><br />

Compulsory school 41 1.15 11.9<br />

Apprenticeship, diploma, etc. 1, 338 37.65 38.25<br />

High school 382 10.75 8.25<br />

University <strong>and</strong> higher vocational tra<strong>in</strong><strong>in</strong>g 1, 793 50.45 41.65<br />

<strong>The</strong> percent <strong>in</strong> Switzerl<strong>and</strong> comes from the Federal Office of Statistics <strong>for</strong> 2015.<br />

<strong>The</strong> category apprenticeship,... <strong>in</strong>cludes all the categories between compulsory school <strong>and</strong> baccalaureate.<br />

26


Figure A.4: Household gross <strong>in</strong>come <strong>in</strong> the sample versus at the national<br />

level<br />

Note: <strong>Rebound</strong> st<strong>and</strong>s <strong>for</strong> the sample of the survey, SHP <strong>for</strong> the Swiss Household<br />

Panel. <strong>The</strong> SHP provides data on the gross <strong>in</strong>come at the household level.<br />

27


Figure A.5: <strong>Direct</strong> RE<br />

Density<br />

0 4.622<br />

Median<br />

0 .2 .4 .6 .8 1<br />

<strong>Direct</strong> RE (n=4’808)<br />

Note: <strong>The</strong> average direct RE is 12.2%. In the histogram the zero-rebound are<br />

dropped, they count <strong>for</strong> about half of the observations.<br />

28


Figure A.6: Qualitative question: changes <strong>in</strong> behaviour<br />

0 500 1,000 1,500 2,000 2,500 3,000 3,500<br />

Increase temp. Sooner/later <strong>in</strong> the year Air<strong>in</strong>g Less attention No decrease if absent<br />

Maybe<br />

Not:Note: Over 10,665 answers (3 answers per respondent). For the last category:<br />

over 7,793 answers.<br />

Yes<br />

29


15<br />

14.57<br />

Figure A.7: Energy Intensity <strong>in</strong> kWh/CHF<br />

13.90<br />

10<br />

10.24<br />

kWh/CHF<br />

6.96<br />

5<br />

2.61<br />

0<br />

1.75 1.61 1.36 1.21<br />

0.92<br />

0.29<br />

Electricity<br />

Air travel<br />

<strong>Heat<strong>in</strong>g</strong><br />

Car fuel<br />

Food <strong>and</strong> beverages<br />

Public transport<br />

Leisure <strong>and</strong> eat<strong>in</strong>g out<br />

Electrical appliances<br />

Other goods<br />

Other services<br />

Clothes <strong>and</strong> shoes<br />

Note: Embodied energy from LCA data. Source: Eco<strong>in</strong>vent (Zürich), <strong>and</strong> Ivan<br />

Tilov (University of Neuchâtel).<br />

30


0<br />

Density<br />

10.67<br />

Median<br />

Figure A.8: <strong>Indirect</strong> RE<br />

0 .2 .4 .6 .8 1 1.2 1.4<br />

Note: <strong>The</strong> average <strong>in</strong>direct RE is 24.7%.<br />

<strong>Indirect</strong> RE (n=10’290)<br />

31


Figure A.9: Total household rebound effect<br />

Density<br />

0 5.204<br />

Median<br />

0 .2 .4 .6 .8 1 1.2 1.4 1.6 1.8 2 2.2 2.4<br />

Total microeconomic RE (n=10’290)<br />

Not:Note: <strong>The</strong> average total household rebound effect is 36.7%.<br />

32


References<br />

Ayd<strong>in</strong>, E., Kok, N., <strong>and</strong> Brounen, D. (2014). Energy efficiency <strong>and</strong> household<br />

behavior: <strong>The</strong> rebound effect <strong>in</strong> the residential sector. Technical report,<br />

Work<strong>in</strong>g Paper.<br />

Azevedo, I. M. (2014). Consumer end-use energy efficiency <strong>and</strong> rebound<br />

effects. Annual Review of Environment <strong>and</strong> Resources, 39(1):393–418.<br />

Belotti, F., Deb, P., Mann<strong>in</strong>g, W. G., Norton, E. C., et al. (2015). Twopm:<br />

Two-part models. Stata Journal, 15(1):3–20.<br />

Best, H. <strong>and</strong> Mayerl, J. (2013). Values, beliefs, attitudes: An empirical study<br />

on the structure of environmental concern <strong>and</strong> recycl<strong>in</strong>g participation. Social<br />

Science Quarterly, 94(3):691–714.<br />

Carson, R. T., Flores, N. E., Mart<strong>in</strong>, K. M., <strong>and</strong> Wright, J. L. (1996). Cont<strong>in</strong>gent<br />

valuation <strong>and</strong> revealed preference methodologies: compar<strong>in</strong>g the<br />

estimates <strong>for</strong> quasi-public goods. L<strong>and</strong> economics, pages 80–99.<br />

Chan, N. W. <strong>and</strong> Gill<strong>in</strong>gham, K. (2015). <strong>The</strong> microeconomic theory of the<br />

rebound effect <strong>and</strong> its welfare implications. Journal of the Association of<br />

Environmental <strong>and</strong> Resource Economists, 2(1):133–159.<br />

Chitnis, M., Sorrell, S., Druckman, A., Firth, S. K., <strong>and</strong> Jackson, T. (2013).<br />

Turn<strong>in</strong>g lights <strong>in</strong>to flights: Estimat<strong>in</strong>g direct <strong>and</strong> <strong>in</strong>direct rebound effects<br />

<strong>for</strong> UK households. Energy Policy, 55:234–250.<br />

Chitnis, M., Sorrell, S., Druckman, A., Firth, S. K., <strong>and</strong> Jackson, T. (2014).<br />

Who rebounds most? Estimat<strong>in</strong>g direct <strong>and</strong> <strong>in</strong>direct rebound effects <strong>for</strong><br />

different UK socioeconomic groups. Ecological Economics, 106:12–32.<br />

Druckman, A., Chitnis, M., Sorrell, S., <strong>and</strong> Jackson, T. (2011). Miss<strong>in</strong>g carbon<br />

reductions? Explor<strong>in</strong>g rebound <strong>and</strong> backfire effects <strong>in</strong> UK households.<br />

Energy Policy, 39(6):3572–3581.<br />

33


Fowlie, M., Greenstone, M., <strong>and</strong> Wolfram, C. (2015). Do energy efficiency<br />

<strong>in</strong>vestments deliver? evidence from the weatherization assistance program.<br />

Technical report, National Bureau of Economic Research.<br />

Gonseth, C., Thalmann, P., <strong>and</strong> Vielle, M. (2015). Impacts of global warm<strong>in</strong>g<br />

on energy use <strong>for</strong> heat<strong>in</strong>g <strong>and</strong> cool<strong>in</strong>g with full rebound effects. Technical<br />

report, Elsevier.<br />

Greene, D. L. (2012). <strong>Rebound</strong> 2007: Analysis of US light-duty vehicle travel<br />

statistics. Energy Policy, 41:14–28.<br />

Grijalva, T. C., Berrens, R. P., Bohara, A. K., <strong>and</strong> Shaw, W. D. (2002). Test<strong>in</strong>g<br />

the validity of cont<strong>in</strong>gent behavior trip responses. American Journal<br />

of Agricultural Economics, 84(2):401–414.<br />

Haas, R. <strong>and</strong> Biermayr, P. (2000). <strong>The</strong> rebound effect <strong>for</strong> space heat<strong>in</strong>g<br />

empirical evidence from Austria. Energy policy, 28(6):403–410.<br />

Lenzen, M. <strong>and</strong> Dey, C. J. (2002). Economic, energy <strong>and</strong> greenhouse emissions<br />

impacts of some consumer choice, technology <strong>and</strong> government outlay<br />

options. Energy Economics, 24(4):377–403.<br />

Loomis, J. B. <strong>and</strong> Richardson, R. B. (2006). An external validity test of<br />

<strong>in</strong>tended behavior: compar<strong>in</strong>g revealed preference <strong>and</strong> <strong>in</strong>tended visitation<br />

<strong>in</strong> response to climate change. Journal of environmental plann<strong>in</strong>g <strong>and</strong><br />

management, 49(4):621–630.<br />

Madlener, R. <strong>and</strong> Hauertmann, M. (2011). <strong>Rebound</strong> effects <strong>in</strong> German residential<br />

heat<strong>in</strong>g: do ownership <strong>and</strong> <strong>in</strong>come matter?<br />

Nadel, S. (2012). <strong>The</strong> rebound effect: Large or small? American Council <strong>for</strong><br />

an Energy-Efficient Economy.<br />

Nadel, S. (2016). <strong>The</strong> potential <strong>for</strong> additional energy efficiency sav<strong>in</strong>gs <strong>in</strong>clud<strong>in</strong>g<br />

how the rebound effect could affect this potential. Current Susta<strong>in</strong>able/Renewable<br />

Energy Reports, 3(1-2):35–41.<br />

34


Schleich, J., Mills, B., <strong>and</strong> Dütschke, E. (2014). A brighter future? Quantify<strong>in</strong>g<br />

the rebound effect <strong>in</strong> energy efficient light<strong>in</strong>g. Energy Policy, 72:35–42.<br />

Small, K. A. <strong>and</strong> Van Dender, K. (2007). Fuel efficiency <strong>and</strong> motor vehicle<br />

travel: <strong>The</strong> decl<strong>in</strong><strong>in</strong>g rebound effect. Energy Journal, 28(1):25–52.<br />

Sorrell, S. <strong>and</strong> Dimitropoulos, J. (2008). <strong>The</strong> rebound effect: Microeconomic<br />

def<strong>in</strong>itions, limitations <strong>and</strong> extensions. Ecological Economics, 65(3):636–<br />

649.<br />

Sorrell, S., Dimitropoulos, J., <strong>and</strong> Sommerville, M. (2009). Empirical estimates<br />

of the direct rebound effect: A review. Energy policy, 37(4):1356–<br />

1371.<br />

<strong>The</strong> Economist (2015). Energy efficiency: <strong>The</strong> <strong>in</strong>visible fuel.<br />

Thomas, B. A. <strong>and</strong> Azevedo, I. L. (2013a). Estimat<strong>in</strong>g direct <strong>and</strong> <strong>in</strong>direct<br />

rebound effects <strong>for</strong> US households with <strong>in</strong>put–output analysis. Part<br />

2: Simulation. Ecological Economics, 86:188–198.<br />

Thomas, B. A. <strong>and</strong> Azevedo, I. L. (2013b). Estimat<strong>in</strong>g direct <strong>and</strong> <strong>in</strong>direct<br />

rebound effects <strong>for</strong> U.S. households with <strong>in</strong>put-output analysis part 1:<br />

<strong>The</strong>oretical framework. Ecological Economics, 86:199–210.<br />

Tilov, I., Farsi, M., <strong>and</strong> Voll<strong>and</strong>, B. (2016). Socio-economic determ<strong>in</strong>ants of<br />

households’ energy dem<strong>and</strong>: a holistic view. Work<strong>in</strong>g paper <strong>in</strong> progress.<br />

Voll<strong>and</strong>, B. (2016). Efficiency <strong>in</strong> Domestic Space <strong>Heat<strong>in</strong>g</strong>: An Estimation<br />

of the <strong>Direct</strong> <strong>Rebound</strong> Effect <strong>for</strong> Domestic <strong>Heat<strong>in</strong>g</strong> <strong>in</strong> the U.S. IRENE<br />

Work<strong>in</strong>g Papers 16-01, IRENE Institute of Economic Research.<br />

Whitehead, J. C., Pattanayak, S. K., Van Houtven, G. L., <strong>and</strong> Gelso, B. R.<br />

(2008). Comb<strong>in</strong><strong>in</strong>g revealed <strong>and</strong> stated preference data to estimate the<br />

nonmarket value of ecological services: an assessment of the state of the<br />

science. Journal of Economic Surveys, 22(5):872–908.<br />

35


Yu, B., Zhang, J., <strong>and</strong> Fujiwara, A. (2013). <strong>Rebound</strong> effects caused by the<br />

improvement of vehicle energy efficiency: an analysis based on a Sp-off-<br />

RP survey. Transportation Research Part D: Transport <strong>and</strong> Environment,<br />

24:62–68.<br />

36

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!