Energy Policy 63 (2013) 775787

Contents lists available at ScienceDirect

Energy Policy
journal homepage: www.elsevier.com/locate/enpol

The inuences of nancial and non-nancial factors on energy-saving

behaviour: A eld experiment in Japan
Kenichi Mizobuchi a,n, Kenji Takeuchi b
a
Department of Economics, Matsuyama University, 4-2, Bunkyo, Matsuyama, Ehime, 790-8578, Japan
b
Graduate School of Economics, Kobe University, 2-1, Rokkodai, Nada, Kobe, Hyogo 657-8501, Japan

H I G H L I G H T S

We studied the effects of (non-)nancial factors on household electricity use.

Financial factors had a signicant effect; non-nancial ones were inconclusive.

Externalities had a positive effect across time and households on saving energy.

There is a heterogeneous treatment effect in nancial-reward intervention.

Before the experiment, participants underestimated the marginal costs of saving electricity.

art ic l e i nf o a b s t r a c t

Article history: This study examines the inuences of nancial and non-nancial factors on electricity-conservation
Received 3 July 2013 behaviour. A random sample of 236 Japanese households participated in the eld experiment and the
Accepted 15 August 2013 participants were offered two interventions, such as monetary rewards, depending on their reduction in
Available online 14 September 2013
electricity consumption and comparative feedback. The average saving rates of the (i) reward-
Keywords: intervention group (5.9%) and the (ii) reward with comparative feedback group (8.2%) are statistically
Electricity-saving behaviour larger than those of the (iii) control group (1.7%). Our study demonstrates the following. First, our
Household econometric analysis conrmed a signicant response by households to nancial incentives but a more
Economic incentive inconclusive response to the treatment that provided non-nancial, additional information. Second, we
found a positive inuence of treatment externalities across time and households on energy saving. Third,
there is a heterogeneous treatment effect in the reward-intervention group, with the households having
a high New Ecological Paradigm (NEP) score being more likely to respond to the reward programme and
save electricity than those that do not. Finally, and most interestingly, differences in responses to the
questionnaire before and after the experiment suggest that the participants had underestimated the
marginal costs of saving electricity before they actually started to do so.
& 2013 Elsevier Ltd. All rights reserved.

1. Introduction
The Great East Japan Earthquake of March 2011 and subse-
quent radiation leak from the Fukushima nuclear plant have
triggered widespread concerns about the safety of nuclear plants.
Since then, most of the nuclear power reactors in Japan have been
shut down.1 However, replacing nuclear power with an alternative
source of electricity in the short term presents great difculty.
Thus, massive attempts have already been implemented to cut the
peak demand for electricity. The household sector will need to
play a key role in these attempts, but signicant energy saving has
been less than successful in this sector.2
Many studies have examined effective intervention for
encouraging household energy-saving behaviour, such as informa-
tion provision, public campaigns, goal setting, feedback, compara-
tive feedback and reward (Winett et al., 1978; Becker, 1978;
McClelland and Cook, 1980; Midden et al., 1983; Brandon and
Lewis, 1999; Abrahamse et al., 2005; Darby, 2006; Abrahamse
et al., 2007; Petersen et al., 2007; Ehrhardt et al., 2010; Allcott,
2011b). Abrahamse et al. (2005) reviewed 38 eld studies in social
and environmental psychology and evaluated the effectiveness of

n
Corresponding author. Tel.: 81 89 926 7140.
2
E-mail addresses: kmizobuc@cc.matsuyama-u.ac.jp (K. Mizobuchi), In 2011, the electricity-saving rates of large and small customers of the
takeuchi@econ.kobe-u.ac.jp (K. Takeuchi). industrial sector were 29% and 19% higher than the rates of the previous year,
1
In the interests of safety, as of July 2013, 52 out of Japans 54 nuclear power respectively. However, the rate of electricity saving among households was only 6%
reactors had been shut down. (Tokyo Electric Power Company, 2011).

0301-4215/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.enpol.2013.08.064
776 K. Mizobuchi, K. Takeuchi / Energy Policy 63 (2013) 775787

the above-mentioned interventions. However, it is not clear from selected Japanese households. We attempted to separate the
these studies which form of intervention is most effective or treatment effects and other externalities that inuence electricity
which has the most enduring effect. One of the main reasons for saving across time and households. We also examined the tests for
this fact is that each study focussed on only a few factors. In order heterogeneous treatment effects. Moreover, we examined the size
to learn about policies that can effectively and efciently reduce and changes of the marginal costs of electricity saving from the
energy consumption, various data and insights need to be inte- pre- and post-questionnaire surveys.
grated. In particular, one needs to understand, rst, the motiva- This article is organised as follows: Section 2 discusses previous
tions of consumers in implementing certain types of energy- studies on household energy-saving behaviour. In Section 3, the
saving behaviour. Second, a careful analysis is required of the eld experiment, in which we used subsidies as an economic
various factors that inuence household behaviour; excluding incentive for electricity-conservation behaviour within households,
crucial factors will lead to biases in the assessment of a particular is described and Section 4 discusses the empirical analysis. Section 5
factors contribution to household behaviour (Van den Bergh, analyses the marginal costs of electricity-conservation behaviour
2008). Therefore, an extension to the broadest possible range of and discusses some limitations of our study, and section 6 concludes
policy instruments would require considering a great many the discussion.
factors, such as nancial, non-nancial and demographic.
Traditionally, nancial and non-nancial approaches differ;
hence, research efforts tend to be conned only to each academic
2. Factors of inuence on household energy-saving
area. Economists tend to examine the inuence of external
behaviour
conditions such as price, income and other socioeconomic
characteristics upon pro-environmental behaviour.3 Thus,
Over the past several decades, many scholars have emphasised
econometric studies of pro-environmental household behaviour
the importance of nancial or non-nancial factors in encouraging
that include non-nancial variables relating to attitudes, knowl-
household energy-conservation behaviour (Stern, 1992; Dwyer
edge, perceptions and values are rare. However, studies by
et al., 1993; Abrahamse et al., 2005; Darby, 2006; Allcott et al.,
psychologists that include these variables exclude economic vari-
2007; Van den Bergh, 2008; Ehrhardt et al., 2010). This section
ables. Recently, several studies have considered both nancial and
gives a brief overview of relevant studies that have used nancial
non-nancial factors in analysing households pro-environmental
or non-nancial factors such as rewards, comparative feedback,
behaviour (Clark et al., 2003; Kotchen and Moore, 2007).
social norms, environmental concerns, goal setting and some
Within traditional economic theory, individuals maximise their
socioeconomic variables with a specic focus on these factors
own utility, subject to budget constraints, and they have little
inuence on households energy-saving behaviour.
incentive to strive towards a public good, that is, pro-environ-
To evaluate the effects of taxes and subsidies for household
mental behaviour; hence, they will choose a free ride instead.
energy usage, many previous studies have focussed on the
However, this theoretical prediction is rarely shown empirically.
behavioural impacts of energy prices by a degree of price elasticity
Thus, economists have started to consider the role of non-nancial
with time-series, cross-section and panel data (Silk and Joutz,
factors, such as warm-glow altruism (Andreoni, 1990) and pater-
1997; Vaage, 2000; Halvorsen and Larsen, 2001). Most of these
nalistic altruism (McConnell, 1997), in motivating individuals
studies showed statistically signicant price elasticities. However,
to contribute to public goods. In the discipline of psychology,
the degree of elasticity uctuates, depending on the data and the
Fransson and Garling (1999) have reviewed many studies examin-
estimation method (Espey and Espey, 2004). In addition, some
ing the relationship between non-nancial variables and pro-
eld studies are based on an intervention through economic
environmental behaviour. The inuence of some non-nancial
rewards (e.g., money or prizes) to examine household energy-
factors, such as attitudes, knowledge, perceptions and values, has
saving behaviour (Winett et al., 1978; Midden et al., 1983;
been demonstrated in terms of pro-environmental behaviour. By
McClelland and Cook, 1980; Petersen et al., 2007). These studies
means of a natural experiment, Guagnano et al. (1995) proved the
indicate that nancial rewards have been successful in reducing
hypothesis that attitudinal factors and external conditions act in
household energy consumption.5 However, two problems remain
combination to inuence pro-environmental behaviour.4
in these previous studies. One is low statistical reliability because
As a result, recent developments in economics (psychology)
of the small number of households involved in the experimental
emphasise the need to consider internal (external) inuences.
groups (fewer than 20). Another is an identication problem, that
However, as we mentioned above, what is lacking is a systematic
is, the treatment effects of rewards could not be isolated because a
integrated analysis of interventions or instruments: non-nancial
combination of other interventions, such as information and
(internal) factors, such as attitude, knowledge, motivation and
individual and comparative feedback, was used in the study.
perception, and nancial (external) factors, such as prices
Feedback is widely used to encourage energy conservation
(a subsidy or tax) and incomes.
(Matsukawa, 2004; Darby, 2006; Ehrhardt-Martinez et al., 2010;
This study aims to evaluate the inuences of both nancial
Gleerup et al., 2010; Jessoe and Rapson, 2012). Fischer (2008)
and non-nancial factors on encouraging electricity-conserva-
reviews a considerable number of feedback studies. In general,
tion behaviour. We conducted a eld experiment based on an
feedback is given in terms of a households own energy savings, in
intervention study from October to November 2011 in randomly
order to conrm the effectiveness of its efforts to save energy;
feedback becomes more effective, in particular, if its frequency is
3
Traditional econometric studies that analyse residential energy conservation increased. Another type of feedback wherein participants are
behaviour tend to focus on the behavioural impacts of energy prices by a degree of provided with information on the energy savings of other study
price elasticity. Dubin and McFadden (1984) were among the rst to investigate
household electricity demand with econometric approaches, using micro-level
5
data. They investigated the inuences of price, income, electrical appliances and Winett et al. (1978) studied the effect of certain energy-saving interventions,
other socioeconomic variables on household electricity demand. Espey and Espey including a nancial incentive (i.e., a reward for reducing electricity consumption),
(2004) provide an overview of econometric analyses of residential electricity involving 107 single-family households in Texas. The electricity conservation by the
demand. group with a nancial incentive was signicantly higher than that by other
4
They also showed, specically, that external conditions affect the strength of intervention groups and the control group. Moreover, if households were in
attitudebehaviour relationships. Thus, strong positive external conditions increase competition with each other, the inuence of the rewards was more effective
the likelihood of attitudes that give rise to pro-environmental behaviours. (McClelland and Cook, 1980; Petersen et al., 2007).
 K. Mizobuchi, K. Takeuchi / Energy Policy 63 (2013) 775787
participants is referred to as comparative feedback (Midden
et al., 1983; Siero et al., 1996; Allcott, 2011b). This type of feedback
has been found to be more effective than individual feedback in
promoting energy conservation (Brandon and Lewis, 1999). Com-
parative feedback may be more effective because of the sense of
competition or because people derive satisfaction from conform-
ing to social norms. In the former case, when participants can
compare their savings levels among themselves, they may try to
reduce energy more than other participants. Such competition
may be especially effective when the contest is also accompanied
by a reward. The latter condition is also important; if households
have less information about their energy-use behaviours, the
social comparisons may facilitate social learning about their
optimal level of energy use, as documented in other contexts by
Cai et al. (2009) and Munshi and Myaux (2006). Comparative
feedback demonstrates relevant social norms in favour of energy
conservation, that is, it becomes clear that others are actively
engaged in energy conservation as well. Some academic work has
shown that providing information on the social norm induces
people to conserve energy (Schultz et al., 2007; Nolan et al., 2008;
Allcott, 2011b). For example, Allcott (2011b) examined non-price
energy-conservation programmes using data from randomised
natural eld experiments in 600,000 households across the United
States, run by a company called OPOWER. This company mails
Home Energy Report letters that compare a households energy
use to that of similar neighbours and provide energy-conservation
tips. Allcott showed that the average energy reduction was 2.0%.
From some of the above studies, we can say that the effect of
(comparative) feedback is amply demonstrated. Hence, in our
experimental study, we deal with comparative feedback as evok-
ing a sense of competition by combining feedback with the
nancial-rewards treatment and not make a feedback treatment
group alone. However, in the multivariate analysis, separate from a
sense of competition, we employ another variable that relates to a
social norm (see Sections 3 and 4).
Environmental concern is also considered as one of the non-
nancial motivating factors regarding pro-environmental behaviour.
For assessing the degree of environmental concern, the New Ecolo-
gical Paradigm (NEP) scale has been developed as an instrument in
the social and behavioural sciences for measuring attitudes about the
environment (Dunlap et al., 2000). The NEP scale has also been used
in the economic literature on households green electricity pro-
gramme (Kotchen and Moore, 2007; Ek and Sderholm, 2008).
Goal setting, which is sometimes used to promote energy
saving, sets energy-saving targets that households can strive for,
such as saving 10% (Abrahamse et al., 2007). The efcacy of goal
setting is strengthened if combined with a promise to save (Katzev
and Johnson, 1983) or with feedback (McCalley and Midden,
2002). Moreover, in the case wherein goal setting was combined
with feedback, a high-level goal appeared to be more effective in
reducing energy demand than a low-level goal (Becker, 1978).
Other variables typically included in previous studies are
household income; age and education level of household mem-
bers; house size; a measure of outside temperature; and owner-
ship of certain appliances (e.g., electric appliances). In most
settings, however, these are treated as the control variables
(Abrahamse et al., 2005).
Here, most of the previous studies in the design of the
treatment programmes mentioned above (reward, feedback, etc.)
did not consider externalities across households and time. For
example, some days or weeks, households may have many guests,
and these guests will increase energy use in those experimental
periods. Alternatively, some households may go out of town for a
month, and this decreases energy use, even if they do not respond
to the treatment. Therefore, these externalities that affect energy
usage should be separated from the treatment effects. Our
777
empirical study tries to show the existence and the size of these
treatment externalities.
This study integrates both nancial and non-nancial factors to
identify key internal and external variables that determine house-
hold electricity-saving behaviour. Our analysis uses data from 236
randomly selected households in a eld experiment. Financial
variables consist of reward, household income and standard socio-
demographic elements. Non-nancial variables consist of com-
parative feedback, social norms, a modied version of the NEP
scale to measure environmental concerns and goal setting. In the
next section, we present our research strategy in greater detail.
3. Field Experiment
3.1. Design
Our eld experiment lasted for 8 weeks, from October to
November 2011. Each enrolled household was entitled to receive
a reward for reducing its electricity consumption to a level below
their previous years consumption for the same period. The
electricity consumption of a household was commonly measured
in kilowatt-hours (kW h), and the reduction of electricity con-
sumption (Reduction) was calculated as follows:
C x C p
Reduction%  100 1
Cp
where Cx and Cp represent the electricity consumption during the
experiment and during the same period of the previous year,
respectively. If Reduction was lower than 0% (Reduction o0),
households could receive a reward (nancial incentive). The amount
of the reward was 200 yen (about $2.0) per 1% of reduction in
electricity consumption (We have assumed the following exchange
rate: $1100 yen). For example, if actual Reduction was 10%,
households were given 2000 yen (200 yen  10).6
On the other hand, we also examined the effectiveness of
competitive consciousness, in which each household received
information about the amount of electricity saving by other
households (e.g., Reduction values). We conducted the comparative
feedback twice: the rst one informed households of the target
levels of electricity saving by means of pre-experiment question-
naires in late September and the second one provided information
on the values of Reduction from the rst month (i.e., October) in
early November.7 Combined with the above reward programme
and the comparative feedback, we can examine the effectiveness
of competitive consciousness.
Households were assigned randomly to three groups
(Group 1) Reward, (Group 2) Reward with Comparative Feed-
back and (Group 3) Control to examine the effectiveness of
nancial incentives and of competitive consciousness. We use a
simple non-parametric approach to compare the three groups in
Section 3.3. We also use a parametric approach in the next
section.
We administered two questionnaires, one before and one after
the eld experiment. The responses from the subjects offer valuable
information on the factors that play a key role in reducing energy
use. In the pre-experiment questionnaire, we asked participants
about the following aspects: their attitude towards environmental
6
According to the Family Income and Expenditure Survey of Japan, the average
electrical bill of Japanese household for 2 months is about 20,000 yen. Therefore,
we set a 200-yen reward per 1% of electricity consumption reduction in our study.
7
The rst comparative feedback is not a real value of Reduction, and there may
be a difference between the target value target and real values. However,
participants can compare others target reduction levels with their own. Therefore,
in terms of comparison with other participants, we deal with it as the rst
comparative feedback.
778 K. Mizobuchi, K. Takeuchi / Energy Policy 63 (2013) 775787
issues, their level of environmental concern,8 attitudes towards
electricity conservation, their intended target level of electricity
saving as well as their level of condence in meeting this target and
the types of home electrical appliances in the household. Finally, we
elicited social norm, socioeconomic and demographic details. In our
study, the degree of social norm is measured by the question, Do
you consider how often people close to you are attempting to
employ electricity-saving behaviour? Comparison with neighbours
or other close persons will construct peoples social norms and will
induce people to conserve energy (Nolan et al., 2008; Allcott,
2011b).9 In the post-experiment questionnaire, we asked partici-
pants, inter alia, about changes in their attitudes towards electricity
conservation, the challenges in conserving electricity and the
appropriateness of economic incentives. Table 1 represents the
schedule of our eld experiment for three groups.
3.2. Respondents and Procedure
We conducted the study in Matsuyama, a city in Ehime
Prefecture, which has approximately 215,000 households in wes-
tern Japan. In August 2011, more than 1000 households in
Matsuyama City, which are registered in the research agency of
the Matsuyama Chamber of Commerce and Industry, were ran-
domly selected and were sent letters informing them about the
study and inviting them to participate in our eld experiment.
Aside from a broad description that the study was about an
experiment in energy saving we did not divulge any details
about the eld experiment in the initial invitation letter. We
accepted the rst 300 applications received.10 Of the households
invited to participate in this study, 236 returned a completed pre-
experiment questionnaire. All respondents were older than 20
years (the oldest was 81). Compared to a national sample (Popula-
tion Statistics, 2009), male respondents were overrepresented,11
and respondents in the category of 55 through 69 years were
slightly underrepresented. In eld-experiment or questionnaire-
survey studies about environmental problems, a tendency exists
towards overrepresentation from high-income households and
underrepresentation from low-income households, that is, the
studies are biased for high incomes (Poortinga et al., 2003).
Fig. 1 represents a comparison of the income distributions in our
study and in a national sample.12 The gure shows that both high
incomes (over 6 million yen) and low incomes (under 4 million
yen) are slightly underrepresented and that incomes of 46
million yen are overrepresented. The average income of Japan is
about 5.5 million yen. Therefore, the households in our study are
not biased towards high incomes but have a slight tendency
towards a higher concentration of average incomes.
8 13
Respondents were asked to what extent they agreed or disagreed with 15
statements comprising the modied New Ecological Paradigm (NEP) scale (Dunlap
et al., 2000). The modied NEP scale is commonly used in the psychology
literature; it aims to capture the following ve facets of environmental concern:
limits to growth, anti-anthropocentrism, the fragility of the balance of nature,
rejection of the idea that humans are exempt from the constraints of nature and the
possibility of an eco-crisis or ecological catastrophe. We list in Appendix A all the
statements inherent in the NEP scale; we also verify the reliability of our NEP scale
using item-total correlation and Cronbachs alpha.
9
Responses were made on a ve-point Likert scale (always 5, usually4,
sometimes 3, rarely 2, never 1), and high score responses mean a high level of
attention to social norms (and vice versa). However, we also know that the way of
capturing social norms with this alternative is weak compared to those of other
studies. We discuss this in section 4.
10
We discuss the sample selection bias of our experimental study in section 5.
11
Most respondents were the householders. This might be the reason for male
overrepresentation.
12
Ministry of Health, Labor and Welfare: National Livelihood Survey 2009.
http://www.mhlw.go.jp/index.shtml.
As described earlier, participating households were entitled to
receive subsidies based on their value of Reduction, which was
calculated according to eq. (1). Shikoku Electric Power Co. was the
electricity supplier for all participating households. Participants
could access their historical data on electricity consumption per
month by logging onto the Shikoku Electric Power website, using
their unique identication numbers. We collected these consump-
tion data for the previous year from the households (printed Web
page) and used them as the measurement criteria (baseline
electricity consumption). Additionally, for tracking ongoing con-
sumption (electricity consumption during our experiment), every
household was asked to send a copy of its monthly electricity-
consumption record from Shikoku Electric Power Co.
3.3. Results
Before proceeding to the results of experimental treatment, we
need to check the randomised assignment of households and
conduct t-tests of the balance of the three randomised groups on
the exogenous covariates (Miguel and Kremer, 2004). Table 2 shows
the average values of electricity consumption in 2011 and 2010 and
of Reduction; weather variables and other characteristics of the
households in each group; and the results of the t-test used to
examine the differences between the three groups.13 For the ideal
randomised assignment, each of the three groups should have
similar characteristics except for Reduction, which shows the treat-
ment effect. Most of the variables are similar within the three
groups. There are no statistically signicant differences across the
households in groups 1, 2 and 3, that is, Electricity_2011 and
Electricity_2010, NEP scale, Target, Age, Married, Number of children
(under 12 years old), Living with parents, Family size, Homeowner-
ship and Heating degree days. However, despite randomised assign-
ments, several variables are statistically partially different across the
three groups. Social norm of households in Group 3 and Income
level of households in Group 2 are signicantly larger than for the
households in the other two groups. Considering all these results,
our randomised assignment is less than perfect but it is satisfactory
from the viewpoint of fundamentals.
Fig. 2 shows the distribution of electricity conservation, in
terms of Reduction, in the sampled households at the end of
8 weeks.14 About 70% of the households successfully reduced their
electricity consumption, a result that could have arisen from a
number of causes. One of them may be the treatment effects, such
as economic incentives or competitive consciousness; for example,
the economic incentives offered in this experiment might be
higher than the marginal cost of the electricity-saving behaviour
or the temperature of the target year may have been cooler than
that of the previous year.15 Moreover, this result does not show the
The number of households in group 1 is about twice as that in the other two
groups; there are two reasons for this. First, as we will elaborate in the following
section, we examine the heterogeneity of the nancial treatment, which was not
considered sufciently in previous studies. Therefore, we need an adequate sample
size. Second, we also want to discuss the degree of the marginal cost of electricity
conservation for the household based on the nancial treatment group. Therefore,
in this study, the sample size of group 1 is substantially larger than those of other
groups. However, in a randomised experiment, one should set a similar number of
households in each group for examining the treatment effects. We state this as the
limitation of our study in section 5.
14
Because of an inadequacy or a lack of electricity-consumption data, 28
participants dropped out of the eld experiment, reducing the total number of
households to 208.
15
One probable reason is a rise in the awareness of the need to save electricity
after the Great East Japan Earthquake in March 2011. Because most of the nuclear
power reactors have been shut down in the interest of safety (52 of Japans 54 in
March 2013), massive electricity-saving measures have been implemented in Japan.
From the pre-experiment questionnaire survey, 63.9% of households perceived a
change in their awareness of interest in electricity saving (high interestinterest).
 K. Mizobuchi, K. Takeuchi / Energy Policy 63 (2013) 775787 779

Table 1
Schedule of our eld experiment.

Contents

August Selecting the participated household (all groups)

Early September Implementation of a pre-questionnaire (all groups)
Late September Informing of intended target level of electricity saving
(Reward and Feedback group, rst comparative feedback)
October 1 Start the experiment (all groups)
Early November Informing of electricity saving rate of October
(Reward and Feedback group, second comparative feedback)
Nobember 30 End the experiment (all groups)
December Implementation of a post-questionnaire (all groups)

All group represents 1) Reward, 2) Reward with Comparative Feedback, and 3) Control groups.

50 the previous year. Here, we simply compared the differences of the

Respondents average electricity saving within three groups statistically.
45 The third column of Table 2 shows the average electricity-
Japan
saving rate (i.e., average Reduction) of the three groups: (Group 1)
40
reward, (Group 2) reward with feedback and (Group 3) control.
35 The average saving rate is highest in the reward with feedback
group (  8.2%), followed by the reward group ( 5.9%) and the
30 control group (  1.7%). Our study investigates the treatment effect;
hence, we test the following null hypothesis:
%

25
20
15
10
5 second hypothesis, the average Reductions of Group 1 and
0 From these results, we might conrm the treatment effects of
<4 4-6 6-8 8-10 10-15 > 15
million million million million million million
yen yen yen yen yen yen
Fig. 1. Income distribution.
pure treatment effects of reward and competitive consciousness,
and we must consider the possibility of the presence of treatment
externalities across time and households, as we discussed in
section 2. For example, some households may have left the city
for a part of the experimental period; if so, they could have
decreased electricity use without responding to the treatment. We
also discuss the degree of such externalities in the next section.
As Fig. 2 shows, in particular, the greatest percentage of
households (18.8%) was in the 510% |Reduction| bracket, followed
by the 1015% bracket (17.3%) and the 15% bracket (15.4%).
However, the percentage of households that saved more than
15% was much smaller, and it decreased as the values of |Reduction|
increased. Additionally, about 28.8% of the households did not
conserve electricity. Fig. 2 shows the distributions of electricity
conservation for three groups. All of the distributions are similar to
Fig. 2, that is, the distribution skewed to the left. However, we can
conrm the differences of the share of under 0 between each
gure, and it increased from Group 1 to Group 3 (30.4% o38.5%)
and Group 2 to Group 3 (17.0% o38.5%). We infer that this
tendency was the inuence of the treatment effect.
The average Reduction of the households in this experiment is
5.4%, indicating a decrease in energy consumption compared to
(footnote continued)
From this result, we can conclude that the awareness for electricity saving has
become high in general and that it may slightly affect the electricity-saving
behaviour of our experimental households.
H 0 : i  j 0 i j; i; j i; ii; iii
where i indicates the average Reduction of ith groups. The fourth
to sixth columns of Table 2 show the p values of three hypotheses,
that is, Group 1 versus Group 3, Group 2 versus Group 3 and
Group 1 versus Group 2. From the p value of the rst and the
Group 2 are found to be signicantly larger than that of Group 3.
Group 1 and Group 2 in our experiment. However, as we men-
tioned above, these differences within three groups include both
treatment and some other factors that inuence energy-saving
behaviour. Moreover, other treatment externalities across time
and households also affect this behaviour. Thus, we should
examine more precise analyses.
4. Empirical Analysis
This section examines the effectiveness of the treatment effect
of our experiment and tries to identify it with both some observed
characteristics and the treatment externalities across time and
households. We rst set an econometric model and estimate the
treatment effect to see whether nancial and non-nancial inter-
ventions affected decisions about the electricity-saving behaviour
of households. We also try to examine to what degree the
externalities also affected our study.
4.1. Model and Data
We set the following parametric model for analysing treatment
effects:
Y 2011;i 1 Y 2010;i Reward Ri Competitive Compi 2 X i i;t ; 2
where, Y 2011;i and Y 2010;i are the households treatment and pre-
treatment electricity consumption, respectively. Ri and Compi are
indicators for whether the household is part of one of the two
treatment groups, such as Group 1 and Group 2. Xi denote demo-
graphic variables and other information from the treatment
survey. Finally, i,t is an idiosyncratic error which includes the other
externalities. Here, notice that i,t has an it subscript: it varies
across household and time. For example, in our experimental
780 K. Mizobuchi, K. Takeuchi / Energy Policy 63 (2013) 775787

Table 2
Average Characteristics of Households, Pre-Treatment.

Group 1 Group 2 Group 3 Group 1 Group 2 Group 1

103 53 52 - Group 3 - Group 3 - Group 2

Electricity_2011 706.883 804.491 745.75  38.87 58.74  97.61

(0.606) (0.544) (0.227)
Electricity_2010 759.417 882.849 767.231  7.81 115.62  123.43
(0.919) (0.253) (0.158)
Reduction  5.876  8.196  1.721  4.15nn  6.48nn 2.32
(0.080) (0.009) (0.274)
Social Norma 3.097 3.151 3.423  0.33nn  0.27nn  0.05
(0.002) (0.034) (0.652)
NEP 57.515 57.17 56.673 0.84 0.5 0.34
(0.488) (0.728) (0.800)
b
TARGET 3.456 3.264 3.481  0.02  0.22 0.19
(0.919) (0.387) (0.353)
Gender 0.165 0.566 0.481  0.32nn 0.09  0.4nn
(0.000) (0.387) (0.000)
Age 43.553 42.547 40.923 2.63 1.62 1.01
(0.204) (0.428) (0.513)
Married 0.709 0.714 0.635 0.07 0.12  0.05
(0.181) (0.093) (0.270)
Number of Children 0.534 0.524 0.404 0.13 0.09 0.04
(0.181) (0.284) (0.375)
Living with parents 0.767 0.690 0.714 0.06  0.01 0.07
(0.234) (0.441) (0.184)
Income c
1.806 2.245 1.712 0.09 0.53nn  0.44nn
(0.600) (0.018) (0.026)
Family Size 3.126 3.321 2.962 0.16 0.36  0.19
(0.450) (0.125) (0.320)
Homeownership 0.689 0.679 0.615 0.07 0.06 0.01
(0.370) (0.498) (0.899)
Size of homed 3.68 3.943 3.654 0.03 0.29  0.26
(0.917) (0.334) (0.300)
Heating degree days 10.606 11.668 11.638  1.03 0.03  1.06
(0.173) (0.975) (0.207)

nnn nn n
, , represent the statistical signicance at the 1%, 5%, and 10% level, respectively. p-value in parenthesis.
a
Responses to Likert Scale (always 5, usually 4, sometimes 3, rarely 2, never 1).
b
Over 30% 8, 25-30% 7, 20-25% 6, 15-20% 5, 10-15% 4, 5-10% 3, 1-5% 2, under 0% 1.
c
Over 15 million yen 6, 10-15 million yen 5, 8-10million yen 4, 6-8 million yen 3, 4-6 million yen 2, under 4 million yen 1.
d
The number of rooms (over seven rooms 7, six rooms 6, ve rooms5, four rooms4, three rooms 3, two rooms2, one room2).

period, some households might have had a large number of
guests, which would increase electricity use. Alternatively, other
households may have left the city for a while, which would decrease
electricity use. Table 3 presents a statistical summary of the above
variables as well as dependent variables.
4.2. Estimation results
The test of heteroscedasticity was conducted based on a
Lagrange multiplier statistic, 40.56, and this statistic has a limiting
chi-squared distribution with 13 degrees of freedom. The sample
value exceeds the critical value of 22.36 (p 0.95, d 13); hence,
the null hypothesis of homoscedasticity could be rejected. To
adjust for heteroscedasticity, we use Whites covariance matrix.
We estimated econometric model (2) in ordinary least squares
with robust standard errors. The estimation results of average
treatment effects are reported in the rst column of Table 4. The
rst and second columns in the table simply compare means of
Y2011 and Reduction, that is, there are no controls for Y2010 or X.
In the third column, we estimate column 1 with controls for
Y2010. In the fourth and fth columns, we added X controls in two
types of congurations. We also estimate in the fth columns
conguration with the Reduction as the dependent variable in the
nal column.
This table may show the effect of reward and compara-
tive feedback on the electricity-saving behaviour of a household.
The estimated coefcients of RE ward and Competative are signi-
cantly different from zero for all columns, and each estimated RE
ward and Competative is similar between columns 3, 4 and 5 with
Electricity_2011 as the dependent variable. In the previous section,
we showed the statistical difference between treatment groups
(i.e., Group 1 and Group 2) and the control group (i.e., Group 3) by
a non-parametric approach. However, this approach could not
identify the treatment effect and the other characteristics or other
externalities across time and households. Here, our parametric
approach could show the presence of the treatment effect on
encouraging electricity-saving behaviour. On the other hand, the
comparison of estimated coefcients between the comparative
feedback with reward and reward groups is not statistically
signicant. Thus, similar to the result in section 3.3, we just
showed there is the incremental impact of the competitive
consciousness treatment on energy-saving behaviour compared
with the nancial-reward treatment.
The coefcient of Social norm in column 4 is negative and
statistically signicant, thereby showing that others behaviour
may have an impact on ones own electricity-saving behaviour.16
16
This impact could stem from two sources. First, overall household behavior
in electricity saving may inuence peoples views concerning personal responsi-
bility. Specically, if others participate in electricity saving, the individual may
experience a loss in his or her self-image as a morally responsible person if he or
she does not do so. Second, there is also the presence of explicit prescriptive social
 K. Mizobuchi, K. Takeuchi / Energy Policy 63 (2013) 775787
Fig. 2. Distribution of Electricity Conservation.
If close friends and family stress the importance of saving electricity,
the individual may deduce that others also believe it is important.17
(footnote continued)
norms stemming from, for instance, family members and close friends. This
inuence may well constitute a complement to the above perceptions of others
contributions.
17
Nyborg et al. (2006) presented a theoretical framework in which individual
responsibility for pro-environmental behaviour depends on beliefs about others
behaviour. Using household survey data, Ek and Sderholm (2008) showed the
positive inuence of social norms on participating in the Swedish green-electricity
781
However, the estimated coefcients of Social norm are not sig-
nicant in the remaining columns (5 and 6). One of the reasons
might be the way we captured the social norm of a household. We
asked a simple question: Do you consider how often people close
to you are attempting electricity-saving behaviour? on the survey
and used it as the proxy variable. We think this question may have
been rather weaker than that of previous studies, and this fact may
have affected our results.
Two other non-nancial factors, such as NEP score and Target,
have no inuence on electricity-conservation behaviour. The
positive effect of the NEP score on electricity saving has mani-
fested in other pro-environmental behaviours, such as participa-
tion in the green-electricity market, in some previous studies
(Clark et al., 2003; Kotchen and Moore, 2007; Ek and Sderholm,
2008). However, electricity-saving behaviour has an aspect not
only of public good being pro-environmental but also of
private good, such as saving ones own money. This consideration
may be the cause of the insignicant effect of the NEP score.
Another possible reason is that households with strong environ-
mental preferences may already have lower pre-experiment elec-
tricity use and, assuming all households are trying to save post
Fukushima, less ability to generate electricity savings, and this
inuence might be presented as statistically insignicant regard-
ing the NEP parameter.
Family size and Homeownership have a positive and signicant
effect on electricity saving. Presumably, in larger families (with
children) it is more difcult to control energy consumption. The
size of the home is negative and signicant. This result may
indicate the possibility that there is large electricity-saving poten-
tial in a house that has many rooms. Finally, the coefcient of
Cooling degree day is positive and signicant, and the positive sign
indicates that if the temperature of the target year is relatively
cooler, the amount of time the air conditioner is used will
decrease.18,19
4.3. The presence of externalities across time and households
As we showed in the previous subsection, the treatment effects
of competitive consciousness and reward could reduce a house-
holds electricity consumption signicantly. However, the Reduc-
tion values of each group in Table 2 could not show the pure size of
the treatment effect because they include the other externalities
across time and households. We have already described the effects
of X variables in the previous section. The remaining thing is the
other externalities.
The variable i,t in estimation model (2) indicates the other
externalities, except for other variables in this model, which could
inuence the electricity consumption. For example, we have a
visitor, or go outside for some days or weeks. The rise and fall of
the number of electricity appliances could not be controlled across
time (between 2010 and 2011) in our experiment;20 this also
affects the electricity consumption. Moreover, post Fukushima
households might have signicant nancial and non-nancial
(footnote continued)
market, and Nolan et al. (2008) and Allcott (2011b) also showed its positive
inuence on U.S. household energy-saving behaviour.
18
Actually, the average temperature within the experimental term was lower
by 0.9 1C than it was the previous year.
19
In the post-study questionnaire, we asked participants to indicate a factor
that was important for electricity saving. Thirty-ve percent of households felt that
co-operation from family members was the most important factor. Willingness to
act was identied as an important factor by 31%, while 15% of households believed
that it was temperature. Only a small percentage of respondents identied
information on energy saving (4%) as an important factor.
20
Changes in the number of persons and physical movements were controlled
for across time in our experiment.
782 K. Mizobuchi, K. Takeuchi / Energy Policy 63 (2013) 775787

Table 3
Summary and standard statistics of the variables included in the analysis.

Variable Mean Std. Dev. Maximum Minimum

Dependent Variables
Electricity_2011 (kW h) 741.471 439.311 2571.000 137.000

Independent Variables
Electricity_2010 (kW h) 792.822 467.723 2628.000 178.000
Reward (dummy) 0.495 0.501 1.000 0.000
Reward with Comparative Feedback (dummy) 0.255 0.437 1.000 0.000
Social Norma 3.192 0.675 5.000 1.000
NEP-score 57.216 7.695 75.000 30.000
Targetb 3.413 1.338 8.000 0.000
Gender (respondent of survey) (male 1, female 0) 0.346 0.477 1.000 0.000
Age (household representative) 42.639 10.802 80.000 21.000
Incomec 1.894 1.103 6.000 1.000
Family size 3.135 1.228 6.000 1.000
Homeownership (yes 1, no 0) 0.668 0.472 1.000 0.000
Size of homen 3.740 1.448 7.000 1.000
Heating degree day 11.135 4.672 21.600 4.700
Reduction (%)  5.428 13.607 69.000  42.800

d
The number of rooms (over seven rooms7, six rooms6, ve rooms 5, four rooms 4, three rooms 3, two rooms 2, one room 2).
a
Responses to Likert Scale (always 5, usually 4, sometimes 3, rarely 2, never 1).
b
Over 30% 8, 2530% 7, 2025% 6, 1520% 5, 1015% 4, 510% 3, 15% 2, under 0% 1.
c
Over 15 million yen 6, 1015 million yen 5, 810million yen 4, 68 million yen 3, 46 million yen 2, under 4 million yen 1.

Table 4
Estimation results.

Model

(1) (2) (3) (4) (5) (6)

Dependent Variable Electricity 2011 Reduction Electricity 2011 Electricity 2011 Electricity 2011 Reduction

Electricity 2010 0.914 nnn 0.911 nnn 0.892nnn  0.007nnn

[0.026] [0.024] [0.025] [0.002]

Financial factor
nnn
Reward 706.884  5.876nnn  31.728n  38.127nn  39.193nn  4.718nn
[18.998] [19.104] [19.856] [2.294]

Non-nancial factors
Comparative Feedback (with Reward) 804.491nnn  8.196nnn  46.892nn  50.481nn  56.404nnn  7.325nnn
[22.106] [21.040] [21.525] [2.542]
Social Norm  18.101n  15.443  1.235
[9.868] [11.251] [1.461]
NEP-Score 0.754 1.389 0.047
[1.034] [1.082] [0.109]
Target 6.439 3.719 0.184
[7.436] [7.534] [0.910]

Socio-Economic Variables
Gender 7.396 1.637
[14.886] [1.827]
Age  0.082 0.075
[0.882] [0.088]
Income 1.610 0.428
[8.651] [0.847]
Family Size 21.473nnn 2.548nnn
[8.002] [0.875]
Homeownership 47.787nn 6.580nn
[19.447] [2.992]
Size of Home  14.875n  1.766n
[7.785] [1.068]

Weather Variable
Heating Degree day 3.004n 0.493nn
[1.729] [0.213]

Constant 44.787nnn 43.689  54.165  10.962

[16.731] [83.737] [88.716] [9.596]
Number of Observations 208 208 208 208 208 208
Adjusted R-squared  0.724 0.021 0.942 0.942 0.945 0.092
S.E. Regression 576.794 13.463 106.146 105.989 102.971 12.967
Log Likelihood  1616.491  834.918  1263.401  1261.552  1251.857  820.873

1. nnn, nn, n represent the statistical signicance at the 1%, 5%, and 10% level, respectively.
2. Standard error in parenthesis.
 K. Mizobuchi, K. Takeuchi / Energy Policy 63 (2013) 775787 783

The households in Group 3 did not need to save electricity

because they were not given any treatment in the experimental
period. Therefore, if they went about their normal daily routine,
the value of Reduction would be zero. However, as we showed
above, we can consider that the externalities may affect electricity
usage. We might naively think that the average value of the
treatment externalities may be considered as the value of Reduc-
tion in Group 3 (i.e.,  1.72%). However, it is not actually that
simple because the electricity saving in this group might be caused
by the difference of X variables, for example, household character-
istics or Heating degree days, which were statistically signicant
in Table 4. Therefore, we divide the households in Group 3 into
saved (Reduction o 0) and unsaved (Reduction Z 0) households and
examine the difference of households characteristics in Table 4.
We set the former and latter subgroups as Group 3_1 and Group
Fig. 3. The Relationship between Residual and Reduction.
3_2, respectively.
Table 5 shows the comparison of the characteristics between
Table 5 Group 3_1 and Group 3_2. From this table, we can conrm that
Comparision of Characteristics between Saved and Unsaved in Control Group.
none of the differences of characteristics, except for Size of home,
Group 3_1 Group 3_2 Group 3_1 is statistically signicant. Here, from Table 4, the characteristics
that inuence electricity consumption were Own, Family size, Size
- Group 3_2 of home and Heating degree days. Size of home has a positive
inuence on electricity-saving behaviour. Therefore, the average
Electricity_2011 735.13 761.43  26.30
(0.856)
Reduction of Group 3 (i.e., 1.72%) might include the contribution of
Electricity_2010 820.29 688.9 131.39 a larger house. For measuring the size of treatment externalities,
(0.340) we must exclude the effect due to Size of home. Here, the
Reduction  10.59 11.37 estimated coefcient for Size of home is 1.766 from Table 4,
and the difference of its average size between Group 3_1 and
NEP 57.13 56 1.13
(0.557) Group 3_2 is 0.78. Therefore, the average contribution of Size of
Social Norm 3.42 3.43  0.01 home to electricity saving in Group 3_1 is about 1.377. Therefore,
(0.953) we should subtract it from the average Reduction of Group 3_1 (i.e.,
Age 41.84 39.57 2.27 10.59) and recalculate the average Reduction of Group 3. From
(0.504)
Gender 0.48 0.48 0.00
this calculation, the new Reduction of Group 3 is about 0.901,
(0.958) which accounts for about 52% of the initial average Reduction in
Income 1.61 1.86  0.25 Group 3; we may regard this as one of the treatment externalities
(0.473) across time and households in our experiment. However, it is not a
Own 0.68 0.52 0.16
signicant change from zero but an incremental one.
(0.281)
Family Size 2.87 3.1  0.23 In the designs of reward or feedback programmes, as with our
(0.543) study, most such previous studies have not considered the
Size of Home 3.97 3.19 0.78n inuences of externalities i,t. Thus, for example, there is a
(0.067) possibility that a household might not conserve electricity at all
Heating degree days 11.09 12.44  1.35
(0.305)
in response to the treatment but that its electricity use would still
Number of Obs. 31 21 decrease owing to a small i,t. Such a case would result in an
overestimation of the treatment effect. Moreover, from a political
1. nnn, nn, n represent the statistical signicance at the 1%, 5%, and 10% level, viewpoint, it is important to understand the existence or size of
respectively.
externalities when we examine the threshold of the reward level.
2. p-value in parenthesis.
For example, we should try to nd some optimal level of the
electricity-conservation threshold that is high enough to make it
incentives to save electricity even outside the experiment. Therefore, sufciently unlikely that somebody who had zero treatment effect
these were also included in i,t. Here, the Reductions of Group 1 and would be able to claim the reward but not so high as to discourage
Group 2, which are shown in Table 2, include the treatment effect and anybody from trying to conserve energy.
treatment externalities. That is, if the effect of the latter is large, we From the results of our randomised experiment, the other
may over- or underestimate the treatment effect. This subsection externalities result in saving electricity incrementally. Therefore,
examines the existence of treatment externalities and the size of it by the size of the Reduction of the two treatment groups (i.e., Group
using the sample of the control group. 1 and Group 2) might be a little small.
Fig. 3 represents the scatter diagram that shows the relation-
ship between estimated i,t of model (5) in Table 4 and the value of 4.4. Testing for heterogeneous treatment effects of nancial reward
Reduction per households. The x-axis and the y-axis denote the
values of residual and Reduction, respectively. From this gure, we Finally, we test the heterogeneous treatment effects by inter-
can conrm that not every plot gathers around the point of origin acting the treatment dummy with the X variables. This result can
but the data may be located in/around the upward-sloping curve. bring out the incremental responsiveness of rich people, old
Thus, for example, by virtue of not being home for several days, people and people with large families to the treatment qualita-
one household could decrease electricity more than by the reduc- tively. Here, a competitive consciousness treatment combines
tion from only their response with the reward programme. There- nancial rewards and comparative feedback and, therefore, this
fore, we might say that the other externalities that we mentioned can be regarded as a heterogeneous treatment effect of nancial
above may affect electricity conservation in no small measure. reward. That is, under the nancial-reward treatment, we have
784 K. Mizobuchi, K. Takeuchi / Energy Policy 63 (2013) 775787

Table 6
Estimation results (test of heterogeneous treatment effect).

Dependent Variable (1) (2)

Electricity 2011 Electricity 2011

Electricity 2010 0.890nnn 0.891nnn

[0.024] [0.024]

Financial factor
Reward 367.199nn 336.359n
[150.623] [173.414]

Non-nancial factors
Comparative Feedback (with Reward)  52.569nn  51.090nn
[20.374] [21.105]
n
Social Norm Reward  21.976  30.311
[20.120] [21.608]
n
NEP-Score Reward  4.674nn  5.158nn
[1.899] [2.181]
n
Target Reward  20.062  18.852
[15.593] [14.372]

Socio-Economic Variables
Gender n Reward 29.800
[30.579]
n
Age Reward 2.192
[1.441]
n
Income Reward  7.504
[17.400]
n
Family Size Reward  2.392
Fig. 4. The difculty of electricity conservation.
[14.878]

Constant  283.965nn  266.527n

[136.246] [149.833]
Number of Observations 208 208
Adjusted R-squared 0.947 0.947
S.E. Regression 100.937 101.180
Log Likelihood  1246.088  1244.385
1. nnn, nn, n represent the statistical signicance at the 1%, 5%, and 10% level,
respectively.
2. Standard error in parenthesis.
3. In both estimation models, we control for social norm, NEP-score, target, gender,
age, income, family size, homeownership, size of home, and heating degree days
that were included in Table 4.
conrmed that the households who can compare their electricity
consumption with other households are likely to save electricity.
Thus, in this subsection, we focus on the heterogeneous treatment
effects of nancial reward. To test for heterogeneity, we include
the key X variables in the estimation model (2) as treatment
interaction terms. Table 6 shows the estimation result. Because of
space limitation, Table 6 shows the coefcients of interaction
terms mainly and excludes the representation of control variables.
From this table, we can conrm that the households who have a
high NEP score are likely to respond to the treatment of rewards
and to conserve more electricity than those who have a relatively
low NEP score. We believe that households with strong environ-
mental preferences may want to save more electricity than house-
holds with weak environmental preferences do. Therefore, if the
marginal costs of electricity conservation decrease because of the
nancial rewards, households with high NEP score may react
strongly to them, by saving more electricity. From these results,
we can say that it is effective for household electricity-
conservation policy to combine an increase in environmental
awareness (i.e., high NEP score) with the reward programme
rather than merely using one approach.
5. Discussion and Limitations
This section discusses two aspects concerning the marginal
cost of saving electricity: (1) a lack of understanding of the
difculties inherent in conserving electricity and (2) the size of
the marginal cost of saving electricity. Fig. 4 shows the partici-
pants evaluation of the difculties associated with electricity
conservation before and after the experiment. Interestingly, while
before the experiment the majority of respondents (60.9%) felt
that the level of electricity conservation was neither easy nor
difcult, at the end of the experiment, the share of neither easy
nor difcult decreased to 33.5%; on the other hand, 54.2% of the
participants felt that saving electricity was difcult or very
difcult.21 This change in response indicates a limited initial
understanding beforehand of the challenges associated with con-
servation behaviour. This result may also suggest that pre-
experiment responses concerning electricity-saving behaviour
have an upward bias, because participants expect to save more
electricity before actually trying to do so. In other words, people
might calculate the marginal cost of actual energy saving less
precisely than they had forecast in the survey.22 Therefore, using
only a stated intention to evaluate household energy saving may
not be appropriate, given the likelihood of overestimation. More-
over, such overestimation also indicates the importance of an
intervention such as our reward programme in reducing the
actual marginal cost.
The marginal cost of energy saving during the experiment
exceeded the incentive amounts. As described earlier, most of the
households underestimated the actual marginal cost of energy
saving; however, their actual efforts to reduce consumption may
have helped them realise the greater marginal cost of conserva-
tion, resulting in some participants ceasing or reducing their
conservation efforts. To conrm this, for Group 1, we asked the
question, Would you conserve electricity if the incentive amounts
21
We also conducted a statistical test to determine whether a difference
existed in the average difculty of saving electricity before and after the experi-
ments; we did this for both groups 1 and 2. We conrmed a signicant difference in
Group 2 at the 1% signicance level but found no signicant difference in Group 1.
22
Allcott (2011a), using data captured from 2100 individuals using the Vehicle
Ownership and Alternatives Survey (VOAS), showed that most people do not
accurately calculate the energy cost related to the purchase and use of their own
vehicles.
 K. Mizobuchi, K. Takeuchi / Energy Policy 63 (2013) 775787
were doubled? In the post-experiment survey, nearly 71% of the
households that failed to reduce their consumption returned
negative responses. This result supports the notion that the actual
marginal cost of saving electricity is quite high.
These results carry important implications for policymakers
who target energy conservation among households: before intro-
ducing nancial incentives to promote conservation, it is necessary
to study the marginal cost of electricity saving and ensure that the
related incentives exceed this cost. Moreover, the estimated
marginal cost of energy saving, as per the survey results, may be
underestimated, making it difcult to achieve the policy objective.
Alternatively, it would be worthwhile to combine economic incen-
tives with other interventions that target lowering of the marginal
cost. For example, if households are unfamiliar with efcient
electricity-saving measures, their marginal cost may increase but
acquiring additional information on efcient energy-saving proce-
dures may reduce the marginal cost of saving electricity.
Although these ndings offer important insights, this study has
several limitations, which must be acknowledged and possibly
overcome in future research. The rst is the possible sample
selection bias. Our study accepted only the rst 300 applications
from among randomly selected households. Even if the sample
selection of the households were random, it is highly likely that
participants who are relatively quick to express a desire to enrol in
this experiment or survey would have higher levels of environ-
mental awareness and would therefore be inherently motivated to
save energy. Furthermore, such participants may view the experi-
ments and/or the survey as an encouragement to conserve energy.
Ek and Sderholm (2010) conrm that participants who volunta-
rily enrol in energy-saving experiments typically have high levels
of environmental awareness and are inherently motivated. Con-
versely, households that are not interested in saving energy or that
have low levels of awareness may not participate in the experi-
ment or survey at all.23 When most of the subjects in an
experiment already have high levels of environmental awareness,
generalisation of the results becomes questionable. In fact, the bias
is greater when the sample size is small. Randomisation is
essential in an experimental study; but in our study, we cannot
exclude the possibility that the participants were biased, in that
they were highly motivated.
Second, the sample size of our study (236 participants) is not
large enough to derive policy implications from the study ndings.
One of the chief reasons for limitation in the sample size was the
use of an observable measure (i.e., a master meter) for validating
the effects of an energy-saving intervention. While an observable
measure can ensure objectivity in measurement, it poses consid-
erable executional difculties whenever the sample size is large or
the experimental period is long. For this reason, it is likely that
many eld experiments regarding the treatment of reward or
feedback programmes and that use observable measures also have
a sample size of o300 (Abrahamse et al., 2005). However, some
recent studies on the treatment of feedback programmes have
been able to use larger sample sizes (Darby, 2006; Allcott et al.,
2007; Ehrhardt-Martinez et al., 2010; Allcott, 2011b; Jessoe and
Rapson, 2012).
Third, the number of households in Group 1 is about twice that
in either of the other two groups. The main purpose of this
approach is to increase the statistical reliability to test the hetero-
geneity of the nancial treatment, as discussed in section 4.4.
However, from the viewpoint of a general randomised experiment,
this is not an appropriate approach and may lead to a biased result.
23
Sardianou (2007) surveyed the energy-consumption patterns of Greek
households and found that individuals with a thorough knowledge of environ-
mental problems and high level of environmental awareness tend to invest the
time needed to reduce energy consumption.
785
Fourth, because our study used comparative feedback collected
twice over a 2-month period, the frequency of the feedback might
be a little low. This might be one reason why we were unable to
obtain a result showing comparative feedback to be effective in
both a non-parametric and a parametric approach. In previous
studies, the effectiveness of feedback was found to increase as the
frequency of the feedback increases (Abrahamse et al., 2005;
Fischer, 2008); thus, verication and comparison are needed when
the frequency changes. These limitations should be addressed in
future research.
6. Conclusion
Through an 8-week eld experiment, this study investigated
the inuences of nancial and non-nancial factors on electricity-
saving behaviour among Japanese households. We randomly
selected 236 Japanese households, divided into three groups:
(i) reward group, (ii) reward with comparative feedback group
and (iii) control group; the rst two are treatment groups. From
the results detailed in sections 3 and 4, we have shown the
effectiveness of a nancial factor in encouraging electricity con-
servation but a more inconclusive response to the non-nancial
additional information treatment. We also examined the inu-
ences of treatment externalities across time and households, a
factor that most similar treatment studies have not considered.
This experiment found these externalities to have a positive
incremental impact on electricity saving; therefore, from this
result, we can see that if we were to fail to account for treatment
externalities, we might have overestimated the response to the
treatment effect. Moreover, we tested the existence of the hetero-
geneous treatment effect in the reward intervention. As a conse-
quence, households with high NEP scores are more likely to respond
to the reward programme and save electricity, compared to house-
holds that do not. This result suggests that in order to save
electricity, policymakers should consider combining a reward
programme with measures that increase environmental awareness.
Finally, the most fascinating nding of our study is that the
households tended to underestimate the marginal costs of elec-
tricity conservation before they actually began to save energy.
Fig. 4 in section 5 shows the change in the perceived difculty of
electricity saving before and after the experiment; this gure
shows that many households felt that the actual work related to
saving electricity was more difcult than they had previously
expected. This result reveals that policymakers should expect the
actual perceived marginal costs of energy saving to be higher than
that shown in the initial survey data; it also indicates the
importance of implementing an intervention programme (e.g.,
reward programme) that both reduces marginal costs and
encourages conservation behaviour.
Acknowledgements
This research is supported by the Japan Society for the Promo-
tion of Science (Grant-in-Aid for Young Scientists (B) #22730217),
which is acknowledged by K. Mizobuchi. K. Takeuchi would like to
thank the Department of Economics at the University of Gothen-
burg for giving him the opportunity of being a visiting researcher.
Appendix A
See Table A1.
786 K. Mizobuchi, K. Takeuchi / Energy Policy 63 (2013) 775787

Table A1
The NEP-scale statements and their response distributions (percentage).

Statement Strongly Partly Unsure Partly Strongly CORR i

disagree disagree agree agree

1. We are approaching the limit of the number of people the earth support. 5.3 6.3 21.6 42.3 24.5 0.51
2. Humans have right to modify the natural environment to suit their needs. 60.6 27.9 8.7 1.4 1.4 0.44
3. When humans interfere with nature it often produces disastrous consequences. 4.8 2.9 16.4 35.1 40.9 0.47
4. Humans ingenuity will insure that we do not make the earth unliveable. 10.6 8.7 34.1 27.9 18.8 0.16
5. Humans are severely abusing the environment. 4.3 1.9 19.7 44.2 29.8 0.52
6. The earth has plenty of natural resources if we just learn how to develop them. 11.5 28.4 27.9 23.1 9.1 0.36
7. Plants and animals have as much right as humans to exist 2.9 5.3 23.6 36.5 31.7 0.41
8. The balance of nature is strong enough to cope with the impacts of modern industrial nations. 18.3 31.3 27.9 16.8 5.8 0.27
9. Despite our special abilities, humans are still subject to the laws of nature. 1.4 2.9 12.0 35.1 48.6 0.26
10. The so-called ecological crisis facing human kind has been greatly exaggerated. 23.1 29.3 29.8 13.0 4.8 0.58
11. The earth is like a spaceship with very limited room and resources. 3.4 4.8 18.8 43.8 29.3 0.43
12. Humans were meant to rule over the rest of nature. 64.9 22.6 8.7 2.9 1.0 0.49
13. The balance of nature is very delicate and easily upset. 1.9 5.8 13.0 40.9 38.5 0.41
14. Humans will eventually learn enough about how nature works to be able to control it. 24.0 24.5 36.5 12.0 2.9 0.21
15. If things continue on their present course, we will soon experience a major ecological catastrophe. 2.9 2.4 12.0 36.1 46.6 0.47

Note: We checked the validity of using the NEP scale in our study. CORRi represents the item-total correlation, and it varies between 0.16 (No. 4) and 0.58 (No. 10). Cronbachs
alphaa coefcient of reliability used to test whether the items are sufciently inter-related to justify their combination in an indexis 0.788. This is similar to the ndings of
Ek and Sderholm (2008), and signicantly higher than the results of Kotchen and Moore (2007). Therefore, we may say that our NEP-oriented scale is reasonably adequate.

References
Abrahamse, W., Steg, L., Vlek, C., Rothengatter, T., 2005. A review of intervention
studies aimed at household energy conservation. Journal of Environmental
Psychology 25, 273291.
Abrahamse, W., Steg, L., Vlek, C., Rothengatter, T., 2007. The effect of tailored
information, goal setting and tailored feedback on household energy use,
energy-related behaviors and behavioral antecedents. Journal of Environmental
Psychology 27, 265276.
Allcott, H., Dean, K., Markus, M., 2007. Community size and network closure.
American Economic Review 97 (2), 8085.
Allcott, H., 2011a. Consumers perceptions and misperceptions of energy costs.
American Economic Review 101 (3), 98104.
Allcott, H., 2011b. Social norms and energy conservation. Journal of Public
Economics 95, 10821095.
Andreoni, J., 1990. Impure altruism and donations to public goods: A theory of
warm-glow giving. Journal of Economics 100, 464477.
Becker, L., 1978. Joint effect of feedback and goal setting on performance: A eld study of
residential energy conservation. Journal of Applied Psychology 63, 428433.
Brandon, G., Lewis, A., 1999. Reducing household energy consumption: A qualitative
and quantitative eld study. Journal of Environmental Psychology 19, 7585.
Cai, H., Chen, Y., Fang, H., 2009. Observational learning: Evidence from a rando-
mized natural eld experiment. American Economic Review 99, 864882.
Clark, C.F., Kotchen, M.J., Moore, M.R., 2003. Internal and external inuences on
pro-environmental behavior: Participation in a green electricity program.
Journal of Environmental Psychology 23, 237246.
Darby, S., 2006. The Effectiveness of Feedback on Energy Consumption. A Review
for DEFRA of the Literature on Metering. Billing and Direct Displays. Environ-
mental Change Institute. University of Oxford Press, Oxford.
Dubin, A., McFadden, D.L., 1984. An econometric analysis of residential electric
appliance holdings and consumption. Econometrica 52, 345362.
Dunlap, R.E., Van, Liere, K.D., Mertig, M.G., Jones, R.E., 2000. Measuring endorse-
ment of the new ecological paradigm: A revised NEP scale. Journal of Social
Issues 56 (3), 425442.
Dwyer, W.O., Leeming, F.C., Cobern, M.K., Porter, B.E., Jackson, J.M., 1993. Critical
review of behavioral interventions to preserve the environment: Research since
1980. Environment and Behavior 25, 275321.
Ehrhardt-Martinez, K., Donnelly, K.A., Laitner, J.P., 2010. Advanced Metering
Initiatives and Residential Feedback Programs: A Meta-Review for Household
Electricity-Saving Opportunities. Report No. E105. American Council for an
Energy-Efcient Economy, Washington, D.C.
Ek, K., Sderholm, P., 2008. Norms and economic motivation in the Swedish green
electricity market. Ecological Economics 68, 169182.
Ek, K., Sderholm, P., 2010. The devil is in the details: Household electricity saving
behavior and the role of information. Energy Policy 38, 15781587.
Espey, J., Espey, M., 2004. Turning on the lights: A meta-analysis of residential electricity
demand elasticities. Journal of Agricultural and Applied Economics 36, 6581.
Fischer, C., 2008. Feedback on household electricity consumption: a tool for saving
energy? Energy Efciency 1, 79104.
Fransson, M., Garling, T., 1999. Environmental concern: Conceptual denitions,
measurement methods, and research ndings. Journal of Environmental
Psychology 19, 369382.
Gleerup, M., Anders, L., Soren, L.P., Mikael, T., 2010. The effect of feedback by text
message (SMS) and email on household electricity consumption: Experimental
evidence. Energy Journa 31 (3), 113132.
Guagnano, G.A., Stern, P.C., Dietz, T., 1995. Inuences on attitude-behavior relation-
ships: A natural experiment with curbside recycling. Environment and Beha-
vior 27 (5), 699718.
Halvorsen, B., Larsen, B.M., 2001. The exibility of household electricity demand
over time. Resource and Energy Economics 23, 118.
Jessoe, K., Rapson, D., 2012. Knowledge Is (Less) Power: Experimental Evidence
from Residential Energy Use. NBER Working Paper No. 18344, August 2012.
Katzev, R.D., Johnson, T.R., 1983. A social-psychological analysis of residential
electricity consumption: The impact of minimal justication techniques.
Journal of Economic Psychology 3, 267284.
Kotchen, M.J., Moore, M.R., 2007. Private provision of environmental public goods:
Household participation in green-electricity programs. Journal of Environmental
Economics and Management 53, 116.
Matsukawa, I., 2004. The effects of information on residential demand for
electricity. Energy Journal 25 (1), 117.
McCalley, L.T., Midden, C.J.H., 2002. Energy conservation through product-
integrated feedback: The roles of goal-setting and social-orientation. Journal
of Economic Psychology 23, 589603.
McClelland, L., Cook, S.W., 1980. Promoting energy conservation in master-metered
apartment through group nancial incentive. Journal of Applied Social Psy-
chology 10 (1), 2031.
McConnell, K.E., 1997. Does altruism undermine existence value? Journal of
Environmental Economics and Management 32, 2237.
Midden, C.J.H., Meter, J.E., Weenig, M.H., Zieverink, H.J.A, 1983. Using feedback,
reinforcement and information to reduce energy consumption in households:
A eld experiment. Journal of Economic Psychology 3 (1), 6586.
Miguel, E., Kremer, M., 2004. Worms: Identifying impacts on education and health
in the presence of treatment externalities. Econometrica 72 (1), 159217.
Munshi, K., Myaux, J., 2006. Social norms and the fertility transition. Journal of
Development Economics 80, 138.
Nolan, M.J., Schultz, P.W., Cialdini, B.R., Goldstein, N.J., Griskevicius, V., 2008.
Normative social inuence is underdetected. Personality and Social Psychology
Bulletin 34, 913923.
Nyborg, K., Howarth, R.B., Brekke, K.A., 2006. Green consumers and public policy: On
socially contingent moral motivation. Resource and Energy Economics 28,
351366.
Petersen, J.E., Shunturov, V., Janda, K., Platt, G., Weinberger, K., 2007. Dormitory
residents reduce electricity consumption when exposed to real-time visual
feedback and incentives. International Journal of Sustainability in Higher
Education 8, 1633.
Poortinga, W., Steg, L., Vlek, C., Wiersma, G., 2003. Household preferences for
energy-saving measures: A conjoint analysis. Journal of Economic Psychology
24, 4964.
Sardianou, E., 2007. Estimating energy conservation patterns of Greek households.
Energy Policy 25. (37782791).
Schultz, P.W., Nolan, M.J., Cialdini, B.R., Goldstein, N.J., Griskevicius, V., 2007. The
constructive, destructive, and reconstructive power of social norms. Psycholo-
gical Science 18, 429434.
Siero, R.W., Bakker, A.B., Dekker, G.B., van den Burg, M.T.C., 1996. Changing
organizational energy consumption behavior through comparative feedback.
Journal of Environmental Psychology 16, 235246.
Silk, J.I., Joutz, F.L., 1997. Short- and long-run elasticities in US residential electricity
demand: A cointegration approach. Energy Economics 19, 493513.
Stern, P.C., 1992. What psychology knows about energy conservation. American
Psychologist 47, 12241232.
 K. Mizobuchi, K. Takeuchi / Energy Policy 63 (2013) 775787 787

Tokyo Electric Power Company, 2011. The supply and demand results for this
summer. http://www.tepco.co.jp/en/press/corp-com/release/11092610-e.html.
Accessed 26 Sept 2011.
Vaage, K., 2000. Heating technology and energy use: A discrete/continuous choice
approach to Norwegian household energy demand. Energy Economics 22,
649666.
Van den Bergh, J.C.J.M., 2008. Environmental regulation of households: An
empirical review of economic and psychological factors. Ecological Economics
66, 559574.
Winett, R.A., Kagel, J.H., Battalio, R.C., Winkler, R.C., 1978. Effects of monetary
rebates, feedback, and information on residential electricity conservation.
Journal of Applied Psychology 63, 7380.