Content uploaded by Luis Mundaca
Author content
All content in this area was uploaded by Luis Mundaca on Jul 01, 2016
Content may be subject to copyright.
- 1 -
Paper presented at the 39th IAEE International Conference ‘Energy: Expectations and Uncertainty’
Bergen, Norway, 19-22 June 2016.
Behaviour, context and electricity use: Exploring the effects of real-
time feedback in the Swedish residential sector
Mats Tedenvall & Luis Mundaca
International Institute for Industrial Environmental Economics at Lund University, Sweden
Abstract
This paper provides an empirical analysis of the effectiveness of real-time feedback technology in the
Swedish residential sector. We take the ‘100Koll’ service provided by one the largest energy
companies in Sweden as a case study. Based on an intervention group (i.e. people using the service)
of 1 753 households, a survey was carried out to explore behavioural, moral, socio-economic and
contextual variables affecting electricity consumption and savings, and the effectiveness of the
100Koll service. Data was collected from January 2011 to April 2015 and both engineering and
econometrics analyses were applied. Results show a fall in consumption in the range of 1.4–1.9%. In
principle, this finding is better explained by socio-economic and contextual factors (e.g. household
size) than behavioural and moral issues. However, those households with greater perceived
behavioural control and a greater sense of moral obligation were the ones that actually reduced their
consumption. It is concluded that the implementation of real-time feedback per se is likely to be
insufficient to foster increased energy efficiency. Complementary policy measures (e.g. energy and
carbon pricing, awareness raising) need to be designed and implemented accordingly.
Keywords: Behaviour; Electricity use; Real-time feedback; Residential sector; Sweden.
- 2 -
1. Introduction
A longstanding argument is that energy efficiency improvements and the resulting economic and
environmental benefits have been hampered by market and behavioural failures (see e.g.
Goldemberg, Johansson, Reddy, & Williams, 1988; Jaffe & Stavins, 1994; Sanstad & Howarth, 1994;
Sutherland, 1991). At the European Union (EU) level, countries have introduced a plethora of policies
to cost-effectively tap energy efficiency potentials.1 To address information-related failures (e.g.
availability of energy use information), the large scale implementation of smart meters (SMs) is now
required at the EU level (via Directives 2006/32/EC and 2009/72/EC). The EU anticipates that the
provision of real-time feedback via SMs will result in a 10% reduction of energy use in the residential
sector (EC, 2011).2 SMs eliminate estimated billing, enable two-way communication between utilities
and users, and provide real-time feedback to residents about their energy use and costs.
In line with Rational Choice Theory (Coleman & Fararo, 1992; Goode, 1997) and the Information-
Deficit Model (Blake, 1999; Owens, 2000), the provision of real-time feedback via SMs assumes that
consumers will behave more rationally if they have access to useful information about their energy
demand and preferences. However, although there is evidence to suggest that real-time feedback
encourages more efficient energy use (see e.g. Darby, 2006; Gans, Alberini, & Longo, 2013a; Schleich,
Klobasa, Gölz, & Brunner, 2013), there is also an emerging body of literature which shows that the
provision of information in itself may be insufficient to lead to behavioural change (see e.g. Bager &
Mundaca, 2015; Fischer, 2008; Owens & Driffill, 2008). Multiple behavioural failures (e.g. bias,
heuristics, illusions, misconceptions) mean that energy users do not behave as rational choice theory
predicts (Mundaca, Neij, Worrell, & McNeil, 2010; Shogren & Taylor, 2008). Behavioural economics
suggests that many other aspects (e.g. values, norms, attitudes, perceived behavioural control and
herustics) can also affect user behaviour (cf. Kahneman, 2003; Kahneman, Knetsch, & Thaler, 1991;
Kahneman & Tversky, 1984; Loewenstein, Read, & Baumeister, 2003; Simon, 1986; Thaler, 1980). The
emerging literature on behavioural economics applied to energy use and decarbonisation argues that
further interventions are needed to exploit beneficial behaviours such as social norms, conditional
cooperation and loss aversion (see e.g. Allcott & Mullainathan, 2010; Bager & Mundaca, 2015; Dietz,
Gardner, Gilligan, Stern, & Vandenbergh, 2009).
Within this context, several studies have been carried out in EU countries (e.g. Denmark, Greece and
the United Kingdom). However, there is very limited knowledge about the effects of real-time
feedback in Sweden, which is the geographical scope of this study. A critical review of existing studies
reveals marginal reductions in consumption (0.04–2.24%), low participation rates, lack of large-scale
trials, and statistically insignificant outcomes (see Pyrko, 2009; Uggmark, 2013). Overall, there is still
limited understanding of the potential effectiveness of real-time feedback, and the underlying factors
affecting energy use in the Swedish residential sector.
This paper provides an empirical analysis of the effectiveness of real-time feedback technology in
Swedish households. Framed by the key theoretical elements of behavioural economics, the study
aims to increase our understanding of how, and to what extent, behavioural, moral and contextual
aspects can affect consumption via the provision of real-time feedback. Behavioural economics
research relies on empirical studies to infer the actual behaviour of individuals, rather than derive
self-evident outcomes from a theoretical analysis. Therefore, our research focused on an
intervention group (i.e. people with access to real-time feedback) of 1 753 Swedish households. A
survey explored multiple variables related to consumption, savings and the effectiveness of the
100Koll service as such. Data was collected from January 2011 to April 2015 and a mix of engineering
and econometrics approaches were deployed for the analysis. The study aims to contribute to
1 Despite significant market and behavioural hurdles, the global energy efficiency market is estimated to be worth at least USD 310 billion/
year, and energy efficiency is now considered the world’s ‘first fuel’ (IEA, 2014, p. 16).
2 A total of 16 Member States will proceed with the large-scale rollout of SMs by 2020 or earlier.
- 3 -
knowledge of behavioural aspects that may be associated with large-scale trials of SMs. However, its
exploratory nature means that our outcomes must be taken with due caution.
The paper is structured as follows: Section 2 outlines the methodology, including the case study,
data, survey and methods. Key findings are presented in Section 3 and the results are compared with
the literature. Finally, Section 4 offers some key conclusions.
2. Methodology
2.1. Case study
The subject of the case study was ‘100Koll’ service provided by the energy company E.ON. The
service allows users to monitor their consumption on a smartphone, tablet or via a web page.
Feedback is provided in (close to) real-time; there is a delay of up to one minute between actual
consumption and its presentation on the display. In order to access the service, the user must install
an optical eye. This is connected to the SM that normally is installed at the property. To transmit data
to the 100Koll database, the optical eye is connected to the user’s Wireless Local Area Network
(WLAN). 100Koll can also monitor the consumption of individual appliances via smart plugs. The plug
can be programmed to switch the supply to the appliance on or off.
Figure 1: The 100Koll smartphone app (middle), the smart plug (left) and the optical eye that is connected to
the communication box (right).
Source: E.ON (used with permission)
Fischer (2008) identifies some characteristics of effective feedback that help to both stimulate energy
efficiency/ conservation measures and are appealing to users. The study argues that feedback should
be based on actual consumption, be provided frequently and in the long term, encourage interaction,
take into account the breakdown of an appliance, offer historical or normative comparisons, and be
presented in an appealing way. 100Koll fulfils many of these requirements as it is based on actual
consumption, which can be monitored anytime, anywhere provided that the user has a smartphone
or tablet. It is provided in close to real-time, shows consumption on an hourly, daily and monthly
basis, while historical consumption is presented on a graph. Actual and historical data reported by
smart plugs is also available. The user can choose to view their data in kWh or cost; however, in the
latter case, they must manually enter the electricity price (in öre/kWh)3. On the other hand, no social
normative comparisons (e.g. with national averages, similar households or households in the
neighbourhood) are provided. Although there is no feedback regarding environmental impacts, the
application includes the animated mascot, Bongo, who becomes happier the more electricity is
3 1 Swedisk Krona = 100 öre.
- 4 -
saved. 100Koll was introduced in Sweden in February 2014 and the number of users has gradually
increased.
2.2. Data collection and survey methods
Consumption data was collected from two groups of E.ON customers: an intervention group that had
installed the 100Koll service in 2014, and a control group that did not use the real-time feedback
service. The initial size of the intervention group was 2 751; however, only the 2 173 users that had
started to use the service before 30 September 2014 were considered. In addition, only users for
whom data for the period January 2011 to April 2015 was available were considered. Buildings that
had different users during the implementation period were excluded, as were users that had moved
house (i.e. used different residential IDs). Data was collected until April 2015, at which time all of the
sample had used the service for 8 to 14 months approximately. The final size of the intervention
group was 1 753.4
The initial size of the control group was 2 048 and was selected by the energy company itself. The
aim was to find a socio-economic and geographical mix that mimicked the intervention group. Users
for whom no data had been recorded in any of the months from January 2011 to April 2015 were
excluded and the final size of the control group was 1 342.
In June 2015, a survey was sent to the 1 753 households in the intervention group. The aim was to
capture the factors underlying their energy consumption and savings. This data provided the input to
an econometric and energy analysis (details in next section). Given a margin of error of 5% and a
confidence level of 95%, the recommended sample size was 316. We received responses from 543
households, corresponding to a 99% confidence level and nearly 4.6% margin of error.
The survey was divided into three main sections: a) behavioural and moral aspects, including
subjective norms, perceived behavioural control, awareness of consequences and ascribed
responsibility (see Table 1); b) socio-economic and contextual factors (e.g. income, education,
household size) (see Table 2); and c) electricity-saving behaviour and the 100Koll. Questions related
to behavioural and moral aspects were constructed with a five point ‘Likert scale’ (Likert, 1932),
where 1 represents “do not agree at all” and 5 represents “completely agree”.
In the third section concerning electricity-saving behaviour and the 100Koll, the following multiple-
choice question was included: “Have you undertaken any of the following measures in your
residence in the last three years?” Options included: new windows, draft-proofing windows,
insulating the roof space, insulating the facade, insulating the roof, installing a heat pump, new
kitchen appliances, lowering thermostats, installing solar panels, changing the heating system,
recovering ventilated air, changing light bulbs, taking shorter showers and other measures. For each
of these measures the respondent could choose “Yes, completely”, “Partially” or “No”. The variables
100K_action and 100K_reduces were used as proxies of how much the 100Koll service had helped
respondents reduce their consumption. 100K_reduces aimed to capture whether users believed that
service could reduce their consumption. In contrast, 100K_action aimed to identify whether they had
taken any energy efficiency actions as a result of the service. These variables were coded as follows:
For 100K_reduces:
• 1 means the 100Koll service did not help to reduce electricity use
• 5 means the 100Koll service did help to reduce electricity use
For 100K_action: (binary)
• 0 means that the 100Koll service contributed (solely or partly) to taking measures to reduce
consumption
• 1 means that other factors helped the user to take action to reduce their consumption
4 Note that Ehrhardt-Martinez et al. (2001) categorized a large-scale study as one that involves more than 100 people.
- 5 -
Table 1: Behavioural and moral variables
Variable
Coding
Interval
Awareness of consequences
(AC)
1 to 5
1 means that energy use does not cause global warming
5 means that energy use has a large impact on global
warming
Ascribed responsibility (AR)
1 to 5
1 means that the respondent does not believe s/he has
any responsibility to act to reduce global warming
5 means that the respondent does believe s/he has a
responsibility to act to reduce global warming
Personal norms (PN)
1 to 5
1 means that the respondent feels little moral obligation
to reduce global warming
5 means that the respondent feels a very high moral
obligation to reduce global warming
Attitudes (AT)
1 to 5
1 means that it is unimportant to reduce consumption
5 means that it is very important to reduce consumption
Perceived behavioural control
(PBC)
1 to 5
1 means the respondent believes that there are very few
opportunities to change their consumption
5 means that the respondent believes that it is easy to
change their consumption
Subjective norms (SN)
1 to 5
1 means that the respondent does not feel any pressure
to reduce their consumption from others
5 means that the respondent feels that there is pressure
from others to reduce their consumption
Table 2: Socio-economic and contextual variables
Variable
Coding
Interval
Living area
1
2
3
4
5
6
0–49 m2
50–99 m2
100–149 m2
150–199 m2
200–249 m2
More than 250 m
2
Household size
1-6
7
1–6 persons
7 or more
Income (Household)
1
2
3
4
0–20 000 SEK*
21 000–50 000 SEK*
51 000–80 000 SEK*
More than 80 000 SEK*
Age (respondent)
1
2
3
4
5
6
18–29 years
30–39 years
40–49 years
50–59 years
60–69 years
70–79 years
Education
1
2
3
4
Elementary school
High school, Vocational training
Qualified vocational school
College/university
(*) SEK: Swedish Krona
2.3. Data analysis methods
Data analysis methods were based on engineering and econometrics approaches. The former was
used to estimate several baselines (i.e. what would have happened in the absence of SMs) in order to
- 6 -
assess the effectiveness of real-time feedback in reducing electricity use. These baselines took into
account climatic correction, control and intervention groups, and different timeframes.
Three baselines (counterfactuals) were developed using different approaches and timeframes. These
approaches aimed to provide conservative estimates of consumption. The first was based on climatic
correction and historical consumption in Swedish households estimated over a three-year period
2011–2013. Climate correction used the method described by the Swedish Meteorological and
Hydrological Institute (SMHI).5 As energy indexes differ as a function of geographical location, each
user was allocated a factor that corresponded to the closest location (in latitude) from a sample of 20
cities spread over Sweden.6 The effectiveness of real-time feedback was estimated by comparing
monthly electricity use post-100Koll installation (i.e. the intervention group) with expected average
electricity use (i.e. other Swedish households).
The second baseline covered the time period just before (October 2013–January 2014) and after
(October 2014–January 2015) the implementation of the 100Koll service. Consumption data was
gathered and compared for both the control and intervention groups. However, this highlighted a
high degree of variability in the control group and the data cast doubt on whether the control group
was truly comparable to the intervention group.7 Consequently, a third baseline was developed. The
approach was similar to the second, but the whole baseline period (January 2011–February 2014)
was used to estimate and compare expected consumption. This data was compared with actual
consumption (October 2014–April 2015) in the intervention group.
Figure 2: Timeframes used to calculate the effectiveness of SMs, showing the baseline period (BLP), the
intervention (IP), and the period just before (JBP) and just after (JAP) implementation.
An econometric analysis and discrete choice models were used to explore the behavioural, moral,
socio-economic and contextual determinants of consumption in the intervention group. The
specification and testing of the model led to the development of different, but complementary,
discrete choice models designed to explain electricity consumption and savings, and other
determinants of the effectiveness of real-time feedback. Building upon the Theory of Planned
Behaviour (TPB) (Ajzen, 1991) and the Value-Belief-Norm (VBN) (Stern, 2000), independent variables
were developed that included attitudes, subjective norms, perceived behavioural control, awareness
of consequences, and ascribed responsibility. Socio-economic and contextual variables such as age,
5 The method consists of three steps. In the first, the base load (unaffected by temperature) is removed. In the second, the remainder is
divided by the energy factor (retrieved from the SMHI). In the third, the base load is added to the corrected value.
6 The places used were Falsterbo, Tomelilla, Hässleholm, Älmhult, Borgholm, Göteborg, Linköping, Norrköping, Järfälla, Österåker,
Enköping, Sundsvall, Härnösand, Sollefteå, Luleå, Boden and Haparanda.
7 Between January 2011 and April 2015 the control group had an average monthly consumption of 926 kWh (SD = 278, Mdn = 955, Max =
2159, Min = 8, N=1 342), compared to the intervention group’s average of 1 281 kWh (SD = 570, Mdn = 1259, Max = 6428, Min = 92,
N=1 753).
- 7 -
education, income, living area and household size were also taken into account. A stepwise
regression analysis and statistical tests and metrics supported the overall quantitative exercise.
First, bivariate tests calculated correlations between variables. Second, partial correlations were
estimated. This step was necessary to identify multicollinearity (i.e. correlations between
independent variables). Third, using multiple regression models, the analysis quantified the
contribution of each variable to variation in consumption or savings. The aim was to identify the
regression model that satisfied the following conditions: highest adjusted R2 (and lowest standard
error), p-values below 0.05 (for independent variables)8, the lowest variation coefficient and no
indication of collinearity. With respect to the latter, Variance Inflation Factors (VIF) were computed
to quantify to contribution of collinearity to the variance of an estimated regression coefficient. A VIF
above 5 was taken as evidence of multicollinearity (Calvo-Mora, Leal, & Roldán, 2006; Mason &
Perreault, 1991). In order to test the internal consistency of a multiple-item construct (i.e. a factor
that is measured with two or more questions) Cronbach’s α (Cronbach, 1951) was used. This method
was applied to three of the moral and psychological factors, which were constructed from the
average of two questions. The method has been applied in earlier work (Abrahamse & Steg, 2009;
Black, Stern, & Elworth, 1985; Botetzagias, Malesios, & Poulou, 2014; Harland, Staats, & Wilke, 1999).
3. Results and discussion
3.1. The effectiveness of real-time feedback
In the first case, historical data from Swedish households and no control group were used to
estimate the baseline. After climate correction, average consumption was estimated at 1 287 kWh
(Mdn = 1091, SD = 805, N=20475) for the period 2011–2013 (i.e. the entire baseline period, BLP).
Actual monthly consumption in the intervention group was 1 269 kWh (Mdn = 1080, SD = 815,
N=1753) during the intervention period. This translates into a difference of approximately 1.4%.
The second approach compared the intervention group with the control group. Consumption in the
four months preceding the system’s implementation (October 2013–January 2014) was compared
with consumption in the four months following its implementation (October 2014–January 2015). For
the control group, consumption was 1 130 kWh pre-implementation and 1 121 kWh post (a
reduction of approximately 0.80%). This can be compared to the intervention group, where
consumption was 1 542 kWh pre-implementation and 1 501 kWh post. Assuming an equal
percentage change for both groups (of 0.80%), estimated consumption for the intervention group is
1 530 kWh; therefore there was an approximately 1.9% reduction in consumption that was due to
the SM technology.
8 Typically, a p-value less than .05 is considered statistically significant (Nuzzo, 2014). Nuzzo highlights, however, that many scientists infer
too much from p-values and that findings must be replicated before claiming that results are robust.
- 8 -
Figure 3: Estimated average monthly consumption (uncorrected for climate differences) for the intervention
group
Under the third approach, the actual average monthly consumption for the control group was 1 165
kWh and the estimated expected consumption was 1 091 kWh (a reduction of approximately of
6.3%). As for the intervention group, estimated consumption was 1 605 kWh. Actual consumption
was 1 480 kWh, reflecting a reduction of 7.7%. Assuming an equal percentage change (of 6.3%) for
both groups, consumption for the intervention group was estimated at 1 503 kWh. Therefore, the
results suggest a reduction of 1.5% due to SM technology.
It is interesting that the second approach found a greater reduction than the first and third
approaches. This may be because the measure only concerns the four first months of the
intervention, while the other measures relate to the full seven months of the intervention. The
finding is consistent with the literature, which shows that the effect of real-time feedback often is
greatest during the period closest to the intervention, and falls as the feedback period lengthens.
Ehrhardt-Martinez (2010) found that average effects were more apparent for shorter (10.1%) than
longer (7.7%) periods of time. Likewise, a study in the Netherlands found that initial reductions in
consumption (7.8%) could not be sustained and after 15 months had fallen to under 1.9% (Van Dam
et al., 2010). In contrast, Fischer (2008) found no clear indication that the initial response to feedback
was greater than after a longer period, and Darby (2006) argued that the effect persists over time.
Our findings seem to be consistent with the few studies carried out in Sweden, which report
reductions in the range of 0.04–2.24% (Pyrko, 2009; Uggmark, 2013). Our results are also in line with
Matsukawa (2004) and Bager and Mundaca (2015), who estimate savings of around 1.5% and 1.6%
respectively. Matsukawa (2004) argues that these low figures may be due to a complex user
interface, together with a lack of helpful information. The study concluded that a lot of cognitive
effort was needed to grasp the information provided by real-time services. Bager and Mundaca
(2015) conducted a meta-analysis of SM interventions9, and estimated that the average change in
consumption was around −1.6% (±9.7%, 1st–3rd quartile: −6.15% to 0.05%, M=−2.9%, N=19).
This lack of effectiveness may also be explained by the accessibility of feedback. To access the
feedback information, the users have to make a conscious decision to open the 100Koll application
9 See e.g. Gans et al. (2013a), van Dam et al. (2010); Raw et al. (2011); Schleich et al. (2013); and Grønhøj & Thøgersen (2011).
Baseline
period
Impl.
period
Intervention
period
- 9 -
on their smartphone or tablet or navigate to the appropriate website. Although this may not seem
like a huge effort, it must compete with a multitude of other services, web pages and applications. A
separate in-home display unit for feedback (e.g. Gans, Alberini, & Longo, 2013b; Mountain, 2006;
Raw & Ross, 2012) and this may present information in a way that is easier to manage (cf. Bager &
Mundaca, 2015).
McKerracher and Torriti (2013) discuss the reasons for the lack of effectiveness and argue that the
‘Hawthorne Effect’ may play a role. Specifically, participants in smaller studies (e.g. < 100) are aware
that their behaviour is being monitored and try to ‘improve’ their performance. The effect dissipates
in larger intervention groups (the case here). The same study analysed 27 interventions and found
that opt-out schemes created a conservation effect of 2.6%; this was compared to opt-in schemes
that generated savings of 4.5%.
Setting aside methodological and contextual aspects, other studies report higher savings than our
estimates. For instance, Darby (2006) concludes that real-time feedback should result in savings in
the range of 5–15%. Another study of 33 trials of an in-home display system concluded that
reductions of 3–5% should be expected for large-scale rollouts of SMs (McKerracher & Torriti, 2013).
Schleich et al. (2013) report 4.5% reductions when email or SMS feedback was provided in Austria,
while Gans et al. (2013) estimate savings in the range of 11–17% for users provided with a key-pad
meter in Northern Ireland. In Denmark, a 3% reduction was estimated for 1 397 users with
exceptionally high consumption who were sent SMS and emails (Gleerup, Larsen, Leth-Petersen,
Togeby, et al., 2010).
3.2. Determinants of consumption, savings and the effectiveness of 100Koll
3.2.1. Descriptive statistics
Of the 543 survey responses, only 226 respondents answered all of the questions. This smaller
sample size increased the margin of error to 6.17%, and reduced the confidence level to nearly 90%.
Descriptive statistics showing the number of responses to each question are shown in Table 3.
The average age of respondents was 48 years and the average living area was 122 m2, for a
household of nearly three persons. Overall, users were aware of the consequences of their energy
use and felt responsible for these consequences. They also had a positive attitude to electricity
saving. Regarding energy-saving norms, pressure came more from respondents themselves than
from others.
- 10 -
Table 3: Variables and descriptive statistics. See Table 1 and 2 for interval values.
Variable
Unit/Coding
N
Mean
Stand. Dev
Min
Max
Dependent variables
Elect_use
kWh
342
14 229
6 513
2 939
53 173
ES_behaviour
Scale (0–30)
543
5.32
4.16
0
29
100K_reduces
Scale (1–5)
497
3.40
1.22
1
5
100K_action
0–1
493
0.28
0.45
0
1
Behavioural and moral variables
AC
Scale (1–5)
501
4.00
0.95
1
5
AR
Scale (1–5)
523
4.04
1.05
1
5
PN
Scale (1–5)
519
3.02
1.07
1
5
AT
Scale (1–5)
522
4.00
0.79
1
5
PBC
Scale (1–5)
530
3.62
1.04
1
5
SN
Scale (1–5)
493
2.53
1.06
1
5
Socio-economic and contextual variables
Living area
Scale (1–6)
503
3.49
0.97
1
6
Household size
Scale (1–7)
503
2.76
1.18
1
7
Income
Scale (1–4)
435
2.34
0.67
1
4
Age
Scale (1–6)
496
3.95
1.27
1
6
Education Scale (1–4) 484 2.76 1.06 1 4
Three of the behavioural and moral factors (SN, AR and PBC), were single-item constructs based on
one question. The other three (AC, PN and AT) were constructed by combining data from two
questions (Table 4). Multiple item constructs are preferable as they average out measurement errors
(Nunnally & Bernstein, 1994) and provide a better representation of a complex concept (McIver &
Carmines, 1981). However, they require more questions to be asked, which may result in a lower
response rate. Cronbach’s α was used to measure the reliability and consistency of these constructs.
The value is normally between 0 and 1. The higher the value the better consistency and reliability of
α is. Using George and Mallery’s (2006) scale, the AC construct is ‘questionable’, the PN construct is
‘acceptable’ and the AT construct is ‘unacceptable’ (see Table 5). The latter raised statistical
concerns, especially as it is significantly lower than the values obtained by Botetzagias et al. (2014)
who used the same two-item construct to investigate electricity saving behaviour in Greece.
Table 4: Survey results for behavioural constructs. A 5-point Likert scale was used in all cases.
Question
Variable
N
Mean
SD
I think that global warming is a problem for the society.
AC
514
4.01
1.08
By saving electricity I contribute to a reduction of global warming.
AC
519
3.91
1.13
I feel guilty when I use a lot of energy.
PN
525
2.88
1.21
I feel like a better person when reducing consumption.
PN
522
3.16
1.20
It is important for me to save electricity to reduce global warming.
AT
523
3.54
1.21
It is important for me to save electricity to reduce my costs.
AT
536
4.46
0.79
- 11 -
Table 5: Cronbach’s α for the three behavioural constructs.
Construct/Variable
Cronbach's α
N (both questions answered)
AC
.659
501
PN
.736
519
AT
.309
522
3.2.2. Bivariate and partial correlations
The results of bivariate correlations are shown in Table 6. The correlation between Elect_use and
100K_reduces and all five contextual variables were statistically significant at either the .05 or .01
level.
Table 6: Bivariate correlations between all variables
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
1. Elect_use
1
2. ES_behaviour
.023
1
3. 100K_reduces
-.158**
.164**
1
4. 100K_action
-.007
.054
.388**
1
5. AC
-.068
.031
.147**
.049
1
6. AR
-.082
.047
.097*
.013
.739**
1
7. PN
-.022
.099*
.280**
.201**
.418**
.322**
1
8. AT
-.074
.090*
.233**
.160**
.623**
.507**
.467**
1
9. PBC
-.089
.214**
.305**
.142**
.187**
.135**
.205**
.261**
1
10. SN
-.039
.074
.212**
.175**
.222**
.204**
.382**
.345**
.268**
1
11. Living area
.379**
.042
-.023
-.052
.009
.019
.018
-.002
-.01
-.064
1
12. Household
size
.223**
.012
.015
-.053
.075
.083
.053
-.021
-.072
-.049
.158**
1
13. Income
.271**
.03
-.071
-.045
-.038
.001
-.036
-.139**
-.059
-.110*
.215**
.214**
1
14. Age
-.112*
-.034
.01
.117*
.005
-.02
.103*
.186**
.119**
.276**
-.061
-.548**
-.118*
1
15. Education
.130*
.001
-.07
-.064
.029
.068
-.05
-.083
-.130**
-.002
.170**
.118**
.313**
-.022
1
* Correlation is significant at the .05 level (2-tailed).
** Correlation is significant at the .01 level (2-tailed).
After controlling for all other variables Living area and Income were statistically significant.
100K_reduces was almost statistically significant with a p-value close to .05. This suggests that those
who believed that 100Koll helped to reduce costs tended to reduce their consumption. Additionally,
there was an insignificant partial correlation (r = −.008, p = .91) between Elect_use and 100K_action,
which described the extent to which the respondent perceived that the 100Koll contributed to
energy efficiency. This result is consistent with the marginal effectiveness of the 100Koll service
described in Section 3.1, which highlighted small savings of 1.4–1.9%. The partial correlation
between Elect_use and ES_behaviour was insignificant (r = .001, p = .99).
Table 7: Partial correlations between Elect_use and dependent variables where significant bivariate
correlations were found. All other variables were controlled for.
Variable
Correlation (r)
p-value
100K_reduces
−.12
.07
Living area
.30
< .001
Household size
.13
.07
Income
.19
.01
Age
−.05
.48
Education
.02
.74
- 12 -
There were significant bivariate correlations between ES_behaviour and 100K_reduces, PN, AT and
PBC. After controlling for all other variables only PBC was statistically significant with an even higher
correlation than the corresponding bivariate correlation (Table 8).
Table 8: Partial correlations between ES_behaviour and dependent variables where significant bivariate
correlations were found. All other variables were controlled for.
Variable
Correlation (r)
p-value
100K_reduces
−.04
.60
PN
.10
.14
AT
.00
.99
PBC
.23
.001
There were statistically significant bivariate correlations between 100K_reduces and all behavioural
and moral variables. After controlling for all other variables, only PN and PBC were statistically
significant (Table 9).
Table 9: Partial correlations between 100K_reduces and dependent variables where significant bivariate
correlations were found. All other variables were controlled for.
Variable
Correlation (r)
p-value
AC
.00
.97
AR
−.08
.25
PN
.27
< .001
AT
.11
.11
PBC
.21
.002
SN
−.03
.67
Finally, results suggested that 100K_reduces may be a proxy for the effectiveness of the 100Koll
service. Statistically significant bivariate correlations were found between 100K_action, all
behavioural and moral factors, Age and PN. Partial correlations showed that only PN was statistically
significant (Table 10).
Table 10: Partial correlations between 100K_action and dependent variables where significant bivariate
correlations were found. All other variables were controlled for.
Variable
Correlation (r)
p-value
PN
.21
.002
AT
.03
.63
PBC
.08
.25
SN
.002
.98
Age
.06
.93
3.2.2. Econometric results
The econometric analysis began with the inclusion of all of the behavioural, moral, socio-economic
and contextual variables and aimed to explain variability in all four dependent variables. The four
initial models included specific predictor variables (e.g. Elect_use as a predictor for 100K_reduces).
The analysis sequentially assessed the unique impact of each independent variable on a given
dependent variable. If a predictor partially explained its behaviour, it was retained, while all other
variables were re-tested to identify whether they were still significant contributors. When a variable
no longer contributed significantly to a given model, it was removed. The final outcome of the
stepwise multiple regression is shown in Table 10.
- 13 -
Table 11: Estimated regression coefficients and related statistics
Dependent
variable
Independent
variables
Unstandarised
β
p-value
Adjusted
R
2
F
p-value
(model)
Elect_use
Living area
Household size
Income
2284
1051
1940
< .001
= .008
< .001
.176
F[3,230]= 17.0
< .001
ES_behaviour
PBC
Education
.87
.51
< .001
= .019
.074
F[2,225]= 9.9
< .001
100K_reduce
s
PBC
PN
.24
.37
= .001
< .001
.167
F[2,230]= 24.1
< .001
100K_action
PN
.11
< .001
.052
F[1,231]= 13.8
< .001
For the first model (Elect_use), three independent variables were statistically significant. Living area
explained 12% of the variance in electricity saving behaviour. A one unit increase of living area
increased annual consumption by an estimated 2 274 kWh.10 The introduction of Income increased
the explanatory value to 16%, while Household size added a marginal 1.7%. The three variables
together explained nearly 18% of variation in Elect_use. No multicollinearity was identified, and the
highest VIF was estimated to be 1.11. The result that socio-economic and contextual factors, rather
than behavioural and moral factors are determinants of consumption confirms findings from other
studies (see e.g. Abrahamse & Steg, 2009; Brandon & Lewis, 1999; Thøgersen & Grønhøj, 2010).
The second model (ES_behaviour) identified two independent variables that were statistically
significant. The two variables PBC (β = 0.87, p < .001) and Education (β = 0.51, p = .02) together
explained 7.4% of variance. The highest VIF was less than five (VIF = 1.11). This result is confirmed by
other studies that show PBC to be an important determinant (e.g. Abrahamse & Steg, 2009;
Botetzagias et al., 2014).
The addition of the remaining behavioural and moral variables did not add any explanatory power.
Although this is in line with, for example, Botetzagias et al. (2014) and Heath and Gifford (2011), it is
inconsistent with other studies that found that at least one moral factor had explanatory power
(Abrahamse & Steg, 2009; Harland et al., 1999). Abrahamse and Steg (2009) suggest that these
differences may be a result of how the situation is framed. If it is presented as a cost-benefit trade-
off, psychological variables are more influential, while studies that highlight environmental or moral
issues give greater weight to moral factors. Another explanation may be that moral factors are
similar to psychological factors (Botetzagias et al., 2014; Kaiser & Scheuthle, 2003).
It is perhaps more surprising that AT was non-significant, both in the light of current theory (Ajzen,
1991) and results from previous studies (Abrahamse & Steg, 2009; Botetzagias et al., 2014; Karlin et
al., 2014). However, Karlin et al. (2014) found that attitudes are primarily associated with
conservation and not with efficiency improvements. Our focus on energy efficiency behaviour may
explain why our results differed from those of other studies.
Education was a statistically significant predictor of behaviour; better-educated respondents
implemented more measures to save electricity. This result confirms previous work, which found that
higher levels of education correspond to a greater willingness to adopt energy efficiency measures
(Nair, Gustavsson, & Mahapatra, 2010; Poortinga, Steg, Vlek, & Wiersma, 2003). A point worth noting
is that the survey included questions related to global warming and electricity use. However, the
current electricity supply mix (mostly nuclear and hydropower) means that consumption makes a
10 Note that the unit of Living area is not square meters but the interval shown in Table 2.
- 14 -
marginal contribution to total greenhouse gas emissions. Therefore, it may not have been
particularly useful to ask about global warming, and instead focus on the environmental
consequences of electricity use (e.g. nuclear waste disposal, environmental impacts of large dams).
The third model explores the determinants of 100K_reduces. This examined the extent to which
users believed that the 100Koll service helped them to reduce consumption. The stepwise regression
found two statistically significant independent variables: PN and PBC. The former explained 13% of
the variance and the latter added 3.7% explanatory power. Together, the two variables PN (β = 0.37,
p < .001) and PBC (β = 0.24, p = .02) explained 16.7% of variance. No multicollinearity was identified
(VIF = 1.33). These results suggest that respondents who believed that they were able to influence
their electricity consumption (PBC) and felt better when saving energy (PN) also believed that 100Koll
helped them to reduce consumption.
Finally, the fourth model examined the determinants of 100K_action, defined as whether 100Koll
triggered energy efficiency actions (or not). Following the analysis, only one independent variable, PN
(β = .11, p < .001), was included in the model. This suggested that personal norms regarding the
obligation to reduce global warming marginally (5.2%) explained whether the 100 Koll service
contributed to decisions about efficiency measures. This result confirmed the finding that people
with high moral values (PN) tended to change their behaviour based on feedback from 100Koll.
4. Concluding remarks
The paper provides a better understanding of the effectiveness and determinants of consumption
based on real-time feedback in the Swedish residential sector. Using the 100Koll service as a case
study, the analysis examined various behavioural, moral, socio-economic and contextual aspects that
potentially frame or drive the effectiveness of real-time feedback and resulting electricity savings.
Results suggest that real-time feedback is ineffective if implemented in isolation. Estimated
electricity savings were relatively low (1.4–1.9%) and these findings were consistent with previous
work. Our results support the claim that responses to, and thus the effectiveness of, real-time
feedback are most apparent when the intervention begins, and decrease in the longer term.
However, unlike earlier work, our savings estimates were lower. Setting aside methodological and
contextual issues, this suggests that the user interface of the 100Koll service needs to be reviewed
and improved to maximise its effectiveness. This is consistent with earlier literature, which stresses
that the effectiveness of real-time feedback is closely related to how information is framed and
presented.
Inconsistencies between our results and earlier work highlight the importance of the characteristics
of the study’s context. These include different geographical boundaries, socio-economic conditions,
size of the intervention group, technological setup, information provided, reliability of the control
group, methods for and uncertainty of baselines, analytical time periods, and climatic conditions.
Discrepancies can be also explained by how real-time feedback is designed relative to the cognitive
abilities of users to understand and operationalise it. Another aspect concerns additional support
services (e.g. energy audits, suggested energy saving improvements, financing). Therefore, it is
unwise to rush to any hasty general conclusion, as outcomes and limitations are very likely to be
context- and study-specific.
From a behavioural and moral point of view, the findings suggest that socio-economic and contextual
factors provide the best explanation for consumption and marginal estimated savings when real-time
feedback is available. In line with the literature, variability in electricity use was relatively better
explained by variables such as living area, income and household size (in particular). Perceived
behavioural control and personal norms were significant predictors of the extent to which Swedish
households considered that real-time feedback would help them to reduce consumption. Moreover,
and despite the relatively low savings, the results indicate that households with greater perceived
behavioral control and moral responsibility were those that actually reduced their consumption.
- 15 -
However, the validity of our results is doubtful, given the limited number of questions making up the
construct. Further research is needed to establish to what extent behavioural and moral
determinants are influenced by real-time feedback.
We conclude that the implementation of SMs per se is likely to be insufficient to foster increased
efficient use of electricity. Our findings suggest the need to adopt and implement other policy
instruments, such as electricity pricing, awareness raising, expert advice and tailored education
campaigns. These complementary measures are particularly critical given the reduced response to
real-time feedback in the long term. Finally, our results confirm the value of the contribution of
behavioural economics as a support for the development of energy and environmental policies.
References
Abrahamse, W., & Steg, L. (2009). How do socio-demographic and psychological factors relate to
households’ direct and indirect energy use and savings? Journal of Economic Psychology,
30(5), 711–720.
Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision
Processes, 50(2), 179–211.
Allcott, H., & Mullainathan, S. (2010). Behavioral science and energy policy. Science, 327(5970),
1204–1205.
Bager, S., & Mundaca, L. (2015). How Smart are Electricity Users with “Smart Metering”? A
Behavioural Economics Experiment. Presented at the 38th IAEE International Conference,
Turkey: International Association for Energy Economics.
Black, J. S., Stern, P. C., & Elworth, J. T. (1985). Personal and contextual influences on househould
energy adaptations. Journal of Applied Psychology, 70(1), 3.
Blake, J. (1999). Overcoming the “value-action gap”in environmental policy: Tensions between
national policy and local experience. Local Environment, 4(3), 257–278.
Botetzagias, I., Malesios, C., & Poulou, D. (2014). Electricity curtailment behaviors in Greek
households: Different behaviors, different predictors. Energy Policy, 69, 415–424.
http://doi.org/10.1016/j.enpol.2014.03.005
Brandon, G., & Lewis, A. (1999). Reducing household energy consumption: a qualitative and
quantitative field study. Journal of Environmental Psychology, 19(1), 75–85.
Calvo-Mora, A., Leal, A., & Roldán, J. (2006). Using enablers of the EFQM model to manage
institutions of higher educationnull. Quality Assurance in Education, 14(2), 99–122.
Coleman, J., & Fararo, T. (1992). Rational Choice Theory. Newbury Park: SAGE Publications.
Cronbach, L. J. (1951). Coefficient alpha and the internal structure of tests. Psychometrika, 16(3),
297–334.
Darby, S. (2006). The effectiveness of feedback on energy consumption, a review for DEFRA of the
literature on metering, billing and direct displays. Oxford, University of oxford.
Dietz, T., Gardner, G., Gilligan, J., Stern, P., & Vandenbergh, M. (2009). Household actions can provide
a behavioral wedge to rapidly reduce US carbon emissions. Proceedings of the National
Academy of Sciences, 106(44), 18452–18456.
EC. (2011). Next steps for smart grids: Europe’s future electricity system will save money and energy -
Press Summary. Brussels. European Commission. Retrieved from http://www.e-
control.at/documents/20903/-/-/633895a3-d5d0-4866-865c-26b785bd1d0d
Ehrhardt-Martinez, K., Donelly, K. A., & Laitner, J. P. (2010). Advanced Metering Initiatives and
Residential Feedback Programs: A Meta-Review for Household Electricity-Saving
Opportunities (No. E105). Washington, D.C.: AmericanCouncilforan Energy-Efficient
Economy.
Fischer, C. (2008). Feedback on household electricity consumption: a tool for saving energy? Energy
Efficiency, 1(1), 79–104. http://doi.org/10.1007/s12053-008-9009-7
- 16 -
Gans, W., Alberini, A., & Longo, A. (2013a). Smart meter devices and the effect of feedback on
residential electricity consumption: Evidence from a natural experiment in Northern Ireland.
Energy Economics, 36, 729–743.
Gans, W., Alberini, A., & Longo, A. (2013b). Smart meter devices and the effect of feedback on
residential electricity consumption: Evidence from a natural experiment in Northern Ireland.
Energy Economics, 36, 729–743.
George, D., & Mallery, P. (2006). SPSS for Windows Step-by-Step: A Simple Guide and Reference, 14.0
update (7 edition). Boston: Allyn & Bacon.
Gifford, R. (2011). The dragons of inaction: pychological barriers that limit climate change mitigation
and adaptation. The American Psychologist, 66(4), 290–302.
http://doi.org/10.1037/a0023566
Gleerup, M., Larsen, A., Leth-Petersen, S., Togeby, M., & others. (2010). The Effect of Feedback by
Text Message(SMS) and Email on Household Electricity Consumption: Experimental Evidence.
Energy Journal, 31(3), 113–132.
Goldemberg, J., Johansson, T. B., Reddy, A., & Williams, R. (1988). Energy for a sustainable world.
New Delhi: Wiley Eastern Limited.
Goode, W. J. (1997). Rational choice theory. The American Sociologist, 28(2), 22–41.
http://doi.org/10.1007/s12108-997-1004-5
Grønhøj, A., & Thøgersen, J. (2011). Feedback on household electricity consumption: learning and
social influence processes. International Journal of Consumer Studies, 35(2), 138–145.
Harland, P., Staats, H., & Wilke, H. A. M. (1999). Explaining Proenvironmental Intention and Behavior
by Personal Norms and the Theory of Planned Behavior1. Journal of Applied Social
Psychology, 29(12), 2505–2528. http://doi.org/10.1111/j.1559-1816.1999.tb00123.x
IEA. (2014). Energy Efficiency Market Report 2014 (p. 224). Paris: OECD/IEA. Retrieved from
http://www.iea.org/publications/freepublications/publication/EEMR2014.pdf
Jaffe, A. B., & Stavins, R. N. (1994). The energy-efficiency gap: What does it mean? Energy Policy,
22(10), 804–810.
Kahneman, D. (2003). Maps of bounded rationality: Psychology for behavioral economics. The
American Economic Review, 93(5), 1449–1475.
Kahneman, D., Knetsch, J. L., & Thaler, R. H. (1991). Anomalies: The endowment effect, loss aversion,
and status quo bias. The Journal of Economic Perspectives, 5(1), 193–206.
Kahneman, D., & Tversky, A. (1984). Choices, values, and frames. American Psychologist, 39(4), 341.
Kaiser, F. G., & Scheuthle, H. (2003). Two challenges to a moral extension of the theory of planned
behavior: moral norms and just world beliefs in conservationism. Personality and Individual
Differences, 35(5), 1033–1048.
Karlin, B., Davis, N., Sanguinetti, A., Gamble, K., Kirkby, D., & Stokols, D. (2014). Dimensions of
Conservation Exploring Differences Among Energy Behaviors. Environment and Behavior,
46(4), 423–452. http://doi.org/10.1177/0013916512467532
Likert, R. (1932). A technique for the measurement of attitudes. Archives of Psychology. Retrieved
from http://psycnet.apa.org/psycinfo/1933-01885-001
Loewenstein, G., Read, D., & Baumeister, R. F. (2003). Time and Decision: Economic and Psychological
Perspectives of Intertemporal Choice: Economic and Psychological Perspectives of
Intertemporal Choice. Russell Sage Foundation. Retrieved from
http://www.google.com/books?hl=en&lr=&id=tAOGAwAAQBAJ&oi=fnd&pg=PR5&dq=the+ec
onomics+of+inmediate+gratification&ots=vNEC1yNN7F&sig=loN5A4L6ePljkvTzQOu0ytQHr34
Mason, C., & Perreault, W. (1991). Collinearity, Power, and Interpretation of Multiple Regression
Analysis. Journal of Marketing Research, 28(3), 268–280.
Matsukawa, I. (2004). The effects of information on residential demand for electricity. Energy
Journal, 25(1), 1–17.
McIver, J., & Carmines, E. G. (1981). Unidimensional scaling (Vol. 24). Sage.
- 17 -
McKerracher, C. ( 1 ), & Torriti, J. ( 2 ). (2013). Energy consumption feedback in perspective:
Integrating Australian data to meta-analyses on in-home displays. Energy Efficiency, 6(2),
387–405. http://doi.org/10.1007/s12053-012-9169-3
Mountain, D. (2006). The impact of real-time feedback on residential electricity consumption: The
Hydro One pilot. Mountain Economic Consulting and Associates Inc., Ontario, 10, 98–105.
Mundaca, L., Neij, L., Worrell, E., & McNeil, M. (2010). Evaluating energy efficiency policies with
energy-economy models. Paper Accepted in the Annual Review of Environment and
Resources, 35, 305–344.
Nair, G., Gustavsson, L., & Mahapatra, K. (2010). Factors influencing energy efficiency investments in
existing Swedish residential buildings. Energy Policy, 38(6), 2956–2963.
Nunnally, J. C., & Bernstein, I. H. (1994). Psychometric theory. New York : McGraw-Hill, cop. 1994.
Nuzzo, R. (2014). Scientific method: Statistical errors. Nature, 506(7487), 150–152.
http://doi.org/10.1038/506150a
Owens, S. (2000). “Engaging the public”: information and deliberation in environmental policy.
Environment and Planning A, 32(7), 1141–1148.
Owens, S., & Driffill, L. (2008). How to change attitudes and behaviours in the context of energy.
Energy Policy, 36(12), 4412–4418. http://doi.org/10.1016/j.enpol.2008.09.031
Poortinga, W., Steg, L., Vlek, C., & Wiersma, G. (2003). Household preferences for energy-saving
measures: A conjoint analysis. Journal of Economic Psychology, 24(1), 49–64.
Pyrko, J. (2009, December). El-info via digitala kanaler. Potential att förändra elanvändning i bostäder
Fallstudie 3 ”EnergiDialog-Privat” hos E.ON Sverige AB. Elforsk. Retrieved from
http://www.elforsk.se/Rapporter/?rid=09_93_
Raw, G., & Ross, D. (2011). Energy Demand Research Project: Final Analysis. AECOM. Retrieved from
https://www.ofgem.gov.uk/ofgem-publications/59105/energy-demand-research-project-
final-analysis.pdf
Raw, G., & Ross, D. (2012, June). Energy Demand Research Project: Final Analysis. AECOM. Retrieved
from https://www.ofgem.gov.uk/ofgem-publications/59105/energy-demand-research-
project-final-analysis.pdf
Sanstad, A. H., & Howarth, R. B. (1994). “Normal” markets, market imperfections and energy
efficiency. Energy Policy, 22(10), 811–819.
Schleich, J., Klobasa, M., Gölz, S., & Brunner, M. (2013). Effects of feedback on residential electricity
demand—Findings from a field trial in Austria. Energy Policy, 61, 1097–1106.
http://doi.org/10.1016/j.enpol.2013.05.012
Shogren, J. F., & Taylor, L. O. (2008). On Behavioral-Environmental Economics. Review of
Environmental Economics and Policy, 2(1), 26–44. http://doi.org/10.1093/reep/rem027
Simon, H. (1986). Rationality in psychology and economics. Journal of Business, S209–S224.
Stern, P. C. (2000). New environmental theories: toward a coherent theory of environmentally
significant behavior. Journal of Social Issues, 56(3), 407–424.
Sutherland, R. J. (1991). Market barriers to energy-efficiency investments. Energy Journal, 12(3), 15–
34.
Thaler, R. (1980). Toward a positive theory of consumer choice. Journal of Economic Behavior &
Organization, 1(1), 39–60. http://doi.org/10.1016/0167-2681(80)90051-7
Thøgersen, J., & Grønhøj, A. (2010). Electricity saving in households—A social cognitive approach.
Energy Policy, 38(12), 7732–7743.
Uggmark, M. (2013, May). Vetenskaplig utvärdering av Sveriges största energisparexperiment på
E.ON - Slutresultat. Retrieved from
http://www.ees.energy.lth.se/fileadmin/ees/Publikationer/Ex5278-MagdalenaUggmark-
EONFinal.pdf
Van Dam, S. S., Bakker, C. A., & Van Hal, J. D. M. (2010). Home energy monitors: impact over the
medium-term. Building Research & Information, 38(5), 458–469.
