See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/319597907

How smart do smart meters need to be?

Article in Building and Environment · September 2017

DOI: 10.1016/j.buildenv.2017.09.008

CITATIONS READS

3 63

13 authors, including:

Nataliya Mogles Jeehang Lee

University of Bath University of Bath
21 PUBLICATIONS 82 CITATIONS 13 PUBLICATIONS 38 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

University of Bath View project

Investigating the sources of cognitive activity in the brain View project

All content following this page was uploaded by Sylwia Hyniewska on 13 September 2017.

The user has requested enhancement of the downloaded file.

Accepted Manuscript

How smart do smart meters need to be?

Nataliya Mogles, Ian Walker, Alfonso P. Ramallo-González, JeeHang Lee, Sukumar

Natarajan, Julian Padget, Elizabeth Gabe-Thomas, Tom Lovett, Gang Ren, Sylwia
Hyniewska, Eamonn O'Neill, Rachid Hourizi, David Coley

PII: S0360-1323(17)30422-5
DOI: 10.1016/j.buildenv.2017.09.008
Reference: BAE 5087

To appear in: Building and Environment

Received Date: 8 July 2017

Revised Date: 30 August 2017
Accepted Date: 6 September 2017

Please cite this article as: Mogles N, Walker I, Ramallo-González AP, Lee J, Natarajan S, Padget J,
Gabe-Thomas E, Lovett T, Ren G, Hyniewska S, O'Neill E, Hourizi R, Coley D, How smart do smart
meters need to be?, Building and Environment (2017), doi: 10.1016/j.buildenv.2017.09.008.

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP
C
AC
 ACCEPTED MANUSCRIPT

How smart do smart meters need to be?

Nataliya Mogles1, Ian Walker2, Alfonso P. Ramallo-González3,4, JeeHang Lee1,

Sukumar Natarajan4, Julian Padget1, Elizabeth Gabe-Thomas2, Tom Lovett1, Gang

Ren1, Sylwia Hyniewska1, Eamonn O'Neill1, Rachid Hourizi1, David Coley4,*

PT
1
Department of Computer Science, University of Bath, UK.

RI
2
Department of Psychology, University of Bath, UK,

SC
3
Faculty of Computer Science, Universidad de Murcia, Spain.
4
Department of Architecture and Civil Engineering, University of Bath, UK,

U
d.a.coley@bath.ac.uk
AN
*
Corresponding author.
M

Graphical Abstract
D
TE
C EP
AC

1
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP

Abstract
C

Governments across the world are investing in smart metering devices that report
AC

energy use to the user with the aim of reducing consumption. However, the

effectiveness of such In-Home Displays (IHDs) has been questioned, since savings are

small. This is possibly because informing the consumer of their consumption in kWh,

or monetary units, fails to give context, or inform of possible actions to reduce

consumption. We investigate, for the first time, the effect of replacing the simple

statement of energy use an IHD gives, with a detailed array of information specifically

2
 ACCEPTED MANUSCRIPT

designed to improve consumer energy literacy and suggest behaviour change through

personalised actionable messages set against a series of psychological value systems

for context, and which can identify potential profligacy. The results from a carefully

controlled field experiment show: 1) value framing and action prompts have a

PT
significant effect on occupants' behaviour, with the mean temperature of homes being

reduced from 22.4°C to 21.7 °C, and a marked reduction in gas consumption—22.0%

RI
overall and 27.2% in high consumers; 2) energy literacy increasing from 0.52 to 1.28

SC
(on a 0-4 scale); 3) it is possible to target potentially profligate households, without

inappropriately messaging others; 4) engagement is high, with 82% of the participants

U
finding the system useful. These results emphasize the necessity of improving energy
AN
literacy when encouraging energy efficient behaviours and point to a new generation

of smart meters with the potential to increase energy literacy, make much greater
M

savings and impact climate change policy.

D

Keywords: energy feedback, In-Home-Displays (IHDs), internal values, action

TE

prompts, energy literacy, smart meter.

C EP
AC

3
 ACCEPTED MANUSCRIPT

1 Introduction
The residential energy sector accounts for 23% of total energy consumption

worldwide, placing it third after industry at 37% and transportation at 28% [1]. In

developed countries the sector is even more important; in the US, for example,

PT
residential consumption represents 25% of total energy use [2] and in the UK 29% [3].

This translates to roughly 12% of UK greenhouse gas emissions [4]. In addition, it is

RI
projected that residential consumption worldwide will grow by an average of 1.4% a

year from 2012 to 2040 [5]. Thus the sector is critical to both national energy policy

SC
and international climate change policy, and many attempts are being made to reduce

U
residential consumption by influencing the behaviour of occupants.
AN
Residential energy consumption is however a multidimensional phenomenon

embedded within a socio-cultural and infrastructure context, and for this reason
M

changing occupant behaviour might be expected to be complex. The focus of the

D

research reported here lies in inducing behavioural change with the help of energy

demand feedback via smart meters or ambient displays. Opinions regarding the
TE

effectiveness of such solutions to date are unfortunately not unanimous [6, 7, 8, 9, 10,
EP

11, 12].

Approaches to energy feedback have so far tended to be one-size-fits-all

C

solutions; however, with new developments in energy data management [13], more
AC

advanced feedback is now possible. The main contribution of the current paper to

digital energy feedback research is twofold: i) in broad terms, it tests the effectiveness

of a novel smarter, building-aware, and more user-personalised digital energy feedback

in an experimental setting for the first time; ii) in more specific terms, it evaluates the

effect of two new approaches to feedback – internal values and tailored action

4
 ACCEPTED MANUSCRIPT

prompts, delivered via a computer tablet. It replaces the normal user- and building-

blind smart meter concept with one that not only reports the energy used but

recommends specific actions and works with the personal values of the user. For

example, it might recommend turning the heating thermostat down one degree-

PT
centigrade, and explain this not just in terms of kWh or financial savings, but with

respect to environmental gain or other personal values. The system tested here is a

RI
simplified version of a future building- and user-conscious approach that would report

SC
complex and tailored information to the user in written or spoken sentences. For

example, “I note that the heating turns off at 9am each weekday, yet your home

U
appears to be unoccupied from 7am; would you like me to change the heating timing?
AN
This might save you £47 per year and the stop the emission of 218 kg of carbon

dioxide”. Or, “I note that your home might be over-ventilated when the heating is on,
M

losing £98 of heat per year. You might like to keep most of your windows closed when

the heating is on”. Such an approach, which we term intelligent smart metering (or
D

ISM), requires knowledge of which values are most likely to prompt the user to act.
TE

These might be financial savings, the reduction of environmental damage, benefits to

future generations, wastefulness, or some other concern specific to the individual.

EP

ISM requires an energy (thermal) model of the building in order to predict

C

accurate savings based on the building and its use, rather than on inaccurate typical
AC

values. To be cost effective, such a model would need to be automatically assembled

from a minimum of sensor information so as not to overly increase the cost of the

smart meter, for example a mix of utility meter data, room temperature, sub-circuit or

high-frequency electricity data to infer occupation and maybe home CO2 concentration

(as a proxy for ventilation rate). The sensors needed and the accuracy of such an

5
 ACCEPTED MANUSCRIPT

approach, which uses inverse modelling to obtain an accurate thermal model from a

time series of data, have been reported by the authors in [14] and [15]. By creating

such a model, the financial, and other, impact of any suggestions, for example turning

down the thermostat by one degree, can be calculated for the specific home and

PT
reported to the user. In addition, inappropriate suggestions, for example suggesting a

reduction in heated temperature in an already under-heated home, can be avoided.

RI
One can imagine other advantages of the ISM approach, for example, the

SC
reporting of the presence of high U-Values in the fabric to utility companies for

targeted intervention.

U
This paper attempts to discover if this new approach can be applied to a group of
AN
homes and if it generates changes in behaviour.
M

2 Background
Some research findings [16] suggest that continuous energy feedback might be
D

an effective driver of energy-related behaviour change. For example, Barbu et al. [17]
TE

suggest that energy feedback provided to users via smart meters could save 5-15% of

total energy costs. Similarly, [18, 19, 20] suggest that energy feedback through
EP

advanced in-home displays (IHDs) could help to save up to 20% of energy costs, either

for electricity or total energy bills. However, reality seems to fall short of such
C

predictions. For example, in [21] a more modest average energy saving of 7% is

AC

reported across multiple utility pilot programs aimed at electrical energy conservation

with the help of IHDs.

Current technological solutions for real time energy feedback suffer from

multiple issues [22, 23], for example: unengaged users; failure to address users’

personal motivations and needs embedded in daily routines and social practices;
6
 ACCEPTED MANUSCRIPT

information comprehension issues caused by abstract numerical information in kWh or

financial costs; and inattention to users’ personal characteristics [24]. It seems clear

that users need something more to motivate and engage them than plain energy

feedback in kWh or cost if we are to get energy reductions of 10% or more. Some

PT
research indicates that intelligent energy feedback that offers different feedback

options might be effective [25].

RI
Chiang et al. [26] have calculated that smart meters can pay back their

SC
installation costs in 4 years if the energy savings are 3% or more, although the

observation that the utility company no longer need to visit the home to read the meter

U
might currently be the main economic driving force for their installation. It is hence
AN
interesting to ask whether we might move the focus from a device that is mainly of use

to the energy company to one that has equal utility to the occupant.
M

The question of what best motivates householders has received considerable

D

attention from energy behaviour researchers, with varying success being noted. For
TE

example, attempts to stimulate savings through social comparison and competition

have not achieved notable success [25, 27, 28, 29]. One possible reason for this is the
EP

so-called boomerang effect [30], whereby when households are told they are using less

energy than average they start to use more when they see what is ‘normal’ or
C

‘permissible’. This suggests that if messages are to be sent to households, only those
AC

where there is some evidence of potential energy profligacy should be targeted with

certain messages, and this is a key aspect of our study, and the one reason for its

success.

The other key aspect is people’s values. Different disciplines, including

economics, psychology, philosophy, sociology and anthropology, have all tried to

7
 ACCEPTED MANUSCRIPT

understand the role of personal values in conservation behaviour. From an economic

point of view, people’s values correspond to long-term preferences, and can be

explained by decision theory [31]. In anthropology, values are ‘cultural worldviews’

and their role in pro-environmental behaviour is studied within climate change risk

PT
perception [32]. In social psychology, Schwartz [33] identified a number of personal

values universal for all cultures and nations. From this perspective, personal values are

RI
defined as superordinate goals that serve as enduring guiding principles in peoples’

SC
lives [33]. Common to all these approaches is the idea that values are conceptually

different from goals, opinions and attitudes: values reflect broad long-term preferences

U
and so provide unity across a broad range of behaviours (a person who above all
AN
values their own self-interest will be self-interested in most settings); goals, opinions

and attitudes, on the other hand, are much more situation-specific and changeable over
M

time [34]. In the current study, we test personal values as motivators to energy-saving

behaviour and focus on altruistic, egoistic and biospheric values, representing,

D

respectively, concern for others, the self and the natural environment. These values
TE

were chosen from the various Schwartz’s personal value sets based on research by

DeGroot and Steg [35, 36], who identified altruistic, egoistic, biospheric and hedonic
EP

(pleasure-seeking) values as the key value orientations highly correlated with pro-
C

environmental behaviour. In this study, hedonic values were dropped from this set
AC

after pilot testing (with an opportunistic sample of 30 UK adults) found much lower

engagement with such messages compared to the other three value orientations (which

pilot participants found useful, and more engaging than energy messages expressed in

standard kWh).

The final facet of this study concerned the nature of the action prompts. The

8
 ACCEPTED MANUSCRIPT

literature hints that energy feedback may be effective when it prompts concrete

personalised actions (for example, “Please switch off unused appliances”, “Adjust your

thermostat setting”), presumably because these do not require knowledge or problem-

solving from the householder, as would a more general action prompt like “Reduce

PT
your energy consumption”. However, at the moment we lack the empirical quantitative

data needed to substantiate these claims [24]. Research on the effect of general action

RI
prompts has shown that it has limited influence on behaviour, but with increased

SC
possibilities for specificity and context detection offered by modern technologies it

might substantially improve the effect of feedback [37]. Some findings also indicate

U
that for the energy information to be useful, it should be tailored to personal contexts
AN
[38]. Thus, specific action prompts, tailored to specific households, might have a more

profound effect on energy related behaviour, and this idea is tested here.
M

Providing information on energy related actions and personal context is directly

D

related to the educational component of energy feedback and the concept of energy
TE

literacy—which implies an in-depth understanding of energy consumption [39] and is

frequently mentioned in pedagogical, educational and environmental policy literature.

EP

Energy literacy has been called “a broad term encompassing content knowledge as

well as a citizenship understanding of energy that includes affective and behavioural

C

aspects” [40], although arguably such a definition conflates knowledge about where
AC

energy is consumed with the motivation to reduce this. Cognitive (knowledge, in-depth

understanding), affective (attitudes, values), and behavioural (social practices)

modules can be distinguished within a complex concept of energy literacy. It is known

that promoting energy literacy can foster a shift in knowledge and perception of energy

and thus will facilitate responsible energy related choices and behaviour [40, 41, 42].

9
 ACCEPTED MANUSCRIPT

To the authors’ knowledge, there are no studies or commercially available

energy feedback systems that use action prompts tailored to users’ behaviour in a

systematic manner. The only work that provides some tailored heating action prompts

is our previous study [43], however we did not aim to test different energy feedback

PT
approaches in an experimental setting. In the work of D’Oca et al. [44] tailored to

users’ contexts, newsletters with action suggestions were sent via email to participants,

RI
though the study only addressed electricity consumption, and action prompts were not

SC
part of an energy feedback display. One work on energy consumption feedback in a

non-residential sector reports the effectiveness of user-tailored context advice and

U
action prompts [45], though this study again was focused only on electricity
AN
consumption and the tailored advice was sent by email not a smart meter.

There are a lot of commercially available devices and applications that provide
M

people with simply energy feedback and give general, not occupant- or building-
D

specific tips and suggestions regarding energy saving issues. In other words, the
TE

existing IHDs are not smart enough to address building context or personal

motivations.
EP

3 Pilot study and research hypotheses

Very few studies test the effect of internal personal values embedded in a digital
C

energy feedback within field experiments. One exception being the previously
AC

mentioned work of Schultz et al [29] which compared feedback based on normative

comparison and financial costs, against a standard feedback showing consumption

alone. We hypothesised that couching energy feedback messages in terms of several of

the personal values cited in the environmental psychology literature would

substantially improve users’ engagement and increase their motivation to save energy

10
 ACCEPTED MANUSCRIPT

since multiple values guide people’s lives [34] and focussing only on one value might

be not effective to increase motivations.

In addition to this, we have included the facet of personalised feedback i.e.

messages that refer, and are valid only for, specific households.

PT
The main research questions that we wanted to answer were:

RI
1. What is the overall effect of digital energy feedback interventions?

SC
2. What is the additional benefit of internal values embedded in energy feedback

messages?

U
3. What is the additional effect of personalised messages with action prompts?
AN
4. What effect has a digital energy feedback on cognitive variables, such as energy

literacy and users’ experience with a digital feedback?

M
D

To answer these questions, we formulated the following hypotheses:

TE

• Hypothesis 1: Digital feedback in general will have an effect on energy related

EP

behaviour compared to a baseline period before the interventions.

• Hypothesis 2: Energy feedback translated according to internal values will have

C

a significant effect on energy related behaviour.

AC

• Hypothesis 3: Personalised messages with action prompts will have a significant

effect on energy related behaviour.

• Hypothesis 4: No matter which behavioural effect of a digital energy feedback is

observed, the digital interventions will influence energy related cognitive

11
 ACCEPTED MANUSCRIPT

variables, such as energy literacy.

4 Method
To answer the research questions defined in Section 3, we designed an energy

feedback system with different types of feedback, we deployed the system in homes in

PT
the UK and performed a field experiment. The experiment had a 2 (Value framed vs.

Non-value framed) by 2 (No action prompt vs. Tailored message with action prompt)

RI
within-subject factorial design resulting in four energy feedback experimental

SC
conditions (see Table 1).

4.1

U
Participants and procedure AN
Our participants were residents of social housing recruited in the first half of

2013 with the help of Exeter City Council. Initially, an information package was sent
M

to the households and later recruiters went door-to-door to identify interested

households. After a consent form was signed by the occupants, the sensors were
D

installed during January-May 2014 in 73 homes. The first phase of the project
TE

established the baseline: sensor data were collected but no energy feedback was

provided to the participants.

EP

For the second phase of the project, an energy feedback experiment was
C

conducted over three months (January 2016 - March 2016) with the help of a new
AC

energy feedback application (iBert) written by the authors and presented to participants

via a tablet computer. The current paper describes the analysis of this second phase of

the project. Due to the length of the study, and the social group involved, some

households dropped out of the study, moved houses or changed contact details, which

resulted in 43 homes participating for the full study period.

12
 ACCEPTED MANUSCRIPT

In this sample, each household had on average 2.8 residents. Each home had a

main contact, and the mean age of these was 50.6 years (with a minimum age of 27

and a maximum of 81), 14 were male and 29 female. Seventy-five percent of

households reported an income lower than the UK national median of £25,600

PT
(identified at a national UK level for 2015, with the London area excluded [46]).

Around 25% of households reported difficulties with paying energy bills. Fourteen

RI
percent of households already had a basic smart meter (could correspond to condition

SC
1 in our study).

The focus of the study was in part the response of occupants to feedback from

U
personalised messages. For this reason, individual basic thermal models of each of the
AN
homes were created. It had already been demonstrated by the authors that the sensors

used can auto-generate a suitable dynamic thermal model of each building based only
M

on the data received from the sensors (without the need for information on the size of
D

the homes or their constructions). This model is a lumped parameter model and the
TE

work has been reported in another paper [15] and shown to produce accurate results.

However, to make sure that no extra noise was added to the experiment and ensure the
EP

results reported were not affected by an auto-generated model which might well

change during the experimental cycle, simple (steady-state) models were created by
C

using floor plans and information about the materials and constructions from the city
AC

council and visual inspection. These models consisted of calculating the mean heat

loss coefficient of each building. In future developments auto-generating models like

those of [15] could easily be used.

The simple model obtained from the building geometry was used by iBert to

investigate the energy savings of various strategies (turning the thermostat down,

13
 ACCEPTED MANUSCRIPT

closing windows, turning the heating off when the building was not occupied, turning

the lights and some other electrical items off when the building was not occupied) that

could be applied to each house if the household breached set action levels of: room

temperature, ventilation rate, having the heating on when building was unoccupied;

PT
and high electricity use when the building was unoccupied. Depending on the

condition applied to the home at the time (see Table 1 and 2), this gave iBert the ability

RI
to inform households of: their current energy consumption; the total consumption over

SC
the last week; their consumption expressed in accordance to other values (such as an

environmental cost); the savings that might occur if they took action; and, most

U
importantly, the specific actions they might take.
AN
Due to page limitations, full details of the electronics, sensors and coding that

lie behind iBert will be described in a later paper. But in outline, the system consisted
M

of sensors that measured real-time utility data, air temperature (in up to three rooms),

radiator temperature (to sense if heating was on), humidity (in up to three rooms),
D

light, motion and CO2 levels, with this data collected approximately every 5 minutes
TE

via Raspberry Pi and transmitted via the home’s broadband or 3G network to a secure

cloud store. These data were then used to see if action levels had been triggered,
EP

calculate any savings that might be made if behaviour were to change in the future, and
C

report back to the occupant via the same connection to the Wi-Fi enabled tablet in the
AC

home as a weekly digest. The frequency with which householders interacted with the

iBert app was recorded directly from each tablet and as well as a record of the tailored

textual messages sent to each house.

Table 1: Energy feedback study design (examples of information presented within

each condition in italics – other non-monetary values, for example environmental
14
 ACCEPTED MANUSCRIPT

impact, were also used.

No Value Value:
No Action C1: Standard display C2: Standard display + values
No personalised (kWh and financial costs)
Your home has used 95 kWh
Your home has used 95 kWh this week; this is equivalent to
this week. £12.

PT
Action C3: Standard display + C4: Standard display +
Personalised + tailored action prompts + tailored action prompts +
values
If you reduced the

RI
thermostat temperature in If you reduced the thermostat
your house one degree you temperature in your house one
would save 11 kWh. degree you would save 11

SC
kWh; this is equivalent to
£1.43.

U
AN
The households were cycled through the four conditions shown in Table 1 with

each condition lasting three weeks. Households were randomly assigned to one of the
M

four conditions according to the Latin-Square counterbalancing design of Table 2. For

example, during the first three weeks, one-quarter of the homes were under condition
D

C1 of Table 1 and in the following three weeks they had condition C2 applied to them.
TE

In conditions C3 and C4, tailored energy feedback messages are generated once
EP

per week, based on energy-related information aggregated over the previous week, and

were displayed during the subsequent week. If the algorithm for identifying energy-
C

wasting events and variables did not detect any potentially profligate energy use in a
AC

particular household, that household did not receive any tailored messages that week.

Table 2: Latin square counterbalancing design for the four experimental conditions: C1
to C4 across the four temporal phases of the experiment.
Baseline Phase 1 Phase 2 Phase 3 Phase 4 After
14/12/15 04/01/16 25/01/16 15/02/16 07/03/16 28/03/16 18/04/16

15
 ACCEPTED MANUSCRIPT

C1 C2 C3 C4
C4 C1 C2 C3
Before iBert After iBert
C3 C4 C1 C2
C2 C3 C4 C1

PT
This results is sixteen textual messages in total: four for personal value neutral

condition C3 and twelve for value-framed condition C4 (egoistic, altruistic and

RI
biospheric values). A household could receive up to four different messages since a

SC
value message is framed according to one of the three personal values and a value is

selected at random. Messages in conditions C3 and C4 are sent only if there are

U
energy-wasting events identified by the intelligent algorithms.
AN
Participants received an incentive of £150 in the form of supermarket vouchers,
M

sent by post in three instalments of £50 (at the beginning of the project, at the end of

year one and at the end of year two after the completion of the project). In addition,
D

they were allowed to keep the tablet once the study has ended. All participants who
TE

completed a post-study phone survey received an additional incentive with a chance to

win a £20 supermarket voucher during a prize draw. After the end of the experiment
EP

the energy feedback app was deactivated and sensor equipment was collected from the
C

participating homes four months after the end of the energy feedback interventions.
AC

Prior to both phases of the study participants were given paper information

sheets, consent forms and surveys that contained among others demographics and the

first seven items from the energy literacy survey described in [40]. This is an updated

version of a validated questionnaire and energy literacy assessment framework

developed in the by DeWaters and Powers [40]. To the authors’ knowledge, it is the

only validated questionnaire that evaluates energy literacy according to a broader

16
 ACCEPTED MANUSCRIPT

definition. The seven items used from this questionnaire addressed only the general

knowledge of energy consumption. (The full energy literacy and user experience

survey used for this work can be found in Appendix A).

As previously mentioned the interactions with iBert were also monitored—

PT
Figure 1. The reader can see that almost all houses had substantial level of interactions

with the tablet. Each interaction with the app consisted of a minimum of three recorded

RI
app related events, such as ‘app started’, ‘app resumed’, ‘app closed’ etc.

U SC
AN
M
D
TE
C EP
AC

Figure 1: Interactions heat map—the colour indicates the number of interactions. The

House ID is the ID of the house within the database.

4.2 The four experimental conditions

C1: Standard energy feedback. For the standard display condition (C1: No

value – No action; see Table 1), we followed the recommendations of the minimal

17
 ACCEPTED MANUSCRIPT

requirements for an In-Home Display (IHD), i.e. we replicated a typical smart meter in

the UK [47]. This displayed: current gas and electricity consumption; cumulative

consumption for the past week; information in kWh; and a visual (i.e. non-numerical)

presentation alongside a numeric one.

PT
C2: Standard feedback with personal values. In a personal value only

RI
condition (C2: Value - No action) the display was the same as in C1, but

SC
complemented the weekly summary information translated into one of the three

internal values: egoistic – expressed in money; biospheric – expressed in trees

U
destroyed; or altruistic – expressed as the cost to society, couched as the number of
AN
minutes with a family doctor that could be provided at the same cost. So, for example,

a household might have the kWh display replaced with information that their energy
M

use in the last week was equivalent in carbon terms as the destruction of 1.7 trees or 20
D

minutes with a doctor. Egoistic, biospheric or altruistic personal values for each
TE

household were chosen randomly.

C3: Standard feedback with tailored actions. This condition was the same
EP

as C1 except households also saw a tailored message box with action prompts in text

form about any potentially energy profligate behaviour (room temperatures > 21°C,
C

occupied CO2 levels of <800 ppm (possibly indicating excessive ventilation), mean
AC

unoccupied electricity consumption in excess of 600W, or heating on when the house

was unoccupied) during the previous week. Action prompts described concrete actions

with low personal costs, e.g. lowering a thermostat setting or leaving windows open

less frequently. A participant could receive from zero to a maximum of four messages.

Messages consisted of four parts: (1) communication of factual information regarding

18
 ACCEPTED MANUSCRIPT

energy related variables [for example, we have noticed that your thermostat was set to

23°C]; (2) evaluation of the factual information [e.g. this value is unnecessarily high];

(3) a motivational component which expressed wasted energy information aggregated

over a longer time period (over a whole winter, over a year) [e.g. in your house this

PT
may imply 520kWh use more per winter]; and (4) an advice part which contained an

action prompt [e.g. we recommend that you lower the thermostat to 21 degrees

RI
centigrade and that you check if you are comfortable at that temperature].

SC
The trigger of 21°C was based on national recommendations [48]. The 800 ppm

was chosen by considering EN 15251. This suggests IDA 2 air quality (medium IAQ)

U
is achieved below 1000 ppm. In an effort to be precautionary, a lower value of 800
AN
ppm was used - this was the median value found during the baseline phase. The 600 W

was also derived from data gathered during the baseline phase. This showed that in all
M

houses consumption was bi-modal: low values during unoccupied periods or the night;
D

and high values during occupied periods. The mean figure that separated the two being
TE

600W.

C4: Standard feedback with tailored actions and values. In experimental

EP

condition C4, households received the same information as in C3, but the kWh part of

the message was translated into one of the three internal personal values just like in
C

C2. An example of an energy feedback message in condition C4 where energy

AC

information is translated into biospheric value concepts might be: “I think a lot of heat

might be escaping from open windows. The escaping heat results in wasted energy.

This requires roughly 12 more trees to compensate for the extra pollution caused by

your home over a winter. This may be because there are too many windows open, they

are open too wide or for too long. Try changing how many windows you open and for

19
 ACCEPTED MANUSCRIPT

how long and you may save energy that over a whole winter is equivalent to planting

12 trees. Try it, and check if this message disappears next week”.

Appendix B shows a screen shot of ibert as seen by a household.

PT
5 Results
In this section, we show the results from the different analyses that

RI
were performed to evaluate the effectiveness of the iBert system.

Analysis 1: Overall effect of iBert

SC
This analysis tests hypothesis 1, asking if digital feedback in general has an

U
effect on energy-related behaviour compared to a baseline period before the
AN
interventions. The purpose of this is to examine if our findings are in line with those

from other EHD trials. A repeated-measures ANOVA was used to evaluate the effect
M

of iBert on internal temperature, electricity consumption and ventilation rate (CO2

levels).
D

We found a significant effect of Time (i.e. study phase) on home internal

TE

temperature, F(2,58) = 3.78, p = 0.029, ηp2 = 0.12. Post-hoc pairwise comparisons

EP

using Holm-Bonferroni p value correction reveal that home internal temperature before

iBert was active is significantly higher than the internal temperature during the iBert
C

experimental phase (d = 0.24, p < 0.001) and the difference between the internal
AC

temperature during the experiment and after the experiment is not significant either (p

= 0.97)—indicating the intervention had lasting impact (Table 3). We understand that

lowering the temperature can lead to larger issues such as comfort problems or even

health risks in vulnerable households, it is for this reason that this analysis is expanded

later to check if this temperature reduction was from all households, or only from ones

20
 ACCEPTED MANUSCRIPT

with initially high temperatures..

Table 3: Mean internal temperature (in ◦C) for households before, during and

after the digital iBert energy feedback interventions.

◦ ◦
Temperature ( C) Standard deviation ( C)

PT
Baseline 20.4 2.07
IBert 19.6 2.50

RI
Phase
After 19.4 2.35

U SC
In addition, we conducted two analyses of covariance to see how often occupants
AN
interacted with iBert and if occupants receiving tailored textual feedback influenced

these results. Frequency of interaction with iBert proved not to be statistically

M

significant. However, change in internal temperature was predicted by whether

householders had received at least one tailored textual feedback message, F(2,56) =
D

3.45, p = 0.04, ηp2 = 0.11. This interaction shows that the difference in internal
TE

temperature between homes with and without a tailored textual feedback is not visible

during the digital interventions phase, but it is after the interventions. In the light of
EP

these results, we conducted a follow up t-test on the differences in internal temperature

C

between with-iBert and after-iBert phases for two groups of households: those who did
AC

not receive any temperature-related tailored textual message and those who did.

This difference is significant, t(29) = 2.29, p = 0.03, which indicates that homes

that received tailored textual messages with action prompts had lowered their internal

temperature after the digital interventions; while homes which did not receive

messages, did not decrease their internal temperature.

21
 ACCEPTED MANUSCRIPT

The overall effect of iBert on electricity consumption and CO2 concentration,

with frequency of app usage and received textual messages as covariates, would

appear not to be statistically significant.

PT
Analysis 2: Post-intervention effect of messages

Temperature-related action prompts were triggered in homes over 21◦C. To

RI
assess whether people responded to these, we conducted four follow-up one-sample t-

SC
tests to compare internal temperatures to 21◦C, before and after the intervention, for the

group who received temperature-related action prompts and the group who did not. For

U
the non-message group (the subgroup whose home temperature was lower than 21◦C
AN
and who therefore were never told to change the temperature even under experimental

conditions C3 and C4), internal temperature was significantly lower than 21◦C both
M

before (mean temperature = 18.7◦C, SD = 1.6, t(24) = -6.99, p < 0.001) and after the
D

experiment (mean temperature = 18.7, SD = 1.9, t(16) = -5.09, p < 0.001) – as would
TE

be expected. The group that received the temperature message because their baseline

temperature was over 21◦C (mean temperature = 22.3◦C, SD = 1.1, t(14) = 4.38, p
EP

<0.001), however, had internal temperatures that essentially settled at 21◦C after the

messages (mean temperature = 21.1◦C, SD = 2.00, t(11) = 0.20, p = 0.80), Figure 3. In

C

summary, people who received temperature-related action prompts saw a reduction in

AC

internal temperatures that put their home temperatures almost exactly at the level that

was suggested; a similar change was not seen in those who received no prompts,

suggesting the effect is likely not the result of weather-related changes. Note that both

groups could still see the standard IHD module of the app after the experiment,

although the tailored messages function was deactivated. No messages were triggered

22
 ACCEPTED MANUSCRIPT

based directly on gas consumption (but rather on internal temperature, heating times

and excessive ventilation), however the use of iBert reduced the mean household gas

consumption by 14 to 29% with a mean of 22.0%.

PT
Analysis 3: Effect of values and action prompts

RI
To evaluate our second and third hypotheses, regarding the effect of internal

values framing and tailored action prompts during the energy feedback iBert phase, we

SC
conducted a two-way repeated measures ANOVA on home internal temperature,

electricity consumption and CO2 level data with the factors Personal Values (present or

U
absent) and Action Prompts (present or absent). Baseline measurements of these
AN
dependent variables, taken before iBert deployment, were used as covariates. After

controlling for the baseline temperature, homes in the value framing conditions (C3
M

and C4) have a lower internal temperature (19.23◦C) compared to homes in the
D

conditions without value framing (C1 and C2) (19.32◦C).

TE

An ANOVA on these internal temperature readings during the iBert phase, with

baseline temperature as a covariate, revealed a significant effect for Value Framing,

EP

F(1,33) = 5.29, p = 0.028, ηp2 = 0.14. The baseline temperature covariate was also
C

significant, F(1,33) = 4.94, p = 0.033, ηp2= 0.13, suggesting the warmer homes at the
AC

start of the study still tended be the warmer homes at the end.

As shown later, iBert had a marked impact on energy literacy. Hence it is

unlikely that action prompts were forgotten between experimental phases. In light of

this, the data were separated into two groups, before the issuing of action prompts and

post issuing. Action prompts were found to have a statistically significant impact under

a pairwise t-test (p = 0.021) on the heating degree day adjusted mean gas consumption
23
 ACCEPTED MANUSCRIPT

(mean before action prompts = 1239W, SD = 379; after 1120W, SD = 348),

contributing 9.6 percentage points to the overall reduction in gas consumption.

In contrast to internal temperature and gas consumption, there were no effects of

Actions or Values on electricity consumption and CO2 concentration.

PT
RI
Analysis 4: Case studies of subgroups with high and low energy

SC
consumptions

To get more insight into the changes in internal temperature, gas and electricity

U
consumption, and CO2 concentration, across all the stages of the study, we examined
AN
the box plots of the data for two subgroups: one being the households that had high

energy use and the other those homes with low energy use.
M

Figure 2 shows that the median temperature of the houses that had a high
D

thermostat temperature (upper graph) went down during the iBert interventions and
TE

stayed low after the study.

C EP
AC

24
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M

Figure 2: Home internal temperatures (in °C) before, across the iBert digital

intervention phase and after the feedback. The graph has been separated as houses
D

likely to receive a personalised message (upper graph) and those that did not (lower).
TE

The blue box represents 50% of the data, the median is represented as a horizontal
EP

(red) line within the box and the blue whiskers represent the 97.5% of the data. The

thick green line is at 21°C—the trigger for receiving a message. The individual dots,
C

and dashed lines, represent individual homes.

AC

The temperatures before the intervention were shown to be statistically different

to 21ºC in the group with temperatures higher than 21ºC in the baseline: with a p-value

rejecting the Null hypothesis (they are equal) of p=0.0039 (<0.05); whereas the

temperatures before and after in the group in which the temperatures were already

25
 ACCEPTED MANUSCRIPT

below 21ºC cannot be considered different (p-value of 0.46). Moreover, a t-test was

used to prove if the hypothesis that the temperatures in the baseline in the houses in

which the starting temperature was larger than 21ºC was statistically different to 21ºC

after the study. The test proved that the temperatures of the group which behaviour

PT
needed to be modified were not statistically different to 21ºC after the study with a p-

value of 0.17 (the null hypothesis mean = 21ºC, can not be rejected) meaning that the

RI
behaviour of the occupants had been modified.

SC
A similar evaluation was done with the CO2 concentration. Houses in which the

CO2 levels were less than 800 ppm were considered as potentially over ventilated, and

U
therefore it was assumed that some energy savings could be achieved by reducing
AN
ventilation.

To evaluate this, we again separated the homes into two groups: the ones that
M

showed high levels of ventilation in the baseline (CO2<800ppm) and those that did not
D

(Figure 3). In this case, all the tests show that the CO2 concentrations were not
TE

changed in the intervention; neither in the homes that showed high levels of CO2 nor in

those that showed low levels. This is possibly because changing the ventilation regime
EP

in a house is not a trivial task. Also, it should be noted that CO2 is not only an indicator

of ventilation, but also an indicator of infiltration occurring through cracks and other
C

defects, and that is not under the control of the occupants.

AC

26
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
Figure 3: Home CO2 concentrations before, across the iBert digital intervention phase
M

and after the feedback. The graph has been separated into houses likely to receive a
D

personalised message (upper graph) and those that were not (lower). The thick green
TE

line is at 800ppm—the trigger for receiving a message.

EP

With respect to electricity, it also seems that the intervention failed to change
C

the behavioural habits of the occupants, with no statistically significant change seen in
AC

the behaviour of big consumers shown in the top part of Figure 4.

The gas consumption however shows a clear indicator of the effect of iBert.

Although iBert contained no direct messaging with respect to gas consumption, three

statements within iBert were designed to reduce gas consumption: change in the

thermostat temperature, change in the operating scheme (turning heating off when

house unoccupied) and change in the ventilation regime. As there was no direct

27
 ACCEPTED MANUSCRIPT

messaging with respect to gas consumption, we cannot separate high and low

consumers based on whether they received a message. A reasonable alternative is to

simply classify those in the top quartile of consumption as high consumers.

Calculating the pre/post iBert household gas consumption for just these high

PT
consumers shows a mean household saving of 27.2%, compared to 22.0% for all

homes.

RI
It is unknown why there was no reduction in electricity use, but the following

SC
are possibilities: (1) for an occupant, reducing electricity consumption is more

complex than simply turning a thermostat down, or ensuring heating is off when

U
leaving the building; (2) rational electricity consumption is more individualistic than
AN
heating consumption, with a multitude of different devices in different homes, this

could lead to high consumers not necessarily being profligate; (3) it is known that
M

people are unsure of which electrical items use the most electricity, and hence which to
D

manage more effectively.

TE
C EP
AC

28
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
Figure 4: Box plots of daily electricity usage after separation of big consumers (top)
M

and small consumers (bottom). The thick green line is at 600 W—the trigger for
D

receiving a message.
TE

Analysis 5: Energy Literacy.

EP

A paired sample t-test was conducted to evaluate the effect of digital energy

feedback interventions on the cognitive component of energy literacy. The results

C

indicated that the mean energy literacy score before the interventions (M= 0.52, SD=
AC

0.71) was significantly lower than the mean energy literacy score after the

interventions (M = 1.28, SD =1.06, t(24) = 3.17, p = 0.004). This confirms our fourth

hypothesis that the energy literacy of participants will improve after digital feedback

that is framed in an educational way.

29
 ACCEPTED MANUSCRIPT

Analysis 6: User experience and engagement

Three months after the end of the energy feedback study, participants were

contacted by phone to arrange for the equipment to be collected and were asked to

complete the same items from the energy literacy survey described in [50,51], again

PT
along with nine questions related to system usability and user experience: four items

adopted from the System Usability Scale (SUS)[52]) which is the most frequently cited

RI
scale for system usability evaluation, one item addressing participants’ preferences for

SC
different components of iBert and four items on the effect of iBert on participants’

behaviour. In total, 65% of the participants completed the post-study survey. The

U
answers were rated on a 3-point ordinal Likert scale. The usability questions were:
AN
“I think that I would like to use iBert frequently”

“I found the system very difficult to use”

M

“I needed to learn a lot of things before I could get going with the system”
D

“How useful did you find iBert overall?”

TE

The results show that the majority of the participants liked to use iBert
EP

frequently, found the system easy to use and in general found iBert useful. Eighty-two

percent of the respondents found live energy consumption data useful, 77% liked the
C

weekly energy summary and 78% the tailored messages; ANOVA analysis showed the
AC

three components of feedback were found roughly equally useful, F(2, 32) = 1.27, p =

0.30. Some 50% of the respondents took energy saving actions suggested by iBert and

roughly 36% of them took other energy saving actions not suggested by the system.

User experience with iBert is summarised in Table 4.

30
 ACCEPTED MANUSCRIPT

Table 4: Self-reported user experience with iBert energy feedback.

Interaction Percentage of
households
Took energy saving actions suggested by iBert 50%
Took additional energy saving actions NOT suggested
by iBert 36%
Paid attention to iBert text messages 78%

PT
Other households members were aware of the iBert
information 86%
Found iBert (partially) useful 82%

RI
SC
6 Discussion
In this section we discuss the findings in the light of the four hypotheses

formulated prior to the study.

U
AN
Hypothesis 1: Digital feedback in general will have an effect on energy related
M

behaviour compared to a baseline period before the interventions.

This hypothesis was confirmed. Homes with high internal temperatures reduced
D

their temperatures and gas consumption was reduced.

TE

Hypothesis 2: Energy feedback translated according to internal personal values

EP

will have a significant effect on energy related behaviour.

This hypothesis was confirmed for home internal temperature measurements.

C

Framing messages in value-based terms led to greater changes in behaviour.

AC

Hypothesis 3: Personalised messages with action prompts will have a

significant effect on energy related behaviour.

This hypothesis was confirmed, with an additional 9.6 percentage point

reduction in gas consumption from the use of action prompts.

31
 ACCEPTED MANUSCRIPT

Hypothesis 4: No matter which behavioural effect of a digital energy feedback

is observed, the digital interventions will influence energy related cognitive variables,

such as energy literacy.

This hypothesis was confirmed. We found a statistically significant positive

PT
effect of iBert interventions on energy literacy. One may argue that the effect of

energy literacy could be attributed to other factors and not the tailored educative

RI
messages per se, e.g. by the Hawthorne effect [53]. This is a phenomenon where

SC
people behave differently when they know they are being observed. Given the fact that

the sensors sets were installed in these homes two years earlier and they had been

U
monitored for a long period of time prior to the energy feedback intervention study, we
AN
believe that the Hawthorne effect was minimal in comparison to other studies as many

of them did not control for this effect.

M

7 Conclusions
D

The main innovative aspect of this study is that it tests for the first time a
TE

combination of two energy feedback strategies, personal values and tailored action

prompts, applied to energy related behaviour in an experimental setting. Value-framed

EP

motivational messages and action prompts tailored to user behaviour and building

characteristics were embedded into an intelligent energy feedback system called iBert.
C

The system gives advice according to individual building context and problematic
AC

energy-related behaviour detection, such as high internal temperature, excessive

ventilation, and heating and electricity usage while the home is not occupied.

Our findings suggest a positive effect of digital feedback on home internal

temperature, and a specific positive effect of internal values and action prompts

incorporated in energy feedback, from digital energy feedback delivered by iBert. This

32
 ACCEPTED MANUSCRIPT

is the first time a smart metering approach has been shown to directly influence room

temperature. Unprecedented gas consumption saving were achieved with much of this

directly attributable to the use of action prompts.

The 22.0% reduction in gas consumption (27.2% in high consumers) found is

PT
substantially greater than the reductions seen in the literature; this provides strong

evidence of the power of personalised feedback, personal values, action prompts and

RI
an active improvement of energy literacy, and could be industry-transformative.

SC
The digital feedback system we have demonstrated in this paper had a marked

positive effect on the knowledge component of participants’ energy literacy. This

U
result is very promising as it indicates that digital feedback (a standard IHD
AN
accompanied by tailored educational messages as in the current study) can improve

general energy understanding. A better understanding of energy use would be a natural

M

first step towards forming new attitudes and foundations for energy-saving behaviour
D

amongst occupants.
TE

In this paper we have evaluated a ground-breaking way of informing occupants

about their energy consumption with the purpose of educating users and lowering their
EP

demand. By being building- and occupant-aware, and only reporting messages when

consumption is potentially high, many of the criticisms of smart metering have been
C

addressed. For the first time a smart metering system has been shown to reduce room
AC

temperatures directly, with the temperature chosen by the occupants identical to that

suggested by the algorithm. The gas consumption savings found in this two-year study

have the potential to be transformative for both the occupants and for climate change

policy.

33
 ACCEPTED MANUSCRIPT

Acknowledgements
This work was done under the auspices of the ENLITEN project, funded by

EPSRC (grant reference EP/K002724/1). We thank Exeter City Council for their

support provided during the project. The internal temperature and energy consumption

PT
data used for the analysis together with the energy literacy and usability surveys data

can be found here: http://doi.org/10.15125/BATH-00256. Ramallo-González would

RI
like to thank the program Saavedra Fajardo (20035/SF/16) funded by Consejería de

Educación y Universidades of CARM, via Fundación Séneca-Agencia de Ciencia y

SC
Tecnología de la Región de Murcia.

U
AN
References
M

[1] IEA. World Energy Statistics and Balances 2016. Technical report,

International Energy Agency, 2016.

D

[2] EIA. Annual Energy Outlook 2016. Technical report, Energy Information
TE

Administartion, US, 2014.

EP

[3] BEIS. Energy Consumption in the UK. Technical report, Department of

Business, Energy and Industrial Strategy, UK, 2016.

C

[4] DECC. 2014 UK Greenhouse Gas Emissions, Final figures. Technical report,
AC

Department of Energy and Climate Change, UK, 2014.

[5] EIA. International Energy Outlook 2016. Technical report, US Energy

Information Administration, 2016.

[6] Sarah Darby. Making it obvious: designing feedback into energy

consumption. In Energy efficiency in household appliances and lighting, pages 685–

34
 ACCEPTED MANUSCRIPT

696. Springer, 2001.

[7] Stavroula Karatasou, Marina Laskari, and Mat Santamouris. Models of

behavior change and residential energy use: a review of research directions and

findings for behavior-based energy efficiency. Advances in Building Energy Research,

PT
8(2):137–147, 2014.

RI
[8] Chris Foulds, Rosalyn A.V. Robison, Rachel Macrorie, Energy monitoring as

a practice: Investigating use of the iMeasure online energy feedback tool, Energy

SC
Policy, 104:194-202, 2017.

U
[9] Thanasis G. Papaioannou, Dimos Kotsopoulos, Cleopatra Bardaki, Stavros
AN
Lounis, Nikos Dimitriou, George Boultadakis, Anastasia Garbi, Anthony Schoofs,

"IoT-enabled gamification for energy conservation in public buildings," 2017 Global

M

Internet of Things Summit (GIoTS), Geneva, Switzerland, 2017, pp. 1-6.

D
TE

[10] Heta Karoliina Kosonen, Amy Ahim Kim, Advancement of behavioral

energy interventions in commercial buildings, Facilities, 35(5/6): 367-382, 2017.

EP

[11] Thorben Jensen, Georg Holtz, Carolin Baedeker, Émile J.L. Chappin,

Energy-efficiency impacts of an air-quality feedback device in residential buildings:

C

An agent-based modeling assessment, Energy and Buildings, 116:151-163, 2016.

AC

[12] Xingxing Zhang, Jingchun Shen, Tong Yang, Llewellyn Tang, Luying

Wang, Yingqi Liu & Peng Xu, Smart meter and in-home display for energy savings in

residential buildings: a pilot investigation in Shanghai, China, Intelligent Buildings

International (In press), 2016.

35
 ACCEPTED MANUSCRIPT

[13] Fernando Terroso-Saenz, Aurora Gonzalez-Vidal, Alfonso P. Ramallo-

González, Antonio F. Skarmeta, An Open IoT Platform for the Management and

Analysis of Energy Data. Future Generation Computer Systems (In press), 2017.

PT
RI
[14] Lovett, T., E. Gabe-Thomas, S. Natarajan, E. O’Neill and J. Padget. 2013.

“‘Just Enough’ sensing to ENLITEN: a preliminary demonstration of sensing strategy

SC
for the ‘energy literacy through an intelligent home energy advisor’ (ENLITEN)

U
project.” In Proceedings of the Fourth International Conference on Future Energy
AN
Systems, 279–280. Berkeley, CA: ACM.

[15] Ramallo-González, A. P., Brown, M., Coley, D. A. 2016, The reliability of

M

inverse modelling for the wide scale characterization of the thermal properties of

buildings, Journal of Building Performance Simulation (In press).

D
TE

[16] Wokje Abrahamse, Linda Steg, Charles Vlek, and Talib Rothengatter. A

review of intervention studies aimed at household energy conservation. Journal of

EP

environmental psychology, 25(3):273–291, 2005.

[17] A.-D. Barbu, N. Griffiths, and G. Morton. Achieving energy efficiency

C

through behaviour change: what does it take? Technical report, EEA, 2013.
AC

[18] Sarah Darby. Smart metering: what potential for householder engagement?

Building Research & Information, 38(5):442–457, 2010.

[19] Tom Hargreaves, Michael Nye, and Jacquelin Burgess. Making energy

visible: A qualitative field study of how householders interact with feedback from

smart energy monitors. Energy Policy, 38(10):6111–6119, 2010.

36
 ACCEPTED MANUSCRIPT

[20] Vine Desley, Buys Laurie, and Morris Peter. The effectiveness of energy

feedback for conservation and peak demand: a literature review. Open Journal of

Energy Efficiency, 2013, 2013.

[21] Ahmad Faruqui, Sanem Sergici, and Ahmed Sharif. The impact of

PT
informational feedback on energy consumptiona survey of the experimental evidence.

Energy, 35(4):1598–1608, 2010.

RI
[22] K. Buchanan, R. Russo, and B. Anderson. The question of energy reduction:

SC
The problem(s) with feedback. Energy Policy, 77:89–96, 2015.

U
[23] Geraldine Fitzpatrick and Greg Smith. Technology-enabled feedback on
AN
domestic energy consumption: Articulating a set of design concerns. Pervasive

Computing, IEEE, 8(1):37–44, 2009.

M

[24] Kathryn Buchanan, Riccardo Russo, and Ben Anderson. Feeding back about

eco-feedback: How do consumers use and respond to energy monitors? Energy Policy,
D

73:138–146, 2014.
TE

[25] Corinna Fischer. Feedback on household electricity consumption: a tool for

EP

saving energy? Energy efficiency, 1(1):79–104, 2008.

[26] Teresa Chiang, Sukumar Natarajan, and Ian Walker. A laboratory test of the
C

efficacy of energy display interface design. Energy and Buildings, 55:471–480, 2012.
AC

[27] Hunt Allcott. Social norms and energy conservation. Journal of Public

Economics, 95(9):1082–1095, 2011.

[28] Sally Dibb, Marylyn Carrigan, Tim Harries, Ruth Rettie, Matthew Studley,

Kevin Burchell, and Simon Chambers. Is social norms marketing effective? a case

study in domestic electricity consumption. European Journal of Marketing,

37
 ACCEPTED MANUSCRIPT

47(9):1458–1475, 2013.

[29] P Wesley Schultz, Mica Estrada, Joseph Schmitt, Rebecca Sokoloski, and

Nilmini Silva-Send. Using in-home displays to provide smart meter feedback about

household electricity consumption: A randomized control trial comparing kilowatts,

PT
cost, and social norms. Energy, 90:351– 358, 2015.

RI
[30] P Wesley Schultz, Jessica M Nolan, Robert B Cialdini, Noah J Goldstein,

and Vladas Griskevicius. The constructive, destructive, and reconstructive power of

SC
social norms. Psychological science, 18(5):429–434, 2007.

U
[31] Adam Corner, Ezra Markowitz, and Nick Pidgeon. Public engagement with
AN
climate change: the role of personal values. Wiley Interdisciplinary Reviews: Climate

Change, 5(3):411–422, 2014.

M

[32] M Douglas. Wildavski. Risk and culture. An Essay on the Selection

ofTechnical and Environinental Dangers, 1982.

D
TE

[33] Shalom H Schwartz. Universals in the content and structure of values:

Theoretical advances and empirical tests in 20 countries. Advances in experimental

EP

social psychology, 25(1):1–65, 1992.

[34] Linda Steg, Jan Willem Bolderdijk, Kees Keizer, and Goda Perlaviciute. An
C

integrated framework for encouraging pro-environmental behaviour: The role of

AC

values, situational factors and goals. Journal of Environmental Psychology, 38:104–

115, 2014.

[35] Judith IM De Groot and Linda Steg. Value orientations to explain beliefs

related to environmental significant behavior how to measure egoistic, altruistic, and

biospheric value orientations. Environment and Behavior, 40(3):330–354, 2008.

38
 ACCEPTED MANUSCRIPT

[36] Linda Steg, Goda Perlaviciute, Ellen Van der Werff, and Judith Lurvink.

The significance of hedonic values for environmentally relevant attitudes, preferences,

and actions. Environment and behavior, page 0013916512454730, 2012.

[37] Jon Froehlich. Promoting energy efficient behaviors in the home through

PT
feedback: The role of human-computer interaction. In Proc. HCIC Workshop, volume

9, pages 1–11, 2009.

RI
[38] Neil Simcock, Sherilyn MacGregor, Philip Catney, Andrew Dobson, Mark

SC
Ormerod, Zoe Robinson, Simon Ross, Sarah Royston, and Sarah Marie Hall. Factors

influencing perceptions of domestic energy information: Content, source and process.

Energy Policy, 65:455–464, 2014.

U
AN
[39] Debby Cotton, Wendy Miller, Jennie Winter, Ian Bailey, and Stephen
M

Sterling. Knowledge, agency and collective action as barriers to energysaving

behaviour. Local Environment, pages 1–15, 2015.

D

[40] Jan DeWaters and Susan Powers. Establishing measurement criteria for an
TE

energy literacy questionnaire. The Journal of Environmental Education, 44(1):38–55,

2013.
EP

[41] Lung-Sheng Lee, Yi-Fang Lee, James W Altschuld, and Ying-Ju Pan.
C

Energy literacy: Evaluating knowledge, affect, and behavior of students in taiwan.

AC

Energy Policy, 76:98–106, 2015.

[42] Mageswary Karpudewan, Jamunah Ponniah, and Ahmad Nurulazam Md

Zain. Project-based learning: An approach to promote energy literacy among

secondary school students. The Asia-Pacific Education Researcher, 25(2):229–237,

2016.

39
 ACCEPTED MANUSCRIPT

[43] Marika Vellei, Sukumar Natarajan, Benjamin Biri, Julian Padget, and Ian

Walker. The effect of real-time context-aware feedback on occupants heating

behaviour and thermal adaptation. Energy and Buildings, 123:179–191, 2016.

[44] Simona DOca, Stefano P. Corgnati, and Tiziana Buso. Smart meters and

PT
energy savings in italy: Determining the effectiveness of persuasive communication in

dwellings. Energy Research & Social Science, 3:131 – 142, 2014.

RI
[45] Andreas Kamilaris, Jodi Neovino, Sekhar Kondepudi, and Balaji Kalluri. A

SC
case study on the individual energy use of personal computers in an office setting and

assessment of various feedback types toward energy savings. Energy and Buildings,

104:73 - 86, 2015.

U
AN
[46] Office for National Statistics. Nowcasting household income in the UK:
M

Financial year ending 2015. Technical report, Office for National Statistics, UK, 2015.

[47] OFGEM. Smart Metering Implementation Programme: In-Home Display.

D

Technical report, Office of gas and electricity markets, UK, 2010.

TE

[48] CIBSE, Guide A: Environmental design, 2006, London: The Chartered

EP

Institution of Building Services Engineers.

[49] DECC. 2014 UK Greenhouse Gas Emissions, Final figures. Technical

C

report, Department of Energy and Climate Change, UK, 2014.

AC

[50] Debby RE Cotton, Wendy Miller, Jennie Winter, Ian Bailey, and Stephen

Sterling. Developing students energy literacy in higher education. International Journal

of Sustainability in Higher Education, 16(4):456–473, 2015.

[51] Jan E DeWaters and Susan E Powers. Energy literacy of secondary students

in new york state (USA): A measure of knowledge, affect, and behavior. Energy
40
 ACCEPTED MANUSCRIPT

Policy, 39(3):1699–1710, 2011.

[52] John Brooke et al. Sus-a quick and dirty usability scale. Usability evaluation

in industry, 189(194):4–7, 1996.

[53] John G Adair. The Hawthorne effect: A reconsideration of the

PT
methodological artifact. Journal of applied psychology, 69(2):334, 1984.

RI
U SC
AN
M
D
TE
C EP
AC

Appendix A Energy literacy and user experience survey

Part 1

1. How much do you feel you know about energy? (Please circle one)

41
 ACCEPTED MANUSCRIPT

A lot (expert)

Quite a bit (informed)

Not much (novice)

PT
Nothing

2. Which of the following sources of information has contributed most to your

RI
understanding of energy issues? (Please choose one option)

SC
Further or higher education

School

U
AN
Books, newspapers or magazines

Friends or family members (including parents)

M

Internet
D

Television iBert system

TE

Other (please specify)

EP

3. The term renewable energy resources means (please circle one answer)

Resources that are free and convenient to use

C

Resources that can be converted directly into heat and electricity

AC

Resources that do not produce air pollution

Resources that are very efficient to use for producing energy

Resources that can be replenished by nature in a short period of time

4. Most of the renewable energy in the UK comes from which of the

42
 ACCEPTED MANUSCRIPT

following sources? (Please circle one)

Solar

Water (hydro/tidal/wave) power

PT
Wind

Landfill gas

RI
Geothermal

SC
Don’t know

U
5. Which of the following actions, if everyone did this all the time, would save the most
AN
energy in the UK? (Please circle the most important )

Turn off lights when they are not in use

M

Turn down the heat in rooms

D

Reduce water consumption

TE

Walk or cycle short distances instead of going by car

EP

Turn appliances off at the plug

6. Which kind of lighting uses the least amount of energy? (Please circle one answer)
C

Standard light bulbs

AC

Low energy light bulbs

Fluorescent lights

LED lights

Don’t know

43
 ACCEPTED MANUSCRIPT

Part 2

1. I think that I would like to use iBert frequently

Agree

PT
Somewhat agree

Disagree

RI
2. I found the system very difficult to use

SC
Agree

U
Somewhat agree AN
Disagree

3. I needed to learn a lot of things before I could get going with the system
M

Agree
D

Somewhat agree
TE

Disagree
EP

4. In which extent did you find the following parts of the iBert system useful:

Live status panel

C
AC

Useless

Moderately useful

Very useful

Weekly energy consumption summary

Useless

44
 ACCEPTED MANUSCRIPT

Moderately useful

Very useful

Text messages with tips and advice/suggestions

PT
Useless

Moderately useful Very useful

RI
5. How useful did you find iBert overall?

SC
Useless

U
Moderately useful AN
Very useful

6. Did you pay attention to the text messages that were displayed by iBert?
M

Yes
D

No
TE

Partially
EP

7. Did you take any energy saving actions suggested by iBert?

Yes
C
AC

No

8. Did you take any additional energy saving actions NOT suggested by iBert during or

after the study?

Yes

No

45
 ACCEPTED MANUSCRIPT

Partially

9. If there are other adults living in your home, did they attend/were aware of the

information provided by iBert?

Not aware

PT
Aware

RI
Partially aware

U SC
AN
M
D
TE
C EP
AC

Appendix B iBert running on an Android tablet

46
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP
C
AC

47
 ACCEPTED MANUSCRIPT
We investigate the effect of replacing the simple statement of energy use a

smart meter gives, with a detailed array of information specifically designed to

improve consumer energy literacy and suggest behaviour change through personalised

actionable messages set against a series of psychological value systems for context,

and which can identify potential profligacy.

PT
The results from a carefully controlled field experiment show: 1) value framing

RI
and action prompts have a significant effect on occupants' behaviour, with the mean

temperature of homes being reduced from 22.4°C to 21.7 °C, and a marked reduction

SC
in gas consumption—22.0% overall and 27.2% in high consumers; 2) energy literacy

increasing from 0.52 to 1.28 (on a 0-4 scale); 3) it is possible to target potentially

U
profligate households, without inappropriately messaging others;
AN
Engagement is high, with 82% of the participants finding the system useful.
M

These results emphasize the necessity of improving energy literacy when

encouraging energy efficient behaviours and point to a new generation of smart

D

meters with the potential to increase energy literacy, make much greater savings and
TE

impact climate change policy.

C EP
AC

View publication stats