Energy olympics

31
EXPERIMENTS IN ENERGY OLYMPICS WEST OF ENGLAND CARBON CHALLENGE 2012-11-07, 1600 Sukumar Natarajan with contributions from Julian Padget, Ian Walker and Teresa Chiang, Nadya Dara, Gokhan Mevlevioglu

Transcript of Energy olympics

Page 1: Energy olympics

EXPERIMENTS IN ENERGY OLYMPICS

WEST OF ENGLAND CARBON CHALLENGE2012-11-07, 1600

Sukumar Natarajan with contributions from Julian Padget, Ian Walker

and Teresa Chiang, Nadya Dara, Gokhan Mevlevioglu

Page 2: Energy olympics

Climate Change

Energy Use / Efficiency /Security

Building Design

Energy and the Design of Environments (EDEn)

COPSE /Prometheus(both using

UKCP09 data)

ENergy Literacy through an

IntelligenT home ENergy advisor

(ENLITEN)

Page 3: Energy olympics

CONTEXT• Domestic:

• Users have control over decisions• Relatively easy to connect cause and effect

• Non domestic:

• Users generally have less control• More tricky to connect cause and effect

Page 4: Energy olympics

Does feedback affect energy consumption?

Question 1: (the type of)

Experiment 2: live (student residence)

Three types of displays were proposed in the literature

Experiment 1: laboratory test

Larger and longer term trials show less than 5% savings from control

Savings from various studies range from 0% to 22%

Page 5: Energy olympics

Does feedback affect energy consumption?

Question 1: (the type of)

Please cite this article in press as: T. Chiang, et al., A laboratory test of the efficacy of energy display interface design, Energy Buildings(2012), http://dx.doi.org/10.1016/j.enbuild.2012.07.026

ARTICLE IN PRESSG ModelENB-3818; No. of Pages 10

T. Chiang et al. / Energy and Buildings xxx (2012) xxx–xxx 3

Fig. 1. Numerical display [29].

Fig. 2. Analogue display [30].

Fig. 3. Ambient design [31].

easy to comprehend due to the consistent exposure in a numberof activities both at home (kitchen timer, alarm clock, TV, etc.)and work (security systems, vending machines, etc.). The presentwork intends to examine presenting energy information with purenumbers.

Analogue displays (e.g. Fig. 2) illustrate the scale of consump-tion usually without numbers, using graphs, charts, dials, columngauges and bars. Compared to numerical displays, these are oftenconsidered easier to read and interpret, especially when makingcomparisons, e.g. between current and target values [24]. Ana-logue displays can be effective in checking readings, evaluatingfuture states and conveying quantitative and qualitative informa-tion simultaneously in a direct and intuitive manner [24]. Andersonand White [25] found through working with focus groups that thedesign of a speedometer dial was particularly preferred for its qual-ities of eye-catching movement, intuitive scale and direction ofchange, and ease of making comparisons. It will be interesting tosee if similar preferences are seen in this study.

Ambient displays (Fig. 3) provide an overall indication of a situa-tion, sometimes make use of peripheral vision, and do not requireusers’ detailed attention [8,22]. No text or numbers are shown;instead, pictures, colours, sounds or flashing lights are used toconvey information. The present work attempts to introduce theinclusion of human faces in the design of user interface prototypes.Human faces are known to have the capacity to attract attention,even when reduced to cartoon form [26,27]. Presumably thanksto their social and biological significance, human faces seem tobe processed differently by the human brain: changes in faces aredetected faster and more accurately than in other objects [e.g. 28].The intention of this work is to find out if the face’s attentioncapturing property makes it a useful cue in the design of energydisplays. Specifically, the study looked at two-dimensional cartoon-like faces.

2.2.2. ColourA large number of studies [e.g. 32–34] on web page design

and screen design have shown that comprehension, distractionand directional attention are heavily influenced by the location,colour, size, orientation, shape and luminance of targets [e.g. 35,36]when people search or read displays. Colour, in particular, hasbeen found to have the capacity to attract attention to targetstimuli due to its highlighting and association effects, and to sepa-rate potential target stimuli from non-target stimuli [37]. It mightalso help to shorten search time more than shapes or numbers[after 5]. As colour is a basic element of visual perception, theprocessing of colour coded information does not require largeamounts of cognitive capacities [after 38,39]. For these reasons,the present work also investigates the effectiveness of coloured(i.e. non black-on-white7) display designs against black-on-whitedisplay designs (details of the experiment are described in Section3.2).

2.3. Users’ preferences

Previous research has shown mixed findings on what informa-tion people preferred to receive and which type of graphical designcould help to serve presenting information better. Researchers inthe USA designed two types of graphical displays to show energyconsumption: bar graphs and distribution graphs. Fig. 4 is one of thedistribution graphic designs that displayed little houses in a neigh-bourhood mapped on an x–y axis graph according to their energyconsumption [40]. This design was ranked most preferred, whilethe same design concept presented in Norway [41] was judged as“childish” and difficult to interpret.

7 The present work studied black-on-white design instead of white-on-blackdesign, as previous studies have shown that dark text on light background (positivepolarity) resulted in better performance compared to light text on dark background(negative polarity) [56,57].

digital

detailedquantitativeinformation

Please cite this article in press as: T. Chiang, et al., A laboratory test of the efficacy of energy display interface design, Energy Buildings(2012), http://dx.doi.org/10.1016/j.enbuild.2012.07.026

ARTICLE IN PRESSG ModelENB-3818; No. of Pages 10

T. Chiang et al. / Energy and Buildings xxx (2012) xxx–xxx 3

Fig. 1. Numerical display [29].

Fig. 2. Analogue display [30].

Fig. 3. Ambient design [31].

easy to comprehend due to the consistent exposure in a numberof activities both at home (kitchen timer, alarm clock, TV, etc.)and work (security systems, vending machines, etc.). The presentwork intends to examine presenting energy information with purenumbers.

Analogue displays (e.g. Fig. 2) illustrate the scale of consump-tion usually without numbers, using graphs, charts, dials, columngauges and bars. Compared to numerical displays, these are oftenconsidered easier to read and interpret, especially when makingcomparisons, e.g. between current and target values [24]. Ana-logue displays can be effective in checking readings, evaluatingfuture states and conveying quantitative and qualitative informa-tion simultaneously in a direct and intuitive manner [24]. Andersonand White [25] found through working with focus groups that thedesign of a speedometer dial was particularly preferred for its qual-ities of eye-catching movement, intuitive scale and direction ofchange, and ease of making comparisons. It will be interesting tosee if similar preferences are seen in this study.

Ambient displays (Fig. 3) provide an overall indication of a situa-tion, sometimes make use of peripheral vision, and do not requireusers’ detailed attention [8,22]. No text or numbers are shown;instead, pictures, colours, sounds or flashing lights are used toconvey information. The present work attempts to introduce theinclusion of human faces in the design of user interface prototypes.Human faces are known to have the capacity to attract attention,even when reduced to cartoon form [26,27]. Presumably thanksto their social and biological significance, human faces seem tobe processed differently by the human brain: changes in faces aredetected faster and more accurately than in other objects [e.g. 28].The intention of this work is to find out if the face’s attentioncapturing property makes it a useful cue in the design of energydisplays. Specifically, the study looked at two-dimensional cartoon-like faces.

2.2.2. ColourA large number of studies [e.g. 32–34] on web page design

and screen design have shown that comprehension, distractionand directional attention are heavily influenced by the location,colour, size, orientation, shape and luminance of targets [e.g. 35,36]when people search or read displays. Colour, in particular, hasbeen found to have the capacity to attract attention to targetstimuli due to its highlighting and association effects, and to sepa-rate potential target stimuli from non-target stimuli [37]. It mightalso help to shorten search time more than shapes or numbers[after 5]. As colour is a basic element of visual perception, theprocessing of colour coded information does not require largeamounts of cognitive capacities [after 38,39]. For these reasons,the present work also investigates the effectiveness of coloured(i.e. non black-on-white7) display designs against black-on-whitedisplay designs (details of the experiment are described in Section3.2).

2.3. Users’ preferences

Previous research has shown mixed findings on what informa-tion people preferred to receive and which type of graphical designcould help to serve presenting information better. Researchers inthe USA designed two types of graphical displays to show energyconsumption: bar graphs and distribution graphs. Fig. 4 is one of thedistribution graphic designs that displayed little houses in a neigh-bourhood mapped on an x–y axis graph according to their energyconsumption [40]. This design was ranked most preferred, whilethe same design concept presented in Norway [41] was judged as“childish” and difficult to interpret.

7 The present work studied black-on-white design instead of white-on-blackdesign, as previous studies have shown that dark text on light background (positivepolarity) resulted in better performance compared to light text on dark background(negative polarity) [56,57].

analogue

illustrate scale of

consumption

Please cite this article in press as: T. Chiang, et al., A laboratory test of the efficacy of energy display interface design, Energy Buildings(2012), http://dx.doi.org/10.1016/j.enbuild.2012.07.026

ARTICLE IN PRESSG ModelENB-3818; No. of Pages 10

T. Chiang et al. / Energy and Buildings xxx (2012) xxx–xxx 3

Fig. 1. Numerical display [29].

Fig. 2. Analogue display [30].

Fig. 3. Ambient design [31].

easy to comprehend due to the consistent exposure in a numberof activities both at home (kitchen timer, alarm clock, TV, etc.)and work (security systems, vending machines, etc.). The presentwork intends to examine presenting energy information with purenumbers.

Analogue displays (e.g. Fig. 2) illustrate the scale of consump-tion usually without numbers, using graphs, charts, dials, columngauges and bars. Compared to numerical displays, these are oftenconsidered easier to read and interpret, especially when makingcomparisons, e.g. between current and target values [24]. Ana-logue displays can be effective in checking readings, evaluatingfuture states and conveying quantitative and qualitative informa-tion simultaneously in a direct and intuitive manner [24]. Andersonand White [25] found through working with focus groups that thedesign of a speedometer dial was particularly preferred for its qual-ities of eye-catching movement, intuitive scale and direction ofchange, and ease of making comparisons. It will be interesting tosee if similar preferences are seen in this study.

Ambient displays (Fig. 3) provide an overall indication of a situa-tion, sometimes make use of peripheral vision, and do not requireusers’ detailed attention [8,22]. No text or numbers are shown;instead, pictures, colours, sounds or flashing lights are used toconvey information. The present work attempts to introduce theinclusion of human faces in the design of user interface prototypes.Human faces are known to have the capacity to attract attention,even when reduced to cartoon form [26,27]. Presumably thanksto their social and biological significance, human faces seem tobe processed differently by the human brain: changes in faces aredetected faster and more accurately than in other objects [e.g. 28].The intention of this work is to find out if the face’s attentioncapturing property makes it a useful cue in the design of energydisplays. Specifically, the study looked at two-dimensional cartoon-like faces.

2.2.2. ColourA large number of studies [e.g. 32–34] on web page design

and screen design have shown that comprehension, distractionand directional attention are heavily influenced by the location,colour, size, orientation, shape and luminance of targets [e.g. 35,36]when people search or read displays. Colour, in particular, hasbeen found to have the capacity to attract attention to targetstimuli due to its highlighting and association effects, and to sepa-rate potential target stimuli from non-target stimuli [37]. It mightalso help to shorten search time more than shapes or numbers[after 5]. As colour is a basic element of visual perception, theprocessing of colour coded information does not require largeamounts of cognitive capacities [after 38,39]. For these reasons,the present work also investigates the effectiveness of coloured(i.e. non black-on-white7) display designs against black-on-whitedisplay designs (details of the experiment are described in Section3.2).

2.3. Users’ preferences

Previous research has shown mixed findings on what informa-tion people preferred to receive and which type of graphical designcould help to serve presenting information better. Researchers inthe USA designed two types of graphical displays to show energyconsumption: bar graphs and distribution graphs. Fig. 4 is one of thedistribution graphic designs that displayed little houses in a neigh-bourhood mapped on an x–y axis graph according to their energyconsumption [40]. This design was ranked most preferred, whilethe same design concept presented in Norway [41] was judged as“childish” and difficult to interpret.

7 The present work studied black-on-white design instead of white-on-blackdesign, as previous studies have shown that dark text on light background (positivepolarity) resulted in better performance compared to light text on dark background(negative polarity) [56,57].

ambient

provide an “indication”

of the situation

☺we wanted to control these quite carefully, so COTS were not an option

Page 6: Energy olympics

How good are we at detecting change in information: speed and accuracy

Experiment 1: The key question is

Page 7: Energy olympics

Let’s do a quick test

Page 8: Energy olympics
Page 9: Energy olympics

Please cite this article in press as: T. Chiang, et al., A laboratory test of the efficacy of energy display interface design, Energy Buildings(2012), http://dx.doi.org/10.1016/j.enbuild.2012.07.026

ARTICLE IN PRESSG ModelENB-3818; No. of Pages 10

6 T. Chiang et al. / Energy and Buildings xxx (2012) xxx–xxx

Fig. 8. Reference image (top); test image (bottom).

had three consumption ranges: low, medium, high, which werederived based on data from an average UK household’s annualelectricity consumption.8 Finally, each display design had a black-on-white version (Fig. 6) and a coloured version (Fig. 7). The overallbackground colour for all test images was white. Red and greenin the colour version were chosen for their associative meanings,representing high and low consumption ranges, respectively, andblack was used to represent medium consumption range. In theblack-on-white version, no colour other than black was used to rep-resent the three consumption levels. Low consumption levels wererepresented by lower numerical values in the numerical design, theleft-hand end of the dial in the analogue design, and a happy face inthe ambient design. Accuracy rate was calculated in percentage ofcorrect answers, and response time was measured in milliseconds(ms).

3.3. Experimental procedure

Participants received oral instructions and had a practice sessionof ten pairs of test images. As shown in Fig. 8, the first test image ineach pair was the “Reference” and was displayed for seven seconds;this was followed by the “Test” image, which was displayed untilthe participant responded. Each pair of images used the samedesign (numerical/analogue/ambient), and the design changed ran-domly from trial to trial. The Reference and Test images differedby a change in the value of one of the five display components.Participants were asked to determine as quickly and accurately aspossible if the consumption value of the changing display compo-nent had increased or decreased. In order to ensure that the timelapse between comprehension and response was minimised, theresponse for “increased” was mapped to the “@” key (next to the

8 For example, to determine what the day-to-day (“Today so far” and “Yester-day”) average consumption range was, the annual average electricity consumptionof 3300 kWh [15] was divided by 365 days to give an average daily consumptionvalue of 9 kWh. The figure was then doubled to give the full range and divided by3 for three daily ranges, i.e. the low range would be 0–6 kWh, the medium rangewould be 6.1–12 kWh, and the high range would be 12.1 kWh and higher. It is worthnoting here that these are realistic, albeit crude, approximations appropriate to theexperiment. While ranges in the real world will no doubt be different, these are notexpected to significantly affect results since participants were focused on changesto displays rather than the values themselves.

Fig. 9. Mean accuracy rates.

Fig. 10. Mean response times.

“Enter” key) and the response for “decreased” was mapped to the“A” key on a standard UK keyboard. Note that the task involvedidentifying which component had changed in addition to detectingthe nature of the change.

3.4. Interviews

Participants were interviewed after the computerised task toexpress their views in the following core questions:

• Which type of display design was the easiest to read?• Did you find the colour version more helpful in seeing the change?• How did you read and compare the information?• Which type of display design would you prefer to have at home?• Would you prefer to have the display design at home in colour?

The interviews were semi-structured, such that discussionstemmed from the core questions. In many instances, more than

Fig. 11. Display design preferences.

Please cite this article in press as: T. Chiang, et al., A laboratory test of the efficacy of energy display interface design, Energy Buildings(2012), http://dx.doi.org/10.1016/j.enbuild.2012.07.026

ARTICLE IN PRESSG ModelENB-3818; No. of Pages 10

6 T. Chiang et al. / Energy and Buildings xxx (2012) xxx–xxx

Fig. 8. Reference image (top); test image (bottom).

had three consumption ranges: low, medium, high, which werederived based on data from an average UK household’s annualelectricity consumption.8 Finally, each display design had a black-on-white version (Fig. 6) and a coloured version (Fig. 7). The overallbackground colour for all test images was white. Red and greenin the colour version were chosen for their associative meanings,representing high and low consumption ranges, respectively, andblack was used to represent medium consumption range. In theblack-on-white version, no colour other than black was used to rep-resent the three consumption levels. Low consumption levels wererepresented by lower numerical values in the numerical design, theleft-hand end of the dial in the analogue design, and a happy face inthe ambient design. Accuracy rate was calculated in percentage ofcorrect answers, and response time was measured in milliseconds(ms).

3.3. Experimental procedure

Participants received oral instructions and had a practice sessionof ten pairs of test images. As shown in Fig. 8, the first test image ineach pair was the “Reference” and was displayed for seven seconds;this was followed by the “Test” image, which was displayed untilthe participant responded. Each pair of images used the samedesign (numerical/analogue/ambient), and the design changed ran-domly from trial to trial. The Reference and Test images differedby a change in the value of one of the five display components.Participants were asked to determine as quickly and accurately aspossible if the consumption value of the changing display compo-nent had increased or decreased. In order to ensure that the timelapse between comprehension and response was minimised, theresponse for “increased” was mapped to the “@” key (next to the

8 For example, to determine what the day-to-day (“Today so far” and “Yester-day”) average consumption range was, the annual average electricity consumptionof 3300 kWh [15] was divided by 365 days to give an average daily consumptionvalue of 9 kWh. The figure was then doubled to give the full range and divided by3 for three daily ranges, i.e. the low range would be 0–6 kWh, the medium rangewould be 6.1–12 kWh, and the high range would be 12.1 kWh and higher. It is worthnoting here that these are realistic, albeit crude, approximations appropriate to theexperiment. While ranges in the real world will no doubt be different, these are notexpected to significantly affect results since participants were focused on changesto displays rather than the values themselves.

Fig. 9. Mean accuracy rates.

Fig. 10. Mean response times.

“Enter” key) and the response for “decreased” was mapped to the“A” key on a standard UK keyboard. Note that the task involvedidentifying which component had changed in addition to detectingthe nature of the change.

3.4. Interviews

Participants were interviewed after the computerised task toexpress their views in the following core questions:

• Which type of display design was the easiest to read?• Did you find the colour version more helpful in seeing the change?• How did you read and compare the information?• Which type of display design would you prefer to have at home?• Would you prefer to have the display design at home in colour?

The interviews were semi-structured, such that discussionstemmed from the core questions. In many instances, more than

Fig. 11. Display design preferences.

Page 10: Energy olympics

Please cite this article in press as: T. Chiang, et al., A laboratory test of the efficacy of energy display interface design, Energy Buildings(2012), http://dx.doi.org/10.1016/j.enbuild.2012.07.026

ARTICLE IN PRESSG ModelENB-3818; No. of Pages 10

6 T. Chiang et al. / Energy and Buildings xxx (2012) xxx–xxx

Fig. 8. Reference image (top); test image (bottom).

had three consumption ranges: low, medium, high, which werederived based on data from an average UK household’s annualelectricity consumption.8 Finally, each display design had a black-on-white version (Fig. 6) and a coloured version (Fig. 7). The overallbackground colour for all test images was white. Red and greenin the colour version were chosen for their associative meanings,representing high and low consumption ranges, respectively, andblack was used to represent medium consumption range. In theblack-on-white version, no colour other than black was used to rep-resent the three consumption levels. Low consumption levels wererepresented by lower numerical values in the numerical design, theleft-hand end of the dial in the analogue design, and a happy face inthe ambient design. Accuracy rate was calculated in percentage ofcorrect answers, and response time was measured in milliseconds(ms).

3.3. Experimental procedure

Participants received oral instructions and had a practice sessionof ten pairs of test images. As shown in Fig. 8, the first test image ineach pair was the “Reference” and was displayed for seven seconds;this was followed by the “Test” image, which was displayed untilthe participant responded. Each pair of images used the samedesign (numerical/analogue/ambient), and the design changed ran-domly from trial to trial. The Reference and Test images differedby a change in the value of one of the five display components.Participants were asked to determine as quickly and accurately aspossible if the consumption value of the changing display compo-nent had increased or decreased. In order to ensure that the timelapse between comprehension and response was minimised, theresponse for “increased” was mapped to the “@” key (next to the

8 For example, to determine what the day-to-day (“Today so far” and “Yester-day”) average consumption range was, the annual average electricity consumptionof 3300 kWh [15] was divided by 365 days to give an average daily consumptionvalue of 9 kWh. The figure was then doubled to give the full range and divided by3 for three daily ranges, i.e. the low range would be 0–6 kWh, the medium rangewould be 6.1–12 kWh, and the high range would be 12.1 kWh and higher. It is worthnoting here that these are realistic, albeit crude, approximations appropriate to theexperiment. While ranges in the real world will no doubt be different, these are notexpected to significantly affect results since participants were focused on changesto displays rather than the values themselves.

Fig. 9. Mean accuracy rates.

Fig. 10. Mean response times.

“Enter” key) and the response for “decreased” was mapped to the“A” key on a standard UK keyboard. Note that the task involvedidentifying which component had changed in addition to detectingthe nature of the change.

3.4. Interviews

Participants were interviewed after the computerised task toexpress their views in the following core questions:

• Which type of display design was the easiest to read?• Did you find the colour version more helpful in seeing the change?• How did you read and compare the information?• Which type of display design would you prefer to have at home?• Would you prefer to have the display design at home in colour?

The interviews were semi-structured, such that discussionstemmed from the core questions. In many instances, more than

Fig. 11. Display design preferences.

Please cite this article in press as: T. Chiang, et al., A laboratory test of the efficacy of energy display interface design, Energy Buildings(2012), http://dx.doi.org/10.1016/j.enbuild.2012.07.026

ARTICLE IN PRESSG ModelENB-3818; No. of Pages 10

8 T. Chiang et al. / Energy and Buildings xxx (2012) xxx–xxx

Table 3Summary of response time and accuracy.

Response time Correct Preference

Design typea Ang > Amb = Nmr Nmr = Amb > Ang Nmr > Amb = AngColourb B/W = Clr B/W = Clr Amb > Ang = Nmr

Amb B/W > Amb Clr

a Amb: ambient, Ang: analogue, Nmr: numerical.b B/W: black-on-white, Clr: colour.

> means significantly greater than, = means marginally different.

found that if users find the product features similar to their previ-ous experience, they are more likely to perform tasks better [51,52].This raises the question of why the dials used in the analoguedesign, which should similarly be familiar to participants from ana-logue clocks, gas and electricity meters and from car gauges, led topoorest performance. Whilst it is possible that analogue displays,whilst common in everyday life, are substantially less commonthan numerical displays, at present the advantage of numbers overdials is not clear, and future work might usefully look in moredepth at analogue representation, including other forms such asgraphs and charts. Despite the special attention-capturing qual-ity of faces seen in earlier research, the emotional faces used inthis work gave no advantage to identifying changes in consump-tion nor did they appeal to many of the participants subjectively.However, for the minority who preferred or performed well withthe ambient design, the simplicity of the facial expressions was saidto be the key feature that helped them see the changes quicker andeasier.

Perhaps surprisingly, given that most participants said they sub-jectively liked colour displays, the presence of colour only slightlyincreased the accuracy rate of all design types, and this was not sta-tistically significant. It was more helpful in reducing the responsetime of the ambient design than in the numerical design, but theresponse time increased in the analogue design. However, evenif participants took longer to respond in the coloured analoguedesign, a higher accuracy rate compared to black-on-white wasachieved like with the other two designs.

Although the presence of colour did not help them perform tasksbetter, a majority of participants believed that colour was helpfulin reading information and seeing changes, and preferred informa-tion to be colour coded, for reasons of extra information, ease ofdistinguishing differences, visual aesthetics and learning experi-ence. Participants who felt colour was unhelpful were unfamiliarwith colour coding and found it “untrustworthy” and “distracting”.They did not find colour coding intuitive and felt that they had tospend more time on re-checking the information, and were stillnot sure if they had made correct answers. Participants pointedout that colour provided extra information which they found wastoo much to take in at a time, therefore they reverted to ignoringthe colour. Some participants were not aware that colour was usedeven though they reported to have normal vision. This could bedue to either mis-reporting of normal vision or the possibility thatnot everyone is receptive to the use of colour. In summary, then,designers might note that whilst colour displays are likely to beaesthetically pleasing to the majority of users, they appear to con-vey no advantage for highlighting changes in energy use and seemto be actively disliked by a minority.

6. Limitations to the study

A number of limitations to the results above have beenidentified:

(a) Sampling bias: The study attempted to recruit participants fromthe general public with as a wide range of socio-economic

backgrounds as possible. However, a majority of the partici-pants had undertaken tertiary education. This would suggestthat the participants are mostly in the middle-income classor higher, and therefore not fully representative of the pop-ulation at large. Further investigation is therefore required torule out a correlation between education, income and taskperformance.

(b) Lab based experiment: The experiment was conducted under“laboratory conditions” and therefore might not adequatelycapture perception in a real-world setting. On the other hand,by removing the distractions people are likely to experiencein their domestic environments, it could be argued that thisstudy provided a useful upper-bound on people’s ability to spotchanges in smart meters. If they were poor at using a certaintype of display when they were entirely focused on the task, itseems unlikely they would work well with that display whentheir attention was divided as in a household setting, althoughthis is, of course, an empirical question.

(c) Task based experiment: In the experiment, participants wereasked to undertake a specific task with the identified goal ofdetermining which one of five displays in a series of test pairshad changed. In a real-world setting, information will changeunpredictably and occupants will not be focused on detectingchanges. Again, however, one might argue that by asking peoplespecifically to focus on spotting changes, the study provided theupper-bound on the ability of these displays to communicatechanges in energy use to people.

(d) Design decisions: Particular design decisions were made inconstructing the experiment. For example, the numerical dis-play design was constructed using a seven-segmented format,which is a very common format found on simple LCD dis-plays. Similarly, the JFreeChart Java library was used to producethe analogue display as it is a freely available resource butled to a specific design of dial being displayed. Such decisionscould affect the perception of information and influence testresults.

Field trials are currently being undertaken to address (b)and (c). Work is also underway to test if education or incomeaffects performance on the computerised test, addressing (a)above. Further work is planned to examine the sensitivity ofthese results to different representations of the same designtype.

7. Conclusions

This paper presents a laboratory study in which three prototypesof display design (numerical displays, analogue displays using dials,and ambient displays using emotional faces) were compared to seewhich best communicated changes in energy use to participants.The role of colour coding was also examined.

Effectiveness (accuracy and response time) and subjective pre-ferences were used to measure how useful each design was. Thenumerical design was the all-round winner in task performance,leading to enhanced communication of energy-use changes com-pared to the analogue and ambient designs. Unexpectedly, thepresence of colour in a smart meter display led to no change inparticipants’ abilities to spot changes in energy use, although sub-jectively people tended to prefer colour displays to black-on-whitedisplays.

Further work will include testing on a larger sample to explorewhy numerical displays are working better, and to see whether it istrue for all people. Work is also planned to look at how effective thedisplay designs used in this work are in realistic residential settings,by displaying live energy consumption data using combinations of

Page 11: Energy olympics

How good are we at reducing energy when presented the same information in different ways?

Experiment 2: The key question is

Page 12: Energy olympics

Recycling

Solsbury Court

Marlborough Court

Outdoor Tennis Courts

West Car ParkP W

East Car ParkP E

P C

P G

P D

P A

P B

P L

P JP K

Copseland

Woodland Court

Lyme Kiln FieldSports Pitches

WestwoodResidences

Norwoo

d Hou

se Wessex

House

Nursery

4S Annexe

3S Annexe

P C

a

Bath Cats and Dogs Home

BaleHaus

EastwoodResidences Eastwood Offices

20/21 22/23

6WS

East building

2 non-participating groups were used as control

Kitchens comparable to each other with 7 students per group

6 kitchens were selected, 2 each on floors 7, 8 and 9

Each “kitchen group” was sub-metered by lights and small power

Student residences with good wifi access, 4 x kitchens per floor

Baseline period for comparison was the average over a period of days immediately before the start of the experiment

Page 13: Energy olympics

colour blind safe

Cumulative consumption from 00:01 to time of

update (say 09:00)

Total Cumulative consumption from

00:01 to 00:00

Average daily consumption

this week

Average daily consumption last

week

Page 14: Energy olympics
Page 15: Energy olympics
Page 16: Energy olympics
Page 17: Energy olympics

GroupGroupGroupGroupGroupGroup

A B C D E F

Weeks

1-2

Weeks 3-4Weeks

5-6

“Winner takes it all” £20 per member of the group that reduced the most over 6 weeks

Page 18: Energy olympics
Page 19: Energy olympics

Plainspeak: there is a 95% probability that these results did not occur due to pure chance

Confidence intervals for the experiment groups did not include zero change, so these results are significant at the 0.05 level.

Plainspeak: the change for these groups from baseline are not reliably different from zero

Confidence intervals for the control groups did include zero change.

Participants preferred this the most, yet ambient seems to perform better!

Page 20: Energy olympics
Page 21: Energy olympics

Question 2: Is competitive ranking enough, on its own, to deliver savings?

We wanted to test if comparing against one’s own savings would be different

Further, this time every person stood to gain financially as rewards were calculated using an arbitrary fixed rate (35p) per kWh saved cumulatively compared to the baseline period.

However, this tends to favour groups starting from a low baseline

In the previous experiment, groups were competing against absolute savings

Page 22: Energy olympics

How good are we at reducing energy when comparing our performance against the

performance of others?

Experiment 3: The key question is

Page 23: Energy olympics
Page 24: Energy olympics

Mean change in consumption (kWh) compared to baseline

Page 25: Energy olympics

Questions 3 & 4: Will any of these ideas work in an office environment? Do rewards matter?

Instead of monetary units, performance was measured in “meters run”

Provide additional context to ranking information

Information was separated by lighting and small power

Applied the idea of self-relative ranking from previous experiment to BH offices

Test if rewards are essential and if (any) savings persist over time.

Information on how to save was given before the start of the competition

Environmental attitudes were measured before and after the competition

Due to technical problems experiment ran for 5 weeks instead of 6

Page 26: Energy olympics

The  effectiveness  of  behavioural  interventions    to  reduce  office  energy  consumption  

 

40/141    

 

Figure  3.2.1.  Building  1  energy  consumption  by  energy  sector  for  11  weeks(26/Apr  –  1/Jul).  

Figure  3.2.2.  Building  2  energy  consumption  by  energy  sector  for  11  weeks  (26/Apr  –  1/Jul).  

However,   it   has   to   be   borne   in   mind   that   described   above   energy   balance   and  

benchmarking  is  valid  only  for  the  obtained  data  from  the  period  from  the  16th  of  April   to  

the  1st  of  July.  The  situation  for  the  whole  year  is  going  to  be  different  as  the  coldest  winter  

months   are   not   considered   in   analysis   when   the   most   amount   of   heat   is   required.  

Therefore,   the  percentage  of   energy   for   heating  would  probably   increase.  For  example,  

Figure   3.2.3   illustrates   the   difference   in   percentage   of   energy   subsectors   in   June   and  

April,  where  it   is  clear  that  heating  is  the  first   largest  consumer   in  April  whereas  in  June  

small  power  consumes  over  a  half  of  all  energy.  

 

Figure  3.2.3.  Comparison  of  total  Buildings  energy  consumption  by  sectors  in  June  and  April.  

Small  power40.54%

Lighting13.13%

Canteen4.79%

Lift0.38%

Server  room14.98%

Printing  room1.17%

Heating25.02% Small  

power46.74%

Lighting23.02%

Canteen4.82%

Lift1.33%

AC0.92% Heating

23.16%

0.00%

10.00%

20.00%

30.00%

40.00%

50.00%

60.00%

Smallpower

Lighting Canteen Lift Serverroom

Printersroom

Heating

June

April

3.2.1   3.2.2  

Proportion of total energy consumption in both buildings by end-use

Page 27: Energy olympics

The  effectiveness  of  behavioural  interventions    to  reduce  office  energy  consumption  

 

53/141    

 

Figure  3.4.1.  Scheme  of  display  screens.  

Lighting   Computer      Floor   Weekday  

average  Weekend  average  

Nominal  Average   Floor   Weekday  

average  Weekend  average  

Nominal  Average  

B1  Ground  Floor  

88.00   22.00   74.80  B1  

Ground  Floor  

263.00   41.00   218.60  

Building  1  

B1  First  Floor   41.00   18.00   36.40   B1  First  

Floor   135.00   72.00   122.40  

B1  Second  Floor  

24.00   0.00   19.20  B1  

Second  Floor  

87.00   24.00   74.40  

B2  First  Floor   69.00   7.00   56.60   B2  First  

Floor   109.00   19.00   91.00  

Building  2  

B2  Second  Floor  

71.00   3.00   57.40  B2  

Second  Floor  

103.00   13.00   85.00  

B2  Third  Floor   64.00   15.00   54.20   B2  Third  

Floor   152.00   55.00   132.60  

Table  3.4.1.  Baseline  values  for  savings  calculation  for  displays.  

3.5.  Questionnaires  on  environmental  awareness.  

In  order  to  determine  the  extent  to  which  the  environmental  awareness  of  occupants  might  

be  influenced  by  the  competition  two  sets  of  questionnaires  have  been  organised,  before  

and  after  the  competition.  The  first  questionnaire  (Appendix  H)  comprises  the  part  on  the  

environmental  awareness  that  has  been  derived  from  Verplanken  et  al  (2008)  where  the  

version  of  Dunlap  et  al  (2000,  cited  in  (Verplanken  et  al,  2008)  was  shortened  and  the  part  

Equivalent of “net earnings” in the previous study

Provide context on “how far away” other groups are

Page 28: Energy olympics

Lighting

Computers

Total

�30%

�20%

�10%

0%

%reductionfrom

baseline

Mean percentage savings and ranges

Page 29: Energy olympics

-40%

-27%

-13%

0%

13%

27%

14/0

7/20

1222

/07/

2012

29/0

7/20

1205

/08/

2012

12/0

8/20

1219

/08/

2012

26/0

8/20

1202

/09/

2012

09/0

9/20

1216

/09/

2012

23/0

9/20

1230

/09/

2012

07/1

0/20

1214

/10/

2012

21/1

0/20

1228

/10/

2012

% sa

vings

from

bas

eline

Computer Use Lighting

End of competition period

The first data point is the average saving after one week

TabletsRemoved

Average savings over all groups split by end-use

Page 30: Energy olympics

The  effectiveness  of  behavioural  interventions    to  reduce  office  energy  consumption  

 

68/141    

P1).  Figure  4.3.1   illustrated  that  the  percentage  of  strongly  agree  people   increased  after  

the   competition,   whereas   before   the   answer   ‘agree’   was   greater.   The   questions   that  

confirm   the   environmental   worries   such   as   about   overpopulation   (Question   2),   human  

activities   consequences   (Questions   4,   6),   and   the   rights   of   other   species   (Question   8)  

were  about   the   same   level  with   the  difference  by   2-­3%   either,   lower   or   higher.  Bear   in  

mind  that  the  proportion  of  response  is  slightly  lower  this  time  it  can  be  assumed  that  the  

general  level  of  the  awareness  has  not  changed  much.  

 

Figure  4.3.1.  Comparison  of  responses  before  and  after  the  competition  on  Question  1.  

In  Building  1  on  the  Ground  and  First  floors  the  amount  of  positive  responses  was  lower  

by   3-­5%,   whereas   on   the   winning   floor   (the   Second)   the   amount   of   positive   answers  

reached  100%,  higher  by  18.75%  than  before  the  competition  (Appendix  P,  Table  P2).  

In  Building  2  the  change  is  slightly  more  significant  (Appendix  P,  Table  P2).  On  the  First  

and  the  Third  floors  awareness  was  higher  by  about  25%,  whereas  on  the  Second  floor  it  

was  lower  by  14%.    

The  questions  that  represent  environmental  attitude  (2,  4,  6,  8)  in  general  received  slightly  

less   positive   response   from   both   buildings   (Appendix   P,   Table   P2).   However,   for   the  

Second  Floor  of  Building  1  (winner)  some  questions  were  answered  more  positively,  such  

as  Question  4   (by  36%)  and  Question  6   (by  21%)  whereas  other  questions   (2,   8)  were  

about  the  same  level.    

0.00%5.00%

10.00%15.00%20.00%25.00%30.00%35.00%40.00%45.00%50.00%

1(Stronglydisagree)

2 3 4 5 6 7(Stronglyagree)

Q1.  I  am  an  environmentally  aware  person

before  competition

after  competition

The  effectiveness  of  behavioural  interventions    to  reduce  office  energy  consumption  

 

69/141    

However,  questions  about  humans’   rights  and  abilities   to  prevent   the  destruction  of   the  

Earth  (3,  5,  7)  in  both  buildings  were  more  optimistic  than  in  the  previous  questionnaires  

and  the  percentage  of  ‘agree’  answers  was  by  8-­12%  higher.  

4.3.2.   Behavioural  change.  

Interestingly,   the   answers   about   behavioural   actions   performed   by   occupants   have  

significantly   improved   in  both  buildings  by  an  average  15.6%  (Figure  4.3.2,  Appendix  P,  

Table  P2).  The  most  significant  was  improvement  in  turning  off  lights  overnight  by  50.25%  

on  the  Ground  floor  of  the  Building  1  and  switching  off  monitors  on  the  Second  floor  of  the  

Building  2  that  grew  from  25%  to  71%  and   the  use  of   ‘sleep’  mode   that   increased  form  

33%  to  71%.  

 

Figure  4.3.2.  The  comparison  of  positive  responses  about  ecological  behaviour.  

Nevertheless,   there   were   a   few   reductions   observed   in   the   performing   of   ecological  

behaviours  such  as  in  adjusting  lighting  level  (Question  9)  on  the  Second  floor  of  Building  

2  and  on  the  First  floor  of  Building  1  (by  15%  and  30%  respectively)  (Appendix  P,  Table  

P2).   The   Second   floor   in   Building   2   reduced   also   the   number   of   positive   response   in  

turning  lights  off  by  25%.  Moreover,  the  First  floor  of  Building  1  decreased  slightly  (by  5%)  

the  amount  of  positive  response  about  turning  off  displays.  

It  has  to  be  borne  in  mind  that  due  to  the  difference  in  number  of  respondents  the  results  

are  not  precise.  However,  the  general  conclusion  might  be  drawn  that  the  improvements  

in  ecological  behaviour  took  place  during  the  competition.  

4.3.3.  Energy  competition  and  occupants’  engagement.  

0.00%

20.00%

40.00%

60.00%

80.00%

100.00%

120.00%

9.  adjustlights

10.  swtich  offcomp

11.  turn  offlights

12.  use'sleep'  mode

13.  switch  offmonitor

Building  2beforeBuilding  2afterBuilding  1beforeBuilding  1after

Page 31: Energy olympics

ANSWERS• Simply putting something in place seems to have an effect•We need to be wary, however, of the Hawthorne effect•What people say and what they do don’t always match• Some designs may be better than others

WHAT WE DON’T KNOW• How do hybrid designs perform?• Are these savings sustained? How can we make them habitual?•What is an appropriate baseline measure?• Are rewards essential? Need they be monetary?