=Paper= {{Paper |id=Vol-1533/paper2 |storemode=property |title=Saving Energy in 1-D: Tailoring Energy-saving Advice Using a Rasch-based Energy Recommender System |pdfUrl=https://ceur-ws.org/Vol-1533/paper2.pdf |volume=Vol-1533 |authors=Alain Starke,Martijn C. Willemsen,Chris Snijders |dblpUrl=https://dblp.org/rec/conf/dmrs/StarkeWS15 }} ==Saving Energy in 1-D: Tailoring Energy-saving Advice Using a Rasch-based Energy Recommender System== https://ceur-ws.org/Vol-1533/paper2.pdf

Saving energy in 1-D: Tailoring energy-saving advice
using a Rasch-based energy recommender system

Alain Starke, Martijn C. Willemsen, Chris Snijders

Eindhoven University of Technology, Human-Technology Interaction Group
P.O. Box 513, 5600 MB Eindhoven, The Netherlands
{a.d.starke, m.c.willemsen, c.c.p.snijders}@tue.nl

Abstract. Although there are numerous possibilities to save energy, conserva-
tion initiatives often do not tailor their content to the consumer. By considering
energy conservation as a one-dimensional construct, where different behaviors
have different execution difficulties, we have set out a Rasch-based energy
recommender system that provides tailored conservation advice to its users.
Through an online choice experiment among 196 users, we found that users
prefer energy-saving measures that fit their Rasch-profile, rather than ones that
fit their conservation attitude.

Keywords: Recommender systems, energy advice, Rasch model, energy effi-
ciency, energy curtailment

1 Introduction

Initiatives that promote energy conservation, such as mass-media campaigns, often
fail to effectively persuade individuals to change their energy-saving behavior [1,2,3].
A main cause for this is that such initiatives do not tailor their content to individual
consumers, e.g. through tailored advice or feedback [3], but are rather general instead
[4]. An additional shortcoming is that providing general information to consumers
leaves them unaware of all the possible conservation measures [2,5].
Recommender systems can overcome these issues by tailoring advice to users
based on their choices, preferences, and behavior [6]. Energy recommender system
research has already pointed out the effectiveness of tailoring choice interfaces to
users’ knowledge levels by adapting the method of preference elicitation [7], and
showed that increased levels of user satisfaction lead to more energy savings [8].
Although such an adaptive interface ensures compatibility with a recommender
system user’s goals and knowledge level [8], it does not personalize the conservation
advice itself. How advice should be tailored is unclear, as researchers disagree of the
dimensionality of energy conservation and are inconsistent in their findings [9].
To tailor energy-saving advice, we will explore how to conceptually differentiate
between energy-saving measures and subsequently perform a user experiment using
an energy recommender system.
2 Dimensionality of energy conservation

The heterogeneity of conservation measures has led to various conceptual differentia-
tions of energy-saving behaviors [9,10]. The most dominant conception of energy-
saving dimensionality is a two-dimensional approach, differentiating between effi-
ciency and curtailment [1,3,5,11]. Efficiency comprises one-time investments in home
equipment, such as installing double-glazed windows, while curtailment involves
reductions in energy-related behaviors, such as lowering one’s thermostat.
A few authors have taken a different view on the dimensionality of energy-saving
behavior [10]. Rather than discerning between behaviors based upon the nature of
their activity, they argue that we must conceptualize energy-saving behavior as goal-
directed behavior [10,12], in which conservation behaviors form a specific class per-
taining to a single goal, saving energy, and that an individual’s willingness to reach
that very goal can be revealed by the behavioral steps that person is willing to take
[12]. For instance, if one commits to execute a behavior carrying large costs, such as
installing a solar boiler (cf. figure 1), one will also be likely to perform a behavior
with fewer costs, such as turning off the lights after leaving a room [10,12].
Fig. 1. An example quantification of a few energy-saving behaviors and the costs they carry,
compared to the distribution in attitudes toward energy conservation of a group of individuals.

The values depicted above were derived using the Rasch model [12], a model com-
monly used in psychometrics [13]. Rasch provides a mathematical formalization of
the theory of goal-directed behavior, equating the costs δ of a behavior i with the con-
servation attitude θ of an individual n in a probabilistic model (cf. equation 1) [12].

ln (1)

By quantifying behaviors in terms of costs and differentiating between them, as well
as gauging the different propensities to save energy of a group of individuals, it is
possible to tailor energy-saving advice based on this common dimensionality.
We have two main research expectations. First, in contrast with curtailment and ef-
ficiency, we expect that energy-saving behaviors form a one-dimensional scale. Sec-
ond, we expect that tailored energy-saving advice, i.e. behaviors that match a user’s
attitude, are perceived as more appropriate than those that are easier or more difficult.

3 Creating a one-dimensional scale of energy-saving measures

We performed a pre-study to fit a one-dimensional scale of conservation measures.
263 participants interacted with our conservation web-tool, containing 88 energy-
saving measures. Each participant had to indicate which measures they already exe-
cuted by either responding ‘yes’, ‘no’, or ‘does not apply’ to each measure presented.
After controlling for misfit persons and items [cf. 13], we fitted a scale of 79 ener-
gy-saving measures with medium to high reliability, ranging in difficulty levels from
-5.73 to 5.49 (M = 0.06; SD = 2.14). In line with our expectations, curtailment and
efficiency measures were mapped onto a one-dimensional scale, with curtailment
bearing less behavioral costs than efficiency measures (Mcur = -0.67; Meff = 1.03).

4 Method - Energy recommender system user experiment

We used the constructed scale in an online energy recommender system to estimate
users’ abilities and recommend them tailored conservation measures accordingly.
Each user had to indicate for 13 semi-randomly sampled energy-saving measures
whether he already executed them, by either responding ‘yes’, ‘no’, or ‘does not ap-
ply’. Using their answers, we estimated user attitudes and provided them two tailored
lists of nine energy-saving recommendations, whose execution difficulty levels were
either 1 logit above, equal to, or 1 logit below the user’s estimated attitudinal level.
To test which relative difficulty level is perceived as most appropriate, each user
had to rank-order both lists in preference order, placing the preferred measures at the
top. Users were only required to rank-order items that they did not already execute.

5 Results

Our web-tool was distributed among the members of the participant database of the
virtual lab at Eindhoven University of Technology. 196 users (51.6% female; Mage =
27.3 years) completed our user experiment.
To test whether users perceived the tailored energy-saving measures as the most
appropriate, we performed multiple rank-ordered logistic regression analyses on the
ranked-ordered lists. Our analyses indicated that the relative difficulty level of a con-
servation measure had a significant effect on a measure’s rank-order position. Contra-
ry to our expectations, we found that relatively easy measures were perceived as the
most appropriate, topping the rank-ordered lists (p < 0.001). This effect was rather
linear: the mean relative difficulty level per ranked position increased while moving
down the list. In other words, users preferred easy measures over more difficult ones.
6 Conclusions & future work

We have demonstrated two things: First, a diverse set of conservation measures can
be mapped onto a one-dimensional scale according to the measure’s difficulty levels.
Second, from the provided tailored recommendations, users perceived the relatively
easy ones as the most appropriate, suggesting that good energy-saving recommenda-
tions should fit a user’s Rasch profile, rather than its attitude. This confirms the va-
lidity of the Rasch model, as individuals tend to exert more low-cost behaviors [12].
Future research must explore two things. First, how Rasch-based, energy-saving
recommendation sets compare to non-personalized baselines, such as a set that con-
sists of the most popular items (i.e. the easiest ones in a Rasch scale). Such a compari-
son should not only be made through choice, but also in terms of user experience
concepts, such as system satisfaction [14]. Second, while we have employed a choice-
support system, future systems must be more persuasive and use conservation nudges.

References
1. Abrahamse, W., Steg, L., Vlek, C., Rothengatter, T.: A review of intervention studies aimed
at household energy conservation. J. Environ. Psychol. 25, 273–291 (2005).
2. Midden, C.J.H., Kaiser, F.G., Mccalley, L.T.: Technology’s four roles in understanding in-
dividuals' conservation of natural resources. J. Soc. Issues. 63, 155–174 (2007).
3. Steg, L.: Promoting household energy conservation. Energy Policy. 36, 4449–4453 (2008).
4. Mckenzie-Mohr, D.: Promoting Sustainable Behavior: An Introduction to Community-
Based Social Marketing. J. Soc. Issues. 56, 543–554 (2000).
5. Gardner, G.T., Stern, P.C.: The Short List: The Most Effective Actions U.S. Households
Can Take to Curb Climate Change. Environ. Sci. Policy Sustain. Dev. 50, 12–25 (2008).
6. Burke, R.: Hybrid Recommender Systems: Survey and Experiments. User Model. User-
adapt. Interact. 12, 331–370 (2002).
7. Knijnenburg, B.P., Reijmer, N.J., Willemsen, M.C.: Each to his own: how different users
call for different interaction methods in recommender systems. Proceedings of the fifth
ACM conference on Recommender systems (pp. 141–148). ACM. (2011).
8. Knijnenburg, B.P., Willemsen, M.C., Broeders, R.: Smart Sustainability through System
Satisfaction: Tailored Preference Elicitation for Energy-saving Recommenders. Full paper
accepted at AMCIS 2014 (2014).
9. Karlin, B., Davis, N., Sanguinetti, A., Gamble, K., et al.: Dimensions of Conservation: Ex-
ploring Differences Among Energy Behaviors. Environ. Behav. 46, 423–452 (2014).
10. Urban, J., Ščasný, M.: Structure of Domestic Energy Saving: How Many Dimensions? Envi-
ron. Behav. (2014).
11. Dietz, T., Gardner, G.T., Gilligan, J., Stern, P.C., Vandenbergh, M.P.: Household actions
can provide a behavioral wedge to rapidly reduce US carbon emissions. Proc. Natl. Acad.
Sci. U. S. A. 106, 18452–18456 (2009).
12. Kaiser, F.G., Byrka, K., Hartig, T.: Reviving Campbell’s paradigm for attitude research.
Pers. Soc. Psychol. Rev. 14, 351–367 (2010).
13. Bond, T.G., Fox, C.M.: Applying the Rasch Model: Fundamental Measurement in the Hu-
man Sciences, Second Edition. Psychology Press (2006).
14. Knijnenburg, B.P., Willemsen, M.C., Gantner, Z., Soncu, H., Newell, C: Explaining the user
experience of recommender systems. User Model. User-adapt. Interact. 22, 441–504 (2012).