Motivate Environmentally Sustainable Thermostat-Use through Goal-Setting, Just-In-Time Recommendations, and Behavior Reflection Christian Koehler Anind K. Dey M-ITI, University of Madeira HCI Institute Campus da Penteada Carnegie Mellon University 9020-105, Funchal, Madeira Pittsburgh, PA 15213 USA christian@m-iti.org anind@cs.cmu.edu Jennifer Mankoff Ian Oakley HCI Institute M-ITI, University of Madeira Carnegie Mellon University Campus da Penteada Pittsburgh, PA 15213 USA 9020-105, Funchal, Madeira jmankoff@cs.cmu.edu ian@uma.pt ABSTRACT systems provide quality of life improvements at the cost of Rising power demands resulting from technological advance- placing strain on limited global resources. As recognition ments is an increasingly important global issue. One way and awareness of this trend have grown, there have been to tackle this problem is to motivate individual behavior increasing calls for citizens to use resources responsibly. In- change, for which the ubiquity of mobile phones offer an vestigations into how energy is consumed in homes provide ideal platform to influence consumption behavior of users. valuable suggestions for how this goal can be achieved and In this paper we explore the possibilities for using timely indicate that temperature regulation systems (such as fur- recommendations, goal-setting, immediate feedback, and vi- naces and air conditioners) are a key target. They are re- sualization of past consumption behavior in order to mo- ported to be responsible for nearly 25% of the consumption tivate people to reduce power consumption resulting from in an average American household [7]. heating/cooling devices. We describe a mobile application Despite the high-energy consumption (and costs) incurred which gives the user direct access to the thermostat and by these systems, studies indicate that there are problems provides feedback everyday on how sustainable the user was with devices intended to increase their efficiency. For in- on the previous day. In addition to this feedback, it gives stance, programmable thermostats are capable of decreasing recommendations to improve the behavior and also offers a the energy use of a temperature regulation system by re- behavior overview. The contributions of this paper include a laxing the maintained temperatures during particular time working system for remote control over the thermostat and periods, such as when homes are empty during the work- a goal-setting, recommendation, and feedback application day. However, 14% of users are reported to not own such a designed to influence a user’s behavior. system and, of those that do possess one, over 40% do not use it [8]. Widespread misconceptions also exist regarding suitable home temperatures. For instance, a 2007 interview Categories and Subject Descriptors study [5] reported that 41% of interviewees believed that H5.m. [Information interfaces and presentation]: e.g., room temperatures should be lower in summer than in win- HCI ter. The authors concluded this results in overly cool rooms in summer wasting a lot of power. Keywords These findings also suggest that, generally, people have poor awareness of how their consumption choices and be- Eco-feedback, Sustainability, Environmental HCI havior affects the environment. Sensing technologies capable of capturing a user’s activity can be combined with digital General Terms displays of consumption (e.g., websites, ambient displays, Design, Human Factors mobile devices) to address this issue and raise awareness of the impact of particular choices. Such systems and devices, 1. INTRODUCTION which are intended to inform users about the current state Technological advancements over the past few decades of their consumption and their impact on the environment, have allowed us to live more comfortable lives at the cost are known as ”eco-feedback technology” [3]. Eco-feedback of consuming increased amounts of energy. Devices like re- technology can be used to motivate users to change their frigerators, air conditioning units, or home entertainment long-term behavior. Research in psychology can inform the development of sys- tems that change a user’s behavior. In a recent survey paper, Copyright is held by the author/owner(s). He et al. [4] discussed a wide range of psychological lit- MobileHCI 2010 September 7-10, 2010, Lisboa, Portugal. ACM 978-1-60558-835-3. erature and developed a ”motivational framework based on the Transtheoretical (or Stages of Behavior Change) Model”. this goal was met the day before, the current week, or the One of their conclusions was that behavior change happens current month. We provide the user with the challenging in stages and that each stage should be supported by quali- goal of reducing at-home temperature to 65°F, because pre- tatively different kinds of feedback. vious work has shown that more attainable goals, although An earlier review of environmental psychology papers per- easier to achieve, have lower rates of success [1].The system formed by Abrahamse et al. [1] evaluated the effectiveness is designed to inform the user and motivate him to change of two intervention strategies - antecedent and consequence his long-term behavior. We intentionally combined goal- strategies. They described antecedent intervention to ”influ- setting with feedback techniques and tailored information ence one or more determinants prior to the performance of to increase the chances that people will reduce energy con- behavior” and considered four techniques: commitment, goal sumption. setting, information, and modeling. One of the conclusions Through a mobile phone application, the user is offered they drew was that presenting the user with one of these in- remote control over the thermostat. This allows error re- terventions alone is not sufficient, but for example combining covery in case the user forgot to set the thermostat before them with other techniques such as feedback proved effec- leaving the house. In addition to supporting manual and tive in influencing behavior. Consequence strategies on the programmatic control of the at-home temperature, we also other hand are based on the assumption that providing users offer automatic control over the thermostat, reducing energy with positive or negative consequences will influence behav- consumption when the user is not at home, and returning ior. They analyzed how different forms of feedback (e.g., the at-home temperature to a desired level when the system continuous, daily, weekly, monthly, or comparative) and the predicts the user is returning home. In doing so, the system presence of real world rewards influence behavior.They con- aims to support the user in setting the right temperatures. cluded that individual feedback combined with comparative The advantage the system offers a user is not only er- feedback provided also energy reduction in the long run. ror recovery through remote control, but also feedback if Rewards on the other hand had only a short-term effect ac- a behavior was environmentally positive or negative. Ad- cording to their analysis. This strongly suggests that mak- ditionally it provides positive reinforcement if the behavior ing users aware of their consumption behaviors is an effec- the day before was positive or timely suggestions on how to tive method for encouraging their reduction. This is the improve the behavior in case it was not positive. approach adopted in this work. To support the automated control of the thermostat, we We suggest that mobile devices are an ideal platform on gather continuous location information using a mobile ap- which to offer motivation to users and convey benefits in- plication. Using the current location and past movement compatible with the functionality of stationary devices. For data we are able to determine the time a user is predicted example, we can use sensors provided by mobile phones (e.g., to return home. We leverage a robust location prediction location sensor, accelerometer, etc.) to infer the user’s cur- algorithm developed by our research group, which aims to rent activity and predict future ones. This functionality can predict future destinations based on prior and current move- be used to provide just-in-time feedback about the outcome ment data ([9]). Using this predicted return time, we can au- of actions and provide live recommendations of alternatives. tomatically control the thermostat as described above. The Integrating all this literature, this paper introduces a mo- location information is not only used for predicting a user’s bile application designed to give users feedback about their return time, but also to calculate how much power was con- past behavior, provide timely recommendations to promote sumed by heating/cooling devices during the time the user behavior change related to temperature regulation power is not at home and while he is at home. consumption, and to support users in the achievement of In addition to location information we also collect inside, the long-term goal of consuming less power in temperature outside, and the thermostat’s temperature setting for the regulation while they are not at home. The application we user’s apartment. This information is used to provide feed- developed allows users to remotely and conveniently control back on the user’s behavior and also to give recommenda- the thermostat in their apartment using their phone. We tions on how to improve an environmentally negative be- believe that the system can be deployed most effectively in havior. To understand a user’s reasoning for changing the the motivations stages of Preparation, Action, and Main- thermostat’s set temperature, we also query the user imme- tenance identified in the Transtheoretical model described diately after a manual change to a higher temperature has above. The remainder of the paper provides a description occurred. of the application and outlines the structure of a study that The system was based on a commercial home automation we intend to run as the next stage of our work. system from Insteon [6] and a custom mobile phone appli- cation written on Google Android. Logically it consists of 2. SYSTEM DESCRIPTION three main parts: the home automation system, the mobile phone, and a central server. The following two sections will We designed a system to help people reduce power con- give a short description of the home automation system and sumption resulting from domestic heating and cooling de- then describe the mobile phone application. vices. It informs the user about his behavior by giving an overview over past consumption in the form of a graph. Studies have shown [1] that providing the user with tai- 2.1 Home Automation System lored information about their consumption can result in re- Home automation - the use of small modules to extend ap- duced energy consumption. Through timely recommenda- pliances with remote control and automation features - is a tions on how to save energy, the system aims to help the commercially available technology. The Insteon system used user’s decision-making process to behave more environmen- in this work allows the remote regulation of home tempera- tally sustainably. In order to further motivate the user, he is ture and also provides support for the calculation of power given a savings goal and the application indicates how well consumption. Calculating the power consumption of heat- ing or cooling devices (e.g., furnaces or air conditioner) is a complex problem, because it not only depends on the length of time a device is active, but also on a set of complex vari- ables including: the efficiency of the heating/cooling device, the volume of the home space, the current inside and out- side temperature, the thickness and material of the walls, the number of walls exposed to outside conditions, and the number and kind of windows in the walls, to name a few issues. Because measuring these data is challenging for in- dividual users the system presented in this paper used a simplified set of calculations using an equation provided by the U.S. Department of Energy: hdd AHC(city) = Cadj ∗ ( hddcity us ) ∗ (1 + S ∗ (TwAvg − Ttyp )) Legend: Figure 1: Screenshot of Graph Overview Screen. AHC(city) = Annual Heating Costs for city Cadj = Adjusted consumption (gas furnace) in mBTU concerns are often raised with novel technology incorporat- hddcity/us = Average number of Heating Degree Days ing automation or context sensing control and can result in S = Savings per degree users developing negative opinions about the systems. By of setback temperature in percent providing a manual override, the system presented in this TwAvg = Set temperature as weighted average paper hopes to avoid this problem. In addition to remote Ttyp = Typical indoor temperature control we also ask the user why he increased the tempera- during heating season ture in the apartment, because we also strive to understand a user’s reasoning it. This equation assumes standard values for some of these 2.4 Graph Overview variables and uses differences between inside and outside The graph overview is one of the core feedback mecha- temperature and the number of hours the user was at home nisms in the application. As illustrated in Figure 1, it and away from home to calculate the power consumption. offers users a consumption overview split into daily, weekly, We believe the results of this calculation to be sufficiently and monthly activity. Each sub-graph includes a goal line indicative of real consumption to serve as effective feedback. highlighting the difference between intended and achieved levels of consumption. This is highlighted using a simple 2.2 Mobile Phone Application red, yellow, and green color scheme. This visualization is The mobile phone application was developed and deployed designed so that users can see at a glance their performance on T-Mobile G1 phones running Google Android. It consists over days, weeks and months. By including these longer pe- of several components and is the core of the system described riods of time, the system aims to convey a sense of mounting in this paper. Its key function is to collect and transmit its achievement and provide motivation to continue with and/or current location to a central server in real time and support improve sustainable behaviors. In order to further support the motivational techniques described in the beginning of this and provide positive reinforcement, a small smiley face this chapter. Using the transmitted location it enables us to is shown if users achieve green behavior for more than half estimate whether or not a user is at home, or whether they of the days in a particular week or month. As described in are likely to return home in the near future. Furthermore we [4], individuals profit from positive reinforcement of their ac- are able to calculate the power consumption while the user tions, which gives them a feeling of success and competence. is at home or out and about in the world. The application The graph system is designed to provide such feedback. provides an interface composed of three main components: a temperature control screen which allows remote control 2.5 Recommendations over the home automation system, a graph overview showing The recommendations screen aims to provide recommen- past behavior, and a recommendation screen. Each of these dations to users on how to achieve consumption goals through components is introduced in the following subsections. behaving sustainably. Once again, to present a clear visual representation we adopted a simple red, yellow and green 2.3 Temperature Control color scheme for this UI. The user is shown a green screen This interface component allows users to view and con- when he behaves sustainably the previous day, by turning trol the current temperature in their homes. This feature down the temperature while away from home to at least 60°F was intended to support error recovery (for instance, by al- and turning down the temperature to 65°F while at home. lowing correction after forgetting to adjust the temperature If the user sets a higher temperature while at home, he is settings prior to leaving home) but also to provide a sense shown a yellow screen. If the temperature is consistently of security and control - to reassure users that they remain high throughout the day, the system will show him a red in control of the system even though it incorporates signifi- screen. For the yellow and red screen we give the user a rec- cant automated elements. Such privacy and lack of control ommendation on how to achieve sustainable behavior, which is in essence an explanation about why his behavior was not give the user information about his environmental behavior sustainable. We also suggest wearing additional clothing to and recommend actions to change environmentally bad be- compensate for cool temperatures at home. havior, we measure the locus-of-control and self-efficacy of In addition to recommendations and behavior indications a user at the beginning, middle, and end of the study. This we also inform users about the environmental impact of the information will tell us if there was a change, for exam- wasteful behavior of leaving temperature regulation systems ple, from external locus-of-control, the state where outside active in empty homes. This is done simply by calculating sources influence events happening in our life’s, to inter- and presenting the number of 60W light bulbs that could nal locus-of-control, the state where the user himself affects have be powered by the wasted energy. events. Measuring self-efficacy will tell us if our system had The recommendations screen is the first screen shown to a self-empowering effect on users, where they increasingly the user upon application startup. Every morning the phone got confident to influence their environmental behavior. notifies the user about the previous days consumption data by flashing the onboard LED and vibrating. The goal of this 4. SUMMARY feedback is to provide positive reinforcement for sustainable In this paper we first gave an introduction into the prob- behavior and to inform users about the impact of their be- lem space of home power consumption and explained some havior on the environment. In this way, we hope to pro- of the problems with the current use of thermostats. Addi- vide an incentive for users to alter unsustainable behaviors. tionally we highlight how research results from psychology Our application informs users about problematic behaviors, can help to influence a user’s behavior and reduce power highlights the impact of these behaviors, and gives specific consumption resulting from heating/cooling devices. recommendations on how to enact change. By showing this We then described a working prototype of a mobile appli- information in the mornings, prior to regular departures to cation designed to influence a user’s sustainable practices. work, we hope to be able influence the user’s behavior in Our prototype allows remote control and automated con- a timely and appropriate fashion. Studies have shown [2] trol over the thermostat, gives the user a goal to achieve, that frequently updated feedbacks with multiple feedback provides daily feedback on how well this goal was met, rec- options such as comparisons of several days or providing ad- ommendations to achieve the goal (or positive reinforcement ditional information sources like recommendations are most in case the goal met), and an overview over past consump- effective. tion behavior. Our immediate next work is to conduct a study of this prototype with 20 people in the fall. 3. STUDY DESCRIPTION In order to test the system and evaluate whether or not 5. REFERENCES it can influence a user’s behavior and awareness of environ- mental issues, a field study of the system is planned. This [1] Abrahamse, W., Steg, L., Vlek, C., and will take place in winter 2010/2011 and will use a group Rothengatter, T. A review of intervention studies of 20 recruited participants residing in a city which expe- aimed at household energy conservation. Journal of riences sub-zero temperatures and significant snowfall for Environmental Psychology. several months. The group will be split into 2 sub-groups, [2] Fischer, C. Feedback on household electricity with each group initially being in a manual condition where consumption: a tool for saving energy? Energy the user has to change the temperature on his own volition Efficiency 1. or an automatic condition where the system automatically [3] Froehlich, J., Findlater, L., and Landay, J. The sets the away-from-home temperature The study will be sep- design of eco-feedback technology. Proc. CHI 2010. arated into two phases and each phase will go for 4 weeks, [4] He, H. A., Greenberg, S., and Huang, E. M. One with a study length of 8 weeks in total. After 4 weeks the size does not fit all: Applying the transtheoretical two sub-groups will switch conditions. model to energy feedback technology design. Proc. CHI Prior to the study, GPS location data will be captured 2010. from each user and used to train the location prediction [5] Karjalainen, S., and VastamLki, R. Occupants model and to calculate baseline consumption data. In order Have a False Idea of Comfortable Summer Season to measure environmental attitudes and temperature com- Temperatures. Proc. Clima. fort levels both before and after the study participants will [6] Smarthome. Insteon Home Automation Systems. be required to fill out a survey every week. The survey is http://www.insteon.net/. split into two categories, a temperature survey that will be [7] US Dept. of Energy. Energy information distributed weekly and a survey to measure environmental administration, US Household Electricity Report, 2005. attitude, locus-of-control, and self-efficacy that will be con- www.eia.doe.gov/emeu/reps/enduse/er01u s.html. ducted at the beginning, middle, and end of the study. The [8] US Dept. of Energy. Energy Information temperature comfort survey will give us data on how the Administration, international energy outlook, 2008. subjective comfort of the user changes through use of our www.eia.doe.gov/oiaf/ieo/enduse.html. system. We believe that a successful application that aims [9] Ziebart, B. D., Maas, A. L., Dey, A. K., and at changing a users environmental behavior also has to con- Bagnell, J. A. Navigate Like a Cabbie: Probabilistic sider a user’s comfort level. A system that disregards an Reasoning from Observed Context-Aware Behavior. individual’s subjective comfort level could be unsuccessful Ubicomp. because the user rejects it. One goal of our system is to change a user’s environmental behavior, thus we need a way to evaluate the impact our two versions of the system have on the environmental awareness of the users. Because we also