upa - home page JUS - Journal of usability studies
An international peer-reviewed journal

Making Energy Savings Easier: Usability Metrics for Thermostats

Daniel Perry, Cecilia Aragon, Alan Meier, Therese Peffer, and Marco Pritoni

Journal of Usability Studies, Volume 6, Issue 4, August 2011, pp. 226 - 244

Article Contents


Introduction

Thermostats have a considerable effect on energy use throughout the U.S. Research on programmable thermostats (PTs), in particular, points to the need for careful and consistent user interface design to realize energy savings in the U.S. (Meier et al., 2010). Recent studies estimate that U.S. residential thermostats control 9% of the nation’s energy use (Peffer, 2011). However, usability issues with modern PT design are leading to errors in operation and wasted energy, with a resulting effect on national energy consumption.

The EnergyStar™ endorsement program for PTs, which had been in place since 1995, was recently discontinued in December 2009 based on these and other results; EnergyStar™ is currently developing usability guidelines for PTs. A key issue in the establishment of these guidelines is the development of a means of measuring usability in thermostats that produces a single number that consumers can use to reliably and repeatedly compare device usability.

We conducted a usability study on five commercially-available residential programmable thermostat interfaces (three touchscreen, one Web, and one button-based), evaluating device usability and effectiveness. Our goal was to devise metrics to evaluate the devices’ usability and the users’ effectiveness at performing common thermostat tasks such as setting heating temperatures and programming weekly schedules. We developed, computed, and tested four novel metrics appropriate to thermostats and similar devices: Time & Success, Path Length, Button Mash, and Confusion. We conducted a statistical evaluation of these metrics and a comparison to standard usability metrics, including the efficiency metric specified by the Common Industry Format for Usability Test Reports (National Institute for Standards and Technology [NIST], 2001), the ratio of the task completion rate to the mean time per task, and Sauro and Kindlund’s (2005) Single Usability Metric (SUM).

We show that all metrics correlated with each other and corresponded with qualitative findings. Finally, we compared the new metrics with standard industry metrics.

One drawback of existing metrics is that many are not normalized to an absolute scale, but vary based on factors such as the maximum time taken by a user to perform a task. By applying the logistic function (Verhulst, 1838), our metrics were all normalized to the scale 0-1 and thus provided an absolute rather than a relative reference. Additionally, our four metrics offer manufacturers and standards organizations several options to compare device usability with a high degree of statistical significance.

We evaluated the metrics with a formal usability study conducted on five programmable thermostats (three touchscreen, one Web, and one-button based) with 31 participants and 295 trials involving five separate tasks.

In our usability test, we found that several of the PT interfaces were complicated and difficult for users to understand, leading to frustrations and major barriers for completing the tasks. Our metrics were able to clearly and objectively distinguish between more usable and less usable PT interfaces.

This paper reports the results of one of the few formal usability studies ever conducted on PTs, as well as the development and evaluation of four novel usability metrics specific to thermostats, appliances, and similar devices. These metrics could be applied to any user interface on a small screen with a relatively small number of buttons or a touchscreen.

Programmable Thermostats

Modern programmable thermostats (PTs) can be scheduled to automatically adjust the indoor temperature for heating or cooling during occupied hours as well as unoccupied or sleeping hours. The adoption of PTs has been strongly supported by the U.S. Department of Energy (DOE), the U.S. Environmental Protection Agency (EPA), and the California Energy Commission (CEC)1. The DOE estimated the average homeowner can save 10% on heating and cooling costs by using a PT to reduce heating and cooling during the night or periods when the house is unoccupied (U.S. Department of Energy [DOE], 2011). The EPA claims homeowners could save about $180 a year with a PT (2009). These predictions are qualified with terms like “effectively used” or “properly setting and maintaining those settings.” However, EPA’s EnergyStar™ program for PTs, which had been in place since 1995, was recently discontinued in December 2009. One of the reasons for this decision is that several recent field studies have shown no significant savings in households using PTs compared to households using non-programmable thermostats; indeed, some studies even showed that homes with PTs used more energy than those relying on manual thermostats (Cross & Judd, 1997; Haiad, Peterson, Reeves, & Hirsch, 2004; Nevius, & Pigg, 2000; Shipworth et al., 2010).

There is increasing evidence that many people do not operate PTs in an optimal manner, leading to unnecessarily high heating and cooling energy use. The user interfaces of many thermostats appear to be a major cause of confusion and errors leading to incorrect settings, failure to override programs, and failure to return to regular schedules after exceptions. Several surveys have shown that approximately half of installed PTs in the U.S. are in “hold” mode, which disables the programmed schedule (California Energy Commission [CEC], 2004; Decision Analyst, 2008). While some studies suggest part of the problem lies in misconceptions about energy in general and how thermostats work in particular (Rathouse & Young, 2004), many reveal that people find PTs difficult to program and to understand (Boait, & Rylatt, 2010; Consumer Reports, 2011; Critchleya, Gilbertsona, Grimsleya, Greena, & Group, 2007; Karjalainen & Koistinen, 2007; Meier et al., 2010; Nevius, & Pigg, 2000).

Most of these studies, however, are qualitative; little quantitative information is available on how people interact with these temperature and environmental controls. To our knowledge, the only comparative usability study on commercially available PTs was conducted by Consumer Reports in 2007 (Consumer Reports, 2011)2. Twenty-five different thermostats were lab-tested to assess their energy performance and their usability. As a result, PTs were ranked according to these criteria, and a series of problems with using thermostats were highlighted. Consumer Reports does not explicitly state what parameters were considered to assess thermostat usability, and it does not appear that quantitative tests were performed.

Manufacturers of PTs have traditionally preferred prescriptive guidelines to a formal usability testing process. One concern is that such usability evaluations may be overly subjective. In order to address this concern, we developed and tested a set of four consistent, normalized metrics specifically designed for devices such as PTs.

Related Work

While there are numerous models to measure and benchmark usability (Tullis & Albert, 2008), there has been little research on whether these metrics can be effectively applied to programmable thermostats and appliances or similar types of embedded devices (computing systems designed to perform only a specific dedicated task).

Babiker, Fujihara, and Boyle (1991) calculated a usability metric that combined objective and subjective measures, yet this metric was specific to hypertext systems.

Smith (1996) derived more robust metrics for measuring efficiency, confidence, and “lostness” (disorientation in an information space) of users of hypertext systems. These metrics offered several ratios comparing the number of different nodes (pages) necessary to complete a task to the actual number of nodes the user accessed. Such ratios, while helpful in normalizing the metrics, possess challenges when applied to thermostats or other embedded devices. Most notably, distinguishing different nodes does not translate easily to systems where the same functions might have different meanings depending on the state of the device.

Otter and Johnson (2000) built upon Smith’s work by adding link weightings to create what they depict as a more accurate hypertext lostness metric. These weightings are specific to the nature of hypertext and again rely on the different node paths.

More recently, Sauro and Kindlund (2005) devised a single summated usability metric (SUM) that combined objective and subjective metrics based on the ISO/ANSI dimensions of effectiveness, efficiency, and satisfaction to form a single score. Testing of this metric was done across several Windows and Web-based platforms, yet was not extended to other devices. The authors did note that their tests were domain specific and that there were possible limitations of the metric when applied to other interfaces or hardware (Sauro & Kindlund, 2005).

Murphy (1998) made necessary distinctions between usability considerations for embedded user interfaces and desktop applications. He noted that the embedded interface functions as a tool with specific ends and that the interface was vying for limited and diverse user attention spans.

The limited work that does reference the usability of embedded devices focuses on prescriptive design principles. Such principles include display size and touchscreen affordances (Murphy, 2001). While prescriptive evaluation methods are a helpful start for the designers of embedded systems, they can also stifle creative processes and these principles must continually be modified to keep pace with ever changing technology.

Given the state of the art of embedded system usability, we determined that there was a need for metrics specific to thermostats and similar devices, and thus designed and conducted a formal usability study on PTs to evaluate such metrics.


1The California Building Standards Code has required the installation of a setback or programmable thermostat in new and renovated residential construction since 1978.

2While PT manufacturers often state that they perform usability tests for their products, they do not disclose results because they consider the user interface a key marketing feature.

 

Previous | Next