Rewarding Energy Engagement Evaluation of Electricity Impacts from CUB Energy Saver, a Residential Efficiency Pilot Program in the ComEd Service Territory Prepared by: Matthew Harding, Assistant Professor of Economics, Stanford University Patrick McNamara, Efficiency 2.0 Abstract This whitepaper evaluates online and printed report savings from CUB Energy Saver, an energy efficiency pilot program created by Efficiency 2.0 and the Citizens Utility Board (CUB) of Illinois to generate energy savings for Commonwealth Edison (ComEd) customers. The CUB Energy Saver program engages customers with advanced energy savings recommendations, feedback and rewards. We have analyzed the program over a 12-month pilot period from July 2010 to July 2011. Through multiple evaluation methods a treatment effect of 6.01 percent is observed for online participants and 1.47 to 1.63 percent for mailer-only participants. Experimental Design and QuasiExperimental Design, as well as other forms of Large-Scale Data Analyses, are explored and employed as the optimal protocol for program evaluation. Efficiency 2.0 provided raw electricity usage and household characteristic data to Matthew Harding, Assistant Professor of Economics, Stanford University, which was accessed through a partnership with CUB and ComEd. All data was scrubbed of personally identifiable information. Table of Contents • I. Introduction (p. 2). Introduces the program by outlining the objectives, the behavioral research in which those objectives are based and how the program will be evaluated. • II. Program Overview (p. 5). Describes how the program was implemented, including marketing and feedback. • III. Evaluation Protocols (p. 12). Sets forth the specific evaluation methods for both online (“engaged”) and mailer-only (“passive”) treatments. • IV. Program Results (p. 16). Provides a detailed analysis of both engaged and passive savings, as well as supporting process metrics. 1 I. Introduction A. Program Objectives Energy efficiency is an oft-discussed method of alleviating many significant issues at the forefront of economics and policy today: climate change, electrical grid stability and consumer welfare. Not only is energy efficiency identified as cost-effective relative to other options, but in many cases provides a positive net present value (McKinsey, 2009), meaning that regardless of any climate problems, energy efficiency has great potential value. Secondly, energy efficiency has the potential to reduce peak-load concerns, which to consumers means a reduced risk of blackouts and brownouts. Because the efficiency of appliances affects products 24 hours a day, lowering baseline usage and peak load simultaneously and reducing maximum load. Finally consumers benefit directly through reduced utility bills, the extra funds of which they can spend however they please. The Illinois Citizens Utility Board and ComEd funded the CUB Energy Saver program in order to help Illinois ratepayers save energy through information and incentives. The CUB Energy Saver pilot was designed to test the impact of incentives, engagement, and feedback across multiple channels including e-mail, direct mail and the use of an online portal on consumer’s energy usage. In addition, the pilot set out to test the scalability of the approach in order to determine its impact in improving the reliability of the grid and reducing pollutants and emissions. B. Evaluation Methods The CUB Energy Saver program is a hybrid opt-out/opt-in program, where mailer-only participants are assigned (opt-out) and online participants actively choose to participate (opt-in). This paper has analyzed both of these components using Experimental Design (ED) and Quasi-Experimental Design (QED) evaluation methods to evaluate the program’s performance. QED is a sub-variant of ED evaluation that is in all ways similar to ED, except for the lack of true random assignment to treatment and control groups. ED programs are experiments in which group assignment is random and measurements are taken both before and after treatment. This allows for the groups to be as similar as possible and accounts for any minute differences that may exist in the variable of measurement, in this case, electricity consumption. Many things can affect the outcome of the experiment such as maturation, the observation effect, and statistical regression (Sullivan, 2010). With appropriately large sample sizes of similar groups these possibilities for bias can be assumed to be non-factors on the outcome of the experiment. By contrast, in QED programs, participants will opt into the program, potentially exhibiting what is generally referred to as “self-selection bias”. Because these 2 households self-select into the program, they are believed to exhibit some form of bias through characteristics like a desire to participate in so-called “green” programs or online information portals that the rest of the population may not have access to. In most cases the lack of an explicit control group is one of practicality where issues of consumer equity mandate that certain programs be made available to all utility customers or budgets limit the scope of the program. Methods to control for this potential bias include Regression Discontinuity, Non-Equivalent Control Groups, and Interrupted Time Series, variations of which are explored in this paper as an optimal evaluation method for this type of program. For a more detailed discussion on any of the aforementioned methods see “Guidelines for Designing Effective Energy Information Feedback Pilots: Research Protocols,” Section 3, pages 13 to 21. Large Scale Data Analysis (LSDA) is employed for supporting or, “process metrics”, that measure non-energy aspects of the program, including online engagement. These process metrics include time on site, e-mail open rate and page-views per visit. Time on site is an important metric because it describes how much time someone spends interacting with the website. E-mail open rate is a very important metric identifying how responsive the users are to the information and indicates the level of trust and engagement they have with the information. Additionally, the CUB Energy Saver program provides monthly bill analysis and custom savings recommendations through e-mail, so the open rate is a strong measure of engagement a user has with the feedback they receive. Page-views per visit is another important metric similar to time on site; using this information one can determine how long users are spending on different types of information. With that and general assumptions about online reading speed one can determine important metrics like the amount of information consumed by the target audience. C. Behavioral Research Erhardt et al. (2010) reviewed 57 studies on the effects of feedback on energy usage, covering over 30 years from 7 countries. Savings ranged from 4 to 12%, with savings increasing as feedback becomes more direct and timely. They found that “Enhanced Billing” and “Estimated Feedback” strategies were potentially the most cost-effective of strategies because of the low cost relative to “Real-Time” feedback opportunities that usually require in-home hardware. In “Enhanced billing” programs, household specific advice is provided using the home’s energy bills to offer custom insights for customers, showing average savings of approximately 3.8%. The “Estimated Feedback” programs usually included more detailed energy audits and were available on an ongoing basis, providing two-way interaction with the household and the utility, exhibiting 6.8% average savings. They argue that advanced meters alone will not provide the greatest energy savings and that “incorporating the best motivational techniques and behavioral approaches 3 will be important to realize optimal savings”. Additionally, they found that most of the savings resulted from behavioral changes and low-cost upgrades like washing in cold water and installing CFLs as opposed to investments in new equipment. On the issue of persistence, they find that longer studies had slightly lower savings, but that the longer programs also had substantially larger sample sizes, which could be the cause of the more modest savings numbers. However, when conducting a within-studies analysis they find that energy savings are persistent but also that the persistence of savings may require continued feedback. They recommend that the best-performing programs will make feedback “convenient, engaging and beneficial for consumers”. Dowling and Uncles (1997) first explored the concept of rewards programs by trying to find out if they actually were effective in achieving their intended goals. These goals can vary from brand recognition to customer retention, with the overall objective of improving their relationship with their customers. They found that the evidence was somewhat mixed on the effectiveness of loyalty programs driving sales and that loyalty programs may be a mistake in terms of cost-effectiveness if introduced into an already competitive market. However, those that “directly enhance the product/service value proposition, or (ii) broaden the availability of the product/service, or (iii) neutralize a competitor’s program, may be worthwhile.” Yi and Jeon (2003) expanded on Dowling and Uncles’ research and argue that loyalty programs are often misunderstood and/or have been misapplied. They find the effects and value of loyalty programs are dependent upon the level of engagement the customer has with the company. Specifically, they find that high-involvement loyalty programs can positively influence brand loyalty but that under low-involvement programs there is minimal effect on brand loyalty. They also addressed the differences between immediate rewards and delayed rewards. Immediate rewards were things like discount coupons or gift certificates where delayed rewards required long-term loyalty to reach a rewards milestone. They found that delayed rewards were only effective for high-engagement loyalty programs and that immediate rewards were effective for both high and low-engagement programs. 4 II. Program Overview CUB Energy Saver includes four primary components: • • • • Direct and community marketing Online engagement Regular email feedback Reward points for saving energy Participation is defined by the receipt of direct mail (“passive participants”) and signing up for the website (“online participants”). A. Program Marketing Direct Marketing A primary method of marketing both the web tools and the value of energy savings in the CUB Energy Saver program was through the use of direct mail. In these mailers, recipients found marketing information about the online rewards program and what they would expect to find when they signed up. They were given 100 rewards points for signing up, which made them automatically eligible for discounts like $10 gift cards at a local restaurant or 20% off online stores. Direct mail has been used widely as an engagement mechanism for various consumer-facing industries, including the utility sector. In this case, the direct mail pieces were targeted at inducing energy savings directly and incentivizing the recipient to sign up online to achieve greater savings. The direct mail pieces delivered to participants were also designed to induce energy savings independent of whether the participant signs up online. Multiple types of mailer designs were delivered and are displayed below. The first mailer was sent in October of 2010, while the second mailer shown was delivered in March and June of 2010. The first mailer lists energy saving actions in a table on the front, displaying both points earned and dollars saved for each action, as well as a secondary set of actions on the back. The front identifies specific redemption opportunities and companies that offer rewards on the website, and the back displays a normative treatment asking the recipient to “team up” with their community to save energy. Within the second version, there are two variants: one that displays rewards redemption opportunities and one that makes normative comparisons. The rewardsoriented variant indicates how many rewards points one would receive by signing up, as well as six different ways to redeem those points. The savings-oriented variant outlines how much the recipient would be saving if they had signed up online with the last mailer, and how much their “most efficient neighbors” who signed up online are saving. Both variants tell the recipient how many of their neighbors signed up online on 5 the front. On the back they display customized savings recommendations and indicate how many of their neighbors have committed to doing those actions on the site. The only difference on the back is the text on the side of the mailer, which indicates how much money could be saved over a year for the savings-oriented mailers or how many points could be earned over a year for the rewards-oriented mailers. 6 7 Community Marketing Another frequently used method of marketing in the CUB Energy Saver program is community marketing, often through local outreach events or direct and indirect contact with potential users via municipalities, employers, service organizations, nonprofits and other formal and informal organizations. At these events, an outreach coordinator will engage with potential users face-to-face, describing the benefits of the online rewards program, such as customized energy saving tips, the ability to earn discounts, and participating in the organization’s efforts to engage similarly situated users. In certain cases, outreach coordinators encourage users to sign up during the event when Internet access is available. An outreach coordinator may also set up more formal presentations to companies, communities or other organizations at Team events. Teams are any group of individuals who decide to join together to save energy as a group, competing against other teams in their area. Teams allow for greater potential savings because of the ongoing influence of social norms outside the realm of the web tools, as well as through the additional content on the web tools related to teams that lets individuals see how their team is performing. Team members often interact in the real world through the Team that brought them together to save. Additionally, members of Teams will see their teams standing relative to the best in their area, which is another normative motivator. Below is an example of a Team page for the City of Evanston employees. 8 B. Online Engagement When a users signs onto the website, they are presented with different tabs to interact with. The “Ways to Save” page outlining various energy saving options can be found immediately below. Clicking on an action brings the user to a page describing that action in greater detail with how-to steps and advice to assist the user in completing the action. The page also provides dynamic fields to customize savings estimates. The “Track Progress” page shows bar graphs outlining usage by end-use and savings todate and the “Rewards” page presents the user with ways to “spend” their points balance by redeeming them for different gift certificates or products. 9 C. Monthly Email Feedback In addition to the on and off-site contact described above, users receive monthly email updates when a bill comes in, informing them whether or not they saved electricity. There is a normative treatment in these e-mails through the thumbs up and associated text outlining how many points they earned if they saved, or a graphic and text explaining that they did not save. In either case they will be see their current points balance and to-date savings since joining, as well as two different energy savings recommendations. These recommendations are adjusted seasonally for optimal savings. If a user receives an e-mail about their May bill, they will see Summer savings opportunities like cleaning their A/C filter and closing their blinds during the day, not just year-round recommendations such as unplugging their coffee maker or to use a drying rack instead of a clothes dryer. 10 D. Rewards Following the recommendations of Yi and Jeon (2003), rewards points are distributed on an immediate and ongoing basis to serve as a motivator for saving energy. For every kWh a user saves in their monthly bill, they are given 2 points, up to 250 points per billing period. Comparing projected and actual usage calculates the individual savings for a given billing period. Users are notified of points through the above e-mail on the day their bill is released, which itself happens the day after usage is read at the meter. These rewards points do not expire and can be redeemed for 116 different items like gift cards to local restaurants, discounts at online retailers or free magazine subscriptions. This immediate feedback is an engagement mechanism for both high and low engagement users; if someone wants to redeem their points immediately, they can do so. If they want to save their points to redeem for a larger item long term, they can do that as well. The rewards page and some sample rewards can be found immediately below. 11 III. Evaluation Protocols A. Impact Evaluation Overview Online Savings One of the challenges of program evaluation consists in finding a valid control group in order to account for unobservable characteristics of adopters. Households which optin are believed to possess characteristics different from their opt-out counterparts because their participation is not random, thereby exhibiting selection bias, which skews the statistical evaluation. While households who don’t adopt are not influenced by the treatment, they are arguably also different from adopting households in an optin program evaluation. Some households who don’t adopt over the period of analysis may adopt at some later stage, while others may not adopt under any circumstances. ET,C = Electricity usage in kWh/day for Treatment and Control groups DACT,C = Daily Average Consumption for Treatment and Control groups n = Number of households m = Month joined y = First month with available bill z = Last month of program S = Program Savings Percentage In order to avoid these issues while evaluating savings for opt-in households, we use a delayed treatment analysis whereby the kWh consumption over a given period for current opt-in households is compared to that of households that opt-in at a later point. This eliminates the potential bias from the propensity of opt-in households to engage in such programs. The aim of this evaluation process is to measure the overall savings achieved by the program at a given time. This measures the aggregate effect without directly estimating the savings for any particular individual. Furthermore, the program accumulates savings from two sources. First, already enrolled households engage in energy savings to a degree, which may vary over time and across households. Second, the overall savings achieved by the program vary with the number of people in the program and 12 the continuous enrollment of new participants contributes to the overall program savings. We implement this method by using the random variation in the timing of adoption. The pre-opt-in billing data from opt-in households make up the control group for prior time periods. We exclude the billing period in which a household joins the site, eliminating potential bias due to pre- and post-treatment effects occurring in the same bill. As an example, assume we have 500 people join the program in each of three months, January, February and March. Those entering the program in January will have their average daily use values ignored for the month of January (per the average daily usage rules outlined above). In February, those who joined in January will become part of the treatment group. Those who join in February will have their bills ignored for the month of February, and become part of the treatment group in March. Those who join in March will have their bills ignored for that month, and become part of the treatment group in April. The February bills of those who join the program in March represent the control group for those who joined the program in January and are part of the treatment group in February. This cycle continues throughout the year eliminating optin bias by comparing households to others exhibiting the same opt-in tendency. Example: Wisconsin Low-Income Energy Assistance Program Evaluation This method has been used widely elsewhere. For instance, a similar method was used in Wisconsin in 2004 to evaluate their Department of Energy’s Low Income Public Benefits program. The primary purpose of the analysis was to determine whether participation in Wisconsin’s Weatherization Assistance Program (WAP) had an effect on arrearage levels for those who were receiving financial assistance for their utility bills through the Low Income Home Energy Assistance Program (LIHEAP). Home weatherization work occurs at a one point of time, so the onset of treatment is both known and fixed. The evaluation challenge comes from the fact that someone could participate in the financial assistance program one year, not participate the next, and once again participate the following year; this leaves no single point in time defining the onset of treatment. Participation in the low-income assistance program was for 12 months at a time, but one could enroll at any time during the calendar year for the 12 months of the program, leading to the creation of what they called a “rolling control group” in which the makeup of the control group is dynamic on a monthly basis. Passive Savings For mailer-only savings evaluation, a difference-of-differences model was used to determine the percentage savings achieved by the treatment group. This was used to account for any small, insignificant differences that may exist between the treatment and control group before the start of treatment. This also ensures the control group is 13 as identical to the treatment group as possible, so that it represents what the treatment group would look like had they not received any mailers. The steps below outline how savings using the difference-of-differences method were calculated. For the differences of differences method, a few steps are needed in the calculation of energy savings for Month X after program initiation. Let YiX denote the energy consumption for the nth household in Month X. 1. For each bill, calculate the daily average consumption (DAC); this is needed to make bills comparable since different bills can have different billing period lengths. number of days in bill for Month X 2. The percentage difference WiX between DAC of Month X with the same month last year (before program initiation), denoted by Month {-X} is calculated for each household 3. The averages, V, across control and treatment households are calculated. 4. The average percentage savings SX in Month X is the difference of these two averages. 5. We then calculate T, the total gross verified energy savings in kWh or therm achieved by treatment households in that month. This is calculated by multiplying S with the total energy consumption E of treatment households in Month {–X}, the baseline used to calculate the percentage savings: . There are some basic billing requirements that make a difference of differences analysis effective. The first is that the treatment and control groups are as statistically similar as possible before treatment so that the difference-of-differences process is 14 only accounting for small, insignificant variations in usage, optimally around 0.1 kWh per day. It cannot account for significant pre-treatment variations of 0.5 kWh or more, as differences that large are likely due to significant variations in other things like home square footage, household members or unobservable characteristics. A second requirement is that the treatment and control groups must be large enough to easily test the difference in observable characteristics pre-treatment, especially usage, as well as to assume unobservable characteristics are similar. B. Baseline Usage Figure 1 outlines the seasonal electricity usage for the online control group through the first year of the program, in sequence. The seasons are defined as they are in the solar calendar, with summer beginning June 22nd and ending September 22nd, etc. The program began in early June, but because the bill that includes the signup date is biased and subsequently discarded, the requisite number of bills available to calculate savings with confidence is not available until June 29th. Therefore, Summer 2010 actually covers the period from June 29th to September 22nd. At least 13 months of bills was required to be included in the evaluation. The treatment group sample size is 2,925 households, while the control group totals 3,382 households. Figure 1 15 IV. Program Results A. Online Savings Figure 2 below shows the 1-year daily savings estimates for the program, from the point at which there were enough bills to calculate savings with confidence at the very end of June. February is henceforth excluded due to irresolvable irregularities in the billing data collected from ComEd during that billing period. Figure 2 Figure 3 outlines the daily usage difference of differences for the treatment and control groups, as well as the daily kWh savings by season. 16 Figure 3 Figure 4 outlines this more specifically with monthly savings estimates. Figure 4 17 Daily, Monthly and Seasonal With the appropriate removal of unknown February bills, weighted year-to-year savings are 6.01%. The average daily usage for the control group was 29.4 kWh/day, with the difference of differences being 1.77 kWh/day. This results in a 95% confidence interval of 1.69 to 1.84kWh/day, or 5.74% to 6.26%. These results are 99.99% significant. The above graphs illustrate that while consumption varies seasonally, significant savings levels are exhibited year-round with fall savings being expectedly lower than the summer and winter. Savings in the Spring of 2011 are higher than expected, though that may be due in part to effective marketing campaigns and public signup events that were not present prior to then. Savings in June 2010 support this conclusion as June 2011 was milder than July or August of 2010, but savings were comparable for both summer periods. March 2011 is the month with the greatest savings, in part because there were a large number of signups in February due to the aforementioned marketing campaigns, and households often see substantial savings in their first month. February is once again excluded due to reliable data being unavailable. B. Printed Report (or “Mailer Only”) Savings In addition to online savings, CUB Energy Saver has also generated passive, “Mailer Only” savings by providing custom energy insights as well as tailored savings opportunities for households. These mailers were distributed to approximately 15,000 households in the ComEd service territory three times over the pilot period. The households received information about the best energy-saving opportunities, incentives for signing up online, and contextual information about their communities. The first mailers were delivered in homes mid-October 2010, the second mid-March 2011, and the third mid-June 2011. Data is available through the end of June. We received a random, non-personally identifiable selection of 20% of available accounts in the ComEd billing database, as well as all retrievable accounts from the treatment group based on name and address information. A total of 15,672 treatment accounts were retrieved, of which 14,855 had 24 or more bills. 60,065 control accounts were leveraged within the same zip code as the treatment group. We conducted two analyses to determine passive savings. The first analysis (Figure 5) restricts the control group by not including participants and control accounts with missing bills. The second analysis (Figure 6) includes these accounts with missing bills. 18 Figure 5 Figure 5 finds a weighted treatment effect of 1.63% and is 90% significant, and Figure 6 finds a weighted treatment effect of 1.47% that is also 90% significant. The treatment group in Figure 5 is approximately 6,873 homes, while the treatment group in Figure 6 contains 15,097. The control group in Figure 5 is just over 25,770 homes, while the control in Figure 6 is about 60,851. The analyses present more similar estimates as the number of control bills increases over time, providing support to the similarity and significance of the two analyses. 19 Figure 6 Though the monthly results are affected by billing availability, especially in the earlier months, the trend of savings over time is mostly similar. Additionally both show statistically significant treatment effects and improvement over time. Since the first mailer was delivered in the middle of October, the second delivered in the middle of March and the third delivered in the middle of June, these savings numbers generally fit the results of similar programs in which the first mailer has a small effect, but isn’t very sticky (“sticky” insofar as the savings effect lasts for a significant period of time after the first mailer is delivered), but improves with further mailers. The second mailer did not arrive until five months after the first, and savings are expected to decrease as the time between the first and second mailer grows. Thereafter, the second mailer savings improves and continues to improve with the third mailer arriving in June. C. Process Metrics There are a multitude of metrics one can use to measure the performance of a website and how users or customers engage with it, but two of the most important ones are ‘time on site’ and ‘bounce rate’. Time on site is a measurement of how long someone stays on a webpage reading, watching videos, filling out forms, etc. The longer the time on site, the longer visitors are using whatever is being provided. Bounce rate is usually 20 very negatively correlated with time on site, as it is the percentage of those who come onto a site and leave without visiting another page. All else being equal, the more pages someone visits, the longer they are on the site. Time on site is a slightly more complex metric, as different industries and types of websites will produce vastly different metrics. Someone can spend a long time on a site because they really want to find something but cannot; these people are motivated but frustrated, and are unlikely to return to your site. At the same time, a short time on site does not necessarily reflect negatively on the site; it very well may be that they’re finding what they want quickly and easily, which will lead them to come back to the site again because they know they can find what they need. Over time an effectively designed site will have time on site shrink slightly through improvements designed to help users effectively navigate the site, but those users come back again and again because they know they can find what they need. So the best metrics for an improving site are decreasing bounce rates, which indicate ease of navigation, and relatively stable if slightly decreased time on site, unless the nature of the site changes significantly. These metrics require better understanding because some sites will by nature have higher bounce rates (links to scientific journals) or longer time on site (Hulu.com for example) by their very nature. Liu et al. (2010) of Microsoft Research analyzed over 2 billion visits to over 200,000 different pages to understand time on site behavior. They found the first 10 seconds on a page were the most important; if someone stayed 10 seconds they often stayed much longer. They observed this behavior follows a Weibull distribution, which is traditionally used in engineering for reliability analysis. The authors also looked further to find differences by site type, and found both page viewing time to be highly dependent upon the nature of the site. Entertainment sites were found to have much longer times than science and education sites. Weinrich and Obendorf (2008) researched how much users actually read when they get to the page they’re looking for. They found that users exhibit a “screen-and-glean” behavior in which they quickly examine a page to see if they found what they wanted. If they believe they did, they will read through the page to get the information they want. They also found that while users expectedly spent more time on pages with more words, every additional 100 words on the page lead to an average of 4.4 seconds more time spent there. Given the average online reading speed is approximately 200 words per minute (WPM)*, they find much of the additional content likely goes unread. 21 Figure 7 E-mail has regularly been used as a way for utility companies to save on customer billing costs, but has not been used widely as a method of engagement with the customer. The potential benefits are substantial. First, e-mail is a much faster method of communication than letters or even phone calls. Second, e-mail communication allows for quick and easy connection both to a utility’s website and to more detailed information regarding someone’s bill and opportunities to save energy. Third, customized billing analysis and recommendations to save energy can be sent the moment a bill comes in rather than through time-consuming and expensive bill generation processes. This keeps the information current and the delivery cheap. Fourth, computer technology allows for more advanced graphics and levels of interactivity directly in the e-mail than is possible through a paper mail piece, increasing both the methods of direct engagement available and the overall potential engagement with the customer. All these potential benefits outline the need to design an e-mail based communication style that both informs and engages, so these benefits can be realized. Bounce Rate and Time on Site Bounce rate and time on site are presented below in Figure 6. The bounce rate for the duration of the program was 41.85% throughout the first year, with no significant differences between new and returning visitors. However the bounce rate has dropped significantly since the program began, with an average of 53.31% in the 1st quarter, 41.36% in the 2nd quarter, 39.31% in the 3rd quarter, and 34.21% in the 4th quarter. The bounce rate drops 64% from the 1st to the 4th quarter, indicating a significant improvement in the users ability to find what they’re looking for. There was no 22 difference in bounce rates through the first year between new and returning users, as the difference between the two goes up and down slightly. Average time on site is the metric designed to measure how long you have someone’s attention and can engage him or her on the information displayed. The average time on site for the program’s first year is 227 seconds, just under 4 minutes. With bounce rate dropping so strongly, one might expect time on site to drop as users are able to find information more quickly. However time on site goes up and down, with an average of 215 seconds in the 1st quarter, 257 seconds in the 2nd quarter, 234 seconds in the 3rd, and 218 in the 4th. The changes in time on site come as the site changes as well; there were significant improvements to the site in the 2nd and 3rd quarters, including both layout changes and a significant amount of additional content. This fits with the divergence between new and returning visitors, as the difference between new and returning users is only 20 and 6 seconds in the first two quarters, while it is 26 and 69 seconds for the 3rd and 4th quarters, respectively. This indicates that while the content on the site is increasing, many users are intentionally returning to the site and engaging with the material and may indicate that they are able to do so in a more intuitive way. Figure 8 Bounce Rate New Users Returning Users Overall 1st Quarter 54.67% 51.06% 53.31% 2nd Quarter 40.29% 42.86% 41.36% 3rd Quarter 40.26% 37.45% 39.31% 4th Quarter 34.14% 34.42% 34.21% 1st Quarter 2nd Quarter 3rd Quarter 4th Quarter Time on Site New Users Returning Users Overall 208 228 215 255 261 257 225 251 234 200 269 218 Email Open Rate The vast majority of the e-mails the users receive from the CUB Energy Saver platform are either monthly savings and rewards updates or regular newsletters, which contain seasonal savings information and other content. The average open rate tracked in Efficiency 2.0’s analytical tool for these e-mails is 53%. However, the actual open rate may be significantly higher, as the software used to track open rates only identifies an e-mail as “opened” if the images in the e-mail are downloaded. Most major e-mail services by default do not display images for a variety of reasons. Typical e-mail open rates vary by industry, but commercial software services which track these metrics typically find open rates to vary between 14 and 29%. As such, an average open rate 23 of 53% is a strong signal of engagement with both CUB Energy Saver and the information it provides. CUB Energy Saver conducted a survey, in which users were asked to identify the things they were doing to save energy since joining the site. The answers were openended and respondents were given $5 Amazon.com gift cards for their responses. The goal of the survey was to identify if users were making behavioral changes or installing new equipment to save energy. “New equipment” can be anything from an efficient refrigerator to insulation, while behavioral changes were defined as recurring actions of no direct cost; the installation of CFLs was classified separately. In all, 83% of respondents identified behavioral changes they made to save energy, frequently actions like turning off lights, leaving the air conditioner temperature setting just a little bit higher, and unplugging appliances. 25% of users made equipment changes in their home, and 80% of those respondents made behavioral changes as well. 41% of respondents said they bought CFLs for more lights or for the first time, 73% of whom also made at least one behavioral change. While there were major changes made, such as the installation of efficient HVAC equipment and new windows, the vast majority of respondents reported they made at least one behavioral change in an effort to save energy. Users also were asked to report on the number of rebates or direct incentives they received both before and after they joined the program. The vast majority of respondents (79%) reported they did not receive a rebate or other incentive from ComEd before or during program participation. Of the 21% of respondents who had received a rebate or incentive, there were no significant differences in the levels of incentives received before or after program participation, with just over 6% of respondents saying they received an incentive or rebate after the program but not before, and just under 9% reporting the opposite. An additional 6% of respondents report receiving incentives both before and after joining CUB Energy Saver. In an additional set of surveys, participants were asked how satisfied they were with ComEd on a 1-7 scale before and after program participation (1 lowest satisfaction; 7 highest). One survey was distributed during the 7th month of the program, and another in the 13th month, both to random sets of users. Satisfaction with the utility improves upon program participation, with a 7.0% increase across utilities by the 7th program month and a 10.6% aggregate increase by the 13th program month. The standard deviation from the mean is relatively low at 1.09, 1.17 and 1.23 for the before, after 6 months and after 12 months. This indicates a large number of moderate increases in satisfaction as opposed to a low number of high increases combined with a large number reporting no increase in satisfaction. 24 Before joining program After 6 months After 12 months 4.68 5.01 5.19 The hypothesis of increased customer satisfaction was also explored by asking participants whether their satisfaction would change if they were no longer able to participate in CUB Energy Saver. They were given three reporting option: that their satisfaction with the utility would strongly decrease, somewhat decrease, or not decrease at all. Approximately 69% of participants stated their satisfaction with the utility would decrease somewhat or strongly if they were no longer able to participate. The remaining 31% responded that their satisfaction with the utility overall would not decrease if they were no longer able to participate in the program. Further research is required to indicate better how this perception of decreased satisfaction changes during different periods of the program treatment. Conclusion CUB Energy Saver has produced significant electricity savings during the pilot period based on demonstrated online engagement and incentives associated with the online and mailer-only components. The use of multiple methods of engagement has been examined and has been found to be a valuable strategy for engaging customers around their energy consumption as well as motivating them to undertake energy saving actions. Further program expansion and evaluation will attempt to illuminate savings differences within household cohorts, marketing strategies and varying incentives to provide more detailed guidance on the drivers of energy savings. 25