EVERY DROP COUNTS: A WATER CONSERVATION EXPERIMENT WITH HOTEL GUESTS.
Water scarcity is an urgent problem worldwide due to climate change, the poor management of water resources, and the overuse of water. (1) In this study, we implement a field experiment with hotel guests to examine the effectiveness of non-price interventions aimed at water conservation. (2) Previously, many randomized controlled trials have been conducted involving the use of behavioral interventions to promote resource conservation, but most of these trials were conducted at permanent residences, so there were private benefits to consumers because of reduced resource use. (3) In this regard, conserving water at nonpermanent residential sites, such as hotels, is particularly challenging. First, water is free of charge for hotel guests. Thus, there is no direct pecuniary incentive for them to save water. Second, because hotels charge guests a fixed price regardless of water usage volume, greater water-saving on the part of guests creates more profits for hotels. This may diminish the hotel's social goal in guests' eyes and therefore resist complying with a business-led water-conservation campaign.
With these two challenges in mind, we implement three behavioral interventions in an experimental study set in a hotel:
(1) Delivering a message reminding guests of water conservation as a social norm by placing a message card in the hotel room,
(2) Augmenting the message card by providing guests an opportunity to make a commitment to water conservation, and
(3) Augmenting the message card by communicating the hotel's social goal to guests by promising that any monetary benefits arising from guests' water-saving will be donated to an environmental charity.
These three interventions are motivated by theories adopted from the behavioral sciences that emphasize the roles that social norms, public commitments, and perceptions about the messenger play in affecting human behavior (see Section II for details). To identify the causal effects of the interventions, we randomized the treatment conditions and control condition (no intervention) across hotel guests during a 2-month period.
To our knowledge, our study is the first to conduct a field experiment on the effects of non-price interventions on water conservation in a nonpermanent residential site. Few studies have tested resource conservation interventions in a context in which users do not pay a variable price for a given resource. (4) Studies in such contexts are difficult simply because of data availability. In the absence of variable pricing, resource use is typically not monitored. To implement our field experiment, we persuaded the owner of a hotel in a tourist-destination city in South Korea to install digital water meters in 66 rooms of that establishment.
Most studies of nonprice interventions in social and environmental programs estimate the treatments' effects on consumption behavior (e.g., did a given treatment reduce water consumption), but do not attempt to measure effects on welfare. (5) To examine how our behavioral interventions influenced consumer welfare, we surveyed guests at check-out and collected information regarding their levels of satisfaction with their stays.
To summarize our main findings, first, we find that all three behavioral interventions induced substantial reductions in the amount of water used by hotel guests. The size of these reductions ranged from 8% to 16.5%. The treatment effects we found were greater than those observed in previous water-saving experiments involving home-based users (Ferraro and Miranda 2013; Ferraro and Price 2013). (6) This is consistent with our hypothesis that hotel guests use water more liberally because it is free of charge. (7) Second, our post-stay survey revealed that customers were not worse off as a result of the behavioral interventions. Although the findings are limited because they are based on subjective responses to non-incentivized survey questions, we believe that the results are important because they suggest that behavioral interventions can not only be effective but also improve overall welfare, even when no monetary incentive is offered.
The remainder of the paper is organized as follows: Section II reviews the literature on energy--and water-conservation experimental studies conducted in the field and highlights the development of this body of work. Section III explains our experimental design and the data collected. Section IV describes the estimation model and presents the empirical results. Section V provides the conclusion.
II. LITERATURE REVIEW
There is a growing body of experimental field research regarding whether nonprice strategies are effective in promoting energy and water conservation. Nonprice strategies have received mounting attention because the price elasticity of energy use is quite low and therefore conventional price interventions are not very effective (Gillingham, Newell, and Palmer 2009). Moreover, pecuniary strategies may be expensive and sometimes even cause unintended, counterproductive effects by diminishing intrinsic motivation (Gneezy and Rustichini 2000). Prior studies--mainly in social psychology and environmental economics--have examined whether behavioral interventions are cost-effective in terms of changing consumer behavior.
List and Price (2013) comprehensively reviewed recent field experiments in economics, emphasizing the efficacy of nonpecuniary strategies that appeal to social norms to encourage resource conservation. In particular, social comparison or comparative feedback has been shown to be effective in changing consumer behavior in various contexts. (8) Allcott (2011) applied the social (neighbor) comparison technique to residential electricity consumers and found that it was more cost-effective than traditional energy conservation programs. Ferraro and Price (2013) also found that social comparison added salience to a pro-social appeal and consequently induced significant reduction in water use.
Furthermore, recent studies have examined whether the impact of a behavioral intervention can be extended over a longer period. Allcott and Rogers (2014) explain that long-term exposure to social comparison affects consumers' habit formation, which is critical for energy consumption, inducing them to replace energy-inefficient appliances with higher-efficiency ones. Bernedo, Ferraro, and Price (2014) find that the effect of a social appeal message that included social comparison persists even 6 years after the message has been communicated. (9)
A theoretical literature has developed to explain the mechanisms of nonprice interventions. Here, we focus on the determinants that are most relevant to the interventions we implemented in our field experiment. First, a growing body of literature shows that social norms play an important role in economic behavior (Elster 1989; Fehr and Fischbacher 2002; Kimbrough and Vostroknutov 2016). In line with this, we assume that hotel guests have internalized water conservation as a norm. Most of us have learned since childhood that water is a limited resource that we should attempt to save for our own sake, as well as the sake of later generations. Still, we tend to have relatively low awareness of such norms in general and to ignore the external impacts of our behavior (Larrick and Soli 2008), particularly when there is a material benefit that conflicts with the norm (in our case, using water freely). In this context, some reminder or psychological cue that renders the norm more salient may trigger conservation behavior (Attari et al. 2010). Moreover, such reminder messages or prompts were found to affect the individual decision-making process in several field experiments involving vaccinations, dental check-ups, and book returns (Altmann and Traxler 2014; Apesteguia, Funk, and Iriberri 2013; Milkman et al. 2011). In our experiment, displaying a message card regarding water conservation provides such a reminder. In economic terms, reminding guests about water conservation as an environmental-protection measure may confer a degree of "moral utility" on saving water, and guests reading this message may be happy to participate in the hotel's campaign (Levitt and List 2007).
Asking guests to make a voluntary commitment regarding water conservation may create guilt if they fail to adhere to such a commitment. Thus, guilt aversion may serve as a primary motivating factor in hotel guests' participation in a water conservation campaign (Baca-Motes et al. 2013; Battigalli and Dufwenberg 2007; Charness and Dufwenberg 2006; Festinger 1957). A utility-maximizing person consumes water until his/her marginal benefit from consuming that water is equal to its marginal cost, which is zero when a guest pays a fixed price regardless of water usage volume. Once the person makes a commitment to water conservation, he or she will intend to behave consistently with that self-promise (Vanberg 2008). In this scenario, although water consumption does not incur any pecuniary costs, it may have moral costs. Baca-Motes et al. (2013) focus on hotel guests, as our study does, and examine the effect of similar commitments on participation in a hotel's towel reuse program.
They find that the act of making a commitment is important and, moreover, that a specific commitment is more effective than a general one. Lastly, the disclosure of a hotel's social goal may also affect guests' willingness to participate in water conservation (Becker-Olsen, Cudmore, and Hill 2006; Gneezy et al. 2010; Sen and Bhattacharya 2001). (10) If a hotel does not convince guests that its motivation in asking for water conservation is to meet a social/environmental goal, some guests may assume that the hotel is motivated by the pecuniary benefits of reduced water usage. Thus, if the hotel makes its environmental-protection objectives explicit, guests will be more likely to believe that the hotel's water-saving efforts have a social benefit.
Allcott and Kessler (2015) raised an important question regarding the success of behavioral interventions. Specifically, they highlighted two hidden or neglected costs of such interventions: one arising from the behavioral change and the other being a direct negative impact on utility, which corresponds to what Glaeser (2006) called "emotional taxes," or a psychological boomerang effect, as we call such a situation in this study. Using an incentivized willingness-to-pay elicitation mechanism, Allcott and Kessler (2015) found that the majority of consumers (59%) were not willing to pay the social marginal cost of the nudge (35% experienced negative utility and did not like to be nudged at all). Thus, along with measurements of program effectiveness, a consumer welfare evaluation is necessary for a full normative assessment of behavioral interventions.
III. EXPERIMENT OVERVIEW
A. Experimental Design
We implemented a field experiment regarding water conservation in partnership with a residential hotel in Gyeongju, Korea, (11) from February 2015 to March 2015. (12) We set up a control condition with no intervention (C) and three treatment conditions (T1, T2, and T3) to examine the effects of (T1) a message reminding guests about water conservation, (T2) a nonenforceable voluntary commitment to conservation on the part of guests, and (T3) the disclosure of the hotel's social goal related to conservation. We selected 66 homogeneous rooms of the hotel's 152 total rooms, assigned interventions randomly to individual rooms, and kept the intervention in place for the entire experimental period of 2 months. (13) Guests were randomly assigned to rooms upon arrival and remained unaware that the experiment was being conducted and their water usage was to be measured throughout their stays. All guests paid the same price per night (about 54 USD), which is the average rate during off-peak season for similar hotels in the same city. None of our guests made room change requests during the entire experimental period.
The control condition (C) was no intervention. Guests assigned to this condition were not exposed to anything related to water conservation. Guests in the first treatment condition (T1: "message") received a message regarding water conservation as follows: "To save water and energy, please use water wisely." Specifically, a message card was attached on the bottom center of the mirror above the vanity unit in the bathroom so that the guests could see it while they washed their hands or face (see Appendix B for details). The message card was expected to prime the social norm of water conservation (Attari et al. 2010; Larrick and Soil 2008).
The second treatment condition (T2) was "commitment." Guests in this treatment were asked to sign a commitment form in response to the question, "Will you participate in the water-saving campaign during your stay?" To avoid having guests in T2 notice any differences in treatment, a hotel clerk delivered the commitment form to the room rather than making it available at check-in after it had been confirmed that guests would not request a room change. Somewhat surprisingly, all guests in T2 agreed to make the requested commitment. On one hand, this may be because we made sure that the hotel clerk brought the commitment form to the room and explained it in a kind fashion. Thus, it was a bit difficult for the guests to refuse in such an intimate face-to-face context. Also, there was absolutely no cost to them from signing the form because in their minds, there would be neither monitoring nor penalties for violation. On the other hand, the high commitment rate could represent strong social capital, which has been correlated with larger responses to nudges (Bolsen, Ferraro, and Miranda 2014; Costa and Kahn 2013). Our results might not generalize to other settings where the commitment rate turns out to be lower. We also displayed the same message card used in T1 at the same position in their bathrooms to remind them of the social norm of water-conservation during their stay.
In the third treatment condition (T3: "social goal"), guests saw a message card including the following statement: "Our hotel has launched a water-saving campaign, and the savings from the campaign will be used to protect the environment." (14) This explicit revelation of the hotel's social goal was expected to convince guests that the campaign was not intended for private benefit. The message card used in T3 was displayed in two places: on the vanity unit in the bathroom and on the TV stand in the living room (see Appendix B for details). However, except for the disclosure of the hotel's intention in T3, the fundamental characteristics of the message cards (i.e., the core information they communicated) were similar across all three treatments.
The experiment was conducted from February 2015 to March 2015. During the experimental period, there were 310 guest parties who did not make multiple or corporate reservations, and we randomly assigned them to one of the 66 rooms that made up the experimental setting (i.e., those with digital water meters and message cards). We did not include those guests who made multiple, group, or corporate reservations (such as guests who were on school trips or taking part in corporate workshops) to minimize potential bias due to unobservable interactions among guests. Also, only four of the 310 guest parties in the sample stayed multiple nights, so we excluded these four parties for the sake of the homogeneity of the sample. The final sample consisted of 306 guest parties. (15) No one in the final sample reserved multiple rooms; that is, all reservations were made individually. (16)
For each guest party, we collected the data on water consumption recorded by the digital water meters (see Figures C1 and C2 in Appendix C) and basic information including the dates of check-in and check-out; the numbers of men, women, and children; and the purpose of travel (family travel or not). There was no separate formal survey for these questions to prevent the guests from noticing that our experiment was being conducted. They were asked in a casual manner at check-in.
As presented in Table 1, we constructed the following variables to represent the characteristics of the guest parties: (i) experimental condition (C, T1, T2, and T3), (ii) total number of guests, (iii) ratio of women to total number of guests and the ratio of children to total number of guests, and (iv) a dummy variable for family travel. As a check for treatment randomization, we conducted a series of t-tests for the null hypothesis that the mean values of the explanatory variables in each treatment condition (TI, T2, or T3) are the same as those of the corresponding variables in C and an F-test for the joint hypothesis. As seen in columns (6)-(9), all p values were greater than 0.10, and thus, the variables were well-balanced across all experimental conditions. In column (1), based on all sample observations, a guest party consisted of about 4.4 people on average, and about 75% of guests were at the hotel for family travel. (17) The ratios of women and children to the total number of guests were 0.38 and 0.28, respectively. In addition to the variables mentioned above, we constructed an indicator for weekends and holidays. We controlled for this time variable to account for any weekend-and-holiday effect, which may have remained even after randomization.
Our dependent variable was daily water use, as associated with a given room. Because the hotel's existing analog water meter measured only the total amount of water used in the entire hotel, we installed digital water meters in each of the 66 rooms used in the study between October 20 and November 1, 2014. These meters enabled us to read the water use per room in tons to an accuracy of five digits after the decimal point. Moreover, to avoid the inconvenience of having to open the bathroom ceiling vents daily to check water use, we connected a signal transmission line to each digital water meter so we could read the water use per room remotely from a computer (see Appendix C for details).
During the experimental period, a dedicated agent checked how much water each room used, based on which we calculated two water use measurements: water use per night (PN) and water use per person per night (PPPN). Water use PN is the total water use volume of a guest party for one night, which is simply the total measured amount of water used per room because all sampled guest parties stayed one night. Water use PPPN is the per-capita water use volume of a given guest party for one night, as computed by dividing water use PN by the total number of guests in a given room. As shown in Table 2, the average water use PN was about 3901, and average water use PPPN was about 891 based on all sample observations. We will discuss water consumption behaviors by treatment group in detail in Section IV.A.
Furthermore, to determine whether our interventions affected guest satisfaction, we asked guests to participate in a post-stay survey at check-out. (18) We attempted to maximize the response rate to avoid selection bias. For example, we asked the hotel's front desk staff to do their best to collect the survey. We also offered survey respondents the chance to enter a lottery to win a digital camera. The survey was very brief, containing three questions related to three areas: (i) overall satisfaction with the hotel, (ii) intention to revisit the hotel, and (iii) willingness to recommend the hotel to others. We asked the respondents to rate their answers on a Likert scale from 1 (Very Poor) to 5 (Excellent). As seen in Table 3, 232 guest parties (76% response rate) participated in the survey. (19)
IV. EXPERIMENTAL FINDINGS
A. The Distribution of Water Consumption
In this subsection, we compare the distribution of water consumption in each treatment condition with that in the control group. As seen in Table 2, in all three treatment groups (T1-T3), both the average water use PN and the average water use PPPN were lower than those of the control group (C). With regard to water use PN, the relative savings of T2 (85.2 1, the equivalent of a 4.5minute shower) was the largest as compared to the C condition, the relative savings of T1 was the second largest (58.7 1, or a 3.1-minute shower), and that of T3 was the smallest (42.11, or a 2.2minute shower). (20) Regarding water use PPPN, the relative water savings of T2 was still the largest (15.21, or a 0.8-minute shower), while that of T3 was the second largest (8.8 1, or a 0.5minute shower), and that of T1 was the smallest (5.3 1, or a 0.3-minute shower).
Also, Figure 1 shows the quantile distributions for water use volume by treatment. In the upper diagrams (A)-(C), the circles and crosses indicate the quantiles for water use PN for the control and corresponding treatment conditions, respectively. Similarly, in the lower diagrams (D)-(F), we show the quantiles for water use PPPN for the control and corresponding treatment conditions, respectively. It is notable that for any given fraction of the sample, its quantile of water use volume in each treatment condition is lower than that in the control condition, meaning that the treatments are effective in reducing water consumption.
B. Average Treatment Effects
We estimate the average treatment effects (ATEs) on water use via the following equation:
(1) [Y.sub.i] = [[beta].sub.0] + [[beta].sub.1]T[1.sub.i] + [[beta].sub.2]T[2.sub.i] + [[beta].sub.3]T[3.sub.i] + [X'.sub.i][gamma] + [[member of].sub.i],
where [Y.sub.i] is either the water use PN or water use PPPN of guest party i. There are three treatment indicators. T[1.sub.i] is an indicator of whether guest party i is in T1. Likewise, 7[2.sub.i] and T[3.sub.i] are indicators for treatments T2 and T3, respectively. Vector [X.sub.i] includes the control variables regarding guest-party characteristics and time-specific variables, such as dummy variables for the total number of guests, the ratios of women and children to the total number of guests, a dummy variable for family travel, and a dummy variable for weekends and holidays. Because the control condition C is excluded in Equation (1), each of the three treatment effect parameters, [[beta].sub.1], [[beta].sub.2], and [[beta].sub.3], indicates the effect of the corresponding treatment relative to the control condition. Lastly, [[epsilon].sub.i] is the standard error term.
Tables 4 and 5 present the estimated ATEs from Equation (1) using water use PN and water use PPPN, respectively. Columns (1) and (2) present the results for Tl with and without the control variables; columns (3) and (4) show the effects of T2 in the same way; and columns (5) and (6) show the effects of T3. Lastly, in columns (7) and (8), we jointly estimate the three treatment effects.
In Table 4, we find that water usage is lower in all three treatments as compared to the control condition. Because the treatments are randomly assigned, the unconditional estimates should represent the causal effects. The estimates in columns (1), (3), and (5) show that T1 decreases water consumption by 591 per night, T2 does so by 851, and T3 does so by 421. The regression-adjusted estimates in columns (2), (4), and (6) are similar, but precision is improved for all three treatments. The results are all statistically significant at the 5% significance level, and the results for the treatment effects are graphically summarized in Figure 2.
The magnitude of the effects is not ignorable. Column (2) shows that T1 saves about 35.21, the equivalent of a 1.9-minute shower, in total water consumption as compared to the average water use in C, which is 442.41 PN. Therefore, T1 creates approximately 8.0% water use savings. Columns (4) and (6) show that T2 saves about 72.91 (16.5%, or a 3.8-minute shower) and that T3 relatively saves about 54.71 (12.4%, or a 2.9-minute shower), respectively. Column (8) shows that all treatments save water in an amount ranging from 32.3 to 72.91 (7.3% to 16.5%, or a 1.7--to 3.8-minute shower). Due to the small sample size, we do not have enough statistical power to discriminate between any two treatments. A comparison between columns (7) and (8) shows that conditioning related to guest characteristics explains considerable variance and reduces our treatment-effect standard errors by a factor of two. An F-test indicates that the treatment effects are jointly significant in the regression specification (p<0.01), while they are not in a simple test of means (p = 0.26).
In Table 5, we find similar results for our alternative measure of water consumption, water use PPPN. Again, the results are presented visually in Figure 2(B).
The magnitude of the effects is also sizable in terms of water consumption per person and per night. The results in columns (2), (4), and (6) show that compared to the average per-capita water use in C (91.21), T1 saves about 7.51 (7.7%, the equivalent of a 0.4-minute shower), T2 about 16.11 (16.6%, or a 0.8-minute shower), and T3 about 11.21 (11.6%, or a 0.6-minute shower). Column (8) shows that all treatments save water in an amount ranging from 6.7 to 16.91 (6.9% to 17.4% or a 0.4- to 0.9-minute shower). Also, both the unconditional and conditional ATEs are jointly significant (the p values are less than 0.01).
To summarize, the three treatments are jointly significantly different from the control, with a pooled treatment effect estimate of 50.1 [+ or -]33.71 PN, but the three treatments are, unfortunately, not statistically significantly distinguishable from one another. To provide guidance for future research, we now present power calculations to indicate how large the sample sizes would need to be for a study that hopes to detect differences between the types of nudges we employed in a hotel water-conservation setting. Following standard conventions, we assume that researchers want to conduct a test with a significance level of 5% and a power of 80% to reject the null hypothesis of equal treatment effects.
In Table 4, the unconditional estimates of the PN treatment effects are -58.71 for T1, -85.21 for T2, and -42.11 for T3. The smallest difference is that between T1 and T3, a difference of 16.61 PN, or about 3.8% of the control baseline water consumption. Similarly, Table 5 shows that regarding the PPPN treatment effects, the smallest difference is again that between T1 and T3, a difference of 3.41, or about 3.5% of the control baseline water consumption PPPN. Suppose, then, that researchers want to be able to detect a 4% difference in water usage between treatments, or 181 PN. Because the root mean square error (RMSE), that is, the square-root of the sum of square-deviations divided by n, is 156.81 PN, researchers would need to collect 252 observations per treatment group and thus 1,008 in total to perform all-pairwise comparisons (Hsu 1996). (21) These numbers remain the same for a minimum detectable difference of 4% (or 3.91) with an RMSE of 15.71 on a PPPN basis.
C. Impacts on Customer Satisfaction
In this subsection, we examine whether our behavioral interventions affect consumer welfare. As Allcott and Kessler (2015) and Glaeser (2006) point out, behavioral interventions, even successful ones, may incur unintended social costs. Behavioral interventions may also have a direct negative impact on some consumers who simply do not like the interventions. Also, there behavioral changes may carry costs for compliers with the interventions. In our case, it is possible that some guests were disturbed by the hotel's water-saving campaign in and of itself. They may not have changed their consumption and yet still felt annoyed by the intervention. Also, some other guests may have agreed with the campaign and attempted to reduce, for example, their shower time. There could be effort costs involved in that kind of behavioral change.
In Table 3, the mean values of their answers were about 2.54 for overall satisfaction, about 2.52 for intention to revisit, and about 2.49 for willingness to recommend. Pearson's Chi-square tests were conducted to ensure that there was no difference between the answers in each treatment group and those in the control group (C), and we present the p values of those tests in columns (6)-(8). Also, the p value for the joint hypothesis is reported in column (9). On average, we find no evidence that average customer satisfaction differed across the control and three treatment conditions regardless of which satisfaction measures are used. Here, we present regression-adjusted estimates. One concern is that as shown in the bottom panel of Table 3, the survey response rate differs across treatments. Specifically, the response rates for the control group (68%), T1 (67%), and T2 (68%) are similar, but that for T3 is much higher (94%). Furthermore, it is conceivable that less-satisfied customers were less likely to respond to our post-stay survey. To address this concern, we first check the differences in descriptive statistics regarding guests' observable characteristics according to response status (i.e., "Response" or "No Response"). Table 6 shows the differences between the "Response" and "No Response" groups and their p values from the mean equality test between the two groups. Significant differences are found in the dummy for family vacation for C and T1. However, the middle panel of Table 6 shows no significant differences in terms of water consumption based on response status. Overall, the results in Table 6 suggest that there is little observable selection in terms of survey response, which relieves our concerns regarding missing data due to a lack of survey responses.
To further check for any potential attrition bias, we employ the weighted linear probability model. (22) Specifically, we use the inverse of the probability of responding to the survey among those in a given experimental condition as each observation's weight and run a weighted regression model (see Chapter 7 in Gerber and Green 2012). (23)
Table 7 presents the results of the weighted linear probability model. For the dependent variable, we create a dummy variable indicating that the guest's score in each category was either "Good" or "Excellent" (there are five responses: "Very Poor," "Poor," "Average," "Good," and "Excellent"). The results below are robust with regard to three measures of customer satisfactions. Column (1) displays overall satisfaction, column (2) displays the intention to revisit the hotel, and column (3) displays willingness to recommend the hotel to others. All three columns use the sample for C, T1, T2, and T3. All columns show that the impacts of our three interventions (T1-T3) on guest satisfaction are not statistically significant. These results confirm that on average, there is no difference in satisfaction associated with the interventions, as explained in Section III.B. However, we cannot completely rule out the possibility that customers who were slightly annoyed by the nudges might not participate in satisfaction survey. For example, if they valued the stay at 50 USD and if they suffered the disutility of 1 USD from the nudge, that would represent only a 2% reduction, and could be difficult to be detected in a survey with a margin of error of 0.24 on a mean of 2.5 (i.e., a 10% margin of error).
V. SUMMARY AND CAVEATS
In this study, we presented the results from a field experiment regarding nonpermanent residential consumers' water conservation. We evaluated the effects of three behavioral interventions: a message regarding water conservation, the solicitation of a nonenforceable voluntary commitment by guests, and explicit disclosure of the hotel's social goal. Our findings suggest that such behavioral interventions can induce a sizable reduction in the amount of water used by hotel guests. Also, we found that customers were not worse off as a result of the behavioral interventions. Our findings suggest that behavioral interventions can be effective not only in terms of encouraging resource conservation but also overall welfare improvement. While prior experimental findings have suggested that behavioral interventions can effectively change people's energy-consumption behaviors, our findings are intriguing given that in our setting, there was no pecuniary incentive or compensation for complying with the interventions.
There are a few caveats associated with our findings. First, our findings, like those of any other experimental findings, are subject to the question of external validity. This is a major concern, particularly for our findings, because the treatments we used in our study could easily be extended to national policies. In this regard, it should be noted that our experiment was conducted at a moderately priced vacation hotel in a tourism city during the off-season. It is possible that customers at more commercial business hotels or luxurious hotels are different in that they demand high-quality services and dislike any external disturbances. Second, our experimental design does not allow us to identify the treatment components precisely. For example, while the social goal treatment turned was ultimately more effective than the message treatment, because we placed one additional message card in the room, it is possible that part of the treatment effect was driven by that additional card. The third caveat relates to our sample size. Due to budgetary limitations, our sample size was small (306 observations), and the experimental period was fairly short (2 months). To obtain more convincing results, we require evidence from large-scale experiments, as well as experiments at various types of sites. In addition, we implemented three interventions, so policymakers will likely want to know which one is the most cost-effective. Unfortunately, due to our small sample size, we do not have the statistical power needed to discern the effects of various interventions. To detect differences between the types of nudges, researchers would need at least 252 observations per treatment group. Needless to say, more studies are warranted.
APPENDIX A: CONDITION RANDOMIZATION, ROOM CHARACTERISTICS, AND FLOOOR PLANS
Caption: FIGURE A1 Floor Plans
TABLE A1 Room-Level Condition Randomization Room No. Floor Code Room Type 114 1 C B 115 1 C B 116 1 C B 117 1 C B 118 1 C B 119 1 C B 123 1 C A 201 2 T3 B 202 2 T3 B 203 2 T3 B 204 2 T3 B 205 2 T3 B 206 2 T3 B 208 2 T3 B 209 2 T3 B 210 2 T3 B 212 2 T3 B 223 2 C C 228 2 C B 229 2 C B 230 2 T2 B 231 2 T2 A 232 2 T2 A 234 2 T1 A 236 2 T1 A 237 2 C A 238 2 T1 A 239 2 T1 A 240 2 T1 A 241 2 T1 A 242 2 T1 A 243 2 C A 244 2 C A 301 3 T2 B 302 3 T2 B 303 3 T2 B 304 3 T2 B 305 3 T1 B 306 3 T1 B 308 3 T1 B 309 3 T1 B 311 3 T1 B 312 3 T1 B 323 3 C C 325 3 C B 327 3 C B 329 3 C B 344 3 C A 408 4 T1 B 409 4 C B 410 4 T2 B 411 4 T2 B 412 4 T2 B 413 4 T1 B 414 4 C B 416 4 C B 420 4 C B 421 4 T1 B 423 4 T1 B 425 4 T1 B 427 4 T1 A 429 4 T1 A 435 4 C A 437 4 C A 438 4 C A 439 4 C A Notes: Control condition C indicates no behavioral intervention. Treatment condition T1 indicates message, T2 commitment, and T3 social goal. Room type A (56 [m.sup.2]) is composed of one bedroom, one living room with a kitchenette, and one bathroom, while room types B (76 [m.sup.2]) and C (93 [m.sup.2]) have two bedrooms, a living room with a kitchenette, and one bathroom.
APPENDIX B: INFORMATION PROVISION BY EXPERIMENTAL CONDITION
Caption: FIGURE B1 Message Card Images
Caption: FIGURE B2 Commitment Form Image
Caption: FIGURE B3 Placement of Message Cards: (A) T1 Placed in the Bathroom, (B) T2 Placed in the Bathroom, (C) T3 Placed in the Bathroom, and (D) T3 Placed in the Living Room.
TABLE B1 The Message Cards Used for the Treatment Conditions English (Translated) Korean (Original) Panel A: T1 To save water and energy, [phrase omitted] please use water wisely. Panel B: T2 To save water and energy, [phrase omitted] please use water wisely. Panel C: T3 Front In this hotel, a water-saving [phrase omitted] campaign has been launched, and the savings from the campaign are used to protect the environment. Many guests who have stayed in the hotel have participated in our campaign, and as a consequence, we achieved the desired water-savings results. Thank you in advance for your cooperation. Rear Please join our campaign! [phrase omitted] Note: Treatment condition Tl indicates message, T2 commitment, and T3 social goal. TABLE B2 The Commitment Form English (Translated) Korean (Original) Please join our campaign! [phrase omitted] Every year, the huge amount of water used by [phrase omitted] hotel guests throughout the country causes a serious waste of energy and resources. In this hotel, for the purpose of environmental protection, a water-saving campaign has been launched. Will you participate in the water-saving [phrase omitted] campaign during your stay?  Yes, I will.  No, thank you.
APPENDIX C: INSTALLATION OF DIGITAL WATER METERS AND SIGNAL TRANSMISSIONS
Caption: FIGURE C1 Installation of Digital Water Meters: (A) Digital Water Meter and (B) Connection between Digital Water Meters and Water Pipes
Caption: FIGURE C2 Installation of Signal Transmissions: (A) Signal Transmission Line and (B) Screenshot of Computer Program to Check Water Use
APPENDIX D: ADJOINING ROOMS WITH DIFFERENT TREATMENTS
In our data, the following four pairs of adjoining rooms received different treatments: (303, 305), (304. 306), (408, 410), and (411, 413). As a robustness check, we further excluded 78 rooms from our sample if guests checked into both sets of rooms in one of these pairs on the same date. Then, based on only 228 rooms, we re-estimated the conditional ATEs, using water use PN as the dependent variable. As found in Table Dl, the results are similar to those in Table 4.
TABLE D1 Results Excluding Adjoining Rooms with Different Treatments Treatments Variables T1 T2 (1) (2) Treatment condition: Message (T1) -35.763 * (19.025) Commitment (T2) -81.545 *** (17.235) Social goal (T3) Characteristics of guest parties: Ratio of women to 86.911 48.482 total guests (89.517) (94.221) Ratio of children to -38.552 -217.083 * total guests (82.089) (115.213) Family vacation -5.645 38.412 (33.670) (32.617) Constant 538.205 *** 601.872 *** (45.737) (51.117) Number of guests Yes Yes dummies Weekend/holiday dummy Yes Yes Observations 93 84 R-squared 0.729 0.804 Treatments Variables T3 All (3) (4) Treatment condition: Message (T1) -36.128 * (20.103) Commitment (T2) -86.594 *** (16.788) Social goal (T3) -52.68 5** -47.840 ** (20.183) (17.973) Characteristics of guest parties: Ratio of women to 162.984 75.666 total guests (142.733) (73.229) Ratio of children to -67.272 -62.400 total guests (128.028) (64.996) Family vacation 37.401 24.439 (35.676) (19.636) Constant 534.158 *** 559.180 *** (73.985) (41.769) Number of guests Yes Yes dummies Weekend/holiday dummy Yes Yes Observations 139 228 R-squared 0.743 0.773 Notes: The clustered robust standard errors by check-in date are given in parentheses. The unit of water use PN is liters. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.
APPENDIX E: ESTIMATING THE PROBABILITY OF RESPONDING TO THE SURVEY
TABLE E1 Predicting Response Statuses Marginal Effects of Logit Regression C T1 T2 T3 Variables (1) (2) (3) (4) Characteristics of guest parties: Ratio of women to -0.023 -0.468 -0.342 1.291 * total guests (1.243) (0.780) (0.734) (0.772) Ratio of children 1.410 -0.537 -0.121 0.164 to total guests (1.113) (0.715) (0.720) (0.397) Family vacation -0.443 ** 0.296 -0.043 0.707 (0.125) (0.214) (0.224) (0.602) Number of guests Yes Yes Yes Yes dummies Weekend/holiday dummy Yes Yes Yes Yes Observations 44 88 79 95 Pseudo R-squared 0.121 0.059 0.109 0.142 Note: The marginal effects of the logit regression for each experimental condition are reported. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.
Allcott, H. "Social Norms and Energy Conservation." Journal of Public Economics, 95(9-10), 2011, 1082-95.
Allcott, H., and J. B. Kessler. "The Welfare Effects of Nudges: A Case Study of Energy Use Social Comparisons." NBER Working Paper No. 21671, 2015.
Allcott, H., and T. Rogers. "The Short-Run and Long-Run Effects of Behavioral Interventions: Experimental Evidence from Energy Conservation." American Economic Review, 104(10), 2014, 3003-37.
Altmann, S., and C. Traxler. "Nudges at the Dentist." European Economic Review, 72, 2014, 19-38.
Angrist, J. D., E. Bettinger, and M. Kremer. "Long-Term Educational Consequences of Secondary School Vouchers: Evidence from Administrative Records in Colombia." American Economic Review, 96(3), 2006, 847-62.
Apesteguia, J., P. Funk, and N. Iriberri. "Promoting Rule Compliance in Daily-Life: Evidence from a Randomized Field Experiment in the Public Libraries of Barcelona." European Economic Review, 64, 2013, 266-84.
Attari, S. Z., M. L. DeKay, C. L. Davidson, and W. B. de Bruin. "Public Perceptions of Energy Consumption and Savings." Proceedings of the National Academy of Sciences, 107(37), 2010, 16054-9.
Ayres, I., S. Raseman, and A. Shih. "Evidence from Two Large Field Experiments that Peer Comparison Feedback Can Reduce Residential Energy Usage." Journal of Law, Economics, and Organization, 29(5), 2013, 992-1022.
Baca-Motes, K., A. Brown, A. Gneezy, E. A. Keenan, and L. D. Nelson. "Commitment and Behavior Change: Evidence from the Field." Journal of Consumer Research, 39(5), 2013, 1070-84.
Battigalli, P., and M. Dufwenberg. "Guilt in Games." American Economic Review, 97(2), 2007, 170-76.
Becker-Olsen, K. L., B. A. Cudmore, and R. P. Hill. "The Impact of Perceived Corporate Social Responsibility on Consumer Behavior." Journal of Business Research, 59(1), 2006,46-53.
Bernedo, M., P. J. Ferraro, and M. Price. "The Persistent Impacts of Norm-Based Messaging and Their Implications for Water Conservation."
Journal of Consumer Policy, 37(3), 2014, 437-52.
Bhanot, S. "Rank and Response: A Field Experiment on Peer Information and Water Use Behavior." Harvard Environmental Economics Program Discussion Paper 15-63, 2015.
Bolsen, T., P. J. Ferraro, and J. J. Miranda. "Are Voters More Likely to Contribute to Other Public Goods? Evidence from a Large-Scale Randomized Policy Experiment." American Journal of Political Science, 58(1), 2014, 17-30.
Brent, D. A., J. H. Cook, and S. Olsen. "Social Comparisons, Household Water Use, and Participation in Utility Conservation Programs: Evidence from Three Randomized Trials." Journal of the Association of Environmental and Resource Economists, 2(4), 2015, 597-627.
Charness, G., and M. Dufwenberg. "Promises and Partnership." Econometrica, 74(6), 2006, 1579-601.
Costa, D. L., and M. E. Kahn. "Energy Conservation 'Nudges' and Environmentalist Ideology: Evidence from a Randomized Residential Electricity Field Experiment." Journal of the European Economic Association, 11(3), 2013, 680-702.
Delmas, M. A., and N. Lessem. "Saving Power to Conserve Your Reputation? The Effectiveness of Private Versus Public Information." Journal of Environmental Economics and Management, 67(3), 2014, 353-70.
Delmas, M. A., M. Fischlein, and O. I. Asensio. "Information Strategies and Energy Conservation Behavior: A Meta-Analysis of Experimental Studies from 1975 to 2012." Energy Policy, 61, 2013, 729-39.
Elster, J. "Social Norms and Economic Theory." Journal of Economic Perspectives, 3(4), 1989, 99-117.
Fehr, E., and U. Fischbacher. "Why Social Preferences Matter--The Impact of Non-Selfish Motives on Competition, Cooperation and Incentives." The Economic Journal, 112(478), 2002, CI-C33.
Ferraro, P. J., and J. J. Miranda. "Heterogeneous Treatment Effects and Mechanisms in Information-Based Environmental Policies: Evidence from a Large-Scale Field Experiment." Resource and Energy Economics, 35(3), 2013, 356-79.
Ferraro, P. J., and M. K. Price. "Using Non-Pecuniary Strategies to Influence Behavior: Evidence from a Large-Scale Field Experiment." Review of Economics and Statistics, 95(1), 2013, 64-73.
Ferraro, P. J., J. J. Miranda, and M. K. Price. "The Persistence of Treatment Effects with Norm-Based Policy Instruments: Evidence from a Randomized Environmental Policy Experiment." American Economic Review, 101(3), 2011,318-22.
Festinger, L. A Theory of Cognitive Dissonance. Stanford, CA: Stanford University Press, 1957.
Florke, M., E. Kynast, I. Barlund, S. Eisner, F. Wimmer, and J. Alcamo. "Domestic and Industrial Water Uses of the Past 60 Years as a Mirror of Socio-Economic Development: A Global Simulation Study." Global Environmental Change, 23(1), 2013, 144-56.
Gerber, A. S., and D. P. Green. Field Experiments: Design, Analysis, and Interpretation. New York: W. W. Norton and Company, 2012.
Gillingham, K., R. G. Newell, and K. Palmer. "Energy Efficiency Economics and Policy." Annual Review of Resource Economics, 1(1), 2009, 597-620.
Glaeser, E. L. "Paternalism and Psychology." Regulation, 29(2), 2006, 32-38.
Gneezy, U., and A. Rustichini. "Pay Enough or Don't Pay at All." Quarterly Journal of Economics, 115(3), 2000, 791-810.
Gneezy, A., U. Gneezy, L. D. Nelson, and A. Brown. "Shared Social Responsibility: A Field Experiment in PayWhat-You-Want Pricing and Charitable Giving." Science, 329(5989), 2010, 325-27.
Goldstein, N. J., R. B. Claldini, and V. Griskevicius. "A Room with a Viewpoint: Using Social Norms to Motivate Environmental Conservation in Hotels." Journal of Consumer Research, 35(3), 2008, 472-82.
Hardin, G. "The Tragedy of the Commons." Science, 162(3859), 1968, 1243-48.
Harding, M., and A. Hsiaw. "Goal Setting and Energy Efficiency." Journal of Economic Behavior and Organization, 107(Part A), 2014, 209-27.
Hsu. J. Multiple Comparisons: Theory and Methods. London: Chapman and Hall, 1996.
Jessoe, K., and D. Rapson. "Knowledge Is (Less) Power: Experimental Evidence from Residential Energy Use." American Economic Review, 104(4), 2014, 1417-38.
Kimbrough, E. O., and A. Vostroknutov. "Norms Make Preferences Social." Journal of the European Economic Association, 14(3), 2016, 608-38.
Laibson, D. "A Cue-Theory of Consumption." Quarterly Journal of Economics, 116(1), 2001, 81-119.
Larrick, R. P.. and J. B. Soil. "The MPG Illusion." Science, 320(5883), 2008, 1593-94.
Levitt, S. D., and J. A. List. "What Do Laboratory Experiments Measuring Social Preferences Reveal about the Real World?" Journal of Economic Perspectives, 21(2), 2007, 153-74.
List, J. A., and M. K. Price. "Using Field Experiments in Environmental and Resource Economics." NBER Working Paper No. 19289, 2013.
Milkman, K. L., J. Beshears, J. J. Choi, D. Laibson, and B. C. Madrian. "Using Implementation Intentions Prompts to Enhance Influenza Vaccination Rates." Proceedings of the National Academy of Sciences, 108(26), 2011, 10415-20.
Nolan, J. M., P. W. Schultz, R. B. Cialdini, N. J. Goldstein, and V. Griskevicius. "Normative Social Influence Is Underdetected." Personality and Social Psychology Bulletin, 34(7), 2008, 913-23!
Olmstead, S. M., and R. N. Stavins. "Comparing Price and Nonprice Approaches to Urban Water Conservation." Water Resources Research, 45(4), 2009, W04301.
Schultz, P. W., J. M. Nolan, R. B. Cialdini, N.J. Goldstein, and V. Griskevicius. "The Constructive, Destructive, and Reconstructive Power of Social Norms." Psychological Science, 18(5), 2007,429-34.
Sen, S., and C. B. Bhattacharya. "Does Doing Good Always Lead to Doing Better? Consumer Reactions to Corporate Social Responsibility." Journal of Marketing Research, 38(2), 2001, 225-43.
Vanberg, C. "Why Do People Keep their Promises? An Experimental Test of Two Explanations." Econometrica, 76(6), 2008, 1467-80.
ATEs: Average Treatment Effects
CSR: Corporate Social Responsibility
EPA: Environmental Protection Agency
PN: Per Night
PPPN: Per Person Per Night
RMSE: Root Mean Square Error
SSR: Shared Social Responsibility
HAILEY HAYEON JOO, JUNGMIN LEE and SANGKON PARK, We would like to thank the Co-Editor, David Reiley, and two anonymous referees for their thoughtful comments. This paper also benefited from discussions with Lorenz Gotte, SeEun Jung, and Tatsuyoshi Saijo. and from valuable comments by seminar and conference participants at Korea University, Asia Meeting of the Econometric Society (Kyoto), Korea Energy Economics Institute, and Korea Resource Economics Association. We are indebted to Kwangsup Cha and Kyungjun Lee for their invaluable support in conducting the experiment. Joo's work was supported by the Sogang University Research Grant of 2013 (201310071.01). Lee's work was supported by Research Resettlement Fund for the new faculty of Seoul National University.
Joo: Assistant Professor, Department of Economics, Sogang University, Seoul 04107, South Korea. Phone +82-2-7058507, Fax +82-2-705-8180, E-mail email@example.com
Lee: Professor, Department of Economics, Seoul National University, Seoul 08826, South Korea. Phone +82-2-8802293, Fax +82-2-886-4231, E-mail firstname.lastname@example.org
Park: Associate Research Fellow, Tourism Policy Research Office, Korean Culture & Tourism Institute, Seoul 07511, South Korea. Phone +82-2-2669-8483, Fax +82-2-26698410, E-mail email@example.com
(1.) For example, most central and southern European countries experienced significant rainfall deficits during the summer of 2015. That same year, South Korea--the site of our experiment--also suffered a severe drought that damaged crops nationwide. Also, as of the time of writing, the U.S. state of California has suffered a record-breaking drought since 2011, forcing state and local governments to take dramatic water-conservation measures.
(2.) Although the total volume of water that hotels and other lodging establishments use is relatively small, it is important in the context of conservation because such facilities are speculated to waste a substantial amount of water. In the United States, the water used in hotels and other lodging businesses accounts for about 15% of total water used in commercial and institutional facilities according to an estimate by the Environmental Protection Agency (EPA). Source: http://www3.epa.gov/watersense/commercial/docs/ factsheets/hotels_fact_sheet_508.pdf (retrieved March 7, 2016).
(3.) To name only a few recent economics papers, see Allcott (2011), Costa and Kahn (2013), Ferraro and Price (2013), Ferraro and Miranda (2013), Allcott and Rogers (2014), and Bhanot (2015).
(4.) The only existing study we found was Delmas and Lessem (2014). They conducted a field experiment in UCLA residence halls with students who did not pay electricity bills. They examined the effect of disclosing information regarding the environmental impact of energy consumption. They found no effect after providing the information privately. However, when such information was made public, there was a significant reduction in electricity consumption.
(5.) The exception is Allcott and Kessler (2015). We will discuss their work in more detail in the following section.
(6.) In a previous home-based experiment, the average water saving rate reached 4.8% (Ferraro and Price 2013).
(7.) For example, Ferraro and Miranda (2013) reported that an average household in their control group consumed 21,790 gal of water from December 2007 to March 2008, or 673.6 1 per day. Because toilets account for about 26.7% of the water consumed in the average U.S. home, showers about 16.7%, and faucets about 15.7%, if we assume that approximately 59.2% of the water use occurs in the bathroom, an average household in Ferraro and Miranda's residential control group used 398.8 1 per day in the bathroom, which was less than the 442.4 1 PN our control group used. Source: https://www3.epa.gov/watersense/our_water/water_ use_today.html.
(8.) Delmas, Fischlein, and Asensio (2013) conducted a meta-analysis of 156 experiments described in 59 papers, including nine papers published in economics journals, on energy conservation from 1975 to 2012. While their analysis found the effects of social comparison to be non-significant, they suggested that this was due to small sample sizes in the focal experiments and called for large-scale field experiments.
(9.) Because the guests in our sample stayed at the hotel for only one night, we could assess only a short-run effect. Regarding the long-run or persistent effect, one may wonder whether our nudged guests continued to save water after they returned home. We suspect such behavior to be unlikely but suggest it as a potential future research topic.
(10.) The concept of a social goal used in our paper is similar to that of shared social responsibility (SSR), originally proposed in Gneezy et al. (2010). SSR is an advanced conception of corporate social responsibility (CSR) because SSR involves the sharing of social responsibility among both firms and consumers. Regarding CSR, please refer to, for example. Sen and Bhattacharya (2001) and Becker-Olsen, Cudmore, and Hill (2006).
(11.) Gyeongju is located on the southeastern coast of the Korean Peninsula and has good accessibility. Via express train, it takes 2 hours to arrive there from Seoul, Korea's largest city, and only 30 minutes from Busan, the second largest city (travel times by car are approximately double those figures). Gyeongju was the capital of the ancient kingdom of Silla from 57 BC to 935 AD, and many of the historical sites there have been well-preserved, making it one of Korea's most popular historical tourist destinations. According to the 2014 Korea National Tourism Survey, around 2.8 million Koreans visited Gyeongju solely for the purpose of sightseeing during that year.
(12.) Our experimental period did not coincide with the city's peak season for tourism, which is summer. The timing decision was not ours but the hotel owner's. Thus, we do not know whether our findings can be extended to the peak season, when water usage tends to be much higher.
(13.) Of the 66 rooms, 26 were assigned to condition C, 20 to T1, and 10 each to T2 and T3. Please refer to Appendix A for details about treatment assignment and floor plans.
(14.) See Appendix B for the full message. The message card used in T3 included another sentence explaining the state of the campaign (i.e., "Many guests who have stayed in the hotel have participated in our campaign ..."). This sentence may have functioned as a normative-component treatment. Thus, the effect of T3 may capture the total effect of the social goal and stated information.
(15.) In Korea, both domestic and foreign travelers tend to stay at a hotel in a city for 1 or 2 days and then move to another hotel in a different city, due to the convenience in traveling between cities. As a result, we examine only single-night stays. However, this is another issue that might affect generalizability. Probably, the effects of such nudges are short-lived, and therefore tourists who spend an entire week at a hotel do not have such large treatment effects on the last day as they do on the first. In our final sample, guests actually checked in during 16 days in February (February 7, 8,9, 10, 11, 12, 13, 14, 15, 16, 17, 19, 20, 21, 22. and 28) and 14 days in March (March 1,6,7,8, 13, 14, 15, 20,21,22,24, 27, 28, and 29).
(16.) As a robustness check, we re-estimated the treatment effects for the sample excluding the adjoining rooms assigned to different conditions and found similar results. The results are presented in Appendix D.
(17.) Our hotel rooms consist of one or two bedrooms, one living room with a kitchenette, and one bathroom as presented in Table Al. There were approximately 10% of guest parties with more than six people staying in a room in our dataset.
(18.) Allcott (2011) points out that while social comparison with neighbors can reduce electricity usage, the welfare effect is ambiguous because the cost of cutting electricity usage is unknown and being compared with one's neighbors may affect consumers' welfare directly. Allcott and Kessler (2015) measured the value of the information obtained via social comparison using an incentive-compatible elicitation method. Such an elicitation method is difficult to implement with hotel guests.
(19.) The response rate is high as compared to those achieved in similar surveys in the literature but still lower than the commitment participation rate in T2. This may be because the survey was conducted at check-out and some guests were time-constrained at that moment.
(20.) The flow rate of a shower head in the hotel rooms was about 19 1 per minute, similar to that of a standard conventional shower head. Source: http://www.gracelinks.org/437/water-saving-tips-in-the-bathroom
(21.) We used the power and sample size analysis tools in PASS software.
(22.) We also computed extreme value bounds based upon the worst and best possible cases: the maximum outcome for all nonrespondents in the treatment groups and the minimum outcome for all nonrespondents in the control group, and vice versa. However, they produced extremely wide intervals and were thus uninformative because the response rate was sufficiently low. We report the estimates in Table 7 for those who are interested.
(23.) We estimate the probability of responding to the survey by using a logit model where the control variables used in our main model are included as explanatory variables. See Table El for the results for the logit estimation.
Caption: FIGURE 1 Quantile Distributions of Water Use Volume by Treatment: (A) Water Use PN: T1, (B) Water Use PN: T2, (C) Water Use PN: T3, (D) Water Use PPPN: T1, (E) Water Use PPPN: T2, and (F) Water Use PPPN: T3
Caption: FIGURE 2 Unconditional and Conditional ATEs: (A) Water Use PN and (B) Water Use PPPN
TABLE 1 Descriptive Statistics for Guest Parties Variables Total C T1 T2 T3 (1) (2) (3) (4) (5) Total number of 4.41 4.59 4.20 4.34 4.58 guests (1.61) (1.48) (1.54) (1.56) (1.75) Ratio of women to 0.38 0.37 0.38 0.38 0.37 total guests (0.11) (0.11) (0.12) (0.11) (0.11) Ratio of children 0.28 0.29 0.28 0.28 0.28 to total guests (0.19) (0.19) (0.19) (0.18) (0.20) Family vacation 0.75 0.77 0.74 0.76 0.73 (0.44) (0.42) (0.44) (0.43) (0.45) Number of rooms 66 26 20 10 10 Number of 306 44 88 79 95 observations p value Variables T1 T2 T3 Joint (6) (7) (8) (9) Total number of .17 .39 .97 .20 guests Ratio of women to .65 .87 .96 .66 total guests Ratio of children .85 .76 .82 .90 to total guests Family vacation .67 .87 .56 .93 Number of rooms Number of observations Notes: Standard deviations are given in parentheses. Control condition C indicates no behavioral intervention. Treatment condition T1 indicates message, T2 commitment, and T3 social goal. The unit of water use is liters. The number of observations for each condition is the number of guest party check-ins. To ensure randomization, we conduct a t-test for the null hypothesis that the mean values of the explanatory variable in each treatment condition (T1, T2, or T3) is the same as that of the corresponding variable in C, and the p values are reported in columns (6)-(8). Also, the p value for the joint hypothesis is reported in column (9). TABLE 2 Mean Water Use Volume Variables Total C T1 T2 (1) (2) (3) (4) Water use PN 390.46 442.41 383.70 357.23 (158.29) (163.63) (151.34) (144.75) Water use PPPN 88.96 97.16 91.82 81.92 (16.39) (21.71) (13.81) (13.30) Number of rooms 66 26 20 10 Number of 306 44 88 79 observations Difference Variables T3 T1-C T2-C T3-C (5) (6) (7) (8) Water use PN 400.28 -58.71 -85.18 -42.13 (167.71) Water use PPPN 88.37 -5.34 -15.24 -8.79 (15.88) Number of rooms 10 Number of 95 observations Notes: Standard deviations are given in parentheses. Control condition C indicates no behavioral intervention. Treatment condition T1 indicates message, T2 commitment, and T3 social goal. The unit of water use is liters. The number of observations for each condition is the number of guest party check-ins. TABLE 3 Summary of Post-Stay Surveys Variables Total C T1 T2 T3 (1) (2) (3) (4) (5) Overall satisfaction 2.54 2.60 2.52 2.37 2.63 (0.08) (0.23) (0.15) (0.17) (0.12) Intention to revisit 2.52 2.50 2.58 2.35 2.60 (0.08) (0.23) (0.15) (0.17) (0.12) Willingness to 2.49 2.50 2.53 2.30 2.57 recommend (0.08) (0.23) (0.15) (0.17) (0.12) Response rate 232 30 59 54 89 (76%) (68%) (67%) (68%) (94%) Number of 306 44 88 79 95 observations p value Variables T1 T2 T3 Joint (6) (7) (8) (9) Overall satisfaction 0.70 0.63 0.68 0.78 Intention to revisit 0.69 0.92 0.65 0.79 Willingness to 0.71 0.88 0.70 0.72 recommend Response rate Number of observations Notes: Standard errors are given in parentheses. Control condition C indicates no behavioral intervention. Treatment condition T1 indicates message, T2 commitment, and T3 social goal. The number of observations for each condition is the number of guest party check-ins. The answers were given on a Likert scale ranging from 1 (Very Poor) to 5 (Excellent). We conduct a Pearson's Chi-square test the null hypothesis that on average, there is no difference between the control and each treatment condition (T1, T2, or T3) with respect to each customer satisfaction measure, and the p values are reported in columns (6)-(8). Also, the p value for the joint hypothesis is reported in column (9). TABLE 4 Results for Water Use PN Treatments T1 T2 Variables (1) (2) (3) Treatment condition: Message (T1) -58.705 * -35.178 ** (29.632) (13.838) Commitment (T2) -85.182 ** (30.397) Social goal (T3) Characteristics of guest parties: Ratio of women to 55.442 total guests (67.446) Ratio of children to -41.476 total guests (62.798) Family vacation 4.677 (21.558) Constant 442.410 *** 189.355 *** 442.410 *** (21.307) (32.319) (21.335) Number of guests dummies No Yes No Weekend/holiday dummy No Yes No Observations 132 132 123 R-squared 0.031 0.735 0.069 Treatments T2 T3 Variables (4) (5) (6) Treatment condition: Message (T1) Commitment (T2) -72.923 ** (17.136) Social goal (T3) -42.128 -52.685 ** (27.283) (20.183) Characteristics of guest parties: Ratio of women to -5.641 162.984 total guests (72.605) (142.733) Ratio of children to -195.212" -67.272 total guests (70.094) (128.028) Family vacation 31.968 37.401 (22.960) (35.676) Constant 239.801 *** 442.410 *** 160.375 ** (34.283) (21.264) (69.931) Number of guests dummies Yes No Yes Weekend/holiday dummy Yes No Yes Observations 123 139 139 R-squared 0.792 0.014 0.743 Treatments T3 All Variables (7) (8) Treatment condition: Message (T1) -58.705 * -32.390 * (29.509) (16.368) Commitment (T2) -85.182 *** -72.939 *** (30.230) (17.705) Social goal (T3) -42.128 -47.090 ** (27.224) (17.886) Characteristics of guest parties: Ratio of women to 47.425 total guests (54.037) Ratio of children to -69.153 total guests (46.665) Family vacation 21.010 (14.114) Constant 442.410 *** 206.505 *** (21.218) (25.545) Number of guests dummies No Yes Weekend/holiday dummy No Yes Observations 306 306 R-squared 0.029 0.777 Notes: The clustered robust standard errors by check-in date are given in parentheses. The unit of water use PN is liters. *, **, and *** indicate significance at the 10%, 5%. and 1% levels, respectively. TABLE 5 Results for Water Use PPPN Treatments T1 T2 Variables (1) (2) (3) Treatment condition: Message (T1) -5.333 -7.487 *** (3.254) (2.497) Commitment (T2) -15.239 *** (3.245) Social goal (T3) Characteristics of guest parties: Ratio of women to 12.974 total guests (14.880) Ratio of children -7.459 to total guests (13.876) Family vacation 0.857 (4.740) Constant 97.155 *** 92.669 *** 97.155 *** (3.536) (7.040) (3.540) Number of guests dummies No Yes No Weekend/holiday dummy No Yes No Observations 132 132 123 S-squared 0.022 0.139 0.162 Treatments T2 T3 Variables (4) (5) (6) Treatment condition: Message (T1) Commitment (T2) -16.130 *** (3.034) Social goal (T3) -8.781 ** -11.222 *** (3.501) (3.631) Characteristics of guest parties: Ratio of women to 1.979 41.037 total guests (15.406) (28.833) Ratio of children -36.445 *** -6.213 to total guests (14.275) (23.681) Family vacation 4.286 6.142 (4.357) (6.920) Constant 100.705 *** 97.155 *** 84.018 *** (7.122) (3.529) (14.081) Number of guests dummies Yes No Yes Weekend/holiday dummy Yes No Yes Observations 123 139 139 S-squared 0.347 0.050 0.184 Treatments All Variables (7) (8) Treatment condition: Message (T1) -5.333 -6.725 ** (3.240) (2.833) Commitment (T2) -15.239 *** -16.869 *** (3.227) (3.041) Social goal (T3) -8.781 ** -9.975 *** (3.493) (3.251) Characteristics of guest parties: Ratio of women to 12.746 total guests (12.914) Ratio of children -11.029 to total guests (9.760) Family vacation 3.365 (2.965) Constant 97.155 *** 96.133 *** (3.521) (6.715) Number of guests dummies No Yes Weekend/holiday dummy No Yes Observations 306 306 S-squared 0.093 0.151 Notes: The clustered robust standard errors by check-in date are given in parentheses. The unit of water use PPPN is liters. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. TABLE 6 Examining Differences in Characteristics among Survey Respondents Difference between "Response" and "No Response" Groups Variables C T1 T2 T3 (1) (2) (3) (4) Total number -0.29 0.31 -0.09 -0.81 of guests (0.48) (0.35) (0.38) (0.74) Ratio of women 0.01 -0.03 -0.01 0.03 to total guests (0.04) (0.03) (0.03) (0.05) Ratio of children -0.04 0.05 0.01 -0.01 to total guests (0.06) (0.04) (0.04) (0.09) Family vacation -0.23 0.18 1 x [10.sup.-3] 0.06 (0.13) (0.10) (0.10) (0.19) Water use PN -61.21 41.37 -2.30 -29.45 (52.75) (34.23) (35.24) (71.05) Water use PPPN -3.00 3.34 -0.09 7.36 (7.09) (3.13) (3.24) (6.69) Number of rooms 26 20 10 10 Number of 44 88 79 95 observations p value for [H.sub.0]: "Response" = "No Response" and [H.sub.1] : "Response" [not equal to] "No Response" Variables C T1 T2 T3 (5) (6) (7) (8) Total number 0.56 0.39 0.82 0.28 of guests Ratio of women 0.86 0.30 0.85 0.58 to total guests Ratio of children 0.56 0.23 0.77 0.92 to total guests Family vacation 0.09 * 0.08 * 0.99 0.74 Water use PN 0.25 0.23 0.95 0.68 Water use PPPN 0.67 0.29 0.98 0.27 Number of rooms Number of observations Notes: The standard errors for differences are given in parentheses. Control condition C indicates no behavioral intervention. Treatment condition T1 indicates message, T2 commitment, and T3 social goal. The unit of water use is liters. The number of observations for each condition is the number of guest party check-ins. To check for differential attrition, we conducted a t-test for the null hypothesis that in a given condition, the mean value of the explanatory variable for those who responded to the survey is the same as that of the corresponding variable for those who did not. indicates a significant difference at the 10% level. TABLE 7 Results for Post-Stay Surveys Dependent Variable: Variables Overall Intention Willingness Satisfaction to Revisit to Recommend (1) (2) (3) Treatment condition: Message (Tl) -0.003 0.079 0.057 (0.111) (0.079) (0.081) [-0.34,0.32] [-0.31,0.35] [-0.32,0.33] Commitment (T2) -0.077 0.021 0.006 (0.124) (0.095) (0.092) [-0.37,0.27] [-0.32,0.32] [-0.34,0.30] Social goal (T3) 0.015 0.093 0.079 (0.114) (0.086) (0.081) [-0.25,0.13] [-0.21,0.16] [-0.23,0.15] Characteristics of guest parties: Ratio of women -0.848 -0.435 -0.519 guests (0.870) (0.759) (0.756) Ratio of -0.186 0.129 0.059 children guests (0.572) (0.491) (0.483) Family vacation -0.183 * -0.281 *** -0.253 ** (0.099) (0.093) (0.091) Constant 0.654 0.393 0.451 (0.461) (0.383) (0.383) Number of guests Yes Yes Yes dummies Weekend/holiday dummy Yes Yes Yes Observations 232 232 232 R-squared 0.071 0.078 0.074 Notes: The table reports the results of the weighted linear probability model. To account for missing answers, we compute the weights by taking the inverse of the logit probability of responding to the post-stay survey for each experimental condition. Extreme value bounds are given in brackets. The clustered robust standard errors by check-in date are given in parentheses. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.
|Printer friendly Cite/link Email Feedback|
|Author:||Joo, Hailey Hayeon; Lee, Jungmin; Park, Sangkon|
|Date:||Jul 1, 2018|
|Previous Article:||URBAN TRANSPORTATION MODE CHOICE AND TRIP COMPLEXITY: BICYCLISTS STICK TO THEIR GEARS.|
|Next Article:||ASSIMILATION IN THE RISK PREFERENCES OF SPOUSES.|