Game-based education promotes sustainable water use

3 In this study we estimate the impact of a game-based educational program aimed at 4 promoting sustainable water usage among 2nd-4th grade students and their families 5 living in the municipality of Lucca, Italy. To this purpose we exploited unique data from 6 a quasi-experiment involving about two thousand students, one thousand participating 7 (the treatment group) and one thousand not participating (the control group) in the 8 program. Data were collected by means of a survey that we speciﬁcally designed and 9 implemented for collecting students’ self-reported behaviors. Our estimates indicate 10 that the program has been successful: the students in the program reported an increase 11 in e ﬃ cient water usage and an increase in the frequency of discussions with their parents 12 about water usage; moreover, positive e ﬀ ects were still observed after six months. Our 13 ﬁndings suggest that game-based educational programs can be an e ﬀ ective instrument 14 to promote sustainable water consumption behaviors in children and their parents.


25
Sustainable water consumption is, in many cases, an instance of prosocial behaviour in a 26 social dilemma (Hardin, 1968): a situation in which a conflict exists between maximizing 27 one's individual benefits and maximizing the benefits of the present and future generations. 28 Individuals who are purely self-interested are less likely to adopt the prosocial behaviors that 29 lead to sustainable water consumption, unless social norms exert sufficient social pressure books, group play, and organized gaming (Orlick, 1983). In particular, the kind of social 39 interactions that come from group play and organized gaming, as well as the time that finding that people who received normative information about similar household in their 46 neighborhoods consumed less water than the control group; moreover, people with already 47 strong personal norms were less affected by the normative information than those with 48 low personal norms. Importantly, children are able to recognize if prosocial norms apply to 49 specific situations (Blake et al., 2015), so that it becomes important that children understand 50 what is sustainable water consumption and can relate their behavior to concrete and specific 51 situations such as water collection or body washing. 52 In this paper we provide evidence regarding the effectiveness of a game-based educational 53 program implemented during the first eleven months of the year 2019 in the municipality 54 of Lucca, Italy. The program was named BLUTUBE: Who brings the water home and was 55 aimed at promoting sustainable water consumption as well as awareness about the municipal 56 water system and its usage. The targets of the program were about 1000 students from 2nd-57 4th grades and their families. The program relied primarily on ludic engagement for the 58 specific objectives of improving students' awareness about the water cycle in nature, the 59 water system of the municipality of Lucca, and the daily usage of water.

60
Our approach to the empirical assessment of the program's impact is based on the quasi-61 experiment methodology (Campbell and Stanley, 2015): we had no possibility to intervene di-62 rectly on the organization of the program, but we were able to implement a simple two-group 63 design (treatment and control) and collect three distinct measurements of target outcome 64 variables over a period of eleven months. In particular, we elicited the students' awareness 65 and their behaviors about water consumption with three waves of surveys administered, re-66 spectively, immediately before the program started, some days after the main activities were 67 over, and after six further months. Responses to this kind of questionnaires have been shown 68 to be a reliable source of information on children's perspectives and perceptions (Danielson 69 and Phelps, 2003).

70
Our main finding is that the program had a positive impact on the awareness of water 71 usage. This effect is primarily driven by an increase in the frequency of self-reported virtuous 72 behaviors regarding water consumption and discussions with parents about water. Moreover, 73 such positive effect appears to be persistent: six months after the end of the main activities 74 of the program the effect is still positive and of appreciable size.  Table 2 in Appendix A). Summary statistics by treatment and control groups for the pre-program survey show 85 that the two groups are not well balanced (see Table 3 in Appendix A): while the difference 86 in the number of students per class is only marginally not statistically significant (Z = 87 −1.95, p = .051), the difference in the measured students' cognitive skills is statistically 88 significant (Z = −2.30, p = .031) as well as the distribution of grades (Z =4.99, p <.0001).

89
These differences are mainly due to the fact that the distribution of students across grades  We also lo oked at the distribution of answers in the pre-program survey for each of the 102 7 questions, testing for statistically significant differences. In four cases we found that the

108
In the light of these results we adopt a two-step strategy. First, we carry out a non-109 parametric analysis of the treatment effect on the aggregate reported behavior. This is 110 possible because, although the treatment and control groups are not perfectly balanced, the 111 aggregate variable comes with similar levels in the two groups for the pre-program survey. 112 We then check the robustness of non-parametric results by running regressions for each wave, 113 including controls for the sample characteristics in order to correct for the lack of sample 114 balancedness. 115 Second, we study the treatment effect on the reported behavior for each of the 7 questions 116 using ordered logit regressions where we pool all data and we control for sample character-    Distributions in the post-program and 6 months after are shifted to the right in the treatment group, with a statistically significant differences between conditions. ES stands for Epps-Singleton test.
respect to the pre-program treatment group (Z = −9.055, p < 0.001) and a statistically 145 significant higher average of about 1.72 with respect to the post-treatment control group 146 (Z = −7.479, p < 0.001). These numbers range from 1.32% to 7.04% of the pre-program av-147 erage, suggesting that the treatment has had a substantial impact between the pre-program 148 and the post-program surveys.
149 Furthermore, Figure 3 shows that there is no appreciable difference between the aggre- the aggregated reported behavior in the treatment group is not significantly different from the control group (Mann-Whitney test, Z = −1.300, p =0.193). In the post-program period and after 6 months, the aggregated reported behavior in the treatment group is significantly higher respect to the control group (Mann-Whitney test, Z = −7.479, p < 0.001 and Z = −5.271, p < 0.001, respectively). The treatment effect is stable after 6 month (Mann-Whitney test, Z =0.165, p =0.869). Error bars represents the 95% confidence interval.
The findings described above rely on the assumption that the lack of balance between 158 treatment and control groups did not bias our estimates. In order to control for such potential 159 problem we run linear regression models where aggregated reported behavior is predicted 160 by the treatment and a number of controls. Importantly, since students came from different 161 schools and classes, and that in one school there is the possibility to have more than one 162 class treated, we are able to control for schools including school fixed effects. In addition, 163 besides a dummy variable for the treatment (which is equal to 1 if the student belongs to 164 the treatment group), we include a dummy for the grade (omitted category is 2nd grade), 165 an index of cognitive skills (fraction of correct answers in logical/mathematical questions), 166 and the number of students in the class. We run similar regressions for the pre-program, the 167 post-program, and the post6-program surveys. Results are reported in Table 1.
168 The dependent variable is the aggregated reported behavior on good/bad practices of water usage. Treatment is equal to 1 if the students are in the treatment group, 0 otherwise. Grade is the students' year group. 2nd Grade is the reference category. Cognitive Skills is equal to 1 if the result obtained in the logical and mathematical questions are higher than the median, 0 otherwise. Students is the number of students in each class. In all cases, we control for school fixed effects. Standard errors (in parenthesis) are clustered at class level. * p<0.05, ** p<0.01, *** p<0.001.
Column (1) reports the results of the school-fixed effect regression in the pre-program suggesting that in the longer run cognitive skills might be a substitute for grade seniority.  Table 3,3o u to f7 196 reported behaviors (Teeth, Hands,a n dP arents)a p p e a rt ob es t a t i s t i c a l l yd i ff e r e n ti nt h e 197 pre-program survey, with the treatment group coming with a higher mean.

198
Looking at the differences between treatment and control groups in the post-program 199 survey, we find that 4 out of 7 variables show a statistically significant difference, with

207
In order to control for potential confounding factors that potentially persisted across the 208 three waves -and which could explain the differences described above -we pool data of Fountain and Waste, when also a positive treatment effect on Shower is found.

221
These results suggest that the program has had a positive effect especially on two di-  effects beyond students' own classes, that is, students in the control group might have been 260 exposed indirectly to the program through their social connections outside their own classes.

261
This latter explanation would imply that the treatment effect is far larger that our estimates 262 indicate. With our data we cannot establish which explanation works better. Additional 263 specific data have to be collected for this purpose.

264
Astandardlimitat ionofquasi-e xpe rime n tisthat,sinc etherandomiz ationprotoc olc an-265 not be managed directly, one cannot conclude about the causal effect of the treatment. We 266 think that such limitation, although not absent, is less severe in our study because the assign-267 ment procedure was largely exogenous to students' and teachers' desires, with constraints 268 for eligibility and required participation that left little room for self-selection. Moreover, 269 we could control for systematic differences in the characteristics of control and treatment 270 groups, such as grade, cognitive abilities, class size, and school.

271
Another limitation of this study is that we could only use self-reported behavior and the program, went to each class participating in the program to give a short talk on the 314 importance of water resources and their consumption as well as to explain the working of 315 the gaming activities (program phases, allocation of game points, publication of rankings) 316 and, in particular, to teach students how to play the board game BLUTUBE. Moreover, 317 each student got its own box of the board game (for playing at home) and each class was 318 also endowed with a copy of the board game (for playing in class).

319
The second phase was titled Bring the water to your mill and lasted 6 weeks during which 320 the students participating in the program had the chance to play as much as they wanted, 321 and accumulate points accordingly, for two distinct rankings: the individual ranking and the 322 class ranking. There were four different ways to obtain points:

323
• playing the board game BLUTUBE at school: each student can play during school time.

324
The teacher records each time a student plays on a scoreboard and each week a picture    This participation protocol allows the applicability and effectiveness of our method of 378 analysis, in that the assignment to the program, although not fully randomized, is to a good 379 extent exogenous to schools, students' and teachers' preferences.  The study includes 28 primary schools. From those schools, 53 classes were directly 387 involved in the program, forming the treatment group. For the control group we selected 388 other 53 classes that were not directly involved in the program, trying to build the best 389 possible counterfactual. This was not an easy task because the total of 106 classes covers 390 about the 90% of the entire population of 2nd-4th grades students in the municipality of 391 Lucca (the overall number of classes being 116). So, together the treatment and control 392 groups represent almost the entire student's population. 393 Students' awareness about the efficient use of water was elicited by means of a paper-394 based survey regarding students' behaviours and habits related to water use and consumption 395 (the original and the English-translated questionnaires can be found in Appendices C.3 and 396 D, respectively). Specifically, the survey contained seven distinct questions about water 397 consumption in familiar circumstances, the extent to which students talk about water with 398 their parents, and the extent to which students eat food containing water (fruit and vegeta-399 bles). These questions are: "How much do you keep the faucet turned on when you brush 400 your teeth?"; "Are you having more often a bath or a shower?"; "Do you drink water more 401 from plastic bottles or from fountains/faucets?"; "Are you eating fruit or vegetables during 402 your meals?"; "When you wash your hands, do you turn the faucet off while you soap your 403 hands?"; "Do you talk with your parents on how the water gets to your house?"; "Do you 404 talk with your parents on how not to waste water?". Answers were recorded using a 1-to-5 405 Likert scale.

406
The survey also contained questions related to relational activities, ludic habits and 407 ludic preferences, that we do not exploit in the following analysis as they were meant for 408 different research purposes. In addition, we tried to measure cognitive skills using logical and  The first survey was collected during February 2019, before the beginning of the program.

413
The parents of students involved signed an informed consent form, with the specific consent 414 for the possibility to link students' answers to their scores in the program. Teachers received 415 only general information about the research project, and specifically no details about what 416 we were trying to elicit. The second survey was administered at the end of the second phase, 417 during the month of May 2019. The survey was identical to the previous one but for the 418 questions aiming at eliciting cognitive skills which we opted to substitute with new ones 419 of comparable difficulty. To ensure consistency, the second survey was administered to the 420 classes involved following the same procedures as in the first wave. Lastly, a third survey was 421 administered six months after, when the program was officially over. This last survey was 422 identical to the previous two but for the questions aiming at eliciting cognitive skills. Also in 423 this case the survey was administered to the classes involved following the same procedures 424 as in the first two waves.