The effect of emotions and framing in a downstream water pollution game

Common-pool resources and other shared resources frequently suffer from overextraction/overuse and associated negative externalities. In this paper we design a framed laboratory experiment on downstream water pollution to investigate (a) the importance of framing in determining the behavior of upstreamers regarding the negative externalities, and (b) the potential of downstreamers to influence the choices of upstreamers using non-monetary sanctions and rewards, alleviating the need for intervention by the local governments and regulatory institutions. Our results show that framing has a significant impact on the behavior of subjects. Subjects behaved more profit-oriented in the self-interest framing and more egalitarian in the empathy framing. In addition, we show that nudging subjects to “walk in the shoes of others” significantly increased empathetic behavior. Lastly, negative emotional feedback is a powerful tool for changing behavior of subjects towards more environmentally friendly and empathetic behavior. Interestingly, positive emotional feedback is counterproductive in that it instead decreases environmentally friendly and empathetic behavior. In general our results indicate that explicit emotional feedback, even though not expressed by everyone, works similarly to the implicit appeal to emotions through framing.

in certain contexts it is impossible to avoid framing without making the task irrelevant.A number of studies examine the effect of framing of instructions on individual behavior in the context of corruption (Abbink and Hennig-Schmidt 2006;Barr and Serra 2009), public goods games (Andreoni 1995;Cookson 2000;Park 2000;Fujimoto and Park 2010;Cubitt et al. 2010Cubitt et al. , 2011)), sequential bargaining games (Brosig et al. 2003), and altruistic giving in a dictator game (Duffy and Kornienko 2010).The existing experiments manipulating the framing of instructions can be separated roughly into two categories: procedural-oriented framing (which compares the different ways of representing the problem, e.g.neutral/off-context vs. contextualized, or the type of payoff presentation) and priming-oriented framing (which suggests a particular value judgment, e.g.selfish vs. altruistic, positive vs. negative, take vs. give).
The effect of procedural-oriented framing is ambiguous.In bribing games, Abbink and Hennig-Schmidt (2006) showed that there is no difference between individual behavior exposed to neutral and framed/contextualized instructions, while Barr and Serra (2009) reported the opposite finding.In public goods games, Cookson (2000) found that changing the presentational variables (e.g., the formulation of the payoff function, the type of comprehension task) has a strong effect on experimental results.In sequential bargaining games, Brosig et al. (2003) reported the existence of hot (ordinary sequential manner, the subjects respond to the choices made by their partner) versus cold effect (strategy method, the subjects submit a complete set of choices before play commences)2 , while earlier research by Brandts and Charness (2000) did not find it.
The effect of priming-oriented framing is more robust.Andreoni (1995) and follow-up studies by Park (2000) and Fujimoto and Park (2010) demonstrated that subjects contributed significantly more to a public good under positive framing (investing in a public good) compared to negative framing (investing in a private good).Along the same lines Cubitt et al. (2010) showed that there is a framing effect when expressing moral judgment about contributions, with free-riding in Give treatment being condemned more strongly than in Take treatment.However, when one controls for contributions, monetary punishment and selfreported emotions are not sensitive to Give versus Take manipulation.In the dictator game Duffy and Kornienko (2010) showed that the subjects give more in an altruistic treatment (dictators are ranked in descending order according to the amount they give) as compared to a control and to a selfish treatment (dictators are ranked in descending order according to the amount they keep for themselves).
This paper contributes to the ongoing discussion in the experimental literature about the framing effects by examining the priming-oriented framing effects in the context of environmental protection.Specifically, we compare two contextualized treatments (empathy framing and self-interest framing) to a context-free neutral wording of instructions in a downstream water pollution game.

Social (dis)approval through reward/punishment and emotions
Rewards and punishments are used to express social (dis)approval and increase fairness and cooperation in social dilemmas and situations involving distribution of income and assets.Experimental papers report that punishment is more effective than rewards in dictator games (Andreoni et al. 2003), public goods games (Dickinson 2001;Sefton et al. 2008;Sutter et al. 2010, but not in Walker and Halloran 2004), and common pool resource games ( van Soest and Vyrastekova 2006).Furthermore, Andreoni et al. (2003) show that both methods are complements rather than substitutes in enforcing the ideals/norms and reaching the specific objectives.This effect is particularly strong when the punishments and rewards are decided and implemented by the group members rather than by an exogenous institution (Sutter et al. 2010).
Actions of individuals, both helping and hurting, are strongly linked to positive and negative emotions of the affected party (Offerman 2002;Bosman and van Widen 2002;Xiao and Houser 2005).Rewards and punishments represent positive and negative reciprocity and as such they are behavioral means to express positive and negative emotions.According to Fehr and Gächter (2002, p.139) negative emotions are the proximate sources of punishment: "Free riding may cause strong negative emotions among the cooperators and these emotions, in turn, may trigger their willingness to punish the free riders."Similarly, "costly punishment might itself be used to express negative emotions" (Xiao and Houser 2005, p.7398); anger (regarding norm violations) and guilt are sufficient to induce punishment (Nelissen and Zeelenberg 2009), and a higher intensity of reported positive/negative emotions increases the likelihood of reward/punishment (Offerman 2002).As positive and negative emotions are expressed in terms of reward and punishment, they in turn may also trigger emotions in a receiving party and cause a second round of reciprocation.
To the best of our knowledge there is no systematic investigation of the impact of expressed positive and negative emotions by the affected party back on the affecting party.This link does not seem to be straightforward.Hopfensitz and Reuben (2009) showed that the effectiveness of negative emotions, expressed by the punishing party, depends on the reaction of those who are punished.Specifically, if the expressed negative emotions lead to anger this may cause retaliation, whereas if shame and guilt are the result, it may restrain the desire of the punished party to fight back (Hopfensitz and Reuben 2009).It is very plausible that the effect of positive emotions will be similarly contradictory.Positive emotions, expressed by the rewarding party, can trigger positive reciprocity in the rewarded party and further increase cooperation and altruistic actions.At the same time positive emotions signal that the rewarding party is satisfied with the outcome, which may encourage the rewarded party to be a little more selfish next time around and test a lower threshold of acceptability.
In this paper we conduct a 2-round game in which after the first round the affected party can express their emotions regarding the outcome/behavior to the affecting party.This also allows testing how much this emotional response will influence the behavior of the affecting party in the second round.

Monetary versus non-monetary rewards and punishments
Several experimental studies demonstrate that non-monetary sanctions and rewards can be as efficient as monetary incentives in inducing cooperation and fair outcomes.Masclet et al. (2003) showed that contribution levels to public goods from both monetary and non-monetary sanctions (expressing a degree of disapproval) are similar, with non-monetary sanctions working better under partner, as compared to stranger-matching.Bochet et al. (2006) found that both face-to-face and verbal communication through a chat room (with anonymity and with no facial expressions) have stronger effects, than monetary punishment, on increasing contributions to public goods.Xiao and Houser (2009) showed in an ultimatum bargaining game that monetary punishment is more effective than informal sanctions (written messages) in promoting fairness.However, their results, taken together with Xiao and Houser (2005), indicate that as a sanctioning mechanism of expressing emotions is overall more beneficial for social welfare then costly monetary punishment, as it eliminates (pecuniary) welfare losses.López-Pérez and Vorsatz (2010) demonstrated that approval/disapproval 3 feedback fosters cooperation in the prisoner's dilemma games as it increases awareness about the other's feelings and because individuals are disapproval-averse.However, the effect of positive and negative feedback is not the same.According to Dugar (2010) costless ratings of disapproval assigned by group members lead to the most efficient equilibrium in the coordination game, whereas ratings of approval cause just the opposite.Finally, Noussair and Tucker (2005) found that monetary and non-monetary punishments are complements rather than substitutes as they appeal to different populations and at different time horizons.
In our experiment we test the effectiveness of non-monetary reward and punishment through a discrete and standardized expression of emotions in inducing a more fair distribution.

Design of the downstream water pollution game
We model the upstream-downstream situation by using a framed laboratory experiment on downstream water pollution.This is a three-player game.The first player is a farmer performing agricultural operations upstream (henceforth called Upstream Farmer or UF) who decides on the usage of conservation technology 4 .This technology (called conservation tillage or CT 5 ) lessens the negative impact of farming on the water quality in downstream rivers and lakes, but is more costly (compared to the alternative intensive tillage or IT 6 ) for the UF.The UF's payoff ( ) in tokens negatively depends on his usage of conservation technology, specifically the amount of land 7 she places under conservation tillage (˕ˠ {Ŵ ŹŴŴ{): ŹŴŴ -Ŷ {ŹŴŴ .˕ˠ { The second player is an individual who is drawing their drinking water from the lakes and rivers downstream (henceforth called Downstream Water User or DWU) and incurs the cost of cleaning the polluted water.The DWU's payoff ( ) in tokens positively depends on the cleanliness of the downstream lake: ŹŴŴ -ŵŴ { ˕ˬ˥IJˬ˩J˥JJ J˦ ˮ˨˥ HI˫˥{ The third player has a dual role: they perform farming operations upstream (thus deciding on the usage of conservation technology) and, at the same time, live on farms tied to rural water supply systems that draw water from the very stream passing through their own farmland, and are, hence, using drinking water from downstream (those individuals are called Upstream Farmer/ Downstream Water User or UF/DWU).The UF/DWU payoff ( È ) in tokens 8 3 A subject could select to send one of the following three messages: ''Your choice was (1) good, (2) neither good nor bad, and (3) bad" (López-Pérez & Vorsatz 2010, p.532). 4 The descriptions of agricultural technologies presented to the subjects in the instructions (Appendix B in the supplementary materials) were significantly modified and simplified to avoid any misunderstanding and ambiguity. 5The CT (minimal tilling of the land) was presented as a relatively lower profit practice.The CT was also presented as the tillage with relatively mild environmental impact: under CT the land is disturbed minimally leading to less soil erosion, lower chemical runoff and overall higher drinking water quality of the downstream rivers and lakes. 6The IT was presented as a relatively higher profit tillage practice for the agricultural operator.The IT was also presented as the tillage with relatively higher negative environmental impact: IT leads to soil erosion and greater chemical runoff, and, thus, significantly reduces the drinking water quality of the downstream rivers and lakes. 7The farmer has an endowment of 500 acres of land. 8Exchange rate $1=70 tokens negatively depends on her usage of conservation technology ˕ˠ È {Ŵ ŹŴŴ{, and positively on the cleanliness of the lake: The cleanliness of the lake is determined as the proportion of land placed under conservation tillage by the two farmers (UF and UF/DWU) 9 : ˕ˬ˥IJˬ˩J˥JJ J˦ ˮ˨˥ HI˫˥ ˕ˠ -˕ˠ È ŹŴŴ -ŹŴŴ ŵŴŴ The game represents a zero-sum game with three players 10 with the structure of the payoffs chosen so that each player gets an equal payoff of 1000 tokens if the farmers choose ˕ˠ ˕ˠ È ŶŹŴ acres, which leads to ˕ˬ˥IJˬ˩J˥JJ J˦ ˮ˨˥ HI˫˥ ŹŴ (for selected combinations of strategies and payoffs see the top half of Table 1).
The Nash equilibrium for both UF and UF/DWU (note, that for the UF/DWU the return on conservation technology is twice the return on the lake cleanliness) is to choose a zero level of conservation technology.In this sense the decision of UF and UF/DWU to use non-zero levels of conservation technology is similar to altruistic giving in a dictator game.

Priming-oriented framing and treatments
We implemented three manipulations: two framing (empathy and self-interest) treatments using loaded language in a contextualized situation and one no-framing treatment using neutral language.In the framing treatments the players were referred to as UF, DWU, and UF/DWU.They were presented with the description of the upstream-downstream situation and asked to make decisions regarding the choice of tillage on their land.In the NEUTRAL FRAME treatment the players were referred to as Player 1, 2, 3, and were presented with a 9 If each of the farmers will choose (independently) to place all their 500 acres of land under CT (˕ˠ ˕ˠ È ŹŴŴ), then the lake cleanliness will be 100%. 10Total social welfare is independent of the payoff distribution: the three players were sharing 3000 tokens.
situation written in a neutral language in which they had to allocate chips between Options A and B.
The two framing treatments differed in how the question was presented to the subject.In the EMPATHY FRAME the participants were choosing the level of conservation tillage which affected the cleanliness of the lake.In addition, they were primed to behave more empathetically.After the instruction and the quiz, but before they started the game, they were reminded that: "The choice of tillage by farmers will greatly affect the water quality of the lake and the payoff for the Downstream Water User.A cleaner lake and higher payoff for the DWU will be assured if the farmers choose to place more land under Conservation Tillage."In the SELF-INTEREST FRAME the participants were choosing the level of intensive tillage (less environmentally friendly technology) which affected the pollution of the lake.Similarly, they were primed to look for their own self-interest and were reminded that: "The choice of tillage by farmers will greatly affect their own profit.
The farmers get higher profit if they choose to place more land under Intensive Tillage."The payoff of the three players in the SELF-INTEREST FRAME was mirroring the presentation in the EMPATHY FRAME and was as follows: Where Hˠ {Ŵ ŹŴŴ{ and Hˠ È {Ŵ ŹŴŴ{ are the amounts of land11 placed by each of the farmers under intensive tillage and the lake pollution calculated as:

˜Jˬˬ˯ˮ˩JJ J˦ ˮ˨˥ HI˫˥ Hˠ -Hˠ È ŹŴŴ -ŹŴŴ ŵŴŴ
The players get an equal payoff of 1000 tokens if the farmers choose Hˠ Hˠ È ŶŹŴ acres which leads to ˜Jˬˬ˯ˮ˩JJ J˦ ˮ˨˥ HI˫˥ ŹŴ (for selected combinations of strategies and payoffs see the bottom half of Table 1).The decisions of the farmers in the EMPATHY FRAME and SELF-INTEREST FRAME treatments can be converted using the formulas: ˕ˠ ŹŴŴ .Hˠ and ˕ˬ˥IJˬ˩J˥JJ J˦ ˮ˨˥ HI˫˥ ŵŴŴ .˜Jˬˬ˯ˮ˩JJ J˦ ˮ˨˥ HI˫˥.For the purposes of the analysis, this conversion has been made and the rest of the paper will be discussing all results in terms of CT and the cleanliness of the lake.

Rounds of the game and emotional feedback
The game was played for two rounds.In each round the UF and UF/DWU12 chose the acreage under CT or IT13 (depending on the treatment).Next, the DWU received information about the level of cleanliness or pollution of the lake (but not the actual CT/IT choices of UF and UF/DWU)14 .
At the end of the first round the DWU was given the opportunity to express her positive or negative emotions by sending a smiley/happy face ☺ or a frowney/unhappy face to both farmers 15 .The faces were in the form of emoticons (Figure 1).Sending each face was costly for the DWU (50 tokens).If the emoticons were sent, they appeared on the screen of both farmers (Figure 2a, b).In order to minimally pollute the data, the payoffs of subjects were not shown to them until after the game (the UF and the DWU could have potentially calculated their earnings, but not the UF/DWU 16 as she was not aware of the cleanliness of the lake).After that, the game proceeded to the second (and last) round, at the end of which the DWU saw the cleanliness/pollution of the lake.* Cleanliness of the lake at 40% is given for reference purpose only.In the SELF-INTEREST FRAME instead of "% cleanliness of the lake" "% pollution of the lake" was displayed, in the NEUTRAL FRAME -"% of chips in Option B".
Although one-shot games are preferable for this type of analysis because they eliminate strategic considerations of players in the repeated game context, we do not believe that this applies to the first round of our game.Neither within one round nor across the two rounds were there any strategic considerations: the decision of the UF affects the payoff of the UF/DWU, but not vice versa; the decision of the UF affects the payoff of the DWU, but not vice versa; finally, the decision of the UF/DWU affects the DWU's payoff, but not vice versa (in the instructions the players were not told that the DWU will be able to send an emotional feedback and, thus, could not have reacted strategically to avoid/solicit it).comparison to each other.This also reflects the real life situation in which the downstream water users can observe only the outcomes of the actions of the upstream farmers, but not their actual actions/choices. 15Faces were identified as smiley/unhappy and frowny/unhappy based on the common usage of emoticons in electronic communications.Google search for a respective image out of the first 24 faces, 21 (smiley) and 22 (happy) clearly resemble ☺; 19 (frowny) and 18 (unhappy) clearly resemble (Appendix A in supplementary materials). 16We specifically did not want to show the cleanliness of the lake to the UF/DWU, as from the cleanliness they could have inferred the CT of the UF, and this knowledge could have affected their decisions in the second round.

Hypotheses
The hypotheses deal with the framing effect, the difference between the decisions of UF and UF/DWU, and the effect of the DWU expressing discrete emotions on the behavior of the UF and the UF/DWU.The framing effect, if present, will affect two decisions: the usage of conservation technology by the farmers and the likelihood to send a smiley or frowney emoticon by the water user.The null hypothesis is that there is neither a difference between the EMPATHY and SELF-INTEREST FRAME nor between the loaded manipulations and the NEUTRAL FRAME.For the farmers' decisions, the alternative hypothesis is that the framing affects farmers' choice of the CT level and the following pattern will be observed: HI˫˥˕ˬ˥IJˬ˩J˥JJ 2 HI˫˥˕ˬ˥IJˬ˩J˥JJ 2 HI˫˥˕ˬ˥IJˬ˩J˥JJ .The rejection of the null hypothesis would be in line with findings reported in similar situations by Andreoni (1995), Park (2000), Fujimoto and Park (2010), Cubitt et al. (2010), and Duffy and Kornienko (2010).As for the likelihood to send an emoticon, the alternative hypothesis is that the loaded framing will result in higher propensity to send a face than the NEUTRAL FRAME, and there will be relatively more smileys in the EMPATHY FRAME and relatively more frowneys in the SELF-INTEREST FRAME.However, in the case of emotional response, it is reasonable to expect that the null hypothesis will not be rejected as the DWU is more likely to be affected by the lake cleanliness and react on that, rather than on the priming or loaded language.This expectation is supported by the findings of Cubitt et al. (2011), who showed that the propensity to punish in the public goods game is driven by the contributions rather than by the framing.
From the self-interest perspective and even accounting for other-regarding preferences and empathy towards the DWU (since both farmers equally affect the lake cleanliness), the decisions of UF and UF/DWU about conservation technology are conceptually equivalent.The null hypothesis is that there will be no difference between the CT chosen by the two farmers within one treatment.The alternative hypothesis is that within one treatment ˕ˠ È 2 ˕ˠ .
The UF/DWU, by the nature of the situation, is better able to "walk-in-the-shoes" of the DWU, while the UF can only imagine how it would feel like.In that sense, the position of the UF/DWU is similar to a situation in which people play non-symmetric games with switching roles17 .The UF/DWU is more likely to engage in self-reflection and empathetic considerations and will have more realistic beliefs on how it feels like being a DWU.The beliefs of the UF, on the other hand, may be less realistic and, thus, their decisions are expected to be less empathetic.In this case we would expect to reject the null hypothesis in favor of the alternative.
A rejection would be corroborated by the difference between bargaining and dictator games in which the players are switching roles and the games in which they do not (see the discussion in Brosig et al. (2003, p.85) comparing their results to Brandts and Charness (2000).
Finally, consider the effect of positive or negative emotions expressed by the DWU on the decisions of the UF and UF/DWU in the second round.The null hypothesis is that the emotional response will have no effect on the choices.The alternative hypothesis is that both smiley and frowney faces will encourage the farmers to increase their CT usage, with frowney being more effective in achieving that goal.As positive and negative emotions are shown to be linked to monetary rewards and punishments (Offerman 2002;Bosman and van Widen 2002) and those monetary rewards and punishments affect behavior, it is reasonable to expect that we will reject the null hypothesis.Along the same lines, as punishments were shown to be more effective than rewards (Andreoni et al. 2003;Dickinson 2001;Sefton et al. 2008;Sutter et al. 2010; van Soest and Vyrastekova 2006), we would expect that expressing negative emotions will increase CT usage more than expressing positive emotions.

Procedures and subjects
Each subject was assigned a 5-digit random number to assure anonymity.The experimental instructions were read to the participants aloud and also presented on their computer screen 18 .In addition, each subject received two handouts: a payoffs table 11x11 with some of the possible combinations of strategies and a list of formulas to calculate the payoffs 19 .Following the instructions the experimenter answered questions and the participants were presented with a quiz checking their understanding of the instructions and the calculation of the payoffs.After all subjects successfully completed the quiz they proceeded to the experiment.
In total, 216 subjects participated in the experiment: 84 in each of the EMPATHY and SELF-INTEREST FRAME and 48 in the NEUTRAL FRAME treatments, resulting respectively in 28 and 16 independent observations per treatment.All subjects were recruited at XXX University (the majority were students, 45% females, age 19-74 years).The experiment was conducted in the Experimental and Behavioral Economics Laboratory.All sessions were computerized and administered using software z-Tree (Fischbacher, 2007).Each session was 70-100 minutes long.The tokens that the participants earned during the experiment (sum of payoffs in both rounds) were converted into dollars ($1=70 tokens) and paid to the participants privately in cash, with average earnings of $28.920 .

Experimental results and discussion
4.1.Decisions of the UF and UF/DWU in the first round 4.1.1.Lake cleanliness and conservation technology The decision of the farmers about the conservation technology led to quite high levels of lake cleanliness in the first round: 46.3%, 34.7% and 29.4% in the EMPATHY, NEUTRAL, and SELF-INTEREST FRAME, respectively.This result is in line with what one would expect given the findings reported in the previous literature, as discussed above.According to the nonparametric Kruskal-Wallis test (χ 2 (2)=7.74,p-value=0.02)there is a significant difference between the treatments.The analysis of each pair of treatments using the Mann-Whitney-Wilcoxon rank-sum test indicated that the difference is significant at 1% between the cleanliness in the EMPATHY and SELF-INTEREST FRAME (p=0.00899) and at 10% between EMPATHY and NEUTRAL FRAME (p=0.0648).
Result 1. Empathy priming leads to higher usage of conservation technology.Levels of conservation tillage varied significantly across the frames for the UF, but not for UF/DWU (Figure 3).According to the Kruskal-Wallis test and the Mann-Whitney-Wilcoxon rank-sum test there is no significant difference between the treatments for the CT levels chosen by the UF/DWU.The Kruskal-Wallis test indicated a significant difference between the treatments for the CT levels chosen by the UF (χ 2 (2)=12.26,p-value=0.0023).Furthermore, the Mann-Whitney-Wilcoxon rank-sum test showed a significant difference between the CT by UF in the EMPATHY and SELF-INTEREST FRAME (p=0.0005) and between the EMPATHY and NEUTRAL FRAME (p=0.0757).
Overall this leads to the rejection of the null hypothesis regarding the absence of a framing effect on the lake cleanliness, with the emphasis that the difference in the lake cleanliness is driven by the UF's behavior.In this game the ratio of UF and UF/DWU is 1:1, and one would expect more pronounced differences between the manipulations the more the ratio in the community is skewed towards UF.Notably, priming for empathy works (as compared to neutral framing), whereas priming for self-interest does not.Individuals behave similarly in the context-free situation and in an environmental context in which they are primed to look for their own self-interest.This suggests that the mere presence of environmental context tempers self-interest considerations.
Result 2. Self-interest considerations and environmental context have diametrical effects on behavior.

Difference between UF and UF/DWU choices
The hypothesis regarding rejected in one treatment and not rejected hardly any difference between the levels of conservation tillage chosen by UF and UF/DWU the EMPATHY FRAME (Mann-Whitney of land placed under the conservation tillage by UF and UF/DWU approximately equal payoffs for UF, UF/DWU and DWU: 1024, 1013, and 963 tokens respectively.In the NEUTRAL FRAME UF), however, not statistically significant ( p-value=0.23).SELF-INTEREST FRAME significant (Mann-Whitney-Wilcoxon rank and UF placing 98 acres under conservation tillage. Result 3. The forced "waking considerations.This result provides further support for the dual (Hayes and Lynne 2004;Lynne 2006 state of others in the community (and the ecosystem within which and as a result are conditioning their other-interest.This result, taken together with the raising the stakes of the upstreamers UF/DWU) inspires their environmental comparable to empathetic priming.regarding considerations.

Expression of emotions by the DWU
Downstreamers demonstrated substantial willingness to (costly) express their emotions: smiley or frowney emoticons were sent in more than 60% of the cases in the environmental

Difference between UF and UF/DWU choices
The hypothesis regarding no difference between the decisions of UF and UF/DWU is not rejected in two others.As is evident from Figure hardly any difference between the levels of conservation tillage chosen by UF and UF/DWU Whitney-Wilcoxon rank-sum test p-value=1).The 238 and 225 acres of land placed under the conservation tillage by UF and UF/DWU, respectively approximately equal payoffs for UF, UF/DWU and DWU: 1024, 1013, and 963 tokens RAME the difference is noticeable (203 for UF/DWU and 144 for significant (Mann-Whitney-Wilcoxon rank-sum test RAME is the only case where the differences are statistically Wilcoxon rank-sum test p-value=0.0369)with UF/DWU conservation tillage.The forced "waking-in-the-shoes-of-others" tempers self further support for the dual-interest and metaeconomic framework Hayes and Lynne 2004;Lynne 2006).This suggests that people are imagining themselves state of others in the community (and the ecosystem within which the economy is embedded) their own internalized pursuit of self-interest with the shared taken together with the aforementioned framing effect upstreamers in the downstream water quality (such as the case with inspires their environmentally conscious and fair behavior by a magnitude comparable to empathetic priming.In that sense salient incentives are as powerful as other Expression of emotions by the DWU Downstreamers demonstrated substantial willingness to (costly) express their emotions: were sent in more than 60% of the cases in the environmental  Positive emotions were expressed when the cleanliness of the lake depending on the treatment (Figure biased in favor of the DWU to trigger their positive emotions of about 20 percentage points of lake cleanliness between the treatments emotions were not expressed.However, there is a consensus regarding what must be punished by negative emotions -a cleanliness of the lake result in being sent. To test for the statistical significance constructed multinomial logit regressions in which we controlled for the cleanliness of the lake (Table 2).The dummies for treatment in the linear form (Model 2) as well as an interaction with cleanliness (Model 3) are not significant.Given this result we cannot reject the null hypothesis on the absence of a framing effect on sending emotional feedback.In line with the findings reported by Cubitt et al. (2011), in our experiment emotional punishments and rewa based on the more immediate payoff specific framing or priming.

% of subjects
framing treatments (consistent with imposing the monetary rewards and punishments in the tick treatment in Andreoni et al. 2003) and 56% of the cases in the neutral scenario.treatment 9 out of 10 emoticons sent were frowneys, while in the 10 were frowneys (Figure 4).expressed when the cleanliness of the lake was 50 Figure 5), suggesting that the distribution of payoff the DWU to trigger their positive emotions through ☺.There is also a spread of about 20 percentage points of lake cleanliness between the treatments in the cases emotions were not expressed.However, there is a consensus regarding what must be punished cleanliness of the lake of on average between 19.7% and 26.8% will To test for the statistical significance of the differences between the treatments we constructed multinomial logit regressions in which we controlled for the cleanliness of the lake (Table 2).The dummies for treatment in the linear form (Model 2) as well as an interaction with el 3) are not significant.Given this result we cannot reject the null hypothesis on the absence of a framing effect on sending emotional feedback.In line with the findings reported by Cubitt et al. (2011), in our experiment emotional punishments and rewa based on the more immediate payoff-relevant information (such as lake cleanliness), rather than 13 monetary rewards and punishments in the in the neutral scenario.In sent were frowneys, while in the 50-70%, payoffs has to be There is also a spread in the cases when emotions were not expressed.However, there is a consensus regarding what must be punished and 26.8% will of the differences between the treatments we constructed multinomial logit regressions in which we controlled for the cleanliness of the lake (Table 2).The dummies for treatment in the linear form (Model 2) as well as an interaction with el 3) are not significant.Given this result we cannot reject the null hypothesis on the absence of a framing effect on sending emotional feedback.In line with the findings reported by Cubitt et al. (2011), in our experiment emotional punishments and rewards are relevant information (such as lake cleanliness), rather than  Cleanliness of the lake is the only variable that affects the expression of emotions (Table 2).Positive emotions (in the form of ☺) are expressed significantly (at 5%-level) less often than no emotions, as is evident from the negative sign in front of INTERCEPT for ☺ in all three models.However, the downstreamers are quite eager to express their negative emotions (positive and significant coefficient in front of INTERCEPT for ).The cleaner the lake is the higher the probability that the DWU expresses positive emotions (in Model 1, for each 1 percentage point increase in lake cleanliness the odds ratio of sending ☺ over not expressing emotions is 1.049).Consequently, the cleaner the lake is, the lower the probability that the DWU expresses negative emotions (in Model 1, for each 1 percentage point increase in lake cleanliness the odds ratio of sending over not expressing emotions is 0.938 21 ).

Smiley
Result 4. The expression of emotions is driven by the outcomes and it is independent of framing.These findings suggest that subjects are very much concerned about fairness and are eager to express directly their emotions even though this expression is costly for them.Positive emotions are expressed when the outcomes are in their favor (higher than a fair/equal share).Negative emotions are expressed when the resulting payoff falls below 30% of the fair payoff.Expression of emotion is motivated by the payoff, rather than by a specific framing.

Effectiveness of emotions: the UF and UF/DWU choices in the second round
The comparison of the effectiveness of positive and negative emotions in promoting environmentally conscious and other-regarding behavior lead to (perhaps) an unsurprising result: the expression of negative emotions worked well and better than the expression of positive emotions.This is in line with findings that punishments are more effective than rewards (Andreoni et al. 2003;Dickinson 2001;Sefton et al. 2008;Sutter et al. 2010; van Soest and Vyrastekova 2006).The cleanliness of the lake in the second round increased on average by 14.7 percentage points in response to .Expressing positive emotions, on the other hand, backfired: the farmers realized that they would be able to squeeze out a little more profit and as a result the cleanliness of the lake decreased on average by 6.0 percentage points in response to ☺.The cleanliness stayed approximately the same (on average 0.6 percentage points reduction) if the emotions were not expressed.Intriguingly, the reaction of the two farmers on emotions was quite different (Figure 6): while the frowney emoticons led to a similar increase in the usage of conservation technology, smiley emoticons negatively affected the UF/DWUs, but not UFs.This rather striking result suggests that positive emotions signaled that the rewarding party was satisfied with the outcome which encouraged the rewarded party (UF/DWU) to yield to the temptation to be more selfish in the second round.

Figure 6. Absolute change in the usage of conservation technolo
The regression analysis (Table 3) provides a mor influence of emotions on outcomes in the second round.The dependent variable in Model 5 is the absolute difference between the cleanliness of the lake in the second and the first rounds.Only negative emotions can explain this difference dummy for is significant at 1% (Model 4), whereas interaction between the treatments and emotions (Model 5) reveals that negative emotions play a significant role only in the SELF significant).Sending a frowning face led cleanliness by 21.8 percentage points cleanliness was 29.4%, this suggests that effective and can, on its own, guarantee that seeming ineffectiveness of the emotional sanctions and rewards in the most likely, driven by the distributions of the already almost equal (1024, 1013, and 963 tokens emotional feedback was unable to the DWU (in the second round the average payoffs in for the UF, UF/DWU, and DWU Models 6-9 are estimated technology by the UF and UF/DWU in the second and the first round variable.As shown above, for both farmers increase in the CT usage only in the UF/DWU was more pronounced: more under CT, while the UF increased CT had a maximum of 500 acres, those numbers are quite high. Result 5. Expressing more fair distributions.nalysis (Table 3) provides a more detailed picture regarding the influence of emotions on outcomes in the second round.The dependent variable in Model 5 is the absolute difference between the cleanliness of the lake in the second and the first rounds.Only negative emotions can explain this difference -the coefficient in front of the is significant at 1% (Model 4), whereas the one for ☺ is not.The analysis of the interaction between the treatments and emotions (Model 5) reveals that negative emotions play ELF-INTEREST FRAME (as the rest of the coefficients are not ending a frowning face led, in this treatment, on average to an increase in the lake percentage points.Coupled with the fact that the first round average 29.4%, this suggests that the expression of negative emotions is extremely n its own, guarantee that the DWU gets their fair share of 1000 tokens.seeming ineffectiveness of the emotional sanctions and rewards in the EMPATHY most likely, driven by the distributions of the average payoffs in the first round already almost equal (1024, 1013, and 963 tokens respectively).And hence in the second round unable to sway the farmers to give up more of their payoffs in favor of the DWU (in the second round the average payoffs in EMPATHY FRAME were 1018, 1016, and 966 for the UF, UF/DWU, and DWU, respectively).
are estimated with the difference between the usage of conservation technology by the UF and UF/DWU in the second and the first round as the dependent , for both farmers a negative emotional feedback led to the SELF-INTEREST FRAME.However, the effect of UF/DWU was more pronounced: ceteris paribus the UF/DWU responded by placing increased CT "only" by 92.3 acres.Given that each of the farmers those numbers are quite high.Expressing negative emotions is effective in moving people towards more fair distributions.e detailed picture regarding the influence of emotions on outcomes in the second round.The dependent variable in Models 4 & 5 is the absolute difference between the cleanliness of the lake in the second and the first the coefficient in front of the not.The analysis of the interaction between the treatments and emotions (Model 5) reveals that negative emotions play (as the rest of the coefficients are not to an increase in the lake the first round average expression of negative emotions is extremely DWU gets their fair share of 1000 tokens.The MPATHY FRAME was, payoffs in the first round which were in the second round payoffs in favor of were 1018, 1016, and 966 the difference between the usage of conservation as the dependent led to a significant However, the effect of on placing 128.1 acres Given that each of the farmers is effective in moving people towards From the actions of UFs and UF/DWUs in the second round we get another confirmation of the power of actually "walking-in-the-shoes" of other people: the UF/DWUs are more responsive to social cues than the UFs in terms of initial decisions and reaction on the DWUs' disapproval.All but one coefficient in Models 6-9 are insignificant and thus definite conclusions cannot be made.Nevertheless, we can make the notable observation that positive coefficients in Models 8 & 9 are larger than the respective coefficients in Models 6 & 7, and all but one negative coefficient are smaller, suggesting higher sensitivity of UF/DWU as compared to UFs.It is reasonable to expect that with a larger sample size our confidence in this conjecture would increase.0.18 0.28 0.06 0.12 0.17 0.23 † -the dependent variable is censored between -100 and 100 ‡ -the dependent variable is censored between -500 and 500 Significance levels: *** -1%, ** -5%, * -10%, ′ -16%. On a larger scale the effectiveness of emotion expression is comparable to the empathetic framing (Figure 7).The non-parametric Kruskal-Wallis test (χ 2 (2)=2.40,p-value=0.30)showed that there is no significant difference between the treatments in the second round.The followup analysis of each pair of treatments using the Mann-Whitney-Wilcoxon rank-sum test indicated that a significant difference persists at the 10% level between the cleanliness in the EMPATHY and NEUTRAL FRAME (p=0.1011), while the formerly significant differences between the EMPATHY and SELF-INTEREST FRAME disappeared.This suggests that explicit emotional feedback, even though not expressed by everyone, works similarly to the implicit appeal to emotions through framing.

Conclusions and implications
Three important implications supports that priming-oriented framing is highly effective in changing the behavior of subjects.Empathy framing leads to a very equitable distribution of income, whereas self leads to more profit maximizing language-decontextualized scenario.(2008, p.312) paper that "the framing effects do not only cause problems; sometimes they m be seen as an asset for the researcher" experiments we should study its impact happen in neutral language and without context.environmental policy should be appealing to the Second, perhaps not surprisingly, the subjects with a higher water contributed much more to the cleanliness of the lake observe this effect despite the identical Nash equilibrium for with the dual-interest and metaecon that the UF/DWUs were placed into the water users and thus displayed more empathetic behavior than the UFs who imagine how it feels.In terms of policy this calls for in negative externalities of their behavior.In contrast to a traditional approach of fully internalizing the externality our results support that even a minor internalization, maybe even symbolic in nature, may suffice to significantly change behavior.
Third, negative emotional feedback (non positively influencing environmental choices and leading to more equitable distributions.As Xiao and Houser (2005, p.

Conclusions and implications
Three important implications can be derived from our study.First, our experiment framing is highly effective in changing the behavior of subjects.Empathy framing leads to a very equitable distribution of income, whereas selfmaximizing behavior.Both of these are different from the neutral scenario.In line with the comment of a referee on the Alpizar et al. (2008, p.312) paper that "the framing effects do not only cause problems; sometimes they m be seen as an asset for the researcher" we argue that rather than avoiding framing in all study its impact as interactions/communications in the field happen in neutral language and without context.In particular our results support that appealing to the empathy of polluters.Second, perhaps not surprisingly, the subjects with a higher stake in the quality of the lake contributed much more to the cleanliness of the lake than those with no stake observe this effect despite the identical Nash equilibrium for both types.This finding is interest and metaeconomics framework (Hayes and Lynne 2004;Lynne 2006 placed into the situation of "walking in the shoes of" downstream and thus displayed more empathetic behavior than the UFs who could only In terms of policy this calls for increased exposure of upstream users to the of their behavior.In contrast to a traditional approach of fully internalizing the externality our results support that even a minor internalization, maybe even symbolic in nature, may suffice to significantly change behavior. feedback (non-monetary punishment) is an effective tool in positively influencing environmental choices and leading to more equitable distributions.As ) noted "the desire to express emotions, and constraints on that demand, are a ubiquitous feature of human social interaction" so it is worth utilizing this desire in addition to the standard policies of monetary sanctions and rewards and putting more interest underlying such decisions.Our findings provide additional derived from our study.First, our experiment framing is highly effective in changing the behavior of subjects.
-interest framing behavior.Both of these are different from the neutralreferee on the Alpizar et al. (2008, p.312) paper that "the framing effects do not only cause problems; sometimes they may than avoiding framing in all in the field do not support that quality of the lake than those with no stake.We This finding is in line Hayes and Lynne 2004;Lynne 2006) in downstream could only creased exposure of upstream users to the of their behavior.In contrast to a traditional approach of fully internalizing the externality our results support that even a minor internalization, maybe even monetary punishment) is an effective tool in positively influencing environmental choices and leading to more equitable distributions.As he desire to express emotions, and constraints on that so it is worth utilizing this desire and putting more additional support for the effectiveness of policies that promote social punishment and public shaming as strategies to achieve lawful and/or cooperative behavior.

Figure 1 .
Figure 1.Example of a decision screen by DWU in EMPATHY FRAME*.

Figure
Figure 2a, b.Smiley and frowney emoticons received by the farmers from the DWU.

Figure 3 .
Figure 3. Conservation Tillage (CT) chosen by the farmers in the first round.

Figure 3 .
Figure 3. Conservation Tillage (CT) chosen by the farmers in the first round.

Figure 4 .
Figure 4. Percentage of DWUs sending an emotional feedback.

Figure 4 .
Figure 4. Percentage of DWUs sending an emotional feedback.

Figure 5 .
Figure 5. Average cleanliness of the lake that triggers emotion expression by treatment.

Figure 6 .
Figure 6.Absolute change in the usage of conservation technology in response to emotional feedback.

Figure 7 .
Figure 7. Average lake cleanliness by round.

Figure 7 .
Figure 7. Average lake cleanliness by round.

Table 1 .
Selected strategies and payoffs of the players.(In each cell the payoffs are written in the following

Table 2 .
Multinomial logit model with the dependent variable "Positive or Negative Emotions" ( Figure 5. Average cleanliness of the lake that triggers emotion expression by treatment.

Table 3 .
Tobit regression with dependent variables "Absolute change in round 2 as compared to round 1".