Abstract
One common fundraising strategy used by nonprofit organizations is providing information to persuade potential donors. We theoretically and empirically analyze how information affects people’s willingness to donate. Our theory suggests when people have different initial beliefs, new information leads to polarization through their understanding and rationalization of social norms. We provide empirical support using an online experiment, demonstrating that environmental and public health information leads to polarization in deforestation prevention donations. Being exposed to information opposite of individuals’ existing beliefs reinforces their current opinions. Our results emphasize that implementing information treatment calls for careful deliberation.
1. Introduction
While there are many reasons why people behave prosocially (see Meier [2006] for a survey of the literature),1 abundant work in economics and psychology has repeatedly emphasized the importance of “nudges” in promoting prosocial behavior; that is, mechanisms that target cognitive heuristics and influence people’s decision-making process without financially incentivizing them or restricting their choice alternatives (Thaler and Sunstein 2008). Many studies have examined the effect of a default setting (i.e., preselect a donation option as the default) (e.g., Andreoni and Payne 2013; Goswami and Urminsky 2016), social norms (i.e., approved choices and actions that most people follow in a community of society) (e.g., Krupka and Croson 2016; Bartke et al. 2017), and social comparison (i.e., providing information on how much a more/less generous donor has contributed) (e.g., Croson and Shang 2008; Shang and Croson 2009; Bartke et al. 2017) in charitable giving. However, the literature on how information affects donation behavior is relatively scarce.2 Our article aims to fill the gap by theoretically and empirically investigating the effect of information on individuals’ environmental donation decisions.
Understanding the effect of information is particularly critical in the realm of environmental donation for two reasons. As public goods, environmental resources are characterized by underprovision because people have incentives to free ride.3 Nonprofit organizations offer mechanisms to promote cooperation, which facilitate collective actions that address the insufficient provision of environmental resources. One common fundraising strategy used by these organizations is to inform and persuade potential donors through information. Given the fierce competition between charitable organizations, limited donor resources, and underfinancing due to free-riding, evaluating whether information can be used to increase fundraising is crucial. Moreover, despite the scientific consensus of the likely causes and effects of many environmental problems, much relevant information is unclear or unknown to the general public. More important, as demonstrated in a large body of economics and psychology literature, information may be interpreted differently by groups with different beliefs, leading to increasing divergence in public opinions and growing skepticism regarding environmental issues, such as climate change (see Hornsey et al. 2016 for a meta-analysis of belief in climate change; Guber 2013; Guber 2013; Unsworth and Fielding 2014; Carmichael & Brulle 2017; Birch 2020). It is therefore important to examine the effect of information, particularly how information interacts with people’s initial beliefs.
We first develop a theoretical model to predict the impact of new information on respondents’ willingness to donate (WTD) to environmental programs. Our model shows that new information interacts with respondents’ prior beliefs through their rationalization of social norms, which then affects donation behavior. The model leads to two testable hypotheses. First, respondents’ WTD conforms to social norms and varies by their initial beliefs. Second, the new information affects respondents’ WTD differently, depending on whether it confirms or contradicts their initial beliefs.
We empirically test these hypotheses through an application of deforestation-prevention donation. Despite the widely documented ecosystem services provided by rainforests, deforestation continues to occur at an alarming rate, mostly in tropical forests in Latin America and Southeast Asia (Curtis et al. 2018). Forest degradation leads to rapid biodiversity loss and is responsible for 15%–25% of global greenhouse gas emissions (Moutinho and Schwartzman 2005; World Wildlife Fund n.d.). In addition, deforestation is also the source of many threats to human health because of pathogen spillovers and outbreaks associated with land use change and population increase in or beneath forests (Guégan et al. 2020). Recent studies have linked deforestation to the increasing severity and frequency of emerging infectious diseases over the past two decades (e.g., Daszak, Cunningham, and Hyatt 2000).4 Although the potential correlation between deforestation and infectious disease outbreaks has been recognized in the scientific community, the general public still lacks relevant information.
We implemented an online experiment with real stakes to test the effect of an information treatment. To alleviate potential hypothetical bias and payment incompatibility issues, we informed the respondents that two winners will be randomly picked from all participants to receive a $100 bonus, and their WTD amount to the nonprofit organization of their choice will be deducted from this bonus. To ensure trust, we provided the donation receipts to the respondents by email. We find that respondents’ donation conforms to the (descriptive) social norm, which is measured by the expected average donation of others. We also show that the information treatment illustrating the potential correlation between deforestation and infectious diseases interacts with participants’ preexisting beliefs on whether deforestation is an issue and leads to polarization between groups. Specifically, the information treatment increases the donation amount in the group agreeing that deforestation is a problem that needs to be addressed and decreases the donation in the disagree group. Taken together, our results show that although information is an important driver promoting people’s environmental donation behavior, it can lead to polarization as an unintended consequence.
Our study contributes to three strands of literature. First, we contribute to the literature on the effect of an information treatment on donation behaviors by focusing on the interactions between information and prior attitudes toward environmental issues (Grieco et al. 2018). Given the critical role of privation donations as a source of funding to provide environmental goods and services, our analysis provides important insight that organizations and institutions can leverage when eliciting private donations to combat underprovision issues of public goods.
Second, our findings add to the recent studies reporting that nudges have only limited effect (e.g., d’Adda, Capraro, and Tavoni 2017) or cause backfiring effects and unintended consequences (e.g., Reijula et al. 2018; Hagmann, Ho, and Loewenstein 2019). We demonstrate that exposing individuals to information that opposes their existing beliefs may reinforce their current opinions, leading to group polarization. Similar to previous studies documenting that additional information intending to promote donation may negatively affect the financial support toward environmental resources (Goff, Waring, and Noblet 2017), our results further emphasize that implementing information treatment calls for careful deliberation and execution because of its interaction with individuals’ prior beliefs.
Third, our study contributes to the broader discussion of polarization in beliefs and attitudes toward environmental issues, such as climate change. One consistent finding in the literature is how people process information based on their preexisting beliefs. Hornsey et al. (2016) conducted a meta-analysis on climate change beliefs and found that values, ideologies, worldviews, and political orientation dominate people’s belief in climate change, rather than education, gender, knowledge, or experience. In addition, many studies show that political biases affect how people interpret climate change information (e.g., Leiserowitz 2006) and argue that it may also be related to selective inattention to climate-related words (Whitman et al. 2018). Past work shows that climate change skeptics report greater justifications for not helping the victims after seeing information linking climate change to natural disasters (Chapman and Lickel 2016). Our findings further confirm that the divide in beliefs affects how people take in new information and consequently affects their behaviors.
2. Theoretical Framework
Inspired by Konow (2000) and Bejarano, Corgnet, and Gómez-Miñambres (2021), we consider a utility maximization model with dissonance costs in environmental donation decisions. Participants understand that two out of all participants will be picked at random to receive a $100 bonus for participating in the study and can choose to donate a portion of the bonus to an environmental program. In addition to the chance of winning the $100, participants derive utility from donating money to nonprofit organizations aligning with their beliefs and from donating more than what they consider the social norm to be. They obtain intrinsic utility from conforming to the social norm and thus want to maintain consistency between actions and the norm. However, when they behave differently from the social norm, they incur a dissonance cost,5 which represents the costs due to shame or guilt associated with deviating from the group norm (Kandel and Lazear 1992; Gawronski and Strack 2012; Bejarano, Corgnet, and Gómez-Miñambres 2021). The baseline utility function for a participant is as follows:
1
where i indicates individual i. Pi is the probability of winning the lottery;6 xi is the respondent’s WTD; αi is the belief parameter that shows how much they support or oppose the cause, βi(> 0) is the comparison parameter that captures the additional utility/disutility from people feeling good or “bad” about themselves when comparing their donation to the social norm 
, and γi(> 0) the dissonance cost parameter representing the cost of deviating from the social norm. Note that these social norms 
are beliefs of each person; in our experiment, we use an incentive-compatible bonus to make sure participants state their true beliefs in the follow-up question after stating their WTD.
Participants choose their donation level to maximize utility, which yields
2
Therefore, for individuals who support the environmental cause or receive utility from behaving more generously (donating more money) than others (αi + βi > Pi), their optimal donation is greater than the social norm: 
. On the other hand, for those who do not support the environmental cause or do not care much about how their donation compares with the average donation (αi + βi > Pi): 
.
Substituting equation [2] into equation [1], we find
3
We define group j as those people who believe in the environmental cause (αj > Pj), we have 
, and 
. That is, those in group j have an incentive to rationalize a higher amount of donation from others because deviating from the norm incurs utility losses. Similarly, we define group k as those people who do not support the environmental cause (αk < Pk), we have 
, which means they have an incentive to rationalize a lower level of social norm to justify their lower donations, although some may be willing to donate more than the social norm depending on their utility from comparison (βk).
Hypothesis 1: Without the information treatment, respondents’ donations are proportional to the social norm: 
, but varies depending on their initial beliefs. The group (j) of participants who support the environmental cause (αj > Pj) have higher WTD than group (k), which does not support the environmental cause (αk ≤ Pk): 
.
When faced with the information treatment, participants may interpret the same information differently and choose to internalize what is consistent with their desired conclusions due to motivated reasoning, which represents a self-serving bias (Klayman and Ha 1987). The information treatment may also make the environmental cause salient and thus incentivize participants to reweigh their choices. Even though participants do not change their expectations of others’ WTD 
,7 they use the new information to rationalize a new norm, which we refer to as the rationalized norm 
, to serve their own interests, based on their initial beliefs about the environmental cause. Previous studies have shown that people think their beliefs are shared by others (e.g., Buchanan 2020), thus they can convince themselves that the rationalized norm is the true norm, although at a cost (Konow 2000). As a result, we predict a wider deviation from the social norm between these two groups when given new information. In other words, new information can lead to group polarization (Sunstein et al. 2017) and could cause backfiring effects in group k that does not believe the environmental cause (e.g. Reijula et al. 2018; Hagmann, Ho, and Loewenstein 2019).
Thus, under the information treatment, participants’ utility function can be written as
4
where comparison utility and dissonance cost depend on how far an individual’s behavior deviates from the rationalized norm 
. There is also a rationalization cost (θi > 0) that participants incur when they rationalize a new norm.
Given equation [4], we use backward induction to solve for new optimal WTD:
5
We plug 
into the utility function to find the optimal rationalized norm:
6
Group j that supports the environmental cause (αj > Pj) has a higher rationalized norm 
, while the other group, group k, that does not support the environmental cause, has a lower rationalized norm 
.
Finally, we can find the relationship between the optimal WTDs with 
or without 
information as well as their relationship to the social norm 
:
7
8
Therefore, we find that 
for group j that believes in the environmental cause (αj > Pj), and 
for group k that does not support the environmental cause (αk < Pk).8
Hypothesis 2: Information treatment can induce group polarization when two groups have different beliefs. WTD of the group (j) that supports the cause will increase, while the WTD of the group (k) that does not support the cause will decrease: 
. Our hypotheses can be summarized in Table 1.
Hypotheses for Information Treatment
3. Experimental Design
We design a donation experiment in which the information treatment is randomly assigned when asking for respondents’ WTD to deforestation-prevention programs. Our study is implemented online through a survey research firm. The firm maintains research panels by recruiting study respondents through social media and other online communities, providing a representative sample that mirrors the general population. Because participants do not contact the survey firm themselves, the recruitment method reduces possible selection bias. Our study was implemented between June 25 and August 4, 2020. It took each participant approximately 25 minutes to finish the experiment. The full survey is presented in Appendix A2.
Payment Vehicle
Respondents are asked to state their WTD some or all of a $100 bonus that will be given to two randomly chosen respondents at the end of the experiment.9 The winners can donate some or all their bonus payments to the nonprofit organization of their choice (out of three alternatives: Rainforest Alliance, Trees for the Future, and World Wildlife Fund). To ensure trust and remind respondents that actual payment will be collected, we create GoFundMe campaigns for nonprofit organizations, so that donation funds are safely and automatically delivered. This also creates a public record verifying that the respondents’ chosen donation actually took place. Finally, we send the donation receipts to the participants through email. Overall, our novel study design ensures that the donation payment is real stakes and thus incentive compatible, which significantly improves the validity of treatment effects evaluations.
Real stakes are generally preferred because they can prevent social desirability bias; that is, respondents provide answers that are more socially desirable than their true attitudes or behaviors. We acknowledge that our design is slightly different from a normal “real stakes” experiment given that a decision is probabilistically implemented. Because the probability of winning the bonus is low and the bonus amount is not high enough to guarantee a large expected value, our study design fits in between a “real stakes” and a “hypothetical stakes” experiment. Previous literature has shown that participants’ behaviors differ between those who receive hypothetical stakes and real stakes (Buhren and Kundt 2015). However, such differences mainly affect the donation levels and have little effect on the treatment effects we are interested in, since the stakes are held constant across the treatments.
Experiment Content
To ensure a clear understanding of the donation mechanism and elicit accurate valuations, at the beginning of the experiment, we inform respondents that two people will be randomly selected to receive a bonus payment of $100 sent to their PayPal accounts. We also provide detailed information on how to register for a free PayPal account (if they do not have one already) to receive the bonus. Following this information, we ask the subjects again whether they are aware that two survey participants will be selected at random to receive a bonus payment. Only those who answer yes are allowed to remain in the survey and be asked for their PayPal accounts.
We have the respondents read important background information on deforestation to offer context on the issue. We specifically provide the definition of deforestation and its negative impact. The background information reads:
Deforestation refers to the process of clearing forested land on a large scale to allow timber production, expansion of agriculture, and the development of roads and urban infrastructure. It threatens the ability of our globe’s forests, especially tropical rainforests, to continue providing valuable ecosystem services. These threatened or diminished services include flood prevention and erosion control, biodiversity preservation, and carbon sequestration. Deforestation also contributes directly to global warming and climate change. According to a recent estimate, deforestation is responsible for 10% of all global warming emissions.
Following the background information and the randomized information treatment, we provide information on the three nonprofit organizations, Rainforest Alliance, Trees for the Future, and World Wildlife Fund, including a description of these groups’ efforts on preventing deforestation and ask respondents for their WTD to deforestation-prevention programs they choose among the alternatives. We suspect that framing the WTD questions also affects individuals’ donation behaviors, given that framing has been found to be a coordination device by affecting people’s beliefs about the behavior of others (Fehr and Schmidt 2006). Following Capraro et al. (2019), we divide participants into three groups (including a control group) and test whether prompting them to think of the moral context of their action and make beliefs about norms salient have explanatory power for their donation behaviors. The first framing treatment, labeled “personal framing treatment” reads: “What do you personally think is the morally right thing to do in this situation?” The second framing treatment, labeled “societal framing treatment” reads: “What do you think your society considers to be the morally right thing to do in this situation?”10
After the WTD question, we ask respondents for their expected donations of other participants and demographic information. The expected donation serves as a measure of what the participants perceive as the descriptive social norm, as explained in the theoretical model section. To ensure incentive compatibility when eliciting expected donations, the respondent whose guess is the closest to the actual average donation from all respondents will receive a bonus payment of $10. We collected related information, such as whether they or any family members own forest land, if they agree that deforestation is an urgent issue that calls for attention and exacerbates global warming and climate change, if they have prior knowledge that deforestation might contribute to the occurrence of pandemics, and if they have previously donated money to nonprofit organizations or volunteered for social or environmental causes. As demonstrated in the theoretical models, participants’ preexisting beliefs of whether deforestation is a problem facilitates our analysis on whether their belief systems interact with the new information and thus change their donation behavior.
4. Data
Our final survey sample consists of responses obtained from 1,200 randomly selected adult residents (i.e., above 18 years old) in the United States, out of a total of 5,463 people who accessed the survey. A portion of the respondents is excluded for the following reasons. First, 364 respondents (6.67%) are under 18 years old and are excluded. Second, in the prescreening consent question, respondents are asked: “Do you commit to carefully reading and providing your thoughtful and honest answers to the questions in this survey?”; 3,214 (58.83%) respondents chose “I will not read carefully and provide my best answers,” while 122 (2.23%) chose “I can’t promise either way.” These respondents did not pass the consent screening and were not allowed to proceed. As a result, 1,763 (32.27%) respondents who chose “I will read carefully and provide my best answers” remain in the sample. We want to note that respondents not consenting is a common phenomenon when social media and online forum outlets are used as survey recruitment channels. Responses are recorded if a participant has clicked the survey, but many lack the patience to answer the questions and thus indicate that they will not read the questions carefully.
Third, to further improve the data quality, we removed respondents who displayed evidence of poor attention and speeding during the survey. Those whose total response time is below one-third of the median total response time (250 seconds) are excluded (495 observations, or 9.06%).
Fourth, we removed 68 (1.24%) observations with missing values for the expected WTD, which is an important variable that measures the social norm in our study. After removing these observations with missing values, the minimum total response time in the sample increased from 250 seconds to 494 seconds, suggesting that those missed questions may be from people speeding through the survey. Last, we balance the age and racial compositions across treatment groups to ensure data quality. After a certain age or racial group has reached the sampling quota in each information treatment group, respondents from these demographic groups are no longer included in the sample, and their answers are not recorded.
Table 2 presents the demographics distributions of the survey respondents and the U.S. adult population. The z-scores obtained from one-sample proportion tests indicate that our survey sample’s gender and marital status distributions are similar to the general adult population. While our sample has lower shares of Hispanics and other races, the overall racial/ethnic composition is generally comparable to the national level. However, respondents in our sample tend to be older and are more likely to hold an undergraduate or graduate degree than the general population. We create sampling weights using the population-level age, education, and income distributions as the target weights to correct for the nonrepresentativeness. All models are estimated based on the sampling weights.
Demographics Distributions of the Survey Participants and the U.S. Adult Population
Table 3 presents the summary statistics of other important attributes of the survey sample. Here we highlight several interesting observations. First, the amount participants expect others to donate, on average, is lower than their own WTD regardless of whether they receive information treatment. Specifically, the average expected donation of others is $37.32, while the respondent’s own donation ranges from $39.51 to $43.12 on average, depending on their treatment group. This is consistent with the “Lake Wobegon effect” when people tend to think that they are more generous than average (Oxford Reference n.d.; Maxwell and Lopus 1994). Second, the average WTD is higher in the treatment group than in the control group. Third, the average expected WTD is similar regardless of information treatment status or whether participants agree/disagree deforestation is a problem, while it is the lowest among the disagree group. Fourth, most respondents (88.50%) agree that deforestation is a problem that calls for attention. However, only 56% of them have prior knowledge about the correlation between deforestation and the increase in occurrences of infectious disease outbreaks. Given the low level of awareness about this information, educating the general public of the potential negative impact of deforestation on the spread of infectious diseases may significantly affect their WTD. Last, a large portion of respondents (74%) have donated money to nonprofit organizations in the past.
Survey Sample Summary Statistics
Figure 1 illustrates the differences between average WTD across information treatments and agree/disagree groups. As expected, the average WTD is the highest among participants who have received information treatment and agree deforestation is a problem, and the lowest among the group that has received the information treatment and disagrees deforestation is a problem.
Average Willingness to Donate by Information Treatment and Agreeing/Disagreeing Groups
5. Empirical Model
Because the donation amount is censored with a lower bound $0 and an upper bound $100, we estimate the following Tobit model to evaluate factors determining respondent i’s WTD using the experiment data to verify the hypotheses predicted by the theoretical models in equation [1] and equation [4]:
9
where WTDi denotes respondent i’s stated WTD. The vector Xi contains individual demographics, including age, gender, race, marital status, income, and education level, along with political party affiliations (i.e., Republican, Democrat, independent, something else). Infoi is a binary variable indicating whether respondent i receives the infectious disease information treatment. Similarly, Personal_ Framingi and Societal_Framingi are binary variables implying whether an individual receives a personal framing or a societal framing, as explained in the experimental design section. Individuals’ inclination toward charitable activities likely has an effect on their WTD, so we include Donated_Moneyi, to measure whether respondent i has previously donated money to nonprofit organizations.
Empathetic individuals are more likely to make charitable donations (Batson and Shaw 1991). To measure participants’ empathy and altruism level, we ask respondents to use the scale from 1 (strongly disagree) to 5 (strongly agree) to indicate to what extent they agree or disagree with the following statement: “Despite how much money I have, I am concerned that there are people who have less money than me.” Concern_lessi is included as a control indicating a subject’s empathy/altruism level.
Given that the collected funds would be used to support deforestation prevention, we suspect that respondents who are forest owners or have family members who own forest land may have a higher WTD. As a result, we include variable Own_Foresti to indicate if a respondent or another family member owns forest land. Preexisting knowledge regarding the possible correlation between deforestation and the spread of infectious disease is a factor differentiating a person’s WTD. The Know_Infect_Diseasei binary variable implies whether respondent i knows previously that deforestation may lead to an increase in infectious diseases. Given the public good nature of deforestation prevention, a person’s expectation of how much money on average the other participants will donate is an important factor affecting a respondent’s own WTD. As a result, we include Expected_WTDi as a control variable.
Respondents’ attitude toward deforestation and its impact is another important factor affecting their WTD. For instance, individuals who agree that deforestation is a problem that calls for attention are likely to donate a larger amount than their counterparts. In fact, we argue that their beliefs of whether deforestation is a problem divide them into a group that supports the environmental cause and a group that does not support the environmental cause, as predicted by our theoretical model. The binary variable Agree_Deforestationi indicates respondents’ beliefs on whether deforestation is a problem that needs to be addressed. εijt is a stochastic error term.
6. Results and Discussion
First, we test the hypotheses derived from the theoretical model in Table 1. Since donation amount is not normally distributed, we perform nonparametric Wilcoxon signed-rank tests and present the test results in Table 4. Note that all tests of significance are performed as one-tailed tests because the hypotheses derived in the theoretical model provide information on the directionality.
The test results in the first row in Table 4 show that the median WTD is greater than the median expected WTD among the deforestation believer group (j), and the median WTD is less than the median expected WTD among the deforestation disbeliever group (k). Results in the second row demonstrate that the new optimal WTD after receiving the information treatment 
is greater than the optimal WTD without information treatment and the social norm 
.
Test Results of the Theoretical Model Hypotheses
Next we estimate the empirical model in equation [9] using the experiment data to further examine the hypotheses. Table 5 presents the Tobit model weighted estimation results, in which sampling weights are used to adjust for age, education, and income. The unweighted estimation results are presented in Appendix Table A1.
Weighted Tobit Models’ Estimation Results for Willingness to Donate (WTD)
Table 5, column (1) presents the estimates based on the subsample without the information treatment to test hypothesis 1 derived from the theoretical model. In addition, to test hypothesis 2 and examine whether and how the information treatment interacts with individuals’ initial beliefs, the full sample is divided into subsamples based on respondents’ beliefs toward deforestation. Columns (2) and (3) present these subsample estimates. Specifically, column (2) shows the results for the group of participants who do not agree that deforestation is a problem, whereas column (3) reports the results for participants who concur that deforestation is an environmental problem that calls for attention.
In our baseline model in equation [1], we predict that without information treatment, respondents choose to donate based on what they perceive as the descriptive social norm, which in our case is what they expect other participants to donate. Individuals’ WTD and their expected WTD of others are highly correlated with a correlation of 0.61, and a person’s WTD increases as expected WTD becomes higher. In addition, the regression results in Table 5, column (1), further indicate that participants’ expected average donation of other participants is positively correlated with their own donation. With a $1 increase in the expected donation of others, a respondent’s own donation increases by $0.89. We thus confirm the first part of our first hypothesis; that is, without the information treatment, respondents’ donation is positively associated with the expected donation of other people.
People with different initial beliefs regarding an environmental issue behave differently when they make donation decisions. In our study, participants differ in their beliefs on whether they think deforestation is a problem that calls for attention. Without any information treatment, these two groups naturally diverge in their WTD because they place different values and importance on this environmental issue. Those who agree that deforestation is a problem have a higher WTD on average than those who do not agree. Specifically, the median WTD among the support group is $36, while it is $25 in the other (do not support) group.
Together with the test results in Table 4, the first hypothesis is confirmed, demonstrating that without information treatment, respondents’ donations are proportional to the social norm but vary depending on their initial beliefs. We acknowledge that it is possible people claim to believe in a norm that is closer to their own behavior so they can justify their decisions. However, our incentive-compatible mechanism aligns people’s beliefs of the social norm with what they state in the question, which alleviates this issue.
When faced with new information, people tend to internalize it with their own interpretations, which can be influenced by their personal beliefs and experiences, to avoid cognitive dissonance (Klayman and Ha 1987; Birch 2020). This self-serving bias can lead to polarization when groups have opposite initial beliefs. Hypothesis 2 states that with information treatment, WTD of the group (j) that supports the cause will increase, while the WTD of the group (k) that does not support the cause will decrease. We examine the impact of the information treatment on these two groups by stratifying the survey responses to two subgroups, as shown in Table 5, columns (2) and (3).
The results of these subsample analyses confirm our hypothesis 2 and reveal that the information treatment has opposite effects on groups with different initial beliefs regarding whether deforestation is a problem. Notably, the coefficient on the information treatment is negatively significant at the 10% level for people who disagree that deforestation is a problem, as shown in Table 5, column (2), but positive and significant at the 10% level in the agree group, as shown in column (3). In particular, the information treatment lowers the disagree group’s average donation amount by $10.60, while it boosts the agree group’s WTD by $4.50. These findings demonstrate that people process new information in a self-serving biased manner where they willingly believe reaffirming evidence but critically scrutinize disconfirming evidence. Concerningly, the reduction in WTD among the group that does not think deforestation is a problem is much greater than the increase in WTD in the agree group, which indicates a stronger backfiring effect and group polarization. There are two possible mechanisms through which the information treatment causes polarization. The first is reactance. Individuals who oppose the environmental cause may feel the information eliminates their freedom of decision making, so they respond negatively by donating less. The second is related to creditability. Participants, particularly those who disagree with certain ideologies or government authority may become more skeptical and reluctant because they perceive the information as false.
In addition to testing our hypotheses, the empirical analysis reveals several interesting findings through a full sample estimation without any stratification, as presented in Table 5, column (4). We find certain individual demographics significantly affect people’s WTD for deforestation prevention. The coefficient estimates in Table 5, column (4), for almost all income levels are positive and highly significant, implying that higher-income people are willing to donate more. For instance, the average WTD of participants whose income is between $25,000 and $49,999 is $6.83 higher than the baseline (those with income less than $25,000). The donation amount slowly increases as income becomes higher, and participants earning $150,000–$199,999 donate additional $21.73 on average, compared with the baseline. The estimate of the Democratic Party is also positively significant at the 10% level, which demonstrates that political party affiliation is an important factor determining an individual’s WTD, corroborating with findings in previous literature (Kotchen, Boyle, and Leiserowitz 2013). Specifically, the average donation among Democrats is $4.82 higher than the donation from Republicans.
We also examine the effect of the framing treatment. Recall that we have two: one personal framing treatment and one societal framing treatment. The coefficient estimate of the personal framing treatment is positive and significant at the 5% level in all models, whereas the coefficient of the societal framing treatment is not. Previous research finds that there is no significant difference between the two (Capraro et al. 2019). Our results, on the contrary, show that different framing treatments have heterogeneous effects. Individuals who have received a personal framing treatment asking them what they personally think is the morally right thing to do are willing to donate an additional $6.62 compared with those who are in the control group, whereas the societal moral nudge that asks participants what they think their society considers to be the morally right thing to do does not influence respondents’ WTD. Our finding confirms that a personal framing treatment serves as a moral reminder, which can be a cost-effective way to promote proenvironmental behaviors.
Other related characteristics influence participants’ donation decisions as well. First, the coefficient of the binary variable indicating whether participants have donated money to nonprofit organizations is positively significant at the 5% level. This suggests that previous donation behavior is positively correlated with respondents’ WTD (Kessler and Milkman 2016). In particular, those who have donated to nonprofit organizations before are willing to donate $5.69 more than their counterparts. Moreover, as a proxy measure for empathy and altruism, the more the participants care about people who make less money than themselves, the higher their WTD. The coefficient of almost all the empathy measurement levels, ranging from 3 (neutral) to 5 (strongly agree) are highly significant and positive at the 1% levels, demonstrating the importance of empathy and altruism in determining donation behaviors. For instance, individuals who strongly agree that they care about those who have less money than themselves are willing to donate $34.82 more than those who strongly disagree with the statement. Furthermore, the positive coefficient on “know the link to pandemics” is statistically significant at the 10% level. This implies that participants with prior knowledge are willing to donate $4.98 more than those without.
Robustness Check
It is possible that certain participants who are more inclined to take online surveys may be different from the average population and thus have different WTD, but it is unclear whether it affects our results and what direction the bias is. In addition, since the deforestation belief question is asked after the WTD elicitation, it hinders our ability to measure how respondents’ beliefs have changed. We acknowledge that there is a possibility that our treatment groups had differences with respect to respondents’ prior beliefs about deforestation. To investigate the robustness of our findings, we bootstrapped the results to observe the reliability of our findings across a larger number of samples. The procedure was performed as follows:
■ We subsampled 2,000 samples of n observations of our sample with replacement for each group: n = 126 for the disagreeing group, n =1,074 for the agreeing group.
■ The two weighted Tobit estimations were run for each new sample, logging key parameters and outcomes of each iteration.
■ We then calculate the bootstrap confidence intervals of our estimates.
In Table 6, we present the original and bootstrapping (2,000 replications) weighted estimation results for key variables. As shown, the two sets of estimation results are similar. The 90% bootstrap confidence interval for the information treatment is (−23.650, −0.390) among the disagreeing subsample and (0.134, 8.601) among the agreeing subsample. These results confirm that our key results are not necessarily a spurious product of the sample and are reliably reproduced across a larger number of repeated subsamples with replacement.
Original and Bootstrapping (2,000 Replications) Subsample Tobit Model Weighted Estimation Results for Key Variables
Considering that a Tobit model comes with distributional assumptions, we have also estimated weighted and unweighted linear regression models without adjusting for $0 donations as a further robustness check of the treatment effect. The linear model estimation results (in Appendix Table A2) are similar to the results of the tobit models, confirming that our treatment effects are robust.
Conclusion
In this article, we develop an intuitive model to theoretically analyze the effect of information treatment on donation decisions. Our theory suggests that when people have different initial beliefs, new information can lead to polarization through people’s understanding and rationalization of social norms. We leverage a real-stakes experiment to empirically explore how a new information treatment affects individuals’ WTD to nonprofit deforestation-prevention programs. Because people have to actually pay for their elicited donation valuations, our survey design alleviates the hypothetical bias issue.
Our results indicate that new information does affect people’s donation behavior. Specifically, an information treatment showing the possible correlation between deforestation and the spread of infectious diseases interacts with participants’ existing belief systems and leads to group polarization. Such information lifts respondents’ WTD among those who believe that deforestation is a problem that calls for attention while lowering the WTD among those who do not believe it. This result highlights the unintended consequences of providing information to potential donors and emphasizes the importance of a careful design of campaign information.
In addition, certain individual characteristics, including income, previous donation behavior, and empathy significantly affect people’s WTD. Those who have previously donated to nonprofit organizations, are in higher-income groups, or are more empathetic tend to donate more than their counterparts. We also find that having prior knowledge regarding the contribution of deforestation to infectious disease risks and a personal framing treatment asking respondents what they personally think is the right thing to do increases people’s donations.
Collectively, these results demonstrate that information has the potential to influence environmental donation behaviors and can be used as a cost-effective tool for promoting prosocial actions. However, without careful design and execution, information treatment can lead to unintended consequences by increasing divergence in opinions and reinforcing self-serving biases. These results have significant implications for environment-related communications. Although difficult, it is better to target the audience and tailor the communications to improve their understanding of the information. Under certain circumstances, it may be better to not provide additional information to avoid unintended consequences. Future research should test the robustness of our findings with different types of information since environmental donation decisions could be affected differently depending on the type of information provided to people.
In the group of disagreeing people, an information treatment may affect individuals differently. In other words, the information treatment effect is heterogeneous not only among the population but also within the subsamples. Some disbelievers may contribute $0 regardless of the information, whereas some are affected by the information and choose to increase or decrease their donations. It is of great interest to provide some reasoning behind this and examine such heterogeneity in future research. We also hope to extend our analysis to investigate the exact causes of polarization and identify effective ways of delivering information to mitigate polarization and bias, particularly on controversial environmental issues. Overall, nongovernmental organizations and government agencies will benefit from considering these conclusions in eliciting private donations. Understanding the significant role new information plays and its interaction with individuals in WTD formation is particularly crucial for policy design and implementation.
Acknowledgments
We thank Grant H. West for his assistance with designing and conducting the online experiment. We are grateful to Editor Daniel Phaneuf, Sarah Jacobson, Joaquín Gómez-Miñambres, and two anonymous reviewers for comments that improved the article.
Footnotes
Appendix materials are freely available at http://le.uwpress.org and via the links in the electronic version of this article.
↵1 There are three main groups of models on why individuals have incentives to behave prosocially. The first indicates that people have prosocial preferences because one’s utility depends on that of others; pure altruism (Becker 1974), impure altruism with a warm-glow motive (Andreoni 1989), and inequality aversion are possible formulations of prosocial preferences (Fehr & Schmidt 1999). The second group focuses on reciprocity, including conditional cooperation (Fischbacher, Gächter, and Fehr 2001) and true reciprocity (Sugden 1984). The third group is based on the importance of social norms acceptance (Alpizar, Carlsson, and Johansson-Stenman 2008) and self-image concerns (Bénabou and Tirole 2006).
↵2 Many studies have focused on the effect of label information (Kallbekken, Sælen, and Hermansen 2013; Stadelmann and Schubert 2018) and presentation of information (Tiefenbeck et al. 2018) on consumer behaviors. The purpose of labels is to improve choices by providing information, which is not a nudge per se. These studies also focus on consumer behaviors instead of prosocial donation behaviors.
↵3 This is because each person benefits from everyone else’s contribution, and each person’s consumption does not diminish everyone else’s consumption.
↵4 A strong correlation has been identified between deforestation and habitat fragmentation and Ebola (Olivero et al. 2017; Rulli et al. 2017), malaria (Olson et al. 2010; MacDonald and Mordecai 2019), and Lyme disease (Li et al. 2012).
↵5 Dissonance costs arise when people experience a discrepancy between their actions (in our case, donation decisions) and their beliefs (i.e., norms) (Festinger 1962).
↵6 Even though the true probability of winning the lottery is objective and does not vary among individuals, respondents may have different estimations of their chances of winning. We acknowledge that by not informing the participants of the total number of participants, they may have a biased estimate of the chance of winning the probabilistic bonus. While our model more closely resembles the experiment design, our key results and predictions would remain unchanged under the assumption of common priors.
↵7 This is because the belief of social norm
is elicited using the incentive-compatible method and Wilcoxon rank sum test to show that does not vary among information treatment and control group (W = 64,048, Wilcoxon rank sum test p-value = 0.3247).↵8 We can also define a subgroup l (αl + βl < Pl) that is a subset of strong disbelievers with strong opinions against the cause, among group k (αk ≤ Pk) of individuals who do not support the environmental cause. Based on equations [6]–[8], we find that subgroup I donates even less than the rationalized lower norm
and less than without information treatment . Therefore, this group diverges further from group j after given the information, which drives the group polarization. This subgroup l corresponds to the disbelievers in the information treatment group in our experiment, who, after the information, still disagree that deforestation is a problem. We thank an anonymous reviewer for pointing this out.↵9 Respondents are provided a base compensation determined by the survey firm for participating in the study.
↵10 Both nudges are followed by the sentence: “With this in mind, please enter the fraction of the $100 that you would like to donate to preventing deforestation” to ensure that respondents understand that they are making the donation.
References
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.