Norm avoiders: The effect of optional descriptive norms on charitable donations

Knowing the descriptive norm concerning others' prosociality could affect your behavior, but would you seek out or avoid such knowledge? This high-powered preregistered experiment explores the effect of both forced and optionally revealed descriptive norms on real monetary donations. These norms were established by learning the proportion of previous participants who had donated to a charitable organization that the respondent now was asked to donate to. For those learning about a norm, participants were more likely to donate if they were shown that a majority donates, compared with if they were shown that a minority donates. For the participants who were asked if they wanted to reveal the norm or not, we found that about half choose to reveal the norm. Those who avoided revealing the norm donated less frequently; both compared with revealers and with those who were forced to view the norm. However, these norm avoiders also donate a higher mean amount. Taken together, this hints at norm avoiders being composed of both altruistic and non-altruistic people, with fewer of those who are undecided. This type of norm avoidance may be more related to information avoidance motives rather than mere curiosity or reactance. The present findings can inspire further research into the motives of norm avoidance.

learning the proportion of previous participants who had donated to a charitable organization that the respondent now was asked to donate to. For those learning about a norm, participants were more likely to donate if they were shown that a majority donates, compared with if they were shown that a minority donates. For the participants who were asked if they wanted to reveal the norm or not, we found that about half choose to reveal the norm. Those who avoided revealing the norm donated less frequently; both compared with revealers and with those who were forced to view the norm. However, these norm avoiders also donate a higher mean amount. Taken together, this hints at norm avoiders being composed of both altruistic and non-altruistic people, with fewer of those who are undecided. This type of norm avoidance may be more related to information avoidance motives rather than mere curiosity or reactance. The present findings can inspire further research into the motives of norm avoidance.

K E Y W O R D S
charitable giving, descriptive norms, information avoidance, optional reveal, prosocial behavior

| INTRODUCTION
Imagine there is a solicitation to collect money for a coworker that is retiring. You personally never knew the person well, but you could spare the money. The solicitor happens to leave the document with previous contributions on your desk. If you have not yet decided on donating or not, perhaps knowing how most of your colleagues behave would guide your own behavior. If you are already settled on donating yourself, would looking at your colleagues' behavior make your own decision less altruistic? Do you look? And if you do, would you follow the example of others?
In order to evaluate ourselves, there is a fundamental need to compare ourselves to what others have done in similar situations (Festinger, 1954). Descriptive norms, which simply describe what others do, have been used in studies to effect prosocial behaviors from towel reuse in hotels (Goldstein et al., 2008) to corruption (Köbis et al., 2015). Even just guessing what the descriptive norm is can increase prosocial behavior (Krupka & Weber, 2009). In the context of charitable giving, a well-researched kind of prosocial behavior, these kinds of descriptive norms have been harnessed to increase donations (see, e.g., Agerström et al., 2016;Alpízar & Martinsson, 2013;Frey & Meier, 2004;Lindersson et al., 2019;Shang & Croson, 2009). However, a question yet to be asked is do people, when given a choice, want to know the norm? What previous studies do not account for is that outside of the laboratory, these kinds of norms are not necessarily forced upon people. In our example above, people may choose to look at the behavior of others, or not. Solicitors, on the other end, could force people to see such information, freely offer it, or hide it. Third parties could also supply information about norms, for those who seek it out. Information-seeking preferences and information avoidance could thus be a crucial, yet missing, component in understanding the impact of descriptive norms on donation behavior.
Consequently, this study investigates what happens when the descriptive norm about charitable donations is made optional.

| Descriptive norms and charitable giving
Studies on charitable giving indicate that donation behavior can be affected by descriptive norms, but also that potential moderators exist (see reviews by van Teunenbroek et al., 2020, andTian &Konrath, 2019). The descriptive norm in previous research have taken different forms, such as displaying the kinds of bills that previous donors have given (e.g., Martin & Randal, 2008), informing about the donation of a single previous donor (e.g., Klinowski, 2020;Shang & Croson, 2009), or using different suggested amounts (e.g., van Teunenbroek, 2016), which limits direct comparisons. Unlike existing reviews, we are here specifically focused on studies where participants are informed about the percentage of others that decide to make a donation (see, e.g., Agerström et al., 2016;Frey & Meier, 2004;Lindersson et al., 2019;Meier, 2007). Donations can be seen as a two-stage process, where the decision to donate and the decision regarding what amount to donate can be affected by separable mechanisms (Dickert et al., 2011). Focusing on norms regarding frequency of donations means that we focus on the chronologically earlier and descriptively simpler norm, namely, the norm of whether to donate or not.
We now turn specifically to studies where the norm describes the frequency of others that are donating. To increase donations, it appears most effective to provide a norm that states somewhere between 64% and 95% of others are donating (Moseley et al., 2018).
For example, in a bi-annual donation request sent out by mail to students, Frey and Meier (2004) tested the difference between displaying a high norm where 64% of previous students had donated and a low norm where 46% of students had donated. Individuals in the high-norm condition donated more often, although the effect was only significant after controlling for previous donor history. More recently, a similar high descriptive norm was used in a field experiment on a university campus. Comparing a high norm of 73% of students donating to a standard appeal, Agerström et al. (2016) found the descriptive norm to increase donations. In a related laboratory study, Lindersson et al. (2019) also found support for the high norm increasing donation behavior, in comparison with presenting no norm.
Common for these studies, and the wider body of research, is that descriptive norms are either shown or not, with no consideration of whether the norm would be sought out, or even attended, when given the option.

| Information avoidance
Quite obviously, experimental studies are most often designed with the expectation that everyone in an experimental condition involving a descriptive norm will view the said norm, as this is what separates the experimental conditions from the control condition. However, in more natural settings, information can often be avoided. Information avoidance here refers to situations where information has no cost and does not require any effort to attain, yet is still avoided (Narayan et al., 2011). There may be a wide range of situations where people avoid information, from receiving medical diagnoses to everyday information regarding one's relationships and finances (Barrafrem et al., 2020;Narayan et al., 2011). Avoiding information can be motivated by wanting to keep consistency between one's actions and beliefs, dodging personal responsibility for negative consequences, or, in the case of norms as information, not wanting to compare unfavorably to others (Golman et al., 2017). For instance, one may not want to ask how much other people have spent on a wedding gift, if one fears negative comparisons by having purchased the cheapest one.
Information avoidance in donation behavior is not always selfishly motivated, however. For example, if one already knows which decision maximizes the payoffs of another person, one could ignore learning which decision maximizes one's own payoff, in order to avoid being tempted to act selfish (Kandul & Ritov, 2017).
In regard to charitable solicitations, people have been shown to sometimes prefer to avoid being asked to donate altogether (Andreoni et al., 2017;Cain et al., 2014). Even after making a donation, it has been hypothesized that people may choose to avoid information about the effectiveness of giving in order to maintain their positive emotions after donating (Niehaus, 2014). Indeed, when given the option to gain relevant information, such as the administrative costs and efficiency of a charitable organization, about half of subjects choose to avoid the information even when it is free (Metzger & Günther, 2019). Such information avoidance could be explained by wanting to preserve the positive emotions gained from acting prosocial. Similarly, people could be motivated to avoid information regarding the descriptive norms of giving.
Those who are not inclined to donate could be driven to avoid information about others' donations, in order to maintain a positive self-image.
People's moral self-image is sensitive to feedback (e.g., Jordan et al., 2015), which gives them a reason to avoid comparison if the norm could indicate that others are donating while they are not. Even if they are donating, seeing such a norm could make their own donation behavior seem less special.

| Descriptive norms as information
Although it cannot be taken for granted that all people seek to know descriptive norms under all circumstances, there are reasons for why people would seek to know them. Mere curiosity could be such a reason. Curiosity can here be conceptualized as a desire for new information and knowledge (Loewenstein, 1994). In general, people are social animals, and many are characterized as being curious about others' beliefs and decisions (Renner, 2006). Specifically, social curiosity can be defined as an interest in how other people think, feel, and behave (Kashdan et al., 2018). However, norms can also be seen as useful information, not just the object of mere curiosity.
From a theoretical perspective, a person can be inclined to learn about and follow a descriptive norm due to both normative and informational conformity (Cialdini & Goldstein, 2004;Claidière & Whiten, 2012). In practice, by asking others about their choices, people can use information about what products they are buying to display their own group membership and conform to expectations, or to receive benefits when publicly showing or telling about one's own decisions (i.e., normative conformity, Bearden et al., 1989). Others' behavior can also be used as an indication of something having desirable qualities and being a good product (i.e., informational conformity, Bearden et al., 1989). This could also be the case for charitable contributions, as others donating to a charity could signal the quality of a charity (Vesterlund, 2003). Descriptive norms can thus constitute relevant information when making a decision to donate and conforming to the norm would first require that one knows it.
Although there are numerous motives for both seeking out and for avoiding descriptive norms in the context of prosocial decisions, actual behavior in the context of charitable donations remains underexplored. At the same time, charitable giving and information search is moving online, making information easily available. Charitable giving made online increased with 518% from 2012 to 2018 (Blackbaud, 2018). Charitable organizations now promote donors sharing their donation decisions or preferred charities on social media (Lacetera et al., 2016), and an increasing number of sites also publish information about charitable organizations as well as guides on how to make informed decisions (e.g., Effective Altruism, GiveWell, GuideStar, and Charity Navigator). Taken together, this means that descriptive norms and other relevant information should become more easily accessible, even outside of the laboratory. With such information available, literally at the palm of one's hand in the case of smartphones, this means that individuals can voluntarily seek out or ignore certain information. Introducing available optional descriptive norms, and studying its effects, can thus be seen as a natural next step in research, which can also increase the external validity of studies.

| Aims
Overall, we aimed to incorporate findings on information avoidance into research on the effects of descriptive norms in donation decisions. The current study aimed at exploring the effects on real donations of making a descriptive norm optional. Our aim in introducing optionally revealed descriptive norms was twofold: to examine the effects on actual donation behavior and to investigate avoidance of the norm itself. Given that avoidance of the norm information would occur, we also aimed to investigate whether such avoidance could be partially explained by factors such as high information avoidance tendencies or low levels of social curiosity. Our reason for including forced norms (non-optional) was to draw comparisons with previous studies. This also allowed us to set up a conceptual replication of previous studies.

| Experimental conditions
We manipulated the existence of a descriptive norm, which had two potential levels. In the two forced norm conditions, participants always viewed a norm. In the two optional norm conditions, participants themselves chose to reveal the norm or not. The norm shown was randomized between two levels: high (83% of others donated) and low (17% of others donated). In the control condition, no norm information was given. Figure 1 gives an overview of the five starting experimental conditions, where the two optional norm conditions include self-selection into a sixth sub-condition, referred to as norm avoiders. Other than being given the option to reveal the norm after reading the charity appeal, participants self-selecting into the subcondition "norm avoiders" saw the same information as participants in the control condition.

| Hypotheses
To structure our hypotheses, we present them in three blocks throughout the study. First, we present hypotheses relating to donation behavior being affected by the high and low norms. Second, we present hypotheses relating to donation behavior being affected by making the norm optional. Third, we present hypotheses relating to predictors of avoiding the norm information. All hypotheses were preregistered with motivations and analyses, with the aim of limiting researcher degrees of freedom (Simmons et al., 2011) and promoting open science. The preregistration (see https://osf.io/vc4kp/) also includes exploratory analyses made here, such as investigating mean amounts of donations made.
1.6.1 | High or low norms and donation behavior H1. We expect to see more people donating in forced high norm than in control. We also expect more people donating in the control condition than in forced low norm. This can partly be seen as a conceptual replication of the comparison of baseline and high-norm conditions in Lindersson et al. (2019).
H2. For participants viewing a norm, we expect a higher number of donations in the high-norm conditions than in the low-norm conditions, both when the norm is forced and optional. This can partly be seen as a conceptual replication of Frey and Meier (2004), with the addition of optional norm conditions beyond forced norm conditions.

| Optional norms and donation behavior
In relation to the effects on donations of making norms optional, we present three hypotheses, followed by a short motivation.
H3. We expect norm avoiders to donate more often than participants in the control condition. This novel prediction was in part based on a related pilot study, in part on theoretical assumptions. Although we acknowledged that there were both prosocial and selfish reasons to avoid the norm, a theoretical assumption was that merely giving people options could itself increase willingness to donate by increasing the agency of the participants in the situation (e.g., Eckel et al., 2017;Kessler et al., 2019). For example, Eckel et al.'s (2017) study suggested that merely giving donors the option to direct some of their donation to a specific cause increased giving, even though very few used the option. Another central assumption was that merely thinking about the norm could increase donations F I G U R E 1 Overview of conditions. The flowchart indicates which steps were common across all conditions and which were unique. The number of participants in each condition shows the number after data exclusions (e.g., Bartke et al., 2017) and the optional information could trigger such thoughts.
H4. For norm revealers, we expect more donations in the condition optional high than in control and more donations in control than in optional low. We thus expected viewers of the norm in the optional conditions to be affected by the norms levels similarly to participants in the mandatory conditions, as can be seen by how H4 mirrors H1.
H5. Collapsing revealers and norm avoiders, we expect more donations in the condition forced high norm than in optional high norm and more donations in optional high norm than in control. The reasoning behind this is that we expected the high norm to have a positive effect and that this effect would be larger in forced high norm, as the number of participants exposed to the norm would be higher than in optional high norm. This also means that we expected the norm level to have a stronger influence on donations than the effect mentioned in H3.

| Predictors for norm avoidance
Finally, based on the previous theoretical background, we expected to find relations between avoiding or revealing the norm and related individual differences. H6 relates to the directions of the relations.
H6. We expect that individuals with higher scores on information avoidance and reactance measures will be less likely to reveal the optional norm. We expect that individuals with higher social curiosity and consumer susceptibility to informational influence (CSII) scores will be more likely to reveal the norm.

| METHOD
The experimental design, sample size, power calculation, detailed hypotheses, hypothesis motivations, and analyses were preregistered and can be found under registrations online (https://osf.io/vc4kp/).

| Participants
A total of 2250 participants were recruited among fluent Englishspeaking US citizens using Prolific. We aimed to recruit 525 participants for each of the optional high-and low-norm conditions and 400 participants for each of the remaining three conditions. The uneven assignment of participants to conditions was due to our aim to compare groups within the optional high-and low-norm conditions.
As decided in our preregistration plan, we dropped any subjects failing the attention check prior to analysis, but report analyses with these subjects in the supporting information if the analyses differ. Out of the 2250 participants, 2227 (50.3% female, M age = 35.81, SD age = 12.53) passed the attention check (corresponding to 99%) and are included in the analysis. Participants were paid a show-up fee of 1 US dollar (USD), and then given another 1 USD as windfall earnings that they could choose to donate from. After deciding on donations, participants were given the remainder of the 1 USD as a bonus payment.
2.2 | Procedure and materials

| Prior to donation
Here we present the procedure in chronological order, together with the relevant materials, complimenting Figure 1 which gives an overview of the differences between conditions. Before taking part of the survey, participants were informed that the survey would concern "personality, financial decisions, and well-being." The decision not to mention the charitable contribution in prior instructions was to avoid self-selection by people to not participate in order to "avoid the ask" (Andreoni et al., 2017). Participants first read a welcome message reiterating the description of the survey, with the following page showing a question regarding current mood. 1 Next, participants were informed that they were given an amount of windfall money, that they could choose to donate or to keep for themselves. The donation appeal followed on the next page.

| Donation appeal and donation
Participants were given a description of the charitable organization St Jude Children's Research Hospital, including an image of a child and a doctor. In terms of framing, the donation request included both the scope of the problem and the positive impact that the organization had. The full donation request is included in the supporting information. After reading the description of the organization, participants continued to the next page, which was different for participants from different conditions. In the optional norm condition, participants were given the option to reveal the norm, with the instruction "Do you want to know if other people taking this survey have donated?" They then answered on the following scale: "Yes, show me how many" or "No, do not show me how many". This decision was a basis for splitting people into groups, where those who chose "No" were categorized as norm avoiders. Participants in other conditions were not given the option and those in forced norm conditions proceeded directly to being shown a norm. The norm shown in both optional and forced norm conditions was randomized between two levels: high ("83% of others taking this survey donated") and low ("17% of others taking this survey donated"). Directly above the norm was the text "Below you can see if other people taking this survey have donated." Participants in the control condition skipped this page entirely. The next page then asked participants if they wanted to donate or not.
Participants who choose to donate were then asked to enter an amount to donate, from 0.01 to 1 USD. This screen was followed by the same mood question as had been shown prior to donation decisions.

| After donation
Having completed the prior parts, participants then proceeded to fill in two question items regarding their beliefs about other participants' donations. The first question asked whether they believed the majority of other people taking this study on Prolific chose to donate or not, and the second asked how much they believed others in their situation give on average. We used the first of these questions as a form Information avoidance was measured using an adapted 2-item scale (Pearson correlation = .41) taken from Howell and Shepperd (2016). It was adapted here as the items "I would avoid learning whether others donate or not to a charity that I donate to" and "Even if it will upset me, I want to know if others are donating to a charity that I donate to".
Consumer susceptibility to informational influence (CSII) was measured using the 4-item (Cronbach's alpha = .86) subscale for informational conformity from Bearden et al. (1989). It was included to investigate potential individual differences related to conforming to the given norm.
The attention check was embedded in a Likert scale, with the statement "This question is an attention check. Please answer by selecting Agree." Gender and age data were also collected. These data are included in Prolific data sets, so participants had already given this information prior to our study.

| Analysis plan
Analyses were performed using SPSS24 and GraphPad Prism 8.3.0.
Overall, we performed χ 2 tests and logistic regression models, where appropriate, when analyzing decisions to donate or not. We followed the same procedure when analyzing decisions to reveal the norm or not. Models and type of analyses, including exploratory analyses, were specified prior to data collection in our preregistration plan. Additionally, we start by reporting how beliefs concerning the norm relate to conditions, as a form of manipulation check. We use two-sided tests throughout.

| RESULTS
About half of participants in the optional low-norm (50.0%) and highnorm (51.9%) conditions chose to reveal the norm. As those who did not reveal the norm viewed the same exact stimuli in both low-and high-norm conditions, we collapsed these two into norm avoiders. Table 1 gives an overview of the percentage of participants that made donations, per condition, as well as the mean donation. We can also find the result of the manipulation check, where participants stated whether they believed the majority of other participants would donate or not. As can be gathered, participants who saw a high norm generally believed the majority of others would donate, while participants who saw low norm did not believe it. Participants in the control T A B L E 1 Percentage of participants donating, per condition, sorted from highest to lowest percentages, and belief that the majority of other participants are donating condition and norm avoiders, who saw no norms, fall between the abovementioned two clusters. Differences between these three clusters are highly significant, but are analyzed post hoc and thus reported in Table S1. We now proceed with our preregistered analyses.

| The effect of high and low norm levels on decisions to donate (H1, H2)
The percentages of participants who choose to donate anything to the charitable organization are shown in Table 1 there was a significant difference, with more donations in the highnorm condition, χ 2 (1, N = 787) = 5.472, p = .019, r = .083. This conceptually replicates the effect found by Frey and Meier (2004) and partly corroborates H2.
We continue investigating the effect of norm level by comparing the high-norm and low-norm conditions, looking both at the optional and forced conditions. Table 2 shows the logistic regression carried out on all participants who viewed a norm, dropping those who either choose to avoid the norm or were in the control condition. Overall we find an effect of norm level in the more complete models, with participants in the high-norm conditions more likely to donate. We also see an effect of gender, with females more likely to donate, in line with previous literature (e.g., Wiepking & Bekkers, 2012). We find no effect regarding whether the norm was presented as being forced or being optional before being revealed. The effect of norm level here further corroborates H2.
We now turn to donation amounts (thus excluding non-donors), which was a preregistered exploratory analysis (meaning we did not hypothesize a direction). This variable was, as expected, not normally distributed, and thus, a log10 transformation was used. Differences in mean donations due to experimental condition were tested using a one-way ANOVA. There was a significant effect of experimental condition on mean donation amounts for the six total conditions (F[5, 1232] = 6.27, p < 0.001, partial η 2 = .025). Post-hoc comparisons using the Holm-Sidak method to adjust for multiple comparisons, reported in full in Note: Logistic regression for all participants viewing a norm. The dependent variable is the decision to donate to charity (1 = donated, 0 = no donation). "Norm level" is a dummy for the type of norm shown (1 = high norm, 0 = low norm). "Optional norm" is a dummy for how the norm was presented (1 = optional, 0 = forced). "Norm level * Optional norm" is an interaction. "Female" is a dummy for gender (1 = female, 0 = male). "Age" is the participant age in years. p < .10. *p < .05. **p < .01. ***p < .001.
F I G U R E 2 Mean donations across conditions for participants who donated. Bars indicate 95% confidence intervals. Conditions are sorted from highest to lowest donation means conditions in relation to means, by giving higher amounts. That norm avoiders stick out this way can also be seen in Figure 2, showing the non-transformed mean amounts.

| The effect of optional norms on donation decisions (H3, H4, H5)
We now turn to investigating the effect that making the norm optional has on donation decisions. Comparing norm avoiders to participants in the control condition, we find a significant difference with norm avoiders donating less frequently, χ 2 (1, N = 909) = 8.756, p = .003, r = .098. This was the opposite direction as expected in H3.
We followed up this result with exploratory pairwise comparisons with the other conditions and found norm avoiders to donate less frequently than all other conditions, except for forced low norm. Specifically, norm avoiders donated less frequently than participants in forced high norm, χ 2 (1, N = 904) = 11.581, p < .001, r = .113, optional high norm, χ 2 (1, N = 781) = 7.567, p = .006, r = .098, optional low norm, χ 2 (1, N = 771) = 6.282, p = .012, r = .090, as well as the already tested control condition. While these exploratory pairwise comparisons were not preregistered, we believe that the results warranted attention and that they show a consistent pattern with norm avoiders donating less frequently than other conditions. Next, we investigate the effects of optional high and low norms on the frequency of donation decisions. Comparing those who chose to view the norm in the optional high-norm condition with the control condition, we find no significant difference, χ 2 (1, N = 668) = 0.014, p = .904. Similarly, we find no significant difference between those who viewed the norm in the optional low-norm condition and participants in the control condition, χ 2 (1, N = 658) = 0.007, p = .933. Thus, we cannot corroborate H4. Comparing conditions optional high norm and forced high norm (including both revealers and avoiders), we find only a borderline significant difference 3 for forced high norm to lead to more donations, χ 2 (1, N = 914) = 3.703, p = .054, r = .064. Comparing conditions optional high norm and control, we find no significant difference, χ 2 (1, N = 919) = 2.163, p = .141. Given this, we cannot fully corroborate H5. Overall, it appears that the optional conditions does not push donating above or below the control condition, except for the norm avoiders, who stick out by donating less frequently but at higher mean amounts.

| Predictors for norm avoidance (H6)
As can be gathered from Table 3, information avoidance was a strong and stable predictor of avoiding to reveal the descriptive norm. This means that participants who were more prone to generally avoid information about others' charitable donations were less likely to reveal the norm. Age and CSII also turn significant as predictors in Note: Logistic regression for all participants in the two optional norm conditions. The dependent variable is the decision to reveal the norm (1 = revealed the norm, 0 = no reveal). "Social Curiosity" is the mean answer on a 7-point Likert-type scale (1 = less curious, 7 = more curious). "Information avoidance" is the mean answer a 7-point Likert-type scale (1 = less information avoidant, 7 = more information avoidant). "Age" is the participant age in years. "Female" is a dummy for gender (1 = female, 0 = male). "CSII" is the mean answer on a 7-point Likert-type scale (1 = less conformist, 7 = more conformist). "Reactance" is the mean answer on a 7-point Likert-type scale (1 = less reactance, 7 = more reactance). Abbreviation: CSII, consumer susceptibility to informational influence. p < .10. *p < .05. **p < .01. ***p < .001.
norm for donations, we found both norm avoidance and norm seeking to be common behaviors. The norm avoiders did stick out compared with the other participants, as they donated less frequently, but also donated higher mean amounts. Turning to those who viewed a descriptive norm, the high norm leads to more donations than the low norm, in line with Frey and Meier (2004) as well as Meier (2007). Both the optional and forced ways of viewing the norms appear to lead to comparable outcomes, with the forced norms potentially leading to stronger effects. Compared with the control condition, the norms showed relatively weak effects on donation behavior overall, despite showing clear effects on beliefs regarding majority behavior.
In relation to previous studies comparing high and low norms of donation frequency, our 8.3% increase in donations between forced high and low norms could be compared with Frey and Meier (2004) finding a raw 2.3% difference between a high and low norm, which increased to 4.6% when controlling for factors such as donor history.
Comparing our forced high norm to the control condition, our 1.5% non-significant increase fall short of studies such as Agerström et al. (2016), whose high norm led to an increase of at least 17%, or used hypothetical donations rather than real monetary giving. Due to these and other differences such as using different percentage rates for the high norm and different charitable organizations to donate to, it appears that conclusions are highly related to context and not highly generalizable.

| Norm avoiders
Given that norm avoiders donated less often, but donated higher mean amounts, the possibility to avoid the norm may have attracted certain types of individuals specifically. The norm avoiders may be composed of both altruistic and non-altruistic participants, but with less people in the space in between these two categories, compared with other conditions. In practice, with some of the individuals in between gone, those remaining either give larger amounts or do not give at all, potentially explaining our results. This could both be due to self-selection to norm avoiding and due to unique effects of being asked about revealing. Our results speak against the hypothesized positive effect of merely being given the option of revealing a norm and speak more for strategic self-selection in using the option. While undecided individuals may seek norms to guide their behavior, both altruistic and non-altruistic participants could have motives to avoid the norm, such as protecting one's self-image and not wanting to become obligated to take certain decisions (Golman et al., 2017;Sweeny et al., 2010).
Specifically, where selfish individuals may be motivated by wanting to avoid becoming obligated to donate, prosocial individuals could have a more complex range of motives for norm avoidance. Prosocial individuals may want to avoid seeing a high norm of donations, as seeing this norm could indicate a less pressing need for their own donation, devaluing it (e.g., Duncan, 2004). Similarly, they may want to avoid seeing a low norm of donations as well, as seeing this norm could indicate a lower quality of the charity, again devaluing their donation. In terms of moral self-image (Jordan et al., 2015), there are reasons to avoid viewing the norm whether one donates regularly or not, as a high norm of donations could indicate that one is less moral than others, or not very special. Regardless of whether they believe the majority donates or not, individuals can be motivated to avoid crowding out effects (e.g., Savary & Goldsmith, 2020). That is, they may not want to reveal a norm, because it opens up the potential for less noble motives, such as reluctant altruism (Reyniers & Bhalla, 2013) rather than pure altruism (e.g., Batson et al., 1989). Even the mere act of seeking to know the norm can imply that one wants to surrender agency to others, thereby both threatening one's own sense of agency, and leading to ambiguous motives for donating. In this sense, ignorance can be a bliss.
That norm avoidance here that did not seem related to social curiosity bodes well for connecting it to the wider literature on information avoidance. It could indicate that descriptive norms are treated as useful information, rather than being seen as interesting merely due to curiosity. 4 The relation between norm avoidance and the adapted information avoidance scale could be a further indication of this or at the very least point to norm avoidance here as being more systematic than a random choice. In a broader sense, given the results here and elsewhere, we may need to question whether more information is always better, or always lead to more prosocial behavior, and if seeking information is always rational.

| Limitations
It should be noted that the effects on donation behavior investigated here appear to be small, yet given the magnitude of total worldwide charitable giving and its potential to save lives, even smaller effects can have real life consequences. The norm presented in the current study hinted at a group identity, in the form of other survey takers like oneself, yet it is possible that more closely related peers could have produced a norm that more participants would follow. Other studies have used temporary membership in an experiment group in comparable ways (e.g., Park & Shin, 2017), but the optimal level of signaling similarity to self to promote following norms is a complex issue (see Tian & Konrath, 2019

| Future research
Further studies will be needed to establish in which contexts norm avoidance occurs. For instance, different forms of prosocial behavior could be affected differently (e.g., Andersson et al., 2020), and norm avoidance could relate to not wanting to know potential harmful consequences of one's actions (e.g., Dana et al., 2006). Norm avoidance may also vary as a function of how salience of the norm interacts with individual differences in conformity or vary in relation to prior beliefs about the norms content interacting with one's own preferred decision. For instance, a prosocial individual could have different motivations to avoid the norm depending on whether they believe the norm is that the majority donates or not. In order to pry apart different motivations and types of norm avoiders, one could look closer at individual differences, for instance by measuring social value orientation (e.g., Murphy et al., 2011) or other measures of prosocial preferences.
Future research could also explore differences between choosing to view optional information, as investigated here, and more actively discarding information that has been given (but not viewed yet). This could create situations more comparable with typical avoidance behavior like avoiding to pick up medical test results. In making studies with higher external validity, one should also aim to include several types of available information, which could realistically be found online. Continuing down this path can lead us closer to an understanding of the role of descriptive norms, in a world of constant competition between pieces information over our limited attentional resources.

| Implications
Based on the current results, there appears to be little advantage for a charitable organization to present descriptive norms about how the majority behaves, with the goal of increasing donations. Showing forced information regarding how others are donating can backfire, leading to lesser number of donations when the norm is that the majority does not donate. In non-experimental solicitation situations, this may be the default situation, as most people do not donate when asked or when seeing a solicitation request. Even when the norm presented is that a majority donates, the effect on donation behavior appears to be very small, at least when the norm is presented as a faceless majority. Giving the option of learning the norm resulted in ambiguous consequences at best. As information becomes increasingly available online, one should be wary of the double-edged nature of norms, in their ability to both decrease or increase prosocial behavior. However, the effects of norms may differ by whether they are found in peer groups, social media, or statistics. This also means that understanding norm-seeking and norm-avoiding behavior may become increasingly important.

FUNDING INFORMATION
This work was supported by the Marianne and Marcus Wallenberg Foundation (grant number: 2014.0187). The funders had no role in study design, data collection, analysis, decision to publish, or preparation of the manuscript.