Introduction
Social norms are rules of behavior that are commonly approved by society, while personal norms represent what people believe to be appropriate behavior for themselves (Bicchieri, Reference Bicchieri2006). If deviations from a norm are likely to be sanctioned by society, individuals are inclined to follow the norm. A popular behavioral intervention based on social norms is a norm-nudge (Bicchieri & Dimant, Reference Bicchieri and Dimant2019), which encourages certain behavior by informing individuals about the actions of others, for example by showing energy conservation behavior of neighbors (Allcott, Reference Allcott2011) or tax compliance rates of fellow citizens (Hallsworth et al., Reference Hallsworth, List, Metcalfe and Vlaev2017). Norm-nudges may prompt people to act the way others are acting, because humans are inclined to model behavior on what others do, or what they believe others do (Bicchieri & Dimant, Reference Bicchieri and Dimant2019).
Norm-nudges are compelling interventions because they are cheap, easy to implement, and less prone to political resistance, compared with traditional interventions such as taxes or regulations (Benartzi et al., Reference Benartzi, Beshears, Milkman, Sunstein, Thaler, Shankar, Tucker-Ray, Congdon and Galing2017). Nevertheless, norm-nudges do not work in all circumstances, and their effectiveness depends on the design of the norm-nudge (Hummel & Maedche, Reference Hummel and Maedche2019). Moreover, there is a risk that a norm-nudge will be ineffective (see, e.g., Chabé-Ferret et al., Reference Chabé-Ferret, Le Coent, Reynaud, Subervie and Lepercq2019; Mackay et al., Reference Mackay, Yamazaki, Jennings, Sibly, van Putten and Emery2019) or even backfire, if not properly tailored to the population and context of interest (Hauser et al., Reference Hauser, Gino and Norton2018). For example, norm-nudges may backfire when they provide information about norm-violating behavior (e.g., tax evasion), which may lower motivations for compliance (Richter et al., Reference Richter, Thøgersen and Klöckner2018). Thus, it is relevant to test different kinds of norm-nudges and empirically assess their effectiveness across contexts.
The aim of this study is to examine the effectiveness of different norm-nudge messages with varying information in increasing individual investments in flood damage mitigation measures. Moreover, this study aims to examine heterogeneity in individual responses to these nudges as well as in the individual investment amounts, including individual characteristics and intercultural differences. We test two empirical norm-nudge frames with a large sample in Spain and the Netherlands and contrast these with a control treatment and a norm-focusing treatment. In the latter, respondents are asked to guess what other respondents would do before making an investment decision relevant to their own payoff. This task has been shown to influence behavior in past work, namely by increasing donations to charity (Bartke et al., Reference Bartke, Friedl, Gelhaar and Reh2017) and encouraging prosocial behavior, such as sharing funds.
Many studies on norm-nudges have focused on applications for health, finances, the environment, and energy (Abrahamse & Steg, Reference Abrahamse and Steg2013; Hummel & Maedche, Reference Hummel and Maedche2019). To our knowledge, previous research has not explored the effect of norm-nudges in the context of natural disaster risk-reduction measures such as investment in flood damage mitigation. Over the last few decades, natural hazards such as floods have increasingly impacted society, and this trend is expected to continue in the coming years due to climate change and population and economic growth in disaster-prone areas (IPCC, 2012; Munich, Reference Munich2018). Floods are among the most costly natural disasters (UNISDR, 2015). Despite the availability of cost-effective measures that limit flood damage to buildings (Aerts et al., Reference Aerts, Botzen, de Moel and Bowman2013), few people in flood-prone areas invest in or implement such measures (Botzen et al., Reference Botzen, Kunreuther, Czajkowski and Moel2019a). This highlights the importance of studying whether norm-nudges can incentivize individuals to invest in cost-effective mitigation measures. Examples of cost-effective individual damage-reducing investments include installing dry floodproofing measures that keep water out of the building during a flood (e.g., flood shields) or wet floodproofing measures that minimize damage when water enters a building (e.g., by applying water-resistant building materials). A recent review showed that flood risk management strategies will be much more cost-effective when including individual-level damage-reducing measures in addition to structural measures from traditional flood risk management, especially under an increased frequency and severity of floods as a result of climate change (Kreibich et al., Reference Kreibich, Bubeck, Van Vliet and De Moel2015).
Investments in individual damage-reducing measures can be considered a public good. For instance, in countries where the government can provide compensation for flood damage, such as the Netherlands, individual investments in reducing flood damage save tax money for compensated victims after flood events. In a previous survey with Dutch homeowners, we elicited social norms with regard to individual flood damage–reducing measures (Mol et al., Reference Mol, Botzen and Blasch2020b). We found that 25% of Dutch homeowners think that their peers would approve if they invested in damage-reducing measures, 50% are indifferent, and the remaining 25% think that their peers would disapprove. Therefore, we believe that investments in flood preparedness are subject to social norms and provide an opportune case for testing social norm-nudges. We focus on descriptive social norm-nudges in this article, because our previous elicitation of injunctive social norms showed that only 10% of Dutch homeowners indicate that their peers think that they should invest in damage-reducing measures. In our experiment, the norm-nudges refer to the flood protection investment behavior of participants in a previous flood risk game (Mol et al., Reference Mol, Botzen and Blasch2020b) and not to the behavior of peers in real life.
Previous research indicates that flood preparedness behavior is driven by the risk-reduction behaviors of others (Grothmann & Reusswig, Reference Grothmann and Reusswig2006; Poussin et al., Reference Poussin, Botzen and Aerts2014). For example, a survey of households in Australia found that perceived social norms had a greater influence on flood insurance purchases than homeowners’ perceptions of flood risk (Lo, Reference Lo2013). In a separate survey, Bubeck et al. (Reference Bubeck, Botzen, Kreibich and Aerts2013) showed a positive relationship between mitigation behavior and having neighbors and friends who implemented flood mitigation measures. However, these studies have not examined the effectiveness of different social norm messages in stimulating individual investments in flood damage mitigation measures.
In this study, we examine the efficacy of several different messages to stimulate flood preparedness measures in a controlled experimental study. As an additional innovation, we compare the impact of social norms on preventive behavior across two countries characterized by different flood risk management regimes. In addition, the populations of these countries differ in the average scores of individualism–collectivism (Pineda et al., Reference Pineda, Hernández-Santaolalla and del Mar Rubio-Hernsández2015), a measure that indicates to what extent people conceptualize themselves in relation to others (Triandis, Reference Triandis1989). Both characteristics may influence the effectiveness of social norm-nudges. We hence assess whether differences in current flood risk management between those countries – with the Netherlands more focused on public flood protection through dikes and Spain on individual protection measures – influence risk attitudes and personal norms for protecting one's home.
Ideally, social norm-nudges are examined in a large-scale field experiment, such as the classical examples on energy conservation and water conservation (Allcott, Reference Allcott2011; Price, Reference Price2014). Such a large-scale field experiment was practically infeasible for the case of flood preparedness, because (1) making substantial investments to make a home flood-proof is a more costly behavior than habitual behavior like energy saving or recycling and (2) there is no obvious field partner, such as a utility company, to measure and stimulate flood preparedness. An online lab experiment is a feasible and less costly alternative that can give directions for future field experiments, for example by identifying the most promising interventions to be tested in the field. Even though the results of lab experiments often correlate well with self-reported behavior in the field (Dohmen et al., Reference Dohmen, Falk, Huffman, Sunde, Schupp and Wagner2011; Dai et al., Reference Dai, Galeotti and Villeval2018), the results should be interpreted with caution (Levitt & List, Reference Levitt and List2007; Lades et al., Reference Lades, Laffan and Weber2020).
Literature review
A growing body of scientific research has identified important aspects of norm-nudge designs (see, e.g., Bicchieri & Dimant, Reference Bicchieri and Dimant2019). One line of research suggests that norm-nudges are effective only if the targeted behavior is interdependent; that is, when individual preferences are conditional on the empirical expectations of the behavior of others (Bicchieri, Reference Bicchieri2016). In contrast, when individuals are primarily motivated by their own basic needs or by their beliefs about what is morally right (i.e., targeted behavior is independent), individuals may expect others to behave in one way while behaving in a different way themselves (Bicchieri, Reference Bicchieri2010). Note that expectations may be normative (what other people would approve of), empirical (what other people do), or both normative and empirical (Bicchieri, Reference Bicchieri2016). In this article, we use the terminology of Bicchieri et al. (Reference Bicchieri, Lindemans and Jiang2014). We focus on interdependent behavior under empirical expectations, or descriptive norms – a preference to do X following the expectation that others do X as well (see Bicchieri & Dimant, Reference Bicchieri and Dimant2019). Note that the term descriptive norm is used slightly differently in the psychological literature, namely as the perception of what is common behavior (Bicchieri & Dimant, Reference Bicchieri and Dimant2019). Alternatively, norm-nudges may be based on injunctive norms, or expectations of what others find appropriate behavior (Cialdini et al., Reference Cialdini, Reno and Kallgren1990), such as Most people think you should not litter (Farrow et al., Reference Farrow, Grolleau and Ibanez2017).
Another important component of norm-nudge design is choosing the appropriate reference network. According to social identity theory, individuals are much more strongly affected by the actions of others if they share a certain group membership, such as gender, neighborhood, or ethnicity (Tajfel, Reference Tajfel1982). For instance, Goldstein et al. (Reference Goldstein, Cialdini and Griskevicius2008) found that referring to a specific reference network in the norm-nudge message other hotel guests who stayed in the same room more effectively promoted towel reuse than a generic message about other hotel guests. Some research suggests that the credibility of the message or message source may alter the effectiveness of norm-nudges. For example, Gifford et al. (Reference Gifford, Lacroix, Chen, Clayton and Manning2018) claimed that mistrust in messages from government officials could prevent citizens from taking action to combat climate change. However, recent experimental evidence on feedback frames to increase proenvironmental behavior did not demonstrate any evidence of a messenger effect (Hafner et al., Reference Hafner, Elmes and Read2017). Note that citizens who believe climate change is real may also mistrust government messages if they think the problem is underestimated. Conversely, individuals may feel threatened in their freedom of choice by the nudge, which may prompt them to act in opposition to the desired behavior. For example, Arad and Rubinstein (Reference Arad and Rubinstein2018) provided respondents with a nudge to increase savings, which increased the number of respondents selecting the savings arrangement. However, when respondents were told the government used a nudge, some respondents opted out of the savings arrangement. A strategy for overcoming this effect is to be transparent about the aim of the nudge, for example by informing respondents that the default option may encourage higher contributions to charity. Recent evidence shows that transparent nudges are judged as more trustworthy (Osman et al., Reference Osman, Fenton, Pilditch, Lagnado and Neil2018) and might be equally effective as traditional nudges that conceal their aim (Bruns et al., Reference Bruns, Kantorowicz-Reznichenko, Klement, Luistro Jonsson and Rahali2018).
Finally, the exact framing of empirical norm-nudge messages may improve their efficacy. For example, Stoffel et al. (Reference Stoffel, Goodwin, Sieverding, Vlaev and von Wagner2019) studied the effect of different quantifiers (a large number and nearly half) on intentions to participate in cancer screening. They found that both verbal quantifiers increased intentions compared with an exact numerical norm message (43%). While most norm-nudge messages are binary (e.g., people pay or do not pay their tax on time, Hallsworth et al., Reference Hallsworth, List, Metcalfe and Vlaev2017), some contexts allow for a continuous approach (e.g., Neighbors used 1,092 kWh on average, Allcott, Reference Allcott2011). However, many cases of norm-nudges use a binary message, even when the exact distribution of this variable is known.
Contributing to the literature on the transparency of nudges (Bruns et al., Reference Bruns, Kantorowicz-Reznichenko, Klement, Luistro Jonsson and Rahali2018), this study tests whether transparently showing the full distribution of choices by previous respondents (i.e., providing the exact percentage of those who chose each option) is more effective than summarizing these choices as a binary message. Furthermore, we measure several individual characteristics that previous research has identified as influencing the effectiveness of social norm messages, such as identification with the reference network (Liu et al., Reference Liu, Thomas and Higgs2019), political identities (Chang et al., Reference Chang, Chen and Krupka2019), and a concern for social comparison (Buunk & Gibbons, Reference Buunk, Gibbons and Guimond2006; Garcia et al., Reference Garcia, Tor and Schiff2013). An additional possible moderator of norm-nudge messages is the extent to which people perceive their relationships with others, which can be measured on a scale ranging from individualist (people conceptualize themselves as individuals) to collectivist (people conceptualize themselves as members of a group) (Triandis, Reference Triandis1989). With regard to social norms, collectivists may be more motivated to follow the behavior of others, implying that people who demonstrate a more collectivist worldview are more inclined to respond to norm-nudge messages (Oh, Reference Oh2013; Baldwin & Mussweiler, Reference Baldwin and Mussweiler2018).
Methodology
We used an experimental study to examine the impact of different norm-nudge messages on individual flood preparedness in two European countries. Following Hafner et al. (Reference Hafner, Elmes, Read and White2019), who argued that the effect of norm-nudge messages on behavioral intentions in real life may apply only to respondents who are in the position to execute the intention, we restricted the sample to homeowners. The design included an incentivized investment game in which respondents were asked to make decisions about investing in cost-effective measures to prevent damage of low-probability floods. To mimic the large consequences of real flood investment decisions, we implemented a random lottery incentive mechanism with high monetary stakes (see Camerer & Hogarth, Reference Camerer and Hogarth1999). Specifically, at the end of the experiment, the software randomly selected one respondent who had the chance to earn up to 650 euro, based on his/her decisions and luck in the game. The payment mechanism was explained at the start of the game to motivate subjects to consider their decisions carefully under high stakes.
Investment game
The investment game was a simplified version of a previous online experiment (Mol et al., Reference Mol, Botzen and Blasch2020b). We used identical parameters to facilitate a comparison of the results. In this game, respondents were asked to imagine owning a house with a flood risk for 25 years. With the hypothetical house comes a savings balance that could be used to make payments in the game, such as purchasing flood damage reduction measures. The currency used in the investment game was ECU (experimental currency units). The game started with the introduction of the parameters: a yearly flood probability of 1%, the maximum damage of 50,000 ECU in case of a flood, and the savings balance of 65,000 ECU. The next page offered five discrete investments with accompanying benefits in terms of reduced damage from flooding. The investment decision was made once, at the start of the game, and damage-reducing measures were effective for the full 25-year period of the game. This one-shot setup of the game was designed to be suitable for the online sample of respondents, accounting for the recommended time span for online surveys.
Figure 1 provides a screenshot of the page in the investment game where the flood risk was realized. This page showed a grid with 100 houses, with the house of the respondent enclosed in a square. The software randomly selected (based on the 1% flood probability) a number of houses that were flooded in the 25 years of the game and highlighted these in blue. In case the house of the respondent was flooded, the 50,000 ECU damage was subtracted from the savings balance. The optimal investment based on expected value calculations was 1000 ECU for a risk neutral (θ = 1) subject with a low time discounting rate (δ = 0.01).
Screenshot of flood risk page.
Experimental treatments
Each respondent was randomly assigned by the software into the control group or one of three treatment groups. In a previous experiment with the flood risk investment game, we compared investments by participants who received a mandatory insurance policy in the game with investments by participants in the control group who did not get any insurance (Mol et al., Reference Mol, Botzen and Blasch2020a). We aimed to power the current experiment such that we would be able to detect an effect size equal to this effect of insurance policy (d = 0.227) that was found in the previous experiment. An a priori sample size analysis with a significance level of 0.05 and power of 80% showed that we would need at least 252 participants per group to detect such an effect. We decided to aim for 250–300 participants in each treatment group. Our budget restricted the number of treatment groups to six. We decided to run all three treatments and the control group in one country (n = 4 × 300 = 1200, the Netherlands) and the most promising treatment plus the control group in the other country (n = 2 × 300 = 600, Spain). This approach was preregistered (https://aspredicted.org/q37kj.pdf). In two treatment groups, we displayed an empirical norm-nudge message with information about decisions of previous respondents. We used the percentages of previous investments in the control treatmentFootnote 1 of Mol et al. (Reference Mol, Botzen and Blasch2020b) to construct these messages. Note that these percentages were based on decisions of a sample of Dutch homeowners. To prevent confounding effects of in-group/out-group preferences, we did not mention the nationality of the reference group, but simply characterized them as homeowners. A third treatment group faced a focusing norm treatment, by eliciting beliefs about others' investment choices before participating in the investment game (Krupka & Weber, Reference Krupka and Weber2009). We did not include an injunctive norm message because results from our previous experiment showed that 90% of Dutch homeowners do not think their peers should invest in flood risk reduction. A message highlighting this information has the potential to backfire such that people are less motivated to invest, particularly if they would otherwise expect that a larger proportion of their peers think they should invest, leading to downward adjustment of beliefs (c.f., Bicchieri & Dimant, Reference Bicchieri and Dimant2019). The game was constructed such that individual decisions could not be observed by other respondents to focus on the effect of the norm-nudge messages in isolation from observability effects. In practice, many flood preparedness measures are taken inside a house and cannot be observed by neighbors either, except for the most extreme case of elevating a house. Figure 2a shows the investment screen in the control treatment.
Screenshots of the four treatments.
Norm-transparent
This treatment showed the full distribution of previous flood preparedness decisions in terms of the percentage of investments of previous respondents in five small text boxes below the five investment options (see Figure 2b). We expected that respondents would be unfamiliar with the flood preparedness decision environment. In other words, few respondents are confronted with flood preparedness decisions on a daily basis,Footnote 2 in contrast to other contexts that have been successfully related to social norms, such as energy conservation. Therefore, we expected respondents to have no (strong) beliefs about others’ behavior in the investment game. The norm-transparent treatment provided new information on mitigation decisions by others, illustrating the informational effect of a descriptive norm (Krupka & Weber, Reference Krupka and Weber2009). All boxes showed flood preparedness decisions by others in percentages. We presented this information as a percentage in the empirical norm-nudge message following Hallsworth et al. (Reference Hallsworth, List, Metcalfe and Vlaev2017), who found in a large natural field experiment that percentage norm messages are more effective than norm messages presented as a fraction (nine out of ten) or a general statement (the majority) in increasing tax compliance. Note that the online study by Stoffel et al. (Reference Stoffel, Goodwin, Sieverding, Vlaev and von Wagner2019) found opposite effects: verbal statements (a large number) were most effective in increasing cancer screening intentions. We believed that the findings of Hallsworth et al. (Reference Hallsworth, List, Metcalfe and Vlaev2017) are more relevant for our flood preparedness context, as they also focus on financial behavior, rather than intentions in the health domain.
Norm-high
In the norm-high treatment, an empirical norm-nudge message was displayed directly below the positive investment options (see Figure 2c). The message emphasized that a large majority of previous respondents had invested in damage-reducing measures, again by showing a percentage (70% of respondents in previous research opted for damage-reducing measures). While previous social norms research has mostly focused on binary outcome variables (e.g., whether or not to donate to charity or play a risky lottery), our setup requires respondents to choose from five discrete investment options. The norm-high treatment highlights the binary first step of the decision (invest vs. not invest) and is an intuitive way to describe the distribution.
Norm-focusing
The final treatment was designed to manipulate the strength of the norm focus (Cialdini et al., Reference Cialdini, Reno and Kallgren1990; Kallgren et al., Reference Kallgren, Reno and Cialdini2000). Krupka and Weber (Reference Krupka and Weber2009) showed in a large lab experiment that the norm focus intervention is effective in increasing prosocial behavior even when respondents believe that others do not behave according to the norm. Recently, Bolton et al. (Reference Bolton, Dimant and Schmidt2020) showed that the mere thought of what other people might do, operationalized with incentivized belief elicitation, leads to the same desired increase of donations compared with a more costly intervention (i.e., with monetary consequences). In line with Krupka and Weber (Reference Krupka and Weber2009) and Bolton et al. (Reference Bolton, Dimant and Schmidt2020), the dependent variable in our experiment cannot be confounded by strategic concerns. We used an incentive-compatible method to elicit beliefs about others' behavior, before requesting that respondents make a decision about their personal investment in the investment game. We asked respondents to estimate the percentage of other respondents investing in damage-reducing measures (1000, 5000, 10,000, or 15,000 ECU). An interactive screen emphasized that the remainder of respondents would not invest (see Figure 2d). We opted for such an explanation to facilitate a comparison of answers with the norm-high treatment, which also showed the percentage of people investing versus not investing. Furthermore, eliciting the full distribution would be a rather complicated task to explain, which could lead to undesirable attrition effects. Belief elicitation was incentivized with an e20 payment on top of the respondent fee for one randomly selected respondent. A large yellow alert marked the transition from the own investment decision to the belief elicitation decision screen. In the control treatment, belief elicitation was conducted after respondents completed the investment game. Figure 3 provides an overview of the experimental treatments.
Overview of experiment per treatment.
The experiment started with a short set of sociodemographic questions. Subsequently, the investment game was introduced through several pages of instructions supported with graphics to facilitate the understanding of respondents with different education levels. As in Mol et al. (Reference Mol, Botzen and Blasch2020b), the investment game was preceded by a test scenario to familiarize respondents with the decision screens. Before finishing the test scenario stage, respondents were requested to answer a few comprehension questions. The instructions were always accessible to respondents throughout the game. Additionally, the experimental software tracked attempts to answer these comprehension questions and the reopening of instructions. These were included in all regression analyses to control for an understanding of the experimental design. All respondents were paid a fixed participation fee of approximately €1, and one participant was randomly selected for a large payment corresponding to the bank balance at the end of the experiment at a conversion rate of 100 ECU = €1, which could range from €100 to €650. Participants were informed at the beginning of the experiment that this random selection would take place after all responses had been recorded. The total number of participants was not communicated upfront. All participants agreed to the informed consent page that explained the payment mechanism and the data storage policy. An email address was provided for anyone who would request further information, but no emails were received. We further paid one participant €20 out of the 57 participants who correctly estimated the share of participants who invested in the game (75%). The average duration of the experiment was 28 min, and the median duration was 12 min. The duration distribution was rather skewed with some extreme values regarding survey length, because the software did not prevent breaks, which allowed subjects to start the experiment and continue later.
The online experiment was distributed by the survey company Panelinzicht in two rounds: in August 2019 to a sample of Dutch homeowners and in October 2019 to a sample of Spanish homeowners. The panel is representative of the population of each country with respect to education, income, and gender. The experiment was translated into the local language of the respondents (Dutch and Spanish) and administered over the Internet using the experimental software oTree (Chen et al., Reference Chen, Schonger and Wickens2016), which allowed for a similar procedure across countries. The experiment started with a selection question to ensure that only homeowners were eligible to participate. The experimental interface was optimized for the screen size of tablets and desktop computers, which was communicated in the invitation email of the panel company. Nevertheless, it was possible to complete the experiment on smaller screens such as smartphones, but this required more effort through zooming and scrolling. The final data set contains 1200 Dutch responses and 605 Spanish responses.
Variables that were part of our hypotheses included trust in the presented information, susceptibility to peer influence, and individualism–collectivism personality scores. These variables were evaluated with survey questions at the end of the investment game. Moreover, we elicited variables that are likely to influence demand for flood damage mitigation investments independent of social norms, such as personal norms, response efficacy of mitigation measures, and risk perception. Table 1 provides an overview of all survey questions administered and the order in which they appeared.
Summary overview of the survey questions.
Notes: Order of variable in parentheses: ‘s’ indicates start survey and ‘f’ indicates final survey.
For example, ‘s7’ indicates that it was the seventh question and appeared in the start survey.
Hypotheses
We formulated hypotheses based on results from the previous literature. All hypotheses were formally preregistered prior to data collection.Footnote 3 Our main hypothesis concerns the effect of empirical norm-nudge messages on investments in damage-reducing measures in the investment game. Norm-nudges may help individual homeowners to act the way others are acting, as humans are inclined to model behavior on what others do. A large body of literature has shown that norm-nudges may be effective to increase environmental-friendly behavior (Allcott, Reference Allcott2011; Abrahamse & Steg, Reference Abrahamse and Steg2013). Furthermore, survey research indicates that flood preparedness behavior is driven by the risk-reduction behavior of others (Grothmann & Reusswig, Reference Grothmann and Reusswig2006; Poussin et al., Reference Poussin, Botzen and Aerts2014) and perceived social normsFootnote 4 (Lo, Reference Lo2013). When the information presented as a social norm-nudge differs from respondents’ a priori expectation as to what others will do, it may lead them to correct their beliefs. We expect that most respondents perceive few others will invest in flood damage mitigation measures and predict that an empirical norm-nudge message will lead to an increase of investments in flood risk protection measures compared with not including a norm-nudge message.
Hypothesis 1: Respondents confronted with an empirical norm-nudge (norm-high and norm-transparent) will invest more in damage-reducing measures than respondents in the control treatment.
To our knowledge, we are the first to test an empirical norm-nudge showing the percentages of previous decisions for each of the five discrete investment options (norm-transparent), as compared with an empirical norm-nudge highlighting the percentage of previous respondents who either have or have not invested (norm-high). Hence, we have no empirical information to hypothesize whether investment in damage-reducing measures will differ between these two norms. We expect that respondents in the norm-transparent treatment will have greater trust in the norm-nudge message than respondents in the norm-high treatment due to the provision of more transparent information in the former condition (see, e.g., Osman et al., Reference Osman, Fenton, Pilditch, Lagnado and Neil2018).
Hypothesis 2a: Respondents in the norm-transparent treatment have greater trust in the norm-nudge message than respondents in the norm-high treatment.
Accordingly, we expect that respondents in the norm-transparent treatment will be more likely to follow the majority of highlighted responses (investing some amount), relative to those in the norm-high treatment, due to greater levels of trust in the message.
Hypothesis 2b: Respondents in the norm-transparent treatment are more likely to follow the majority of the highlighted responses and invest than respondents in the norm-high treatment.
In the norm-focusing treatment, respondents are confronted with a belief elicitation page before they are asked to make an investment decision for their own payoff. The question, ‘What proportion of other respondents would invest in damage-reducing measures?,’ may encourage respondents to think about the norm (Kallgren et al., Reference Kallgren, Reno and Cialdini2000), which could increase investments even when respondents do not believe many others will invest (Bolton et al., Reference Bolton, Dimant and Schmidt2020). Based on the large effect sizes in previous research on norm-focusing and norm beliefs (Krupka & Weber, Reference Krupka and Weber2009; Bartke et al., Reference Bartke, Friedl, Gelhaar and Reh2017), we expect that the norm-focusing treatment leads to the highest investments of all treatments.
Hypothesis 3: Respondents in the norm-focusing treatment will invest the most in damage-reducing measures, relative to all other treatments.
Susceptibility to peer influence (Dielman et al., Reference Dielman, Campanelli, Shope and Butchart1987; Bearden et al., Reference Bearden, Netemeyer and Teel1989) may be another important driver of norm-nudge effects. Susceptibility to peer pressure is commonly studied in adolescents and young adults, where it has been found to drive gambling (Langhinrichsen-Rohling et al., Reference Langhinrichsen-Rohling, Rohde, Seeley and Rohling2004) and delinquent (Prinstein et al., Reference Prinstein, Brechwald and Cohen2011) behavior. Eckel et al. (Reference Eckel, Grossman, Johnson, de Oliveira, Rojas and Wilson2011) found that high-school students who report being highly independent are less likely to conform to normative behavior in decisions about sharing funds. Recent empirical evidence among adults shows that individuals are more likely to follow provided peer information if they are susceptible to informational influence, across several domains, including investment decisions (Hoffmann & Broekhuizen, Reference Hoffmann and Broekhuizen2009), consumer choice (Orth & Kahle, Reference Orth and Kahle2008), vaccination behavior (FitzSimons et al., Reference FitzSimons, Hendrickx, Lernout, Badur, Vorsters and Van Damme2014), and retirement decisions (Verhallen et al., Reference Verhallen, Brüggen, Post and Odekerken-Schrrder2018). In a recent article, Stöckli and Hofer (Reference Stöckli and Hofer2020) examined susceptibility to social influence among a large sample of adult online social media users. The authors found that susceptible individuals are more likely to buy what other users post about and to obtain information about political issues following other uses. Thus, we expect to find the same pattern of results.
Hypothesis 4: The effect of the empirical norm-nudge messages is greater for respondents who demonstrate higher levels of susceptibility to peer influence.
The degree to which empirical norm-nudges affect individuals may further be subject to differences in individualism–collectivism. These differences in self-concept can influence engagement in protective behaviors. For example, Parboteeah et al. (Reference Parboteeah, Addae and Cullen2012) found that collectivists are more likely to support sustainability initiatives. Recent evidence shows that people from individualist cultures respond differently to nudges in the context of vaccination behavior (Betsch et al., Reference Betsch, Böhm, Korn and Holtmann2017), in that they are more willing to be vaccinated. Individuals in southern European countries (such as Spain) generally score higher on collectivism than individuals in the Netherlands (Hofstede, Reference Hofstede2001; De Raad et al., Reference De Raad, Morales-Vives, Barelds, Van Oudenhoven, Renner and Timmerman2016). Therefore, we expect relevant variation in this variable within our sample that may explain heterogeneity in responses to the social norms message. To investigate the cultural differences of empirical norm-nudges on flood preparedness, we will examine respondents' scores on an 11-item individualism–collectivism scale (Cai & Fink, Reference Cai and Fink2002), a revised version of the original INDCOL scale (Triandis, Reference Triandis1989). We expect that respondents with a more collectivist worldview are more strongly influenced by an empirical norm-nudge message, as they are more inclined to follow group behavior.
Hypothesis 5: The effect of an empirical norm-nudge is larger for individuals with high collectivism scores on the individualism–collectivism scale, relative to those with high individualism scores.
Results
In this section, we report the experimental results, beginning with descriptive statistics for each preregistered hypothesis. We then turn our attention to a secondary treatment in the Spanish dataset with regard to intention to search for more information about flood risks. Finally, we report a number of observations from an exploratory analysis of the data.
Table 2 presents descriptive statistics of our sample. Demographic variables are very similar across countries, except for incomeFootnote 5 and home value,Footnote 6 which are in line with the population statistics of each respective country.
Descriptive statistics by country.
Results by hypotheses
Our main hypothesis concerned the effect of empirical norm-nudge messages on investments in damage-reducing measures in the investment game. Figure 4 shows the proportions of each investment level chosen by our respondents, split per treatment. The shaded areas indicate positive investments (1000, 5000, 10,000, or 15,000 ECU), and the remaining white area indicates the proportion of respondents who did not invest anything. We observe almost identical investment levels across treatments and countries. This result is unexpected, given the experimental research on the effectiveness of norm-nudges in the environmental domain (Abrahamse & Steg, Reference Abrahamse and Steg2013; Farrow et al., Reference Farrow, Grolleau and Ibanez2017; van Valkengoed & Steg, Reference van Valkengoed and Steg2019) and the survey research on the importance of perceived social norms in the flood risk domain (Bubeck et al., Reference Bubeck, Botzen, Kreibich and Aerts2013; Lo, Reference Lo2013).
Investments in damage-reducing measures by treatment.
We examined this result more formally with χ 2 tests and probit regressions and discuss the results in detail below. Table 3 provides an overview of all hypotheses and reports two-tailed χ 2 tests to analyze the differences in frequencies of investments with respect to treatments and independent variables of interest. Hypotheses 4 and 5 predicted interaction effects of susceptibility to peer influence and collectivism on the relationship between norm treatments and damage-reducing investments. Therefore, Table 3 reports z-statistics of the interaction term in a probit regression with binary investment as the dependent variable and susceptibility peer influence and collectivism as independent variables.
Results by hypotheses.
Notes: We report W tests for main effects and z-scores of probit regressions for hypotheses predicting an interaction. Support indicated for p < 0.05.
The first two rows of Table 3 show no support for a main treatment effect as predicted by Hypothesis 1; investments do not differ between respondents in the control group and the norm-high group (W = 85,493.5, p = 0.11), nor between respondents in the control group and the norm-transparent group (W = 87,618.5, p = 0.53). Our norm-transparent treatment did not lead to a higher trust in the messenger (W = 47,932, p = 0.27) (Hypothesis 2a), and investments were identical in the norm-transparent treatment and the norm-high treatment (W = 43,589.5, p = 0.3) (Hypothesis 2b). With regard to Hypothesis 3, we find no differences in investments between the norm-focusing group and the control group (W = 181,483, p = 0.8), the norm-high group (W = 98,444, p = 0.06), and the norm-transparent group (W = 93,252.5, p = 0.39). The difference in investments between the control group and the norm-transparent treatment is significantly stronger (z = −2.298, p = 0.02) for respondents with high levels of susceptibility to peer influence, which is in line with Hypothesis 4. However, this result is not found when comparing the control group with the norm-high (z = −0.662, p = 0.51) and norm-focusing (z = 0.707, p = 0.48) treatments. Finally, we find no support for Hypothesis 5; the coefficients of the interaction terms between collectivism and the treatment conditions on investments in damage-reducing measures are not significant (norm-focusing: p = 0.48; norm-transparent: p = 0.16; norm-high: p = 0.5).
The fact that our statistical tests do not support any differences between experimental treatments should not automatically lead to accepting the null. To examine the possibility of a null result in more detail, we conducted a power analysis. A meaningful effect size to compare our results is the effect of insurance policy compared with a control group without insurance (d = 0.227), in a previous experiment that used the flood risk investment game (Mol et al., Reference Mol, Botzen and Blasch2020a). We find that we achieve power in excess of 85% (Hypothesis 2a, norm-transparent vs. norm-high) and up to 93% (Hypothesis 1, norm-transparent vs. control) to find this effect size. The effect sizes in the current experiment range from 0.01 to 0.12 for pairwise comparisons, and a Kruskal–Wallis test confirms that there are no significant differences in average investments across all treatments (p = 0.26). These effect sizes are so small that they are not meaningful: the smallest effect size of interest based on a previous experiment with the same investment game was 0.227, which is substantially larger than what we find here.
Secondary treatment
As soon as the data collection for the Dutch respondents was completed, we conducted a preliminary analysis to determine the most promising treatment condition for the Spanish respondents, as indicated in the preregistration. We hypothesized that the nonsignificant effects of norm treatments on investment decisions, as described above, might be attributed to the cost of this investment decision. In other words, changing intentions following the norm treatments is a first step that many are willing to make, while the next step of changing behavior is more difficult to achieve, especially if it requires a costly investment (e.g., time, money, and effort). In our application, the behavioral change of interest is costly by definition: to make or increase a financial investment in flood preparedness measures. In line with previous research (Dur et al., Reference Dur, Fleming, van Garderen and van Lent2021), we speculated that norm-nudge messages are more apt to influence behavioral outcomes for which there is no monetary cost, such as clicking a link to retrieve more information, but less effective at producing changes in behaviors for which there is a tangible cost, such as investing money or increasing savings. To test this alternate explanation, we constructed a button to open a pageFootnote 7 with more information about flood risk and mitigation possibilities in Spain. We randomly distributed a descriptive social norm-nudge message based on results from a previous survey (Mol et al., Reference Mol, Botzen and Blasch2020b) to half of the respondents: Recent research shows that 68% of homeowners have installed at least one measure to protect their home from flood damage (see Figure 5). We expected more clicks on the link for information from respondents who received the norm-nudge message, compared with respondents in the control group. The results demonstrate that only a very small proportion of our sample clicked the link (63 respondents), and we find no differences in information search between these groups (χ 2 = 0, p = 0.989).
Screenshots of secondary treatment.
Beliefs
Next, we investigate the results of our belief elicitation question. On average, 75% of our respondents (Spanish and Dutch respondents combined) invested at least 1000 ECU (the minimum amount). The correct answer to the belief elicitation question, which asked respondents to indicate the percentage of other respondents investing a positive amount, was thus 75%. Figure 6 shows the distribution of beliefs about other respondents’ investment behavior in our sample, ranging from 0 to 100. The average belief was 46% and the median belief was 50%. A majority of respondents underestimated the correct answer, both in the control treatment (72%) and in the norm-focusing treatment (70%). Furthermore, belief responses indicated a preference for round numbers, such as 10, 50, and 80.
Histogram of beliefs. Note: The blue dotted line indicates the correct answer.
If we assume that beliefs in the norm-transparent and norm-focusing treatments are generally equal to the beliefs elicited in the control treatment, it implies that the information presented in those treatments (namely, that in previous research 70% of homeowners invested a positive amount) would correct beliefs upward for approximately 72% of respondents. This illustrates that there was a gap between information and beliefs, which would allow for an upward correction of these beliefs. Yet, an upward adjustment of beliefs did not, in this case, result in an adjustment of behavior. The absence of a treatment effect can, hence, not be explained by the absence of a gap between beliefs and the descriptive norm. However, the opposite effect may be possible: that the information–belief gap is so large that it actually signals that respondents are not aware of any norm with regard to flood preparedness measures, resulting in overall low norm-sensitivity. This would be in line with the argumentation in Bicchieri (Reference Bicchieri2006).
The belief elicitation question was asked in two of our four treatments: control and norm-focusing. The only difference between these treatments was whether the belief elicitation question was asked before (norm-focusing) or after (control) respondents' own investment decisions were taken. We found no difference in belief distributions across treatments (Kolmogorov–Smirnov test, p = 0.982). The lack of a treatment effect with regard to elicited beliefs suggests that beliefs about others' behavior and investment decisions are made concurrently – it does not matter which question is posed first. Figure 7 shows the relationship between beliefs and one's own investment decisions. The figure demonstrates a positive relationship between investments and elicited beliefs, implying that investments increase with the belief that more people are investing.
Beliefs of other respondents investing by own investment. Notes: The blue dotted line indicates the percentage of respondents who invested in our sample. Each individual observation is indicated with a gray dot, to which a small arbitrary noise has been added to the x coordinate to facilitate readability. The boxplot whiskers indicate the interquartile range, and the middle lines represent medians.
Other correlates of investments
Prior to our next set of analyses, we explored which variables have the most predictive power when it comes to investments in damage-reducing measures. We estimated simple binary correlations between all hypothesized predictor variables, control variables, and the dependent variable. Figure 8 shows the distribution of the 10 variables with the strongest correlations with decisions in the investment game in order of correlation strength.
Histograms of flood belief variables and correlations with investment decisions. Note: The stars indicate significant Spearman's correlations (***p < 0.001).
Type of respondents
An alternate explanation for the absence of norm-nudge message effects is that such messages are not effective for respondents who have already decided they want to invest. In contrast, those respondents with no clear preferences with regard to investing, for example those lacking a strong personal norm to invest, or those without positive experiences with measures already installed at home, could be more sensitive to information about other respondents' behavior. As Sunstein (Reference Sunstein2017) has noted, though nudges may appear to be ineffective at the aggregate level, they may demonstrate effects in distinct subpopulations.
To test this alternate explanation, we constructed two dummy variables to indicate a type of respondent who may be more susceptible to the treatments based on the most important predictors in Figure 8. The strongest predictor of investment in damage reduction is the number of measures already installed at home (see question f18 in Table 1). Therefore, we constructed a dummy of ‘No-measures individuals’ (1 = zero measure installed at home, 0 = at least one measure installed at home). Other important predictors from Figure 8 included personal norm, present bias, response efficacy, and expected water level.Footnote 8 We constructed a ‘Non-investors’ dummy (n = 369) to indicate individuals who do not expect high water levels (expected water level in cm = 0), have low response efficacy (strongly disagree or disagree), and do not have a strong personal norm (strongly disagree or disagree).Footnote 9 After constructing the two dummies to indicate the ‘No-measures individuals’ and the ‘Non-investors’ type of respondents, we conducted probit regressions to assess whether the norm-nudge treatments worked differently for these subsamples. The dependent variable in these regressions was binary investment in protection (in the flood risk investment game), and the treatment dummies were included as explanatory variables. The model was estimated separately for each of the different subsamples (‘No-measures individuals’, ‘Measures-individuals’, ‘Non-investors’, and ‘Investors’). We expected that the treatment is more effective for the noninvestors or those respondents who have not yet installed any measures at home than for the other samples. We expected that the ‘Non-investors’ and the ‘No-measures individuals’ would not be intrinsically motivated, based on the observation that these people do not have a strong personal norm or have not installed any measures at home. Therefore, we expected a larger effect of the treatments for those subsamples, as the treatments are external (they provide information).
Table 4 reports the results of the probit regressions of treatment by type of respondents. Model 1 restricts the sample to ‘Investors’, whereas Model 2 restricts the sample to the opposite set (‘Non-investors’). Model 3 restricts the sample to respondents who installed at least one measure at home (‘Measures’), while Model 4 restricts the sample to respondents who did not install any measure at home (‘No measures’). Across all models, we find no effect of treatment on investment in protection for any of the subsamples. As a robustness check, we ran a probit regression analysis (not reported here) on the full sample with interaction terms, all of which were nonsignificant. In sum, we find no support for the alternate explanation that norm-nudge treatments are more effective for a subsample of the respondents, such as those lacking a strong personal norm to invest, or those without positive experiences with measures already installed at home.
Probit regressions of treatment by type of respondents.
Note: Robust standard errors are given in parentheses.
**p < 0.05; ***p < 0.01.
Personal norms
As a next step in our analyses, we explored the differences between personal norms and social norms. As a complement to social norms, personal norms represent what people believe to be appropriate behavior for themselves (Schwartz, Reference Schwartz1977), or what they feel morally obliged to do (Harland et al., Reference Harland, Staats and Wilke1999). Previous research has shown that personal norms can be powerful determinants of proenvironmental behavior (Bamberg & Möser, Reference Bamberg and Möser2007; Yazdanpanah & Forouzani, Reference Yazdanpanah and Forouzani2015; Farrow et al., Reference Farrow, Grolleau and Ibanez2017). For example, Doran et al. (Reference Doran, Böhm, Pfister, Steentjes and Pidgeon2019) showed that personal norms (moral concerns) are a stronger predictor of policy support to mitigate climate change than consequence evaluations. Huber et al. (Reference Huber, Kip Viscusi and Bell2020) examined 5 years of longitudinal US household data and found that that personal norms are strongly related to recycling behavior. We measured personal norm as a response on a 5-point scale to the statement, I am morally obligated to take measures to reduce flood risk to my home, adapted from Doran and Larsen (Reference Doran and Larsen2016). We find that personal norm is significantly correlated with investment decisions (Spearman's correlation ρ = 0.174, p < 0.001), such that stronger personal norms correspond to higher investments. Note that the results on personal norms are correlational, thus providing limited information about causality.
Discussion
We conducted a high-powered preregistered experiment with homeowners to assess the effectiveness of norm-nudge frames on flood preparedness across countries. We found no evidence of a treatment effect: investments in damage-reducing measures of respondents in the norm-transparent, norm-focusing, and norm-high treatment groups did not differ from investments in the control group. We examined the alternate hypothesis that social norms affect intentions rather than costly behavioral change with a secondary treatment in the Spanish sample, but the results show no difference between the two treatment groups. Furthermore, we analyzed a subset of respondents who were not motivated by individual flood beliefs and personal norms and replicated the null effect of our full sample in this subset.Footnote 10 Several recent examples of studies that do not identify treatment effects of social norm-nudges are in line with our results: in the environmental domain (Chabé-Ferret et al., Reference Chabé-Ferret, Le Coent, Reynaud, Subervie and Lepercq2019; Mackay et al., Reference Mackay, Yamazaki, Jennings, Sibly, van Putten and Emery2019) and the financial domain (Franklin et al., Reference Franklin, Folke and Ruggeri2019). Generally, it has been noted for various domains, including corruption (Köbis et al., Reference Köbis, Troost, Brandt and Soraperra2019) and obesity (Oliver & Ubel, Reference Oliver and Ubel2014), that behavioral approaches such as norm-nudges should not be taken as substitutes but rather as supplements to traditional policies.
A recent article by one of the founding fathers of nudging outlined the main reasons for ineffective nudges and three possible responses (Sunstein, Reference Sunstein2017). We can rule out one of the two main reasons for failing nudges, namely counternudges, which are nudges aiming to promote the opposite behavior from the original nudge, as they were not at stake in our experiment. The second reason would be that decision makers have strong antecedent preferences, which would be hard to change regardless of the strength of the nudge. We assumed that most respondents were unfamiliar with the flood damage–reducing investment decision, which would argue against strong preferences. Nevertheless, strong preferences with regard to risk aversion, for example, could explain our results.
Comparing our results with prior findings on the effects of social norm-nudges, we find that our results differ from a recent meta-analysis of field experiments using social norms to promote proenvironmental behavior that reveals a medium-sized main effect of social norms compared with control conditions (Bergquist et al., Reference Bergquist, Nilsson and Schultz2019). Nevertheless, the authors find that social norms are less effective when communicated explicitly (i.e., by computerized messages) rather than implicitly (i.e., by cues in the environment) and that the influence of social norms is weaker in nonstudent samples than in student samples. Our design used explicit social norms in a nonstudent sample, which, according to Bergquist et al. (Reference Bergquist, Nilsson and Schultz2019), could explain the weak effects.
We are not the first to report a null effect of social norm-nudges: recent publications have revealed similar findings, and in some empirical studies, there was a backfiring effect of norm-nudge interventions (see, e.g., Fellner et al., Reference Fellner, Sausgruber and Traxler2013; Cranor et al., Reference Cranor, Goldin, Homonoff, Moore, Homono and Moore2020; Dimant et al., Reference Dimant, van Kleef and Shalvi2020). For example, Fellner et al. (Reference Fellner, Sausgruber and Traxler2013) and Cranor et al. (Reference Cranor, Goldin, Homonoff, Moore, Homono and Moore2020) examine the effects of a social norm-nudge on compliance in payments of TV license fees and taxes and find zero effects on compliance. Tyers (Reference Tyers2018) uses a social norm-nudge to try and convince consumers to buy carbon offsets for flight tickets and find no effect. The author argues that the main problem is that carbon offsetting is an unfamiliar concept to most participants, in contrast to other domains where norm-nudging has been effective, such as recycling, organ donation, or charitable giving. This explanation appears to also reflect our findings. Even though we have no focus group results to back up the claim that respondents are unfamiliar with the flood preparedness decision, several respondents answered the final open-answer feedback question by indicating that they had never thought about the topic before.
An alternate explanation for null effects of social norm-nudges by Dur et al. (Reference Dur, Fleming, van Garderen and van Lent2021) states that norm-nudges work well on changing intentions but may ultimately fail to change (costly) behavior. In a field experiment in a retail bank, Dur et al. (Reference Dur, Fleming, van Garderen and van Lent2021) find that a social norm-nudge increases intended savings and information search about saving plans, but not actual savings. We find a null effect for both behavior (investments in the flood risk investment game) as well as intentions (information search in the secondary treatment), rejecting this alternate explanation for our results. It should be noted that our study was an online lab experiment, rather than a field experiment that comprises the majority of the literature on social norm interventions. In a study similar to ours, Dimant et al. (Reference Dimant, van Kleef and Shalvi2020) examine the effect of social norm-nudge messages on cheating in an online laboratory experiment and find that their simple norm-nudges are unsuccessful at shifting norms, presumably because a behavioral norm is already in place. Another relevant online lab experiment using norm-nudge messages is Capraro et al. (Reference Capraro, Jagfeld, Klein, Mul and de Pol2019) that examines simple messages promoting personal norms (‘what do you think is the morally right thing to do?’). They show that these messages can effectively increase donation behavior immediately after the nudge, as well as in subsequent choices. These findings are in line with the high correlations between investments in our game and answers to the personal norms statements.
A further potential explanation for the absence of an effect could be that an online lab experiment is an artificial setting that does not perfectly translate into decisions in the field. Nevertheless, findings in Mol et al. (Reference Mol, Botzen and Blasch2020b) demonstrate that the flood risk game is generally well suited to illustrate the behavior of homeowners regarding flood risk preparedness. For example, the behavior of the players of the flood risk game under different flood risk probabilities and flood risk insurance schemes (mandatory vs. voluntary, high vs. low deductible) is very much in line with theoretical predictions. This gives us confidence that the general mechanisms leading to higher or lower investments into flood risk preparedness in the game can be transferred – at least qualitatively – to the real world. One additional result that points in this direction is that the Spearman's correlation between investments in the game and the number of flood damage reduction measures implemented by respondents (survey question f18) is positive and significant (ρ = 0.203, p < 0.001). Furthermore, our finding that flood risk perceptions and the perceived efficacy of flood damage mitigation measures in a real-world context are similar to behavior in the game (see Figure 6) suggests that the lab game may translate into behavior and attitudes in the field. Nevertheless, when interpreting the results, one should keep in mind that the translation of behavior in the lab into relevant domains in the field is never perfect. One way to counter this limitation in future research is to develop a large-scale experiment in cooperation with homeowners’ associations and local governments to test the effect of interventions on real-world behavior.
Finally, Czajkowski et al. (Reference Czajkowski, Zagórska and Hanley2019) compared different variations of descriptive social norm-nudges in a field experiment for household waste sorting with a stated preference approach and found that the willingness to pay for recycling increases with the size of the norm, but that the effect is not monotonic. In other words, high absolute levels of the norm are less effective as they are ‘out of reach’. This finding relates to the difference between beliefs (of the current norm) and the information presented in the nudge.
One limitation of our study is that we did not communicate upfront to participants about the probability of being paid, as this was dependent on the total number of participants. A recent theoretical study has shown that to be fully incentive-compatible, the pay-one mechanism should be transparent about the chances of being paid before the start of the experiment (Azrieli et al., Reference Azrieli, Chambers and Healy2018). Therefore, we recommend that further research should communicate the chances of being paid clearly and early. Another limitation is that we did not elicit beliefs in the norm-transparent and norm-high treatments. We did not do this, because we did not expect to find independent beliefs about investments in the current sample after providing respondents with the percentages of previous investors. Moreover, we argued that investment data from a previous sample only provide an indication about the nature of investment behavior in the current sample.
In retrospect, it would have been interesting to elicit beliefs in the norm-transparent and norm-high treatments to check the consistency of beliefs across treatment groups. Had we found the same distributions of beliefs in the norm-transparent and norm-high treatments as in the control group, we would have inferred that respondents simply ignored our norm-nudge messages. Furthermore, we could have used the beliefs in the norm-transparent and norm-high treatments to test for the ‘norm distance effect’ (Bergquist & Nilsson, Reference Bergquist and Nilsson2018) that suggests that the power of social norms (messages) is larger when behavior is closer to the (perceived) norm. A second limitation is that the cultural differences between Spain and the Netherlands are not extremely large (Pineda et al., Reference Pineda, Hernández-Santaolalla and del Mar Rubio-Hernsández2015). To obtain a more heterogeneous sample with regard to the individualism–collectivism scale, researchers should consider surveying homeowners in more culturally diverse countries, such as Japan and the USA (Hofstede, Reference Hofstede2001).
This study suggests three main takeaways for flood risk communication policies. First, communication to raise risk awareness should take risk-related emotions into account. This recommendation follows from our finding that worry and concern are significant predictors of investments in damage reduction (see Figure 8), which is in line with the previous literature (Kunreuther, Reference Kunreuther2018). Second, informing homeowners about the effectiveness of damage mitigation measures may enhance flood preparedness. This recommendation follows from the strong positive correlation between response efficacy and investments in damage reduction in our study (see Figure 8), confirming previous findings on this topic (Poussin et al., Reference Poussin, Botzen and Aerts2014; Mol et al., Reference Mol, Botzen and Blasch2020a). Third, policy makers should pay particular attention to activating personal norms that were found to be associated with flood risk preparedness (as indicated by the strong correlation of personal norms with investments in flood preparedness in Figure 8). These results are in line with Wenzig and Gruchmann (Reference Wenzig and Gruchmann2018), who showed that personal norms generally have a much larger influence on proenvironmental behavior than social norms, and with Botzen et al. (Reference Botzen, Kunreuther and Michel-Kerjan2019b), who showed that personal norms matter more than social norms in a flood risk context. Schwartz (Reference Schwartz2012) argued that norms need to be activated to be able to influence intention or behavior by being aware of the consequences of actions and feeling responsible. This explanation complements our results in the context of flood risk preparedness, which was framed on the individual level – respondents who feel responsible for their home (i.e., personal norms) invest more in damage-reducing measures than those who do not feel morally obligated to protect their homes. The actions of neighbors and other fellow homeowners (i.e., social norms) may be of lower importance for mitigation decisions in the context of flood preparedness.
We further found a significant negative correlation between present bias and investments in damage-reducing measures. This finding is in line with the previous literature about myopia in the context of preparedness for low-probability/high-impact events, such as floods (Botzen et al., Reference Botzen, Kunreuther, Czajkowski and Moel2019a; Royal & Walls, Reference Royal and Walls2019). Homeowners perceive the high upfront costs of investing in damage reduction to be much higher than the expected benefits, and when they are presented as biased, they care more about costs now than about benefits later. One way to overcome this bias is through offering low-interest loans that spread the investment costs over time (Meyer & Kunreuther, Reference Meyer and Kunreuther2017; Kunreuther & Pauly, Reference Kunreuther and Pauly2018), which could stimulate flood preparedness. If a nudge in the environmental domain proves ineffective, as we show in the current article, it may warrant the use of stronger measures, such as incentives, regulations, and bans, to influence preparedness (Sunstein & Reisch, Reference Sunstein and Reisch2013; Carlsson et al., Reference Carlsson, Gravert, Kurz and Johansson-Stenman2019).
Conclusion
Floods are one of the deadliest and costliest natural disasters worldwide. Fortunately, individual homeowners can invest in several cost-effective measures to prepare their homes for flooding. We attempted to increase investments in flood risk reduction measures in a controlled experiment by subtly nudging respondents (homeowners in the Netherlands and Spain) to consider the social norm of fellow homeowners. In particular, we created different norm-nudge messages by either providing percentages of the population that previously invested in different flood-reduction options (norm-transparent), or the percentage of previous respondents who invested in flood reduction (norm-high). These treatments were contrasted with a control treatment and a norm-focusing treatment, in which respondents' beliefs about normative patterns of flood-reduction investments were elicited. We did not find any evidence of a treatment effect, suggesting that our social norm-nudges do not affect the flood preparedness of respondents in a flood risk investment game.
Our results suggest that the problem of underpreparedness for natural disasters cannot be addressed by providing these social norm-nudges, a finding that differs from previous studies that concluded that social norm-nudge messages can be effective ways to facilitate behavioral change in the environmental domain. Our exploratory analyses reveal that there is a strong correlation between beliefs of others' behavior and one's own investments; however, our treatments did not influence either.
A crucial difference between our study and other successful social norm-nudges from the literature is that the flood preparedness context is not a very familiar one for most respondents. Such unfamiliarity may lead to a large information–belief gap; respondents may not be aware of any norm with regard to flood preparedness measures, which could result in overall low norm-sensitivity. Further research could look into this relationship between familiarity and norm-nudge effectiveness. A positive recommendation for further research would be to develop personal or moral norm-nudge messages.
Acknowledgments
We would like to thank conference participants at NoBeC 2019 in Philadelphia, BEEMA5 in Villanova, and WINK Nudging and Beyond in Utrecht for valuable comments. This research has received financial support from the Alfred P. Sloan Foundation (Grant No. G-2018-11100), the Netherlands Organization for Scientific Research (VIDI 452.14.005), and the Wharton Risk Management and Decision Processes Center.
Open access
