Framing of risk and preferences for annual and multi-year flood insurance

Journal of Economic Psychology 39 (2013) 357–375

Contents lists available at SciVerse ScienceDirect

Journal of Economic Psychology

journal homepage: www.elsevier .com/ locate/ joep

Framing of risk and preferences for annual and multi-year floodinsurance q

0167-4870/$ - see front matter � 2013 Elsevier B.V. All rights reserved.http://dx.doi.org/10.1016/j.joep.2013.05.007

q Wouter Botzen has been responsible for the design, the statistical analysis and the reporting of the results of the choice experiment. Joop deTeun Terpstra have contributed to the overall research strategy in which the choice experiment was embedded.⇑ Corresponding author. Tel.: +31 205983979; fax: +31 20598955.

E-mail address: [email protected] (W.J.W. Botzen).

W.J. Wouter Botzen a,⇑, Joop de Boer a, Teun Terpstra b

a Institute for Environmental Studies, VU University Amsterdam, De Boelelaan 1087, 1081 HV Amsterdam, The Netherlandsb HKV Consultants, Lelystad, The Netherlands

a r t i c l e i n f o

Article history:Received 5 December 2012Received in revised form 21 May 2013Accepted 26 May 2013Available online 1 June 2013

JEL classification:D81G22

PsycINFO classification:31403920

Keywords:CommunicationDecision makingFramingPreference uncertaintyRisk perception

a b s t r a c t

The decision of many individuals in floodplains to not purchase flood insurance may impairthe risk-spreading function of flood insurance markets. This study estimates the effective-ness of risk communication frames and insurance policy conditions in increasing demandfor flood insurance. It is examined how communication interacts with individual framesabout the flood hazard that are rooted in regulatory focus theory. A choice experiment elic-its willingness-to-pay (WTP) for annual and multi-year flood insurance, using of a survey ofa representative sample of 1250 households. The statistical method is a mixed logit modelthat accounts for heteroskedasticity arising from stated choice certainty. The communica-tion frames considerably increase WTP compared with a control group. This effect of com-munication is positively related to an individual’s degree of prevention motivation.Moreover, we find that demand for flood insurance can be increased by introducingmulti-year policies, as long as the contract duration is not too long.

� 2013 Elsevier B.V. All rights reserved.

1. Introduction

A challenge for traditional economic theory is that many homeowners in floodplains do not purchase flood insurance (e.g.Krantz & Kunreuther, 2007; Kunreuther & Pauly, 2004). For instance, many homeowners in the USA do not purchase floodinsurance which has been provided by the National Flood Insurance Program (NFIP) at premiums that are close to, or in somecases even below, the expected value of flood damage (Browne & Hoyt, 2000; Kriesel & Landry, 2004; Michel-Kerjan & Kousky,2010). This can be explained by the attitude of many people that ‘‘a flood will not happen to me’’ (Kunreuther, 1978). Economicexperiments have confirmed that indeed many individuals neglect low-probability risk and do not purchase insurance, whileanother large group has a willingness-to-pay (WTP) that is considerably above the expected value of the loss (Laury, Morgen-McInnes, & Swarthout, 2009; McClelland, Schulze, & Coursey, 1993; Schade, Kunreuther, & Koellinger, 2011). Similar behaviorhas been observed in stated preference surveys of flood insurance demand (Botzen & van den Bergh, 2012a, 2012b).

Boer and

http://crossmark.crossref.org/dialog/?doi=10.1016/j.joep.2013.05.007&domain=pdf

http://dx.doi.org/10.1016/j.joep.2013.05.007

mailto:[email protected]

http://dx.doi.org/10.1016/j.joep.2013.05.007

http://www.sciencedirect.com/science/journal/01674870

http://www.elsevier.com/locate/joep

358 W.J.W. Botzen et al. / Journal of Economic Psychology 39 (2013) 357–375

Flood insurance purchases appear to be inconsistent with standard expected utility theory, which predicts that individ-uals purchase flood insurance as long as the insurance premium is not much higher than the expected value of the loss,under the assumption that individuals are risk averse and fully informed about the risk. Kunreuther and Pauly (2004)augment this standard theory by postulating that consumers incur implicit or explicit transaction costs associated withobtaining information about underlying loss probabilities. Individuals may not spend time on acquiring such information,because this is a costly activity. This relates to ‘bounded rationality’, in the sense that it may be a rational strategy not topursue information because it is hard and time-consuming, and involves subjective attention costs (Conlisk, 1996).Kunreuther and Pauly (2004) show that, if the costs of acquiring information on risk are high and/or the perceived loss prob-ability is low, then individuals can be discouraged from making rational insurance decisions, and they do not insure againstdisasters. A solution may entail providing information about the risk, but empirical evidence on this is largely lacking. Anextensive review of the literature on perception and communication by Kellens, Terpstra, and De Maeyer (2012) concludedthat very few studies have examined the effects of flood risk communication, and none of these assessed the influence of riskcommunication on flood insurance demand.1

This study examines the effect of risk communication frames on flood insurance demand, by explicitly controlling forframes that individuals have about flood risk. Individual risk frames are ‘‘organizing principles that shape in a hidden andtaken-for-granted way how people conceptualize an issue’’ (de Boer, Wardekker, & van der Sluis, 2010), and are in this studyelicited as prevention and promotion motivations related to flood-risk protection. It is the interplay of such individual andcommunication frames of risk that can be expected to influence individual decision making regarding insurance purchases.Communication frames are defined as ‘storylines of a risk aimed at informing individuals’. Our approach to frames refers tomental knowledge structures that capture the typical features of a situation; these may include standards of reference,which are the basis of the well-known framing effects of ‘‘gains’’ versus ‘‘losses.’’ However, the latter type of effects is notpart of this study. We use such storylines of risk since Kunreuther, Novemsky, and Kahneman (2001) show that merely pro-viding probabilistic information on risk does not affect individual risk judgments, while providing descriptions that allowindividuals to place probabilistic information in a useful context can be effective in influencing individual risk perceptions.The storylines applied in this study emphasize either the risks or amenities associated with living in a floodplain or the abil-ity of water managers to protect land from flooding, in the context of either high-probability/low-impact floods in unpro-tected floodplains or low-probability/high-impact floods in protected floodplains. These communication frames provide arealistic description of flood risks in the study area in the Netherlands (see Section 3.1). Similar typologies are applicableto other countries, such as the USA where flood-risk awareness campaigns have been proposed in floodplains with, and with-out, flood protection infrastructure (Ludy & Kondolf, 2012).

Demand for flood insurance has been elicited under various risk communication frames using a choice experiment thatwas part of a survey of a representative sample of 1250 households. An error correction mixed logit model estimates theinfluence on the demand for flood insurance of frames, as well as of the characteristics of the insurance policy, while takingaccount of model error introduced by individual stated choice uncertainty. Contract duration is one of the characteristics ofthe insurance since recently it has been proposed to introduce multi-year flood insurance (MYI) policies for increasing theuptake of flood insurance, in particular, in the USA (Michel-Kerjan, 2010; Michel-Kerjan, Lemoyne de Forges, & Kunreuther,2012). A theoretical model of single and MYI policies for catastrophe risk in a competitive market by Kleindorfer, Kunreuther,and Ou-Yang (2012) shows that risk-averse individuals prefer the price stability offered by MYI policies compared with an-nual insurance contracts, even though MYI policies are priced at a premium mark-up due to the uncertainty of future(re)insurance costs. Our study is one of the first to empirically examine how flood insurance demand relates to the durationof the insurance policy. Moreover, the insights that our study delivers into the effectiveness of risk communication are ofrelevance for the Netherlands, where the introduction of flood insurance is being considered since flood coverage is generallyexcluded from home and contents insurance policies (Botzen & van den Bergh, 2008), and have a broader relevance for coun-tries where the uptake of natural disaster insurance is low (Paudel, Botzen, & Aerts, 2012).

The remainder of this paper is organized as follows. Section 2 explains the adopted psychological framework of decisionmaking under risk. Section 3 discusses the survey. Section 4 describes the methods used to model demand for flood insurance.Section 5 provides the model results. Section 6 discusses the main implications of the findings of this study and concludes.

2. Prevention and promotion motivations for individual decision making under risk

Several prominent academics have called for improving the behavioral foundations of economic choice models that gobeyond standard microeconomic theory, including Nobel laureates McFadden (2001) and Kahneman (2003). Due to the fail-ure of expected utility theory in describing decision making about low-probability risks (Starmer, 2000), it is worthwhileexploring other theoretical frameworks for explaining individual decision making, such as theories that are founded in psy-chology. For example, Protection Motivation Theory has been implemented in a choice model for food to explain preventionof low-probability high-consequence health risk (Scarpa & Thiene, 2011). Another example is Nocella, Boecker, Hubbard, andScarpa (2012), who include psychological constructs of the Theory of Planned Behavior in a choice model for certified ani-

1 For readers interested in communication of health risks we refer to the review articles by Edwards et al. (2000), Ancker, Senathirajah, Kukafka, and Starren(2006), and Dusetzina et al. (2012).

Table 1Characteristics of the promotion and the prevention systems. Source: Higgins (2009) and Scholer et al. (2010).

The promotion systemRelates to advancement, aspirations (ideals) and accomplishmentsSensitive to gains versus non-gainsMore sensitive to positive than negative changes from the status quo

The prevention systemRelates to duties, responsibilities (oughts) and securitySensitive to losses versus non-lossesMore sensitive to negative than positive changes from the status quo

W.J.W. Botzen et al. / Journal of Economic Psychology 39 (2013) 357–375 359

mal-friendly foods which increase animal welfare and can lower health risks. Krantz and Kunreuther (2007) have empha-sized the importance of goals (or motives) in theories of individual decision making under risk and, in particular, flood insur-ance purchases. In psychology, such motivational accounts have focussed on the distinction between individual motivationsto maximize outcomes versus motivations to avoid failure (Atkinson, 1957; Lopes, 1987).

Regulatory focus theory characterizes two coexisting motivational systems that shape decisions: namely, the promotionand prevention systems with features as described in Table 1 (Higgins, 1997, 1998). Prevention-motivated individuals arefocused on achieving progress that restores or returns them to the status quo, while promotion-motivated individualsaim to improve upon the status quo. This distinction between preferences for outcomes that are coded as being higher orlower than a reference point (the status quo) has already been made in prospect theory (Kahneman & Tversky, 1979; Tversky& Kahneman, 1992). Regulatory focus theory goes one step further by explaining the motivations that underlie these pref-erence differences. Promotion and prevention motivations can imply that individuals have different risk attitudes to servetheir underlying motivational system, as Scholer, Zou, Fujita, Stroessner, and Higgins (2010) have shown. For instance, pre-vention-motivated individuals are likely to engage in risk-seeking behavior when this allows them to overcome a loss andreturn to the status quo, while in other situations they engage in risk-averse behavior in order to prevent falling below thestatus quo. This behavior is consistent with evidence in experimental economics showing that the same individual can ex-hibit both risk-seeking and risk-averse preferences in different contexts (Starmer, 2000). Other research finds that theendowment effect (loss aversion) is associated with prevention motivation (Liberman, Idson, Camacho, & Higgins, 1999),and the same applies to the probability weighting function in prospect theory, which induces risk aversion when individualsare faced with a hazard (Kluger, Stephan, Ganzach, & Hershkovitz, 2004).

Both the prevention and promotion systems may guide decisions about purchasing flood insurance, although we are notaware of other studies that have examined this empirically. Prevention-focussed individuals could view insurance as a pro-tective device that helps them to maintain the status quo. Therefore, they may purchase flood insurance with the objective ofpreventing a reduction in their wealth that would occur if they were uninsured and became the victim of a flood. The pro-motion system may motivate individuals to purchase insurance if individuals regard insurance as a means to accomplishadvancement. The status quo right after a flood is a deteriorated financial position, and the payout of insurance may be seenas an advancement upon that status quo. According to this perspective, the purchase of insurance may be seen as an invest-ment that yields a return during times of financial hardship.

3. Survey method

3.1. Description of the survey

The survey was conducted in June 2011 among a sample of respondents in Rotterdam and adjacent municipalities. Thesurvey was administered over the Internet. The respondents were randomly selected panel members of TNS-NIPO (DutchInstitute for Public Opinion and Market Research), which is a large market research agency in the Netherlands. A first qual-itative pilot was performed by professional interviewers using 8 in-depth face-to-face interviews. A second quantitative pilotincluded 36 completed surveys. The purpose of these pilots was to check whether respondents understood the questions,and to trial the online implementation of the survey. The response rate to the final survey was 69%. The overall structureof the survey was as follows. First, respondents received the description of the communication frame. Second, several blocksof questions were shown to measure individuals’ attitudes toward living in the area described in the frame. Third, respon-dents answered the choice experiment (Section 3.2). Fourth, blocks of questions elicit individual characteristics related toprevention and promotion motivations and attitudes towards climate change. The questionnaire concluded with the usualsocio-economic questions.

Every respondent was given one of the six versions of the questionnaire, each of which had a different risk communica-tion frame, as visualized in Fig. 1.2 Each frame consists of a description of flood risk with an emphasis on different aspects of theflood hazard. Descriptions were included in addition to probabilistic information, because it has been shown that individuals areoften insensitive to only probabilistic information about a hazard, and react more to contextual information (Kunreuther et al.,

2 Complete descriptions of these frames are available from the authors upon request.

Unprotected floodplainwith emphasis on:

Protected floodplainwith emphasis on:

risk(Frame 1)

safety(Frame 2)

amenities(Frame 3)

risk(Frame 4)

safety(Frame 5)

No riskcommunication

(Frame 6)

Fig. 1. The risk communication frames of different versions of the questionnaire.

Flood insurance Flood insurance A Flood insurance B

No

flood

insurance

Maximum coverage Coverage€25,000

Coverage€200,000

Contract duration Duration1 year

Duration5 years

Deductible Deductible€0

Deductible€500

Premium Premium€2.50 per month(€30 per year)

Premium€10 per month(€120 per year)

Fig. 2. Example of a choice card.


2001). The type of communication that the survey imitates corresponds to actual practice by certain local authorities, whichsend letters to inform inhabitants of floodplains about the flood hazard. The frames form a storyline throughout the survey,in the sense that specific questions, such as the choice experiment, are explicitly related to one of these frames. While, in prac-tice, a time-lag may exist between the receipt of information about risk and the insurance purchase decision, the experimentthat elicits flood-insurance demand follows the risk communication after several questions in the survey have been answered. Acontrol group received no flood risk communication (Frame 6). The survey questions in the control group were related to ‘‘livingin the river delta’’, which basically applies to the entire Netherlands. A frame contained either a description of the flood prob-ability and potential water levels in a floodplain that is protected by dyke infrastructure (low-probability/high-impact risk) or afloodplain that is located outside of the flood protection system (high-probability/low-impact risk).3 In particular, the frame ofthe flood risk in the protected floodplain described that flood protection can fail during high water levels, which occur on aver-age once in 2000 years, and that flood water levels can be up to 2 or 3 m high during such an event. The frame of flood risk in theunprotected floodplain described that high water levels, which can cover streets with water, occur on average once in 10 years.Moreover, each of these frames emphasized different aspects of living in such an area: namely, risks related to climate change,water safety policy, and the amenities of living near water. These texts are provided in Appendix A. A description of the ame-nities of water was included in the frame of an unprotected floodplain, but not in combination with a protected floodplain, be-cause flood protection infrastructure in the latter essentially aims to keep water out. Each frame was accompanied by fourphotos in order to provide a visualization of the description of the flood hazard (see Appendix A). All texts and photos werecarefully chosen to provide realistic representations and descriptions that the Dutch government could use, and, in fact, hasused in some cases, for the purpose of flood risk communication.

3.2. Choice experiment

Currently, flood coverage is excluded from home and contents insurance policies in the Netherlands.4 Victims of flooddamage may receive partial compensation for flood damage from the Dutch government via the Calamities and CompensationAct (Wet Tegemoetkoming Schade bij Rampen en Zware Ongevallen �WTS in Dutch), but this is uncertain. The introduction tothe choice experiment explained that the government decides to abolish the WTS, and that, instead, it becomes possible to pur-chase flood insurance. Subsequently, respondents were asked to imagine that they had to decide whether or not they would be

3 Placing absolute risk levels in context by relating these to levels of risk in other times could be a useful risk communication strategy. However, such anapproach has not been followed here, because flood probabilities in the Netherlands have remained rather constant over time as these are determined by fixedsafety standards that flood protection infrastructure has to meet by law.

4 Since September 2012 (more than 1 year after the survey was conducted) a new insurance company has introduced a catastrophe insurance policy whichcovers the risk of flooding in the Netherlands. It is, however, unlikely that this insurance will result in a broad insurance coverage for flood risk in theNetherlands because: coverage is restricted to only homeowners who live in protected floodplains; the insurance company has placed a maximum limit on thenumber of policies that can be sold; and relative to the low flood probability in protected floodplains, a high premium is charged (between about €1000 and€2500 per year in areas that are exposed to flooding) for a limited coverage (€75,000 per event). Because of the high costs of private flood insurance in theNetherlands, it has been proposed to introduce a public–private flood insurance arrangement in which the government participates as a reinsurer of extremeflood damage (see for example, Botzen & van den Bergh, 2008). The results of the choice experiment presented in this study provide insights into the potentialdemand for such a flood insurance arrangement.

Table 2Levels of the flood insurance attributes.

Attribute Levels

Coverage €25,000; €50,000; €100,000; €200,000; €400,000; €600,000Duration 1 year; 5 years; 10 years; 15 yearsDeductible €0; €1000; €2000; €3000; €5000Premium €2.50; €5; €10; €15; €20; €30; €50; €80 per month

Fig. 3. Distribution of stated choice certainty.


willing to purchase flood insurance if they owned a house with a ground floor in the area that was described to them (the sto-ryline of the frame), and, if so, which insurance policy they would prefer. This implies that the estimates of flood insurance de-mand that are derived in this study are only applicable to homeowners with a ground floor.5 We did not provide an indication ofthe value of the house that respondents were to imagine owning. Therefore, their own perceptions of this value could affecttheir willingness to pay for higher levels of insurance coverage. On being presented with each choice card, respondents wereasked to indicate their preference between ‘‘flood insurance A’’, ‘‘flood insurance B’’, or an option of ‘‘no flood insurance’’.The insurance policies differed according to the attributes coverage, contract duration, deductible, and premium. Fig. 2 providesan example of a choice card.

Respondents were asked for their preference with respect to one practice choice card, which was carefully explained, aswell as to four choice cards which had different combinations of levels of the attributes. Table 2 provides the levels that theattributes could take on, which were carefully selected after consultation with insurance companies. Contract duration wasbased on recent studies that advocate the introduction of MYI contracts of durations of up to 15 years in the USA, in order toincrease the market penetration of flood insurance (Michel-Kerjan, 2010; Michel-Kerjan & Kunreuther, 2011). However,Aerts and Botzen (2011) have examined the pricing of long-term flood risk in the face of climate change, and concluded thatuncertainty about climate change may result in a premium mark up of MYI compared with annual policies. An importantquestion that is analyzed here is whether consumers are willing to pay such a mark-up. Kunreuther and Michel-Kerjan(2012) examine this question for 2-period flood insurance contracts in the USA using an economic experiment, and concludethat more individuals are willing to purchase a 2-period than a 1-period contract. The choice experiment implemented hereexamines whether this conclusion also holds for the longer contract durations that have been proposed in the literature, andhow preferences for the duration of a flood insurance policy are influenced by risk communication.

The experimental design allocates combinations of levels to the attributes which together make up a choice card that isshown to a respondent. The design was generated using the software Ngene,6 in accordance with the principle of D-efficiency.This principle means that the design is created in such a way that it provides the maximum amount of information that can beobtained, which is expected to result in the smallest variance of the choice model. What is called a ‘point efficient design’ of 48choice cards was generated using prior estimates of the coefficient values obtained from the results of the pilot survey.7 Such astatistically efficient design that includes prior information about coefficient values is more statistically efficient and more ro-bust to model misspecification than statistical designs without prior information, or orthogonal designs that were often used inearlier studies (Ferrini & Scarpa, 2007). The design satisfies the properties of level balance, moderate attribute level overlap, andorthogonality (uncorrelated attributes), and excludes dominant choice options. A commonly applied ‘blocking procedure’ di-vides the total number of choice cards of the choice experiment in such a way that each respondent has to answer only fourchoice cards. In order to optimize the design, different combinations of the attribute values were trialed. Finally, the design withthe minimized D-error that was finally selected was obtained after repeated iterations which were allowed to run uninterrupt-edly for several days.8

5 The descriptions of the frames do not really limit the inferences of the results, since these frames of flood risk in protected and unprotected areas are quitegeneral and broadly applicable to areas in the Netherlands that are exposed to flooding.

6 See: http://choice-metrics.com/.7 A mixed logit panel model with a normally-distributed random parameter for the coverage attribute and a zero mean normally-distributed error correction

term has been fitted to estimate the design, using 250 Halton draws.8 The final D-error is 0.00018.

http://choice-metrics.com/

Table 3Differences between situational and chronic prevention and promotion motivations.

Prevention motivation Promotion motivation

Situational: reaction to the risk described inthe frame

In this situation I would expect to sufferflood damagea

In this situation I would mostly consider the amenities ofwatera

Chronic: general characteristics of anindividual’s behavior

I prefer a secure environment andfinancial security

I am able to handle setbacks and I remain optimisticabout a good ending

a The term ‘‘this situation’’ refers to the area that was described to respondents in the communication frame.


3.3. Eliciting choice certainty

Choice models are derived from random utility theory. This theory assumes that individuals answer choice tasks as ra-tional economic agents, who are able to consistently reveal their pre-defined preferences, and are able to choose an alter-native in a choice task by trading off preferences for the attributes of a variety of alternatives (McFadden, 2001). In practice,this assumption may be violated (Loomes, Orr, & Sugden, 2009). Several studies have shown that many individuals haveuncertain preferences for alternatives, and are uncertain about the choices made in a choice experiment (e.g. Lundhede,Olsen, Jacobsen, & Thorsen, 2009). Such uncertainties, as has been formalized by Li and Mattsson (1995), may arise becauserespondents are uncertain about the exact value of the good for them, or about the meaning and description of the valu-ation task (Shaikh, Sun, & van Kooten, 2007). Accounting for stated choice certainty in the statistical analysis of stated pref-erence studies can improve the fit of the choice model (Brouwer, 2010); influence WTP estimates (Alberini, Boyle, & Welsh,2003); and may mitigate the effects of hypothetical bias (Hensher, 2010). Hung (2009) stressed the importance of control-ling for individual preference uncertainty in assessing flood insurance demand using stated preference methods. Here, theusual methodology to elicit choice certainty is followed. This places a 10-point scale after each choice card, and asks therespondent how certain he/she is about the choice made on a scale from 0 to 10, where 0 means ‘‘not certain at all’’ and 10means ‘‘very certain’’ (Brouwer, Dekker, Rolfe, & Windle, 2010).

Fig. 3 shows the distribution of the respondents’ ratings of choice certainty. The overall mean score is 6.78. Most of therespondents are quite certain about the choices they made in the experiment since about 90% answered the middle categoryof certainty (5), or higher. This suggests that the experiment was not too difficult for many respondents, and that mostrespondents had a high degree of certainty about their preferences for the insurance options.

3.4. Question used to measure prevention and promotion motivations

The questions used to measure prevention and promotion motivations are discussed in detail in de Boer, Botzen, andTersptra (2013), and are briefly described here. Two variables have been constructed that represent different elementsof promotion and prevention motivation: namely, ‘‘situational’’, which directly relates to the risk described in the frame,and ‘‘chronic’’, which relates to general characteristics of an individual’s behavior. These differences and their most impor-tant items are summarized in Table 3. A total of 7 frame-related questions measured situational motivated reasoning interms of prevention focused defensive pessimism and promotion focused optimism. These questions measured motivatedreasoning in terms of prevention focused defensive pessimism (e.g. ‘‘If I lived in a neighborhood outside the dikes, then Iwould make sure that I am well prepared for high water levels’’) and promotion focused optimism (e.g. ‘‘If I lived in aneighborhood outside the dikes, then I think that my house will be very attractive because of the water abundant environ-ment’’). All responses were invited on a 7-point scale, anchored by 1 (‘‘does not apply to me at all’’) to 7 (‘‘applies to mecompletely’’). The participants in the control group answered slightly differently worded questions (‘‘As inhabitant of thisriver delta, I want to make sure that I am well prepared for high water levels’’).

The participants’ chronic motivational orientation was measured by their ratings of eight short portraits. Each portraitconsists of two sentences describing a person in terms of a value or preference that is important to him or her. We assumethat chronic prevention motivations are related to security, safety, stability or obeying rules. Chronic promotion motiva-tions are characterized by portraits of persons who are self-determined, achievement oriented, and open to change. Thefemale version of a prevention item is: ‘‘A safe environment is important for her; she prefers to avoid everything thatis risky.’’ The female version of a promotion item is: ‘‘She is able to handle setbacks very well; she remains optimisticabout a positive outcome.’’ The participants were asked to compare the portrait to themselves and to rate on a 7-pointscale ‘‘how much like you’’ the person is. A set of six items measured general beliefs about climate change as well as spe-cific beliefs about climate change risks at the local level, which were used for creating the climate skeptic variable (de Boeret al., 2013). An example of an important item of this variable is ‘‘The seriousness of climate change has beenexaggerated’’.


3.5. Sample characteristics

A total of 1256 respondents who were equally distributed over the different frames (questionnaire versions) completedthe choice experiment.9 After removing protest responses,10 this resulted in a total of 4940 completed choices. Overall, thesample is close to being representative for the Dutch population aged 25–75. 95% of the respondents were born in the Nether-lands, while this is 90% in the actual Dutch population. 50% of the sample are female. The average age is 50 and all age groupsbetween 25 and 75 years are adequately represented. 26% have a Bachelor’s or Master’s degree, which is similar to the Dutchaverage. The median category of gross household income is €38,800–€51,300, which is close to the Dutch average of €55,700.

3.6. Coding of the variables

Appendix B describes the variables used in the analysis. Different methods of coding categorical variables have been ap-plied. Dummy variables are used for several categorical variables, such as the risk communication frames. Continuous vari-ables are created from categorical variables that represent monetary classes, such as household income. Ordinal qualitativevariables, which are partitioned into J intervals, can be transformed into values on the real axis using an approach proposedby Terza (1987).11 Using Terza (1987), an ordered categorical variable (e.g. categories of experienced difficulty in filling out theexperiment like ‘‘easy’’ or ‘‘difficult’’) can be translated into an underlying numerical variable with a continuous scale (e.g. alatent variable of the degree of difficulty) (see also van Praag, Frijters, & Ferrer-i-Carbonell, 2003). The Terza (1987) transforma-tion has been applied to the education level of the respondent and the degree of difficulty of answering the choice experiment.Principle component analyses have been used to construct the variables representing prevention and promotion motivations, aswell as a variable representing the degree to which an individual is a climate change sceptic, as has been described by de Boeret al. (2013). The Cronbach’s alphas (see Table B1) show that these variables are measured in a reliable way.

4. Methods used for the choice model analysis

4.1. The heteroskedastic mixed logit model

The random utility model assumes that the probability pij of an individual i choosing alternative j equals the probabilitythat the utility of alternative j is greater than, or equal to, the utility associated with an alternative q for every alternative inthe choice set (q = 1 ,. . . , J):

9 TheFrame 6

10 Eighgovernmor livin

11 Thesuprem

12 It shGumbe

pij ¼ prob½ðVijt þ eijtÞP ðVijt þ eijtÞ8q 2 q ¼ 1; . . . ; J; j–q�; ð1Þ

where Vijt and eijt are, respectively, the observed and unobserved components of individual i’s utility associated with alter-native j in period t. It is assumed that eijt follows a type 1 extreme value distribution. The observed utility Vijt can beexpressed as bxijt, where xijt are the observed attribute levels that influence the choice for alternative j with coefficients b.In addition, utility can be made dependent on choice invariant characteristics zi with parameters a. In other words,Vijt = b0xijt + a0zi. A priori it is not known how the communication frames will influence flood insurance demand. These framescould influence the valuation of the attributes of flood insurance or the overall utility level of the flood insurance alterna-tives. Therefore, we will empirically examine whether the frame variables interact significant with the attributes (xijt)and/or whether the frame variables independently influence the overall utility of flood insurance and enter the variablevector zi. The influence of the frames on the utility of insurance may differ between people depending on their degrees ofprevention and promotion motivations. Moreover, the strength of these motivations may influence the individual valuationof the attributes of flood insurance. Therefore, we will empirically examine whether the variables of prevention and promo-tion motivations interact significantly with the insurance attributes (xijt) or the frame variables in the vector zi.

A mixed logit model with an error correction component is employed to fit the choice model. The mixed logit model is avery flexible choice model that allows for unobserved heterogeneity in individual preferences for attributes (McFadden,2001). A normally-distributed zero mean error correction component is added to the mixed logit model, in order to allowfor different variances of the insurance alternatives and the opt out option (Scarpa, Willis, & Acutt, 2007). Choice certaintyis introduced in the model via heteroskedasticity which are additive effects to the indirect utility function. This approachviews choice uncertainty as a source of model uncertainty, and assumes that more uncertain preferences result in moreinconsistent choices (Beck, Rose, & Hensher, 2011).12 This view is supported by an analysis of the determinants of choice cer-

number of completed questionnaires per frame are: 216 for Frame 1; 208 for Frame 2; 209 for Frame 3; 216 for Frame 4; 202 for Frame 5; and 205 for.ty-four protest responses were excluded. These include individuals who state that they: do not accept the abolishment of the current arrangement ofent compensation; do not believe/accept the description of flood risk (the communication frame); or do not accept the scenario of owning a house and/

g on the ground floor.se are characterized by a continuous unobservable ordinal latent index, where each interval is ranked (1 through J) in increasing order according to itsum (Terza, 1987).ould be noted that this approach differs from Lundhede et al. (2009) who include choice certainty via the parameterization of the scale parameter of the

l error, which is a multiplicative effect on indirect utility.


tainty in Appendix C, which shows that stated choice certainty is positively related to the experienced difficulty of the exper-iment and tough choices, while certainty is positively related to income and the education level. In the error correction mixedlogit model, this approach implies that stated choice certainty enters the model directly as heteroskedasticity in the error termsof the model. In other words, the error terms are allowed to depend on stated choice certainty. In this model, individual i’s utilityfor alternative j = 1, . . . , J in choice setting t = 1, . . . ,T is (e.g. Patil, Burris, Shaw, & Concas, 2011):

Uijt ¼ b0xijt þ a0jzi þ eijt þXM

m

djmrmuimt: ð2Þ

Respondents answer four sequential choice cards (so that T = 4), which means that error terms are not independent. Thispanel data structure is represented by the time subscript t, and is explicitly modeled by estimating the same error for choicesmade by the same individual (Train, 2003). The indicator variable djm equals 1 if the error component m appears in utilityfunction j (in this application the utility of flood insurance), and 0 otherwise (in this application the utility of not having floodinsurance). The model estimates the standard deviation rm of effect m; and uimt is the individual effect m for individual i. TheM random individual effects (rmuimt) are normally-distributed with zero means and variances r2

m. The vector b0 containscoefficients bjki of attributes xjik, k = 1, . . . ,K which can be specified to be random as follows:

bjki ¼ qjk þ djkwi þ rjkv jki: ð3Þ

Individual heterogeneity in the means of the randomly-distributed coefficients is introduced by wi, which is a set ofchoice invariant characteristics. qjk is a constant term and djk is a vector of coefficients which produce an individual specificmean. vjki is a normally-distributed random term with mean zero and standard deviation 1. rjk is the standard deviation ofthe marginal distribution of qjk. In the population bjki � Normal ½qjk þ d0jkwi;r2

jk�. Heteroskedasticity is introduced in the dis-

tribution of the random parameters by specifying r2ijk ¼ r2

jk � ðexpðc0jkhiÞÞ2, and in the variance of the error component by

r2mi ¼ r2

m � ðexpðh0mhiÞÞ2. In this application hi is the variable of stated choice certainty. The choice probability of this model

is:

Pijt ¼expðb0xijt þ a0jzi þ

PMmdjmrm expðh0mhiÞuimÞPJ

q¼1 expðb0xiqt þ a0qzi þPM

mdqmrm expðh0mhiÞuimÞ: ð4Þ

The parameters are estimated by conducting maximum simulated likelihood with Monte Carlo integration using 1000Halton quasi-random draws. Individual specific parameters, conditional on choice, for a random coefficient can be estimatedwith the simulation of the equation (Train, 2003):

bn ¼

Rbn

bn

QTt¼1Pðchoice jjxnt;bnÞgðbj/Þ

h idbnR

bn

QTt¼1Pðchoice jjxnt;bnÞgðbj/Þ

h idbn

; ð5Þ

where g(b|/) is the distribution of b in the population; and / is its vector of parameters. It is these parameters that will beused for the estimations of market penetration, and WTP for flood insurance.

Choice models can be used to predict consumer choices for a specific product. The potential market penetration for floodinsurance (D) is estimated using the commonly applied method ‘sample enumeration’, in which the predicted share of thesample who choose (flood insurance) alternative i is calculated with the equation (Ben-Akiva & Lerman, 1985):

DðiÞ ¼ 1N

XN

n¼1

PðijxnÞ; ð6Þ

where N is the number of individuals in the sample; and P(i|xn) is the probability (Eq. (4)) that an individual n chooses i withspecific attribute values xn on the basis of the utility specification of the choice alternatives and the parameter estimates ofthe error correction mixed logit model that accounts for heteroskedasticity arising from choice certainty.

Elasticities and marginal effects (ME) are estimated as they have a more intuitive interpretation than coefficients (b) inchoice models. Elasticities indicate the percentage change in the probability of choosing to purchase flood insurance withrespect to a 1% increase in the level of an attribute of that alternative or explanatory variable. The ME is the change inthe choice probability for flood insurance, given a unit increase in an explanatory variable. Both elasticities and MEs are com-puted for continuous variables using probability weighted-sample enumeration (see Louviere, Hensher, & Swait, 2000).

4.2. Model specification

The utility of flood insurance depends on the attributes. Dummy variables are used for contract durations of 5, 10 and15 years to capture their non-linear relationship with the utility of flood insurance (see Section 6). The influence of the riskcommunication frames on the utility for flood insurance has been examined by testing all possible interactions of the attri-butes with the dummy variables representing each communication frame, and by including these variables independently inthe utility of flood insurance. The dummy variable of the control group has been excluded from the model. In other words,


the frame variables capture the differences between utility for flood insurance in the communication frames relative to thebaseline of no communication.

It has been examined how individual characteristics can explain heterogeneity in preferences by testing all possible inter-actions of all explanatory variables with the attributes of flood insurance, as well as with the variables of the frames. The finalmodel includes only significant variables. It turned out that several variables of characteristics of the respondents’ house13

and the geographical characteristics of the respondents’ living area related to flood risks14 are statistically insignificant. Thissuggests that, as intended, respondents based their decision to purchase flood insurance on the characteristics of flood riskas described in the frame, and on the exposure to the flood hazard as described in the choice experiment, i.e. owning a housewith a ground floor in the area described in the frame. Variables representing socio-economic characteristics, apart from thepsychological characteristics, appeared to be insignificant and are, therefore, excluded from the final model.15 Unobserved het-erogeneity for preferences for the attributes is estimated by specifying random distributions for the coefficients of the attri-butes.16 Except for coverage, the standard deviations of the coefficients of the attributes turned out to be insignificant,which means that these parameters can be specified as fixed in the final model. The resulting utility specifications are:

13 Therespond

14 Thethe clos

15 Insi16 Foll

distribu17 Thi

Uflood insurance ¼ b1 � coverageþ b2 � coverage� CPREVM þ b3 � coverage� CPROM þ b4 � 5year durationþ b5

� 10year durationþ b6 � 15year durationþ b7 � deductibleþ b8 � premiumþ b9 � premium

� climateskeptic þ b10 � premium� Frame1þ c1 � Frame1þ c2 � Frame1� CPREVM þ c3 � Frame2

þ c4 � Frame2� CPREVM þ c5 � Frame3þ c6 � Frame3� CPREVM þ c7 � Frame3� SPREVM þ c8

� Frame4þ c9 � Frame4� CPREVM þ c10 � Frame5þ c11 � Frame5� CPREVM þ c12 � Frame5

� SPREVM

Uno flood insurance ¼ a1 � constant; ð7Þ

CPREVM and SPREVM stand for, respectively, chronic and situational prevention motivations, and CPROM stands for chronicpromotion motivation.

5. Results

5.1. Estimation results of the choice model

Table 4 shows the results of the error correction mixed logit model (Eq. (4)) with (right columns) and without (left col-umns), the correction for choice uncertainty which is modeled as heteroskedasticity in the variances of both the error com-ponent and the random coefficient of coverage. Choice certainty is included as a dummy variable that equals 1 forrespondents who stated a certainty level of 5 or higher.17 The pseudo-R2 statistic is 0.28 which indicates a good fit for thistype of models. The standard deviation of the error component is statistically significant in both models, and implies a consid-erably larger variance of the utility specification of the insurance alternatives than of the opt out. This is in line with others whofind that the variance of utility of hypothetical alternatives is larger than the opt out, which has been called the ‘status quo’effect in choice experiments (Botzen & van den Bergh, 2012b; Hess & Rose, 2009; Hu, Cox, & Pan, 2009).

The significant standard deviation of the random coefficient of coverage indicates that unobserved preference heteroge-neity exists for this attribute. The parameters of the heteroskedasticity in the variances of the error component and the ran-dom coefficient of coverage are statistically significant and negative. This means that these variances are smaller for higherratings of choice certainty (ratings P5). This reflects the intuitive implication that larger choice uncertainty adds unex-plained variance (error) to the model. In other words, the error correction term indicates that additional variance is associ-ated with the cognitive effort of evaluating the two hypothetical insurance options relative to the status quo, and theobserved heteroskedasticity arising from stated choice certainty shows that this additional variance is smaller for individualswith a high choice certainty. The model without heteroskedasticity (left columns) can be rejected in favor of the heterosked-astic model (right columns) on the basis of the AIC and the Vuong test (p-value <0.01). Therefore, the remainder of this paperdiscusses the results of the latter model, unless stated otherwise. Marginal effects and elasticities are provided in Table 5,and will be discussed next, alongside the coefficient estimates in Table 4.

The coefficients of the attributes of the experiment are all statistically significant (Table 4). The utility of flood insurance ispositively related to coverage, which is valued more by individuals who can be characterized as having high levels of chronic

se characteristics of the respondent’s house include variables concerning: the floor level on which the respondent currently lives; whether or not theent owns his/her house; the value of the house (if the respondent is a homeowner); and the rent (if the respondent is a tenant).se geographical variables have been created based on the respondent’s zip code and Geographical Information Systems (GIS), and represent: distance toest main river; distance to the coast; elevation of the house relative to potential water level; classification of land use (urban, green urban or rural area).gnificant socio-economic characteristics are: gender; income; education level; age; and the number of household members.owing common practice (Train, 2003), normal distributions have been specified for the coefficients of coverage and deductible, while uniformtions have been specified for the dummy variables of the contract duration.

s specification results in the best model fit out of the many possible specifications that were estimated.

Table 4Coefficients and t-statistics of the choice models.

Variabele Error correction-mixed logitmodel

Error correction-mixed logit modelwith heteroskedasticity

Coefficient t-Statistic Coefficient t-Statistic

Coverage 0.0016*** 9.71 0.0017*** 9.72Coverage � chronic prevention motivation 0.0005*** 2.94 0.0005*** 3.05Coverage � chronic promotion motivation 0.0004** 2.24 0.0003** 2.015-year duration 0.2156*** 2.92 0.2016*** 2.6110-year duration 0.3372*** 5.31 0.3352*** 5.4215-year duration �0.1805** �2.50 �0.1824*** �2.64Deductible �0.2386*** �15.68 �0.2365*** �16.10Premium �0.0329*** �20.78 �0.0329*** �24.14Premium � climate sceptic �0.0032** �2.55 �0.0031*** �3.13Premium � Frame 1 0.0068** 2.16 0.0068*** 2.71Frame 1 2.9520*** 6.18 2.8800*** 5.98Frame 1 � chronic prevention motivation 0.7860** 2.19 0.9937*** 2.66Frame 2 3.1356*** 6.46 3.0611*** 6.47Frame 2 � chronic prevention motivation 1.0209*** 2.96 1.0596*** 3.10Frame 3 3.1725*** 6.42 3.2414*** 6.63Frame 3 � chronic prevention motivation 1.1001*** 2.91 1.0825*** 3.29Frame 3 � situational prevention motivation 1.1269*** 2.69 1.0986*** 2.72Frame 4 2.8654*** 6.14 2.8143*** 6.02Frame 4 � chronic prevention motivation 1.2372*** 3.87 1.3807*** 4.48Frame 5 2.3810*** 5.15 2.2586*** 5.09Frame 5 � chronic prevention motivation 0.7173** 2.08 0.6551* 1.88Frame 5 � situational prevention motivation 0.7425** 1.93 0.7979** 2.05Constant �1.5682*** �5.06 �1.6764*** �5.50Standard deviation of the error component 3.4895*** 16.69 4.7646*** 12.49Standard deviation of the coefficient of coverage 0.0021*** 5.63 0.0054*** 4.33Heteroskedastic variance in the error component (choice certainty P5) n.a. �1.2451*** �4.73Heteroskedastic variance in the coefficient of coverage (choice certainty P5) n.a. �0.3383*** �3.91Number of observations 4940 4940McFadden R2 0.28 0.28AIC 1.5925 1.5876Log likelihood �3908 �3894

n.a. Stands for not applicable. Estimations are performed with Limdep software.* Statistical significance at the 10%.** Statistical significance at the 5%.*** Statistical significance at the 1%.


prevention and chronic promotion motivations. The utility of flood insurance relates nonlinearly to the duration of the con-tract: compared with a yearly contract, the utility increases if the contract lasts for 5 years, and the utility is even higher for a10-year contract, but is lower for a 15-year contract. The deductible has a negative influence on the utility of insurance. Thereduction in insurance demand that follows from an increase in the deductible of €1000 is much larger than the increase indemand that results from a similar expansion of coverage, as is reflected by the ME values in Table 5. A strong preference forlow deductibles has also been observed in actual flood insurance markets, such as in the USA (Michel-Kerjan & Kousky,2010). The premium is highly significant and has a negative influence on insurance demand. Individuals who are climatesceptics have a stronger reaction to the premium and, thereby, a lower utility of flood insurance.

The coefficient of an interaction variable of Frame 1 with the premium is positive and significant, which implies that theprice elasticity is lower in Frame 1 (�0.25 compared with �0.31 in the other frames). The price elasticity is relatively inelas-tic, which is in line with estimated price elasticities in actual flood insurance markets (Browne & Hoyt, 2000). The other attri-butes have a similar influence on the utility of insurance for all frames. The effects of the frames work through the level of theutility of insurance, as is reflected by the significant positive coefficients of the frame variables in the utility specification offlood insurance. This implies that, although the frames hardly influence the individual preferences for the attributes, theframes do have a significant positive influence on the overall preference for flood insurance compared with the no floodinsurance alternative. In other words, flood risk communication increases the overall value placed on flood insurance.

The five communication frame variables interact significantly with chronic prevention motivation. This means that thepositive effect of the communication frames on flood insurance demand is stronger for individuals who have a high (general)level of prevention motivation. Moreover, the effects of Frames 3 and 5 on flood insurance demand are stronger for individ-uals with a high level of situational prevention motivation, which reflect motivations that are specifically related to the com-munication frame. Of the psychological variables, chronic prevention motivation has the largest influence on flood insurancedemand. Fig. 4 illustrates that the average probability of choosing to have flood insurance increases from 0.78 to 0.9 ifchronic prevention motivation increases from the lowest to the highest category. The finding that the effects of communi-

Table 5Marginal effects and, if applicable, elasticities (between parentheses).

Marginal effect (ME) or elasticity

Coverage 0.0009 (0.15)Duration 5 years 0.009Duration 10 years 0.01Duration 15 years �0.008Deductible �0.01 (�0.18)Premium; average over all frames �0.02 (�0.30)Premium; Frame 1 �0.01 (�0.25)Premium; exclusive of Frame 1 �0.02 (�0.31)Frame 1 0.14Frame 2 0.14Frame 3 0.13Frame 4 0.13Frame 5 0.11Chronic prevention motivationa 0.01Chronic promotion motivationa 0.001Climate sceptica �0.0008Situational prevention motivationa 0.003

a The ME of this principle component variable has been constructed in threesteps. First, the z-score data are ordered and divided into 7 equally-sized per-centiles, and a variable has been constructed that indicates to which percentileeach z-score belongs (1 = lowest score, and 7 = highest score). Second, the averagez-score per percentile is calculated. Third, the ME shown in the table is calculatedas the change in the proportion of choices for flood insurance, when changing theaverage z-score of the middle percentile (4) to the subsequent higher category (5).


cation are stronger for individuals with a high degree of prevention, but not promotion, motivation may be explained by thefocus of the communication frames on the threats posed by the flood hazard, which overlaps more with an individual (prior)focus on prevention than promotion.

5.2. The influence of communication frames on demand curves for flood insurance

Potential market penetrations of annual flood insurance with coverage of €100,000 and a deductible of €5000 are approx-imated by predicting for every frame the percentage of choices to purchase flood insurance in the sample under varyingmonthly premiums of between €10 and €80 per month. These particular coverage and deductible values were chosen forthe predictions because consultations with insurance companies revealed that these values are being considered for the de-sign of flood insurance policies in the Netherlands. It should be realized that these market penetrations are only applicable tohouseholds who own a house with a ground floor. Fig. 5 shows the demand curve for the frame without risk communication.The results indicate that about 68% of the respondents are willing to purchase the flood insurance for a premium of €10,while this declines to 39% if the premium is €80. Even without risk communication, many homeowners are interested in pur-chasing flood insurance, but there is also a non-negligible share that chooses not to insure.

Fig. 6 shows the demand curves for flood insurance in the risk communication Frames 1–5. Note that the vertical axes ofFigs. 5 and 6 differ for the purpose of readability. The interpretation of these curves is as follows: the demand curve in Frame1 shows the percentage of respondents who are willing to purchase flood insurance for a certain premium, if they have re-ceived communication about the flood risk as provided in Frame 1 (unprotected area with emphasis on risk), etc. Overall,differences in demand curves between the three frames of unprotected areas are small. Frame 1 of an unprotected floodplain

Fig. 4. The influence of chronic prevention motivation on the probability of choosing to purchase flood insurance. Notes: The average choice probabilityrelated to this principle component variable has been constructed in three steps. First, the z-score data are ordered and divided into 7 equally-sizedpercentiles, and a variable has been constructed that indicates to which percentile each z-score belongs (1 = lowest score, and 7 = highest score). Second, theaverage z-score per percentile is calculated. Third, the average proportion of choices to purchase flood insurance is calculated per average z-score, using Eq.(6).

Fig. 5. Demand curve for flood insurance in the frame without risk communication (Frame 6). Notes: The dotted curves represent 95% confidence intervals.


with emphasis on risk results in the highest demand curve, while the demand curve in Frame 2 with emphasis on safety isonly slightly lower. The demand curve in Frame 3 is slightly lower than in Frame 2 for premiums up to €40, and is similar forhigher premiums. The demand curve of flood insurance in the protected floodplain with emphasis on risk (Frame 4) is lowerthan the demand curves of the frames of an unprotected floodplain, and the lowest demand curve is of the protected flood-plain with emphasis on safety (Frame 5). Nevertheless, the demand curve in Frame 5 is substantially higher than the demandcurve without risk communication in Fig. 5. This shows that the overall effect of flood risk communication is to considerablyincrease flood insurance demand, even if communication is about a low-probability/high-impact risk in an area with floodprotection infrastructure.

5.3. WTP for flood insurance

The estimation results of the choice model (Table 4) are used to assess changes in an individual’s overall utility of a floodinsurance policy per frame, characterized by specific levels of the attributes, compared with having no flood insurance. Thesechanges in utility can be translated into changes in consumer surplus, (CS) according to:

18 Res19 The

DEðCSnÞ ¼1an

lnXJ1

j¼1

eV1nj

!� ln

XJ0

j¼1

eV0nj

!" #; ð8Þ

where an is the marginal utility of income, which equals the negative of the marginal utility of price; and V1nj and V0

nj repre-sent utility with, and without, flood insurance based on the utility specification of the choice model (Eq. (7)). The marginalutility of price (a) in specification 7 is a = b8 + b9 � climate skeptic + b10 � Frame 1. The (sample average) improvement in con-sumer surplus of flood insurance compared with no flood insurance has been calculated per frame in order to give an esti-mate of maximum WTP for flood insurance per communication frame. This section reports the WTP estimates of the choicemodel that accounts for heteroskedasticity arising from stated choice certainty. These estimates differ between �1% and +6%from the WTP values of the choice model that does not account for this type of heteroskedasticity.

The WTP results are shown in Table 6. All differences between mean WTP values are statistically significant.18 The WTP isthe highest in Frame 1 that corresponds to communication about an unprotected floodplain with emphasis on risk. The differ-ences between Frames 2, 3 and 4 are small, while communication about the protected floodplain with emphasis on safety re-sults in a lower WTP (Frame 5). Overall, the WTP values for flood insurance are substantially higher in the frames withcommunication compared with the frame in which no communication about the flood hazard is applied (Frame 6).

Table 7 shows how the WTP for flood insurance relates to the contract duration of the insurance policy. All differencesbetween mean WTP values are statistically significant.19 The WTP for a 5-year policy is between 4% and 5% higher than theWTP for a yearly policy in Frames 1–5, and it is about 7–8% higher for a 10-year policy in these frames. These WTP differencesare more pronounced in the frame without communication (Frame 6): the WTP is 8% and 14% higher for, respectively, a 5- and10-year policy compared with a yearly contract. The WTP for a 15-year policy is between 3% and 4% lower in Frames 1–5, whilethis is 6% in Frame 6. The finding that MYI policies can increase demand is in line with results of Kunreuther and Michel-Kerjan(2012) who conduct a web-based experiment with real monetary stakes among 450 adults in the USA to examine demand fortwo-period versus one-period flood insurance. One of their main findings is that demand for the 2-period contract remains high-er than demand for the 1-period contract, even if the premium for a 2-period contract is between 5% and 10% higher. This isconsistent with the results in Table 7 of the experiment conducted here for contract durations of 5 up to 10 years, but, in addi-tion, Table 7 shows that the relationship between WTP and contract duration reverses for very long contracts (>15 years).

ults of t-tests show that all mean differences are statistically significant at the 1% level.results of the t-tests show that all mean differences are statistically significant at the 5% level.

Fig. 6. Demand curves for flood insurance in the risk communication frames of an unprotected floodplain with emphasis on risk (Frame 1), safety (Frame 2),or amenities of water (Frame 3), and communication frames of a protected floodplain with emphasis on risk (Frame 4), or safety (Frame 5). Notes: Thedotted curves represent 95% confidence intervals.

Table 7Maximum WTP for a flood insurance policy with €100,000 coverage and €5000 deductible for different years of contract duration, per frame (in € per month).

Type of risk communication Unprotected floodplain Protected floodplain No risk communication

Emphasis on Risk Safety Amenities Risk SafetyFrame number 1 2 3 4 5 6

Policy with duration of:5 years 156 (2.47) 128 (1.98) 134 (2.26) 122 (2.02) 108 (1.82) 54 (1.02)10 years 160 (2.51) 132 (2.01) 138 (2.29) 126 (2.05) 111 (1.86) 57 (1.06)15 years 144 (2.35) 118 (1.89) 124 (2.17) 113 (1.92) 99 (1.72) 47 (0.91)

Notes: Standard errors are shown between parentheses below the corresponding WTP values.

Table 6Maximum WTP for a yearly flood insurance policy with €100,000 coverage and €5000 deductible, per communication frame (in € per month).

Type of risk communication Unprotected floodplain Protected floodplain No risk communication

Emphasis on Risk Safety Amenities Risk SafetyFrame number 1 2 3 4 5 6

Sample average WTPStandard error

150(2.41)

123(1.93)

129(2.22)

117(1.97)

103(1.77)

50(0.96)


6. Discussion and concluding remarks

The failure of many individuals in floodplains to purchase flood insurance, even if premiums are close to actuarially fairlevels, poses a challenge for traditional microeconomic theory, and may impair the risk-spreading function of flood insurancemarkets. Risk communication has been advocated as a solution for the low uptake of natural disaster insurance. In addition,the introduction of multi-year insurance (MYI) coverage for flood damage has been proposed as a solution for increasing thelow uptake of flood insurance (Michel-Kerjan & Kunreuther, 2011). The effectiveness of these solutions for increasing floodinsurance demand has been hardly researched empirically. This study has examined the influence on demand for annual andmulti-year flood insurance of communication and individual frames about the flood hazard. For this purpose, a unique dataset was collected by conducting a survey of a representative sample of 1250 Dutch households. Demand for flood insurance


was elicited by means of a choice experiment, which has been analyzed with a mixed logit model with an error correctioncomponent. The choice model has a firm psychological foundation by applying regulatory focus theory, which explains indi-vidual decision making under risk using prevention and promotion motivations. This study delivers the following five maininsights, which will be discussed next.

First, flood risk communication results in a substantial increase in the potential market penetration and the WTP for floodinsurance compared with the control group who received no communication. This finding underpins the important role thatflood risk communication can play in efforts to increase the demand for flood insurance. For example, in the USA the volun-tary take up of flood insurance provided through the National Flood Insurance Program (NFIP) is low. A possible explanationprovided for this low demand by Michel-Kerjan and Kunreuther (2011) is that the language that is commonly used to com-municate risk, like the ‘‘100 year return-period flood’’, is ineffective in creating risk awareness among individuals. Thedescriptions and visualizations of the flood hazard used in the communication frames in the surveys conducted for this paperare likely to have a greater influence on individual demand for flood insurance than providing only probabilistic information.In July 2012 the US Congress enacted new legislation that calls for studies to reform the NFIP, because the current pro-gramme faces several problems, including operating losses and a low uptake of flood insurance. In light of the present find-ings, the design of an effective flood-risk communication campaign should be an integral component of such a reform. Futureresearch could examine how the effects of communication on individual risk awareness last over time.

Second, the observed effect of communication is the largest for the description of an unprotected floodplain that faces ahigh-probability/low-impact flood risk combined with a description that emphasizes risk. Nevertheless, the effect of com-munication is also large if flood hazard information is provided about a protected floodplain that faces a low-probability/high-impact flood risk. This is an important finding because most floodplains in the Netherlands are protected by dykeinfrastructure (Aerts & Botzen, 2011). In the USA, areas are mapped out of the floodplain if flood protection infrastructureis in place, which has contributed to low perceptions of flood risk, as well as insufficient levels of flood preparedness insuch areas (Ludy & Kondolf, 2012). The results of this study suggest that providing information about residual flood riskin protected floodplains can be effective in stimulating demand for flood insurance, despite high flood protection levels.It should be noted that differences in demand between the protected and unprotected floodplains are not the result ofa traditional framing effect in which the same risk is presented in different ways, since levels of flood risk in these twoframes are different. Only small differences can be observed in the influence on flood insurance demand of risk commu-nication frames that emphasize different aspects of the flood hazard, such as risk, safety, and amenities of water. Govern-ments that aim to develop floodplains may find it more desirable to communicate about the latter two aspects. The resultsimply that putting emphasis on these more attractive aspects of water does not necessarily result in much lower riskawareness than if risks are emphasized, and this can, therefore, be effective in stimulating demand for insurance againstflood risk.

Third, the effect of communication on flood insurance demand is considerably larger for individuals who are character-ized as having a high degree of prevention motivation. In this respect, chronic motivations had a larger influence on floodinsurance demand than situational motivations. This finding suggests that demand is largely driven by motivations thatare related to general characteristics of an individual’s behavior, while the influence of motivations that directly relate tothe flood risk described in the frame is weaker. While both prevention and promotion motivations are positively relatedto individual preferences for flood insurance coverage, prevention motivations interact positively with the communicationframes. This implies that the descriptions adopted in the communication frames are especially effective in raising the riskawareness of prevention-focussed individuals, while this is less the case for promotion-focussed individuals. Future researchcould explore how to create communication frames that are better suited for the latter.

Fourth, consumer demand for flood insurance can be increased by introducing multi-year insurance (MYI) policies, aslong as the contract duration is not too long. Relative to an annual flood insurance policy, the WTP for flood insurance is high-er if the contract duration is 5 years, and it is even higher for a 10-year policy, but the WTP is lower for a 15-year policy.These results imply that individuals value the price stability that MYI can offer up to contract durations of 10 years, whilethe disadvantage that MYI cannot be easily cancelled looms larger for longer contracts. These findings suggest that introduc-ing MYI policies that cover flood damage with contract durations up to 10 years may increase demand for flood insurance,which confirms theoretical predictions by Kleindorfer, Kunreuther, and Ou-Yang (2012). This finding is of practical impor-tance, as is revealed by an analysis by Michel-Kerjan et al. (2012) of flood insurance policy tenure of the entire portfolioof the NFIP between 2001 and 2009. That analysis indicates that median tenure is only between 2 and 4 years, which sug-gests that many individuals only hold onto their flood insurance policy if they experience several loss events and get a returnon their policy. Michel-Kerjan et al. (2012) propose the introduction of MYI policies for flood damage as a solution for theproblem that individuals tend to act myopically and commonly cancel annual policies. The empirical results presented hereshow that consumer demand for MYI contracts up to 10 years exceeds demand for annual flood policies, which provides asolid basis for further exploring the practical feasibility of introducing MYI for flood risk.

Fifth, an interesting finding for methodological reasons is that choice model fit is improved by eliciting respondents’ cer-tainty about their preferences for flood insurance stated when presented with each choice card of the experiment, and byaccounting for these as heteroskedasticity in the model. The effect of accounting for this heteroskedasticity is a change inWTP values between �1% and +6%, which is not very large. An analysis of the determinants of stated choice certainty findsthat the overall difficulty that respondents’ experienced in filling out the choice experiment, as well as tough choices in achoice card, reduced certainty, while certainty is positively related to income and the education level. These insights are


consistent with the choice model results which reveal that respondents’ certainty enters the model as a source of error thatadds to the model variance. The convenient method proposed in this paper for eliciting preference certainty for insuranceand for including this certainty in the statistical analysis could be useful for future studies that elicit consumer demand.The flood-risk communication frames do not influence respondents’ certainty about their preferences for flood insurance,but these frames do influence their demand for flood insurance. This suggests that we can exclude the possibility that theeffects of communication on insurance demand arise because the frames made respondents more uncertain about their pref-erences for flood insurance.

In conclusion, the estimated demand curves show that many Dutch household are interested in purchasing flood insur-ance, at least if the premium is not too high. Nevertheless, there are also a substantial number of homeowners that willchoose not to purchase flood insurance, even if premiums are low. The results of this study have shown that providing ade-quate information about flood risk or introducing multi-year contracts that offer price stability may be an effective way tostimulate flood insurance demand.

Acknowledgements

We are very grateful to Jeroen Aerts and Susan van’t Klooster for our collaboration in designing the questionnaires, andtheir suggestions for the paper. We also thank Anneloes Klaassen and Lisanne van Thiel from TNS NIPO for their assistancewith administering the surveys. Alfred Wagtendonk helped in preparing the GIS data. Elisabeth Gsottbauer, Erwann Michel-Kerjan and two anonymous referees provided useful comments on the paper. This research was carried out in the context ofthe Dutch National Research Programme ‘‘Knowledge for Climate (KvK)’’, and has been co-funded by The Netherlands Orga-nisation for Scientific Research (NWO). The usual disclaimer applies.

Appendix A. Description of the frames about risk, safety, and amenities of water

The text of the risk frame was accompanied by the photos shown in Fig. A1 and reads as follows:

‘‘Floods are part of all time periods and usually cause substantial damage. Moreover, the climate is changing, whichincreases the water in rivers and causes sea level rise. According to the experts, the Netherlands is insufficiently protectedagainst the consequences of climate change. As a result, flood risk in residential areas in unprotected floodplains canincrease in the future.’’

The text of the safety frame was accompanied by the photos shown in Fig. A2 and reads as follows:

Fig. A1. Set of photos shown in the frame about an unprotected floodplain with the emphasis on risk.

Fig. A2. Set of photos shown in the frame about an unprotected floodplain with the emphasis on safety.


‘‘In the Netherlands we have already known for a very long time that water entails risks. The first dykes were constructedmore than 1000 years ago. Moreover, the government continuously works on water safety. Recently, the new Delta Com-mission has advised how the Netherlands can also be protected against the water in the far future (until the year 2100).’’

The text of the amenities of water frame was accompanied by the photos shown in Fig. A3 and reads as follows:

Fig. A3. Set of photos shown in the frame about an unprotected floodplain with the emphasis on the amenities of water.

Table B1Variables used in the statistical analysis.

Coverage Attribute choice experiment, maximum insurance coverage of flood damage in 1000 €

Deductible Attribute choice experiment, deductible of the insurance in 1000 €

5-year duration Attribute choice experiment, dummy variable, 1 = insurance contract duration is 5 years10-year duration Attribute choice experiment, dummy variable, 1 = insurance contract duration is 10 years15-year duration Attribute choice experiment, dummy variable, 1 = insurance contract duration is 15 yearsPremium Attribute choice experiment, monthly insurance premium in €

Chronic preventionmotivation

Principal component variable (alpha = 0.70), degree of chronic prevention motivation of the respondent

Situational preventionmotivation

Principal component variable (alpha = 0.82), degree of situational prevention motivation of the respondent

Chronic promotionmotivation

Principal component variable (alpha = 0.61), degree of chronic promotion motivation of the respondent

Situational promotionmotivation

Principal component variable (alpha = 0.65), degree of situational promotion motivation of the respondent

Climate sceptic Principal component variable (alpha = 0.74), degree to which the respondent is sceptical about climate changeFrame 1 Dummy variable, 1 = communication frame of an unprotected floodplain with emphasis on riskFrame 2 Dummy variable, 1 = communication frame of an unprotected floodplain with emphasis on safetyFrame 3 Dummy variable, 1 = communication frame of an unprotected floodplain with emphasis on amenities of waterFrame 4 Dummy variable, 1 = communication frame of a protected floodplain with emphasis on riskFrame 5 Dummy variable, 1 = communication frame of a protected floodplain with emphasis on safetyDifficulty Latent variable created from a categorical variable of experienced difficulty in answering the choice experiment (1–5),

1 = very easy and 5 = very difficultUtility difference Utility difference between the best and worst alternative in the choice cardChoice set number Continuous variable, number of the choice card on which the choice certainty was statedIncome Continuous variable, yearly gross household income in thousands of euroAge Age of the respondent in yearsWomen Dummy variable, 1 = respondent is a womenEducation Latent variable created from a categorical variable of the education level (1–4), 1 = low education, 4 = high education

Table C1Panel ordered probit model results of determinants of certainty in choice.

Variable Coefficient t-Statistic

Difficulty �0.7449*** �17.61Utility difference 0.0850*** 5.27Choice set number 0.0654*** 4.77Chronic prevention motivation 0.1002** 2.45Chronic promotion motivation 0.0727* 1.81Situational prevention motivation �0.0837* �1.88Situational promotion motivation 0.1032** 2.45Climate sceptic 0.0091 0.81Age 0.0015 0.62Women 0.0053 0.07Education 0.1624*** 3.81Income 0.0002*** 2.54Frame 1 0.0664 0.46Frame 2 0.0720 0.50Frame 3 �0.2141 �1.48Frame 4 0.1845 1.32Frame 5 �0.0488 �0.34r2 0.5697*** 39.58Number of observations 4940Log likelihood Chi2 452***

Log likelihood �8864

Estimations are performed with STATA software.* Statistical significance at the 10%.** Statistical significance at the 5%.*** Statistical significance at the 1%.


‘‘Although water entails risks, it also provides opportunities. Many people experience water as a source of relaxation,space, freedom, and refreshment. Living near the water is becoming increasingly popular. Partly because of climatechange, construction companies have explored new ideas to make residential areas more beautiful and attractive. Thesewater-abundant areas in unprotected floodplains are already under construction.’’


Appendix B. Description of the variables

See Table B1.

Appendix C. Random effects ordered probit model of the determinants of stated choice certainty

There are only a few studies which have examined the sources of stated choice certainty in choice experiments (Olsen,Lundhede, Jacobsen, & Thorsen, 2011). Table C1 shows the results of a random-effects ordered probit model of the factorsof influence on choice certainty. Dummy variables of the frames are insignificant, which suggests that, as intended, respon-dents did not perceive the frames as too extreme or conflicting with their own prior beliefs in the sense that the frames cre-ated uncertainty about their preferences for flood insurance. The degree of difficulty that the respondent had in making achoice between the alternatives in the experiment has a significant and negative influence on stated choice certainty. A var-iable that reflects the utility difference between the best alternative and the worst alternative relates significantly and pos-itively to stated choice certainty,20 which confirms the Wang (1997) hypothesis that tough choices on the cards reduce choicecertainty. In line with Brouwer et al. (2010) and Olsen et al. (2011), preference certainty increases with the choice card number(1–4), which suggests that a learning effect occurs. Education and income influences stated choice certainty positively.

References

Aerts, J. C. J. H., & Botzen, W. J. W. (2011). Climate change impacts on pricing long-term flood insurance: A comprehensive study for the Netherlands. GlobalEnvironmental Change, 2(1), 1–7.

Alberini, A., Boyle, K., & Welsh, M. (2003). Analysis of contingent valuation data with multiple bids and response options allowing respondents to expressuncertainty. Journal of Environmental Economics and Management, 45, 40–62.

Ancker, J. S., Senathirajah, Y., Kukafka, R., & Starren, J. B. (2006). Design features of graphs in health risk communication: A systematic review. Journal of theAmerican Medical Informatics Association, 13(6), 608–618.

Atkinson, J. W. (1957). Motivational determinants of risk-taking behavior. Psychological Review, 64, 359–372.Beck, J. M., Rose, J. M., & Hensher, D. A. (2011). Consistently inconsistent: The role of certainty, acceptability and scale in automobile choice. Working paper

ITLS-WP-11-07. Institute of Transport and Logistics Studies, The University of Sydney.Ben-Akiva, M., & Lerman, S. R. (1985). Discrete choice analysis: Theory and application to travel demand. Cambridge: The MIT Press.Botzen, W. J. W., & van den Bergh, J. C. J. M. (2008). Insurance against climate change and flooding in the Netherlands: Present, future, and comparison with

other countries. Risk Analysis, 28(2), 413–426.Botzen, W. J. W., & van den Bergh, J. C. J. M. (2012a). Risk attitudes to low-probability climate change risks: WTP for flood insurance. Journal of Economic

Behavior and Organization, 82(1), 151–166.Botzen, W. J. W., & van den Bergh, J. C. J. M. (2012b). Monetary valuation of insurance against flood risk under climate change. International Economic Review,

53(3), 1005–1025.Brouwer, R. (2010). A mixed approach to payment certainty calibration in discrete choice welfare estimation. Applied Economics, iFirst, 1–14. http://

dx.doi.org/10.1080/00036840903035977.Brouwer, R., Dekker, T., Rolfe, J., & Windle, J. (2010). Choice certainty and consistency in repeated choice experiments. Environmental and Resource Economics,

46, 93–109.Browne, M. J., & Hoyt, R. E. (2000). The demand for flood insurance: Empirical evidence. Journal of Risk and Uncertainty, 20, 291–306.Conlisk, J. (1996). Why bounded rationality? Journal of Economic Literature, 34(2), 669–700.de Boer, J., Botzen, W. J. W., & Tersptra, T. (2013). Improving flood risk communication by focusing on prevention-focused motivation. Risk Analysis (in

press).de Boer, J., Wardekker, J. A., & van der Sluis, J. P. (2010). Frame-based guide to situated decision-making on climate change. Global Environmental Change, 20,

502–510.Dusetzina, S. B., Higashi, A. S., Dorsey, E. R., Conti, R., Huskamp, H. A., Zhu, S., et al (2012). Impact of FDA drug risk communications on health care utilization

and health behaviors: A systematic review. Medical Care, 50(6), 466–478.Edwards, A., Hood, K., Matthews, E., Russell, D., Russell, I., Barker, J., et al (2000). The effectiveness of one-to-one risk-communication interventions in health

care: A systematic review. Medical Decision Making, 20(3), 290–297.Ferrini, S., & Scarpa, R. (2007). Designs with a priori information for nonmarket valuation with choice experiments: A Monte Carlo study. Journal of

Environmental Economics and Management, 53, 342–363.Hensher, D. A. (2010). Hypothetical bias, stated choice experiments and willingness to pay. Transportation Research Part B, 44(6), 735–752.Hess, S., & Rose, J. M. (2009). Should reference alternatives in pivot design SC surveys be treated differently? Environmental and Resource Economics, 42,

297–317.Higgins, E. T. (1997). Beyond pleasure and pain. American Psychologists, 52, 1280–1300.Higgins, E. T. (1998). Promotion and prevention: Regulatory focus as a motivational principle. In M. P. Zanna (Ed.). Advances in experimental social psychology

(Vol. 30, pp. 1–46). New York, NY: Academic Press.Higgins, E. T. (2009). Regulatory fit in the goal-pursuit process. In G. B. Moskowitz & H. Grant (Eds.), The psychology of goals (pp. 505–533). New York:

Guilford Press.Hu, W., Cox, K. B. K., & Pan, M. (2009). Economic values of Dolphin excursions in Hawaii: A stated choice analysis. Marine Resource Economics, 24, 61–76.Hung, H. C. (2009). The attitude towards flood insurance purchase when respondents’ preferences are uncertain: A fuzzy approach. Journal of Risk Research,

12(2), 239–258.Kahneman, D. (2003). Maps of bounded rationality: Psychology for behavioural economics. American Economic Review, 93(5), 1449–1475.Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47, 263–291.Kellens, W., Terpstra, T., & De Maeyer, P. (2012). Perception and communication of flood risks: A systematic review of empirical research. Risk Analysis.

http://dx.doi.org/10.1111/j.1539-6924.2012.01844.x.Kleindorfer, P. R., Kunreuther, H. C., & Ou-Yang, C. (2012). Single-year and multi-year insurance policies in a competitive market. Journal of Risk and

Uncertainty, 45, 51–78.

20 This variable has been constructed in two steps. First, the choice model is estimated, and then used to estimate the utility for every alternative in a choicetask for each individual. Second, a variable was created that reflects the utility difference between the best alternative and the worst alternative.

http://refhub.elsevier.com/S0167-4870(13)00062-7/h0005














http://dx.doi.org/10.1080/00036840903035977

http://dx.doi.org/10.1080/00036840903035977


























http://dx.doi.org/10.1111/j.1539-6924.2012.01844.x




Kluger, A. N., Stephan, E., Ganzach, Y., & Hershkovitz, M. (2004). The effect of regulatory focus on the shape of probability-weighting function: Evidence froma cross-modality matching method. Organizational Behavior and Human Decision Processes, 95, 20–39.

Krantz, D. H., & Kunreuther, H. C. (2007). Goals and plans in decision making. Judgment and Decision Making, 2(3), 137–168.Kriesel, W., & Landry, C. (2004). Participation in the National Flood Insurance Program: An empirical analysis for coastal properties. Journal of Risk and

Insurance, 71(3), 405–420.Kunreuther, H. C. (1978). Disaster insurance protection: Public policy lessons. New York: Wiley & Sons.Kunreuther, H. C., & Michel-Kerjan, E. O. (2012). Demand for multi-year insurance: Experimental evidence. Working Manuscript. Center for Risk

Management and Decision Processes. The Wharton School, University of Pennsylvania.Kunreuther, H. C., Novemsky, N., & Kahneman, D. (2001). Making low probabilities useful. Journal of Risk and Uncertainty, 23(2), 103–120.Kunreuther, H. C., & Pauly, M. (2004). Neglecting disaster: Why don’t people insure against large losses? Journal of Risk and Uncertainty, 28(1), 5–21.Laury, S. K., Morgen-McInnes, M., & Swarthout, J. T. (2009). Insurance decisions for low-probability losses. Journal of Risk and Uncertainty, 39, 17–44.Li, C., & Mattsson, L. (1995). Discrete choice under preference uncertainty: An improved structural model for contingent valuation. Journal of Environmental

Economics and Management, 28, 256–269.Liberman, N., Idson, L. C., Camacho, C. J., & Higgins, E. T. (1999). Promotion and prevention choices between stability and change. Journal of Personality and

Social Psychology, 77, 1135–1145.Loomes, G., Orr, S., & Sugden, R. (2009). Taste uncertainty and status quo effects in consumer choice. Journal of Risk and Uncertainty, 39, 113–135.Lopes, L. L. (1987). Between hope and fear: The psychology of risk. In L. Berkowitz (Ed.). Advances in experimental social psychology (Vol. 20, pp. 255–295).

San Diego: CA Academic Press.Louviere, J. J., Hensher, D. A., & Swait, J. D. (2000). Stated choice methods: Analysis and application. Cambridge: Cambridge University Press.Ludy, J., & Kondolf, M. (2012). Flood risk perception in lands ‘‘protected’’ by 100-year levees. Natural Hazards, 61, 829–842.Lundhede, T. H., Olsen, S. B., Jacobsen, J. B., & Thorsen, B. J. (2009). Handling respondent uncertainty in choice experiments: Evaluating recoding approaches

against explicit modelling of uncertainty. Journal of Choice Modelling, 2(2), 118–147.McClelland, G. H., Schulze, W. D., & Coursey, D. L. (1993). Insurance for low-probability hazards: A bimodal response to unlikely events. Journal of Risk and

Uncertainty, 7(1), 95–116.McFadden, D. (2001). Economic choices. American Economic Review, 91(3), 351–378.Michel-Kerjan, E. O. (2010). Catastrophe economics: The National Flood Insurance Program. Journal of Economic Perspectives, 24(4), 165–168.Michel-Kerjan, E. O., & Kousky, C. (2010). Come rain or shine: Evidence on flood insurance purchase in Florida. Journal of Risk and Insurance, 77, 369–397.Michel-Kerjan, E. O., & Kunreuther, H. C. (2011). Redesigning flood insurance. Science, 333, 408–409.Michel-Kerjan, E. O., Lemoyne de Forges, S., & Kunreuther, H. C. (2012). Policy tenure under the U.S. National Flood Insurance Program (NFIP). Risk Analysis,

32(4), 644–658.Nocella, G., Boecker, A., Hubbard, L., & Scarpa, R. (2012). Eliciting consumer preferences for certified animal-friendly foods: Can elements of the Theory of

Planned Behavior improve choice experiment analysis? Psychology and Marketing, 29(11), 850–868.Olsen, S. B., Lundhede, T., Jacobsen, J., & Thorsen, B. (2011). Tough and easy choices: Testing the influence of utility difference on stated certainty-in-choice

in choice experiments. Environmental and Resource Economics, 49(4), 491–510.Patil, S., Burris, M., Shaw, W. D., & Concas, S. (2011). Variation in the value of travel time savings and its impact on the benefits of managed lanes.

Transportation Planning and Technology, 34(6), 547–567.Paudel, Y., Botzen, W. J. W., & Aerts, J. C. J. H. (2012). A comparative study of public-private catastrophe insurance systems: Lessons from current practices.

Geneva Papers on Risk and Insurance, 37, 257–285.Scarpa, R., & Thiene, M. (2011). Organic food choices and Protection Motivation Theory: Addressing the psychological sources of heterogeneity. Food Quality

and Preference, 22(6), 532–541.Scarpa, R., Willis, K. G., & Acutt, M. (2007). Valuing externalities from water supply: Status quo, choice complexity and individual random effects in panel

kernel logit analysis of choice experiments. Journal of Environmental Planning and Management, 50, 449–466.Schade, C., Kunreuther, H., & Koellinger, P. (2011). Protecting against low-probability disasters: The role of worry. Journal of Behavioral Decision Making.

http://dx.doi.org/10.1002/bdm.754.Scholer, A. A., Zou, X., Fujita, K., Stroessner, S. J., & Higgins, E. T. (2010). When risk seeking becomes a motivational necessity. Journal of Personality and Social

Psychology, 99(2), 215–231.Shaikh, S. L., Sun, L., & van Kooten, G. C. (2007). Treating respondent uncertainty in contingent valuation: A comparison of empirical treatments. Ecological

Economics, 62(1), 115–125.Starmer, C. (2000). Developments in non-expected utility theory: The hunt for a descriptive theory of choice under risk. Journal of Economic Literature, 38,

332–382.Terza, J. V. (1987). Estimating linear models with ordinal qualitative regressors. Journal of Econometrics, 34(3), 275–291.Train, K. E. (2003). Discrete choice models with simulation. Cambridge: Cambridge University Press.Tversky, A., & Kahneman, D. (1992). Advances in prospect theory: Cumulative representation of uncertainty. Journal of Risk and Uncertainty, 5, 297–323.van Praag, B. M. S., Frijters, P., & Ferrer-i-Carbonell, A. (2003). The anatomy of subjective well-being. Journal of Economic Behavior and Organization, 51(1),

29–49.Wang, H. (1997). Treatment of the ‘‘Don’t-Know’’ responses in contingent valuation: a random valuation model. Journal of Environmental Economics and

Management, 32, 219–232.









































http://dx.doi.org/10.1002/bdm.754














Documents

Framing of risk and preferences for annual and multi-year flood insurance