Risk Analysis, Vol. 31, No. 10, 2011 DOI: 10.1111/j.1539-6924.2011.01616.x

Emotions, Trust, and Perceived Risk: Affective andCognitive Routes to Flood Preparedness Behavior

Teun Terpstra

Despite the prognoses of the effects of global warming (e.g., rising sea levels, increasing riverdischarges), few international studies have addressed how flood preparedness should be stim-ulated among private citizens. This article aims to predict Dutch citizens flood preparednessintentions by testing a path model, including previous flood hazard experiences, trust in pub-lic flood protection, and flood risk perceptions (both affective and cognitive components).Data were collected through questionnaire surveys in two coastal communities (n = 169,n = 244) and in one river area community (n = 658). Causal relations were tested by meansof structural equation modeling (SEM). Overall, the results indicate that both cognitive andaffective mechanisms influence citizens preparedness intentions. First, a higher level of trustreduces citizens perceptions of flood likelihood, which in turn hampers their flood prepared-ness intentions (cognitive route). Second, trust also lessens the amount of dread evoked byflood risk, which in turn impedes flood preparedness intentions (affective route). Moreover,the affective route showed that levels of dread were especially influenced by citizens nega-tive and positive emotions related to their previous flood hazard experiences. Negative emo-tions most often reflected fear and powerlessness, while positive emotions most frequentlyreflected feelings of solidarity. The results are consistent with the affect heuristic and thehistorical context of Dutch flood risk management. The great challenge for flood risk man-agement is the accommodation of both cognitive and affective mechanisms in risk communi-cations, especially when most people lack an emotional basis stemming from previous floodhazard events.

KEYWORDS: Affect heuristic; emotions; experience; flood hazard; risk perception; trust

1. INTRODUCTION

Floods have devastating effects all over theglobe, both in terms of material damage(1) as wellas in lost lives.(2) Because global warming will putlow-lying, densely populated deltas and river areacommunities at even greater risk,(3) the improvementof public flood protection as well as citizens floodpreparedness will become a worldwide challenge.Global warming is one of the driving forces behindthe debate on flood risk management in the Nether-

Address correspondence to Teun Terpstra, HKV Consultant,The Netherlands; t.terpstra@hkv.nl and t.terpstra@utwente.nl.

lands, which is on the verge of shifting from a primar-ily prevention-based approach toward a broader riskmanagement approach that includes citizens disas-ter preparedness.(4) Unfortunately, however, only afew international studies have addressed the issue ofcitizens flood preparedness behavior.(5)

This article focuses exclusively on Dutch citi-zens flood preparedness intentions. Without the ex-isting river and sea flood defenses, two-thirds ofthe Netherlands would be permanently flooded. Asthese areas are inhabited by 60% of the population(about 9 million people), many citizens depend heav-ily on the flood defenses for their safety. Due to itshigh flood protection standards, the Netherlands is

1658 0272-4332/11/0100-1658$22.00/1 C 2011 Society for Risk Analysis

Affective and Cognitive Routes to Flood Preparedness Behavior 1659

currently also one of the best protected deltas in theworld.(6) Since the last severe flood occurred over55 years ago (in 1953, claiming 1,836 casualties), mostcitizens have no personal recollection of flooding. Itis often assumed that the high levels of public floodprotection and the lack of flood experiences havelowered citizens flood risk perceptions, which in turnkeeps them from preparing for potential flood dis-asters.(7) Thus far, however, these assumptions havenever been tested. This article therefore investigatescitizens flood preparedness intentions and focuseson three underlying determinants: citizens flood riskperceptions, their trust in public flood protection, andtheir previous flood hazard experiences.

2. THEORY AND EXPECTATIONS

2.1. Effects of Hazard Experiences

The personal experience of a disaster makespeople uniquely aware of their vulnerability to adisasters consequences. Disaster experience is there-fore often assumed to increase peoples risk percep-tions.(8) For instance, both Grothmann and Reuss-wig(5) and Siegrist and Gutscher(9) came to thisconclusion in regards to flood hazards. However,studies in other hazard domains have shown oppo-site effects.(10) To resolve these inconsistencies, Lin-dell and Perry(11) suggested that the effect of hazardexperience depends on how people interpret theirexperiences or what they have learned from them.Indications of how people interpret their flood ex-periences were recently presented by Siegrist andGutscher.(12) These authors found that roughly 2035% of the flood victims interviewed mentioned feel-ings of uncertainty and insecurity, fear and shock,and helplessness as the worst outcomes of their floodexperience, while among nonvictims, hardly any re-spondents anticipated these emotions as the worstoutcomes of future flooding. According to Siegristand Gutscher,(12) these negative emotions were a keyfactor in explaining why flood victims had taken sub-stantially more precautionary actions against futurefloods than nonvictims. These results are generallyconsistent with the findings of two recent experimen-tal studies. Keller, Siegrist, and Gutscher(13) aimed toincrease negative affect by presenting subjects withpictures of flooded houses. Although no manipula-tion checks were performed, these authors suggestedthat the presentation of emotion-laden images con-tributed to a higher level of perceived flood risk inthe subjects. Vastfjall, Peters, and Slovic(14) manip-

ulated negative affect associated with the 2004 EastAsia Tsunami. By reminding their Swedish subjectsof the tsunami, they indeed elicited negative affect,which in turn resulted in more pessimistic expecta-tions about the future. Interestingly, none of thesesubjects had been directly involved in the disaster,but the Swedish media covered the tsunami exten-sively, since 600 Swedish tourists were killed or lost.

The notion that negative emotions increase riskperceptions (and potentially adaptive behavior) isconsistent with research on affect and decision mak-ing(15) (the affect heuristic) and is coined therisk-as-feelings hypothesis.(16) In other words, peo-ple may interpret their hazard experiences differ-ently, depending on whether these experiences evokenegative emotionsor as Slovic, Finucane, Peters,and MacGregor(17) described negative affect, (sub-tle) feelings of badness (p. 312). If negative af-fect is experienced when making risk judgments, theperceived risk should increase. Moreover, accord-ing to the affect heuristic, positive affect (feelings ofgoodness) should decrease perceived risk. For in-stance, experiments by Finucane, Alhakami, Slovic,and Johnson(18) showed that an increased negativeaffect associated with nuclear technology resulted inhigher risk judgments, while positive affect had theopposite effect.

Unfortunately, few studies have investigatedpositive emotions associated with disasters. Rather,studies have focused on negative emotions and ad-verse health impacts. These studies have found thatthe intensity of negative emotions increases withthe severity of personal disaster consequences (e.g.,property losses, injuries, stresses in social relations).Interestingly, such negative emotions may developinto more severe health problems (e.g., posttrau-matic stress disorder), particularly when people lackthe economic, social, and psychological resourcesneeded to cope with these experiences.(19) Moreover,in the field of trauma research, increasingly morestudies have shown that stronger coping abilities andresilience against health impairments are related tothe experience of positive emotions.(20,21) Accordingto the broaden-and-build theory,(22) positive emo-tions contribute to the ability to cope with stressand negative life experiences because they stimulatethought and increase the number of perceived cop-ing behaviors, thereby adding to ones physical, in-tellectual, social, and psychological resources.(23) Re-garding disasters, Fredrickson, Tugade, Waugh, andLarkin(24) supported the link between positive emo-tions (e.g., gratitude, interest) and resilience (e.g.,

1660 Terpstra

life satisfaction, optimism) in the context of the9/11 terrorist attacks. Moreover, increases in cop-ing resources may be enduring, which makes peoplemore resilient when dealing with future events.(25)

Vazquez, Cervellon, Perez-Sales, Vidales, and Ga-borit(26) investigated positive emotions among earth-quake survivors in refugee camps in El Salvador. Inaddition to negative emotions, almost 75% of the in-terviewees recalled moments of happiness that couldbe attributed to either being alive or feeling ac-companied. Community and social activities largelycontributed to positive emotions and the ability tocope with the difficult circumstances.(27) Moreover,the majority of survivors indicated that they hadgained something positive (e.g., regarding social rela-tions and personal skills) and felt better able to copewith future events. Similar findings were reportedby Tang,(28) who investigated posttraumatic growthamong Thai survivors of the 2004 East Asia Tsunami.Regression analyses showed that the seeking and/orreceiving of social support were the most importantpredictor of positive adaptation (perceived positivechanges in, for instance, relations with others and ap-preciation of life).

In summary, the affect heuristic predicts thatpositive and negative emotions attached to naturalhazard experiences should influence risk perceptionsand, possibly, preparedness behavior. Indicators ofthe effects of negative emotions have recently beenshown by Siegrist and Gutscher.(12) Positive emo-tions have been investigated in the context of trauma(e.g., abuse, violence) and crisis (e.g., 9/11 attacks)in relation to peoples coping abilities and resiliencein dealing with future events. However, at the timeof this research we are unaware of any studies thatinvestigated the effect of positive emotions on theperceived risk of natural hazards and subsequent pre-paredness behavior. Moreover, in addition to experi-encing positive and negative emotions after disasters,people may also be left emotionally unimpressed.This article will explore whether it is useful to dis-tinguish between people who experience positive af-fect, those who experience negative affect, and thosewho are left emotionally unimpressed when evalu-ating the effects of flood hazard experiences in theNetherlands.

2.2. Trust in Public Flood Protection

Despite the high flood protection standards inthe Netherlands, no flood protection system is per-fect. Relatively recently, in 1993 and 1995, high river

discharges unexpectedly threatened several uprivercommunities that nearly escaped flood disaster. Typ-ically, citizens lack the expert knowledge to judgethe uncertainties that cause such unexpected events.Only when citizens trust the risk experts will theybe able to tolerate these uncertainties and live rel-atively unconcerned behind the flood defenses.(29)

Trust serves to reduce the complexity of a situa-tion.(30) These authors argued that, especially whenpeople lack knowledge about a hazard, their riskjudgments are based on the degree to which theytrust the responsible risk managers. That is, whentrust is high, risk perception is low, and vice versa.(31)

Trust shares conceptual similarity with affect.Affective responses are generated quickly and au-tomatically and are experienced as a feeling state,defining whether something is perceived as good orbad.(15) Thus, both trust and affect reduce the com-plexity of risk judgments because neither requires aconsideration of all of the pros and cons related tothe risks. Poortinga and Pidgeon(32) suggested thattrust and affect share similarities because they re-flect more general attitudes toward risk, which inturn drive more specific risk judgments. Althoughwe expect that trust and affect both predict morespecific attitudes toward flood preparedness, we alsoforesee that more cognitive evaluations of flood riskplay a role. First, this would be consistent with re-search showing that many decisions are informed bythe integrated effect of affective and cognitive pro-cesses.(17) Second and less obvious, however, com-pared to other types of risks, flooding risk may bea special case. For most technological risks (e.g., anuclear power plant, genetically modified food), onecannot actually see from the outside what constitutesthe risk and what has been done precisely to reducethe likelihood of a mishap. Therefore, positive feel-ings and trust in risk managers are highly importantbecause these are virtually the only means throughwhich people accept their own lack of knowledgeabout a risk. However, flood probability is reducedby the construction of flood defenses. These flood de-fenses (e.g., a dike or dam) are grossly visible in thelandscape and thus provide some information aboutthe quality of risk management. In the Netherlands,people can physically walk on dikes and dams, andcharacteristics such as their magnitude, height, andstate provide a yardstick for estimating the providedprotection level. Thus, although lay people lack theexpertise needed to calculate the actual protectionlevel provided by flood defenses, they may deriveflood likelihood based on the amount of trust that is

Affective and Cognitive Routes to Flood Preparedness Behavior 1661

inspired by their observations. Moreover, public andmedia communications often boast about the qual-ity of the Dutch flood defenses. Therefore, trust maynot only relate to affective responses to risk (e.g.,dread), but trust may also predict how people as-sess flood likelihood (a more cognitive evaluationof flooding risk). Grothmann and Reusswig(5) con-firmed this relation among citizens from the Germancity of Cologne, which is located about 175 km up-stream from the Dutch border along the river Rhine.They found that those citizens withmore faith in pub-lic flood protection revealed lower flood risk percep-tions (a conjoint measure of perceived flood proba-bility and severity) and had taken less precautionarymeasures. We too expect to find that perceptions offlood risk (especially flood likelihood) are decreasedby a high level of trust in the flood defenses, which inturn keeps citizens from preparing for potential flooddisasters.

2.3. Aims and Expectations

This article investigates the predictors of floodpreparedness among Dutch citizens. We focus specif-ically on how affect associated with previous floodhazard experiences, trust in public flood protection,and cognitive and affective evaluations of flood riskshapes flood preparedness intentions. Fig. 1 presentsthe expected relations among these variables.

First, the model reflects our view that higher riskperceptions increase citizens flood preparedness in-tentions (H1a, H1b, and H1c). Second, we expectthat higher levels of trust have a direct negative im-pact on citizens risk perceptions (H2a, H2b, andH2c) and their flood preparedness intentions (H2d).Finally, we predict that citizens affective interpre-tations of their flood hazard experiences directlyinfluence their flood risk perceptions (H3a, H3b, and

H3c), their trust in flood protection (H3d), and theirflood preparedness intentions (H3e). Specifically, wepredict that positive affect increases trust and de-creases perceived risk and preparedness intentionsdue to an optimistic bias. Negative affect results inthe opposite effect. In addition to these direct effects,we will also test for mediated effects. Our model pro-poses two mediating variablestrust in flood pro-tection and flood risk perceptions. Although directempirical evidence is scarce, Grothmann and Reuss-wig(5) provided indirect evidence by investigatingcitizens flood damage mitigation behavior throughhierarchical regression analyses. Flood experiencecontributed most to explaining the variance in pre-cautionary behavior since it was entered first in theregression analyses. In subsequent separate steps,risk perception and trust added little unique explana-tory power (36%). However, because these vari-ables did show significant, mediocre correlations withprecautionary behavior (0.210.39), the almost com-plete absence of unique effects of trust and perceivedrisk on precautionary behavior suggests partial andpossibly full mediation. We will therefore test forthe potential mediation effects of trust and perceivedrisk. The model displayed in Fig. 1 will be tested us-ing structural equation modeling (SEM). SEM is cur-rently regarded as the most sophisticated method forevaluating causal relations among multiple depen-dent variables based on cross-sectional data, due toits incorporation of both latent and observed vari-ables and the inclusion of measurement errors.(3335)

In addition, we will validate a three-way classifi-cation of emotions (negative emotions, neutral emo-tions, and positive emotions) by comparing thesethree groups to citizens who lack any flood hazardexperiences. In line with the affect heuristic, citizenswho are left emotionally unimpressed (neutral emo-tions) should reveal the same levels of perceived risk,

Preparedness intention

Trust in flood protection

Perceived likelihood

Perceived dread

Perceived consequences

H3d

H3a

H3bH3c

H2b

H2a

H2c

H2d

H3e

H1a

H1c

H1b

Feelings associated with previous experience

Fig. 1. Path model of predicted causaleffects.

1662 Terpstra

trust, and preparedness intentions as inexperiencedcitizens. Citizens with negative emotions should re-veal higher risk perceptions, less trust, and lower pre-paredness intentions, as compared to the inexperi-enced respondents, while the opposite is expected forcitizens with positive emotions.

2.4. The Present Studies

Predictions were tested in two studies. Study 1was conducted in a coastal community that recentlyexperienced a heavy storm. Study 2 was performedto replicate causal relations in two different flood riskcontextsin the impact area of the 1953 flood disas-ter at the Dutch coast and in the river area that cameclose to experiencing floods in 1993 and 1995. More-over, Study 2 aimed to gain further insight into whichemotions are evoked by previous flood hazard expe-riences.

3. STUDY 1: THE EFFECTS OF STORMEXPERIENCES

3.1. Method

3.1.1. Sample and Procedures

Data were collected in the northern end of theNetherlands at the Wadden Sea coast. Early Novem-ber 2006, about 1 month before data collection, aheavy storm (force 10 on the Beaufort-scale) causedstorm water levels near the study area that, accordingto statistics, occur 15 times every 1,000 years1 (1.5%probability per year). Despite the relatively high wa-ter levels, there was no imminent risk of flooding be-cause the existing dike protects against water levelsreached during storms that occur once in every 4,000years (0.025% probability per year). Still, this eventgainedmuch (inter)national media attention, primar-ily because a hundred horses in an unprotected floodplain needed rescue.

A random sample of 7,000 household addresses,gathered from a telephone book, received a let-ter explaining our research and inviting them toparticipate in our Internet survey.2 The letter con-tained the Internet address and a password for tak-ing the questionnaire. Data were collected between

1 From the website of the Royal Dutch Meteorological Institute,www.knmi.com, accessed on November 4, 2006.

2 In 2008, 86% of Dutch households were connected to the Inter-net, which is among the highest rates of Internet penetration inEurope (Statistics Netherlands, 2008).

December 8, 2006 and January 10, 2007. Respon-dents were thanked for participation in the invita-tion letter, on the website, and again at the end ofthe survey. In total, 826 citizens participated in thesurvey (response rate 11.8%). For the purpose ofthis study, we selected a subsample of 244 respon-dents who recalled a personal experience of stormevents in the study area (but who reported no otherflood hazard experiences). In addition, we also se-lected respondents who failed to recall any floodhazard experiences (n = 228). This latter, inexperi-enced group serves exclusively as a reference groupfor comparison with the experienced respondents. Ofthe 472 respondents (244 with experience of stormevents, and 228 without flood hazard experiences),338 (72%) respondents were males, and the meanage was 50.4 years (SD = 12.8).

3.1.2. Measurements

Emotions related to storm events were measuredby asking those who reported personal experiencesof storm events in the study area: Could you indicatethe type of feelings you experience now, when re-calling what you experienced at that time? We pro-vided five choices: very negative feelings, rathernegative feelings, neither negative nor positivefeelings, rather positive feelings, and very pos-itive feelings. All other questionnaire items werelabeled using the prefixes not at all, hardly,somewhat, quite, and very (five-point Likert-type scales). In addition, the questionnaire mea-sured trust in flood protection (four items, Cronbachsalpha = 0.90), risk perceptionsby three constructs,perceived dread (three items, Cronbachs alpha =0.94), perceived flood likelihood (one item), and per-ceived flood consequences (three items, Cronbachsalpha = 0.79)and flood preparedness intentions(three items, Cronbachs alpha = 0.86). Table Ipresents all item contents.

3.1.3. Analyses

The proposed path model of Fig. 1 was evalu-ated using SEM in AMOS 17 (SPSS Inc., Chicago,IL, USA), including both latent and observed (i.e.,indicator) variables. Prior to analysis, the data werechecked for normality. The subsequent analysis wasperformed in two steps (for another recent ap-plication of this procedure see, for instance, Pe-ters(36)). First, to ensure that the questionnaire itemsmeasured their intended constructs correctly, the

Affective and Cognitive Routes to Flood Preparedness Behavior 1663

Table I. Latent Variables, Items, and Factor Loadings in Study 1

Latent Variable Item Loading

Emotions related to stormevents

Could you indicate the type of feelings you experience now, when recalling what youexperienced at that time?

Trust in flood protection I am confident that the flood defenses along Wadden Sea coast are maintained well. 0.81I have confidence in the technological skills of flood risk managers. 0.88I have confidence in the strength of the flood defenses along Wadden Sea coast. 0.89I am confident that there are sufficient, properly qualified people working with the watermanagement authorities.

0.74

Perceived dread Could you indicate how you feel now, at this moment, when you think of the risk of flooding inyour area?

I feel frightened. 0.90I feel worried. 0.83I feel restless. 0.97

Perceived flood likelihood How likely do you find major flooding in your area within the next 10 years? Perceived flood How severely do you feel a floods consequences will affect you personally? 0.61consequences In case of flooding, how likely are you to become a victim (damage to possessions/wounded)? 0.87

How likely does it seem to you that a flood will have fatal consequences for you personally? 0.69Flood preparedness To what extent are you interested in information about flood preparedness? 0.76intentions To what extent do you intend to search for information about flood preparedness? 0.84

To what extent do you intend to prepare for flooding? 0.80

measurement model was validated by performingfirst-order confirmatory factor analysis (CFA). Basedon goodness-of-fit indexes, factor loadings, and mod-ification indexes it was decided whether the modelneeded adjustment.(33) Second, if the measurementmodel was satisfactory, AMOS was subsequentlyused to estimate the regression paths of Fig. 1, simul-taneously.

Mediation effects were evaluated followingBrown.(37) Imagine a model predicting Y1 fromX1 and the potential mediator Y2. Mediation thenrequires compliance with three conditions. Thereshould be a significant direct effect of X1 on the me-diating variable Y2 (condition 1) and a significant di-rect effect of the mediator Y2 on Y1 (condition 2).Third, when controlling for the mediating variableY2, the significant direct effect of X1 on Y1 shouldbecome less significant (partial mediation) or disap-pear (full mediation). In AMOS, condition 3 can beverified by assessing the total effect that is composedof a direct and indirect effect. In case the total effectof X1 on Y1 is significant, but the direct effect be-comes less significant or drops below significance dueto a significant indirect effect, then there is evidenceof partial or full mediation, respectively.

A problem often encountered in performingSEM is a lack of consensus on the criteria for accept-ing or rejecting a model. Based on state-of-the-artknowledge, Kline(34) recommends reporting at leastthe (1) model chi-square, (2) the Steiger-Lind root

mean square error of approximation (RMSEA) withits 90% confidence interval (90%CI), (3) the Bentlercomparative index (CFI), and (4) the standardizedroot mean square residual (SRMR). The model chi-square is often divided by the model degrees of free-dom (2/df), which is less sensitive to fluctuations insample size. This test statistic should be below 5 andpreferably below 2 in order to accept the model. ACFI value (which varies from 0 to 1) above 0.90 andpreferably above 0.95 indicates a model fit. RMSEAvalues should be as low as 0.08 and preferably 0.06 orlower in order to accept the model. Finally, SRMRvalues below 0.10 are considered as favorable.

The classification of respondents in three emo-tions groupsnegative, nonspecific, and positiveemotionswas verified by comparing their scalemeans to those of the inexperienced respondents.To this end, we performed a multivariate analysis ofvariance and post hoc tests, using SPSS 15 (SPSS Inc.,Chicago, IL, USA).

3.2. Results

3.2.1. Analysis of the Proposed Model (Respondentswith Flood Experience)

Measurement model. Because emotions asso-ciated with previous storm weather experiencescould only be measured for respondents who indi-cated such experiences, the subsequent analysis only

1664 Terpstra

Table II. Model Fit Indexes Study 1

Model Analysis Items/Constructs n 2 df 2/df RMSEA (90% CI) CFI SRMR

1 Measurement model(first-order CFA)

13/4 244 102.02 59 1.73 0.05 (0.030.07) 0.98 0.05

2 Structural model (pathanalysis)

13/4 + 2 single indicatorvariables

244 134.35 77 1.74 0.06 (0.040.07) 0.97 0.04

applies to those respondents (n = 244). Using a first-order CFA, the measurement model estimated theextent to which the observed items loaded onto theirrespective latent variables. To this end, all latent con-structs but no observed error variances were allowedto covary with one another. Constructs were scaledby fixing one factor loading to one of each of thefour latent variables, which ensured the identificationof the model. All other factor loadings were left un-constrained. The CFA included four latent constructsand 13 measured variables with accompanying er-ror termstrust (four items), dread (three items),perceived consequences (three items), and prepared-ness intention (three items). Note that emotions at-tached to previous experiences and perceived floodlikelihood were omitted from this validation proce-dure, since these variables were measured by oneitem only. Results indicated satisfactory factor load-ings on all constructs (see Table I) and a good fit (Ta-ble II, Model 1), indicating that questionnaire itemsloaded on the intended latent constructs while cross-loadings were absent.

Structural model. Next, we evaluated the pro-posed path model of Fig. 1. In addition to thefour latent variablestrust (four items), dread (three

items), perceived consequences (three items), andpreparedness intention (three items)the model in-cluded emotions attached to previous experiencesand perceived flood likelihood. The causal structurewas specified by drawing the regression paths iden-tical to Fig. 1. The model revealed a good fit (seeTable II, Model 2). The results of this procedure aredepicted in Fig. 2.

First, the analysis supported the predicted effectsof perceived dread (H1a; = 0.20, p < 0.05) andperceived likelihood (H1b; = 0.46, p < 0.001) onpreparedness intention, but rejected the predicted ef-fect of perceived consequences on preparedness in-tention (H1c; = 0.09, ns).

Second, the model supported that trust wasa significant predictor of perceived dread (H2a; = 0.29, p < 0.001) and perceived flood likeli-hood (H2b; = 0.46, p < 0.001)but the modelrejected that trust predicted the perceived conse-quences (H2c; = 0.12, ns). In addition, theAMOSoutput indicated that the total effect of trust on pre-paredness intention was significant ( = 0.18, p