Risk Analysis, Vol. 34, No. 8, 2014

DOI: 10.1111/risa.12181

You Have Been Framed! How Antecedents of Information Need Mediate the Effects of Risk Communication Messages T. Terpstra,1,∗ R. Zaalberg,2 J. de Boer,3 and W. J. W. Botzen3

This study investigates the processes that mediate the effects of framing flood risks on people’s information needs. Insight into the effects of risk frames is important for developing balanced risk communication that explains both risks and benefits of living near water. The research was inspired by the risk information seeking and processing model and related models. In a web-based survey, respondents (n = 1,457) were randomly assigned to one of three communication frames or a control frame (experimental conditions). Each frame identically explained flood risk and additionally refined the message by emphasizing climate change, the quality of flood risk management, or the amenities of living near water. We tested the extent to which risk perceptions, trust, and affective responses mediate the framing effects on information need. As expected, the frames on average resulted in higher information need than the control frame. Attempts to lower fear appeal by stressing safety or amenities instead of climate change were marginally successful, a phenomenon that is known as a “negativity bias.” Framing effects were mediated by negative attributes (risk perception and negative affect) but not by positive attributes (trust and positive affect). This finding calls for theoretical refinement. Practically, communication messages will be more effective when they stimulate risk perceptions and evoke negative affect. However, arousal of fear may have unwanted side effects. For instance, fear arousal could lead to lower levels of trust in risk management among citizens. Regular monitoring of citizens’ attitudes is important to prevent extreme levels of distrust or cynicism. KEY WORDS: Affect heuristic; flood risk; framing; information seeking; mediation analysis; risk communication; risk perception; trust

1. INTRODUCTION

docked in or near the city center. When ship size gradually increased during the 20th century, larger ports were built on new locations and the older traditional docks became abandoned. The open space and uninterrupted vistas of the river is a key feature of these areas, which makes them potentially attractive locations for living, working, and leisure. Although these old docks are often elevated, they are not protected by dikes. Thus, their location outside the embanked area also makes them potentially vulnerable to floods. Moreover, the frequency of inundations may increase due to climate change. When designing these areas, all kinds of measures can be taken to minimize the risk of damage and casualties, such as elevating buildings and main

Many of the world’s ancient cities have been founded at the mouth of major rivers because in those times rivers were (and still are) vital for transport and trade. In Europe, the ports of Hamburg (Germany), Antwerp (Belgium), and Rotterdam (the Netherlands) have flourished and now rank among the world’s major ports. Originally, ships 1 HKV

Consultants, 8203 AC Lelystad, the Netherlands. UR, 6708 PB Wageningen, the Netherlands. 3 Institute for Environmental Studies, VU University Amsterdam, the Netherlands. ∗ Address correspondence to T. Terpstra, P.O. Box 2120, 8203 AC Lelystad, the Netherlands; [email protected]. 2 Wageningen

1506

C 2014 Society for Risk Analysis 0272-4332/14/0100-1506$22.00/1 

Framing Flood Risk roads. Nevertheless, it is important that future residents are aware of the potential risk of floods, which may be regarded as unwelcome but not necessarily life-threatening events. Local governments face the challenging task of developing balanced risk communication that presents these areas as attractive living locations but at the same time does not neglect the flood risks. This is important because risk communication that explains or emphasizes the risks could lead to fearful responses among inhabitants, inhibiting sell rates. Conversely, communication that stresses high quality of flood risk management or positive sides of living near water (e.g., a nice view) could lead to lowered levels of risk awareness among potential buyers, causing them to perceive the authorities as responsible for their protection and for covering the damage in case of flooding. A recent literature review(1) indicated that hardly any studies tested the effects of risk communication on citizens’ perceptions of and their behavioral responses to flood risks. This article investigates the responses of citizens to various forms of risk communication in the light of the risk information seeking and processing (RISP)(2) model and related models using an experimental research design. More specifically, we study the extent to which flood risk framing influences citizens’ need for additional information about flood risks. All the communication frames (i.e., experimental conditions) contained realistic information explaining the flood risks, but varied from one another with regard to emphasis on either climate change effects (frame 1), high-quality flood risk management (frame 2), or positive aspects of living near water (frame 3). Through the experimental research design, we test the cognitive and affective processes that underlie (i.e., mediate) the effects of risk communication on information need, as proposed by RISP. The next section delivers the theoretical background. 2. THEORETICAL BACKGROUND Communication about risks confronts people with things, circumstances, or forces that pose a threat to people or what they value.(3) Risk communication stimulates changing thoughts about risks. While perceiving a threat is generally recognized as a cognitive process, emotions such as fear are considered necessary to fully understand human responses to risks. These cognitive and emotional processes together motivate adaptive coping behavior.(4–6) Adaptive coping behavior is a strategy to deal with a per-

1507 ceived threat, such as placing the fuse box on a higher floor or learning evacuation routes. 2.1. Information Need In the process of adaptive coping, information seeking is often an important first step. In case of a perceived threat, people who perceive to have insufficient information may seek information to clarify a perceived threat’s significance and to select an appropriate action.(7) In the RISP model, information insufficiency is seen as a motivational factor that drives people to overcome a perceived gap between the information held and the information needed. When a person’s current knowledge (information held) is below a certain sufficiency threshold (information needed), s/he is expected to perceive information insufficiency, which in turn drives him/her to close the gap by seeking additional information. Thus, perceiving insufficiency implies an information need, which in turn drives a person to seek information.(2,8,9) Therefore, stimulating citizens’ need for risk-related information could be an important step in risk communication campaigns that aim to increase risk awareness, information seeking, and, ultimately, adaptive behavior. So far, research that tests the effects of risk communication messages on information need seems to be lacking. That is, previous surveys have at least partially supported RISP’s hypotheses that information need is correlated with constructs such as risk perceptions, affective responses, and (inversely) with trust in risk management. But none of these studies actually investigated the extent to which these constructs mediate the effects of risk communication messages on information need. Therefore, the relative contribution of underlying cognitive and affective processes is still unclear. Investigating such cause and effect relations requires an experimental approach. This will be the method of this article. The next sections provide the theoretical background for our research model (see Fig. 1). We first discuss the empirical relations between information need (the dependent variable) and its proposed antecedents (i.e., risk perceptions, affective responses, and trust) based on nonexperimental (i.e., cross-sectional) survey data and existing theoretical frameworks. Subsequently, we will discuss the empirical evidence regarding the effects of risk communication messages on risk perceptions, affective responses, and trust in the domain of environmental hazards, based on experimental research designs.

1508

Terpstra et al. Unmediated model

Contrast coded communication frames

c

Information need

Mediated model Contrast coded communication frames

a

c’

Mediators 1. Risk perception 2. Trust 3. Negative affect 4. Positive affect

Information need

b

Fig. 1. Research model.

2.2. Predicting Information Need from Risk Perception, Affective Responses, and Trust: Results from Cross-Sectional Surveys Griffin et al.’s(2) introduction of the RISP model has attracted attention from researchers in various domains. Consistent with ideas about the interplay between affect and cognition, the RISP model incorporates affective and cognitive factors in order to predict people’s information needs. According to RISP, people’s information needs increase with the intensity of their affective responses, such as fear or anger. In turn, people’s affective responses are expected to increase when they perceive higher risks and have lower trust in risk management institutions. Several studies have applied (variations of) RISP to global warming,(10) industrial risks,(11,12) and flood risk.(13,14) Although these studies were all inspired by RISP, they sometimes differed with respect to the conceptualization of predictors and their relations with information need. For instance, Kellens et al.(14) tested the effects of risk perception (i.e., perceived likelihood and impact), trust, and affective responses (i.e., worry) on information needs related to flood risks at the Belgian coast. Instead of viewing risk perception, trust, and affect as unique predictors of information need, these authors incorporated these variables in a single risk perception scale in order to predict information need. As predicted, a path analysis showed that higher levels of their perceived risk indicator contributed to higher information need. Also Ter Huurne and Gutteling(12) constructed a composite risk perception scale comprising perceived danger, perceived severity of consequences, and affective responses. In contrast to Kellens et al.,(14) the

composite scale did not surface as a significant predictor of information need in a regression analysis. Similarly, Griffin et al.(13) tested the overall effects on information need of various predictors, including risk perception, trust, affective responses (i.e., anger), attribution of harm, and personal efficacy. Although the overall effect of these predictors on information need was statistically significant, from their regression analysis it remained unclear how these variables contributed individually. In contrast, Kahlor(10) and Ter Huurne et al.(11) reported effects of individual variables. These authors performed path analysis and found that—as proposed by RISP—higher levels of perceived risk predicted stronger affective responses (i.e., worry), which in turn contributed to higher information need. Kahlor(10) also noted that the affective response mediated the effect of perceived risk on information need. In addition, results of Ter Huurne et al.(11) indicated that lower levels of trust contributed to a stronger affective response. Since mediation analysis was not performed, it remained unclear to what extent risk perception and trust indirectly influenced information need. Nevertheless, correlations reported by Ter Huurne and Gutteling(15) supported the notion that higher trust is related to lower levels of information need. Finally, beyond the domain of environmental hazards Kuttschreuter(16) studied people’s information needs in response to food risk messages. Correlations indicated that higher risk perceptions, lower trust in food safety, and higher affective responses were related to higher information need. 2.3. Effects of Communication Messages on the Antecedents of Information Need: Results from Experimental Research Designs In this section, we provide an overview of the effects of persuasive risk communication messages on the proposed antecedents of information need. In the context of environmental hazards, people’s cognitive responses (i.e., risk perceptions) to persuasive communication have been studied in relation to seismic risk,(17) tornado risk,(18) and flood risk.(19) These experimental studies confirmed that fear appeal messages (i.e., the deliberate arousal of fearful feelings) enhanced cognitive appraisal of the risk (e.g., perceptions about damage), which in turn contributed (along with other variables) to higher coping intentions and more behavior change. In particular, the experiment conducted by Kievik and Gutteling(19) showed that subjects who were exposed to high

Framing Flood Risk (compared to low) fear appeal messages were more likely to search for information about flood risks, which suggests a higher information need. Finucane et al.(20) studied the mediating role of affective responses in judging risks and benefits of nuclear technology. When communication messages portrayed the risks as high (or benefits as low), the subsequent experience of negative affect caused subjects to perceive benefits of nuclear technology as low (or risks as high). Conversely, when benefits were portrayed as high (or risk as low), the subsequent experience of positive affect caused subjects to perceive risks of nuclear technology as low (or benefits as high). Differences in emotional states are also associated with differences in how people process information about risks. Negative emotions such as fear are associated with effortful, detail-oriented cognitive processing of information (systematic processing) while positive emotions such as happiness or joy are associated with heuristic processing of information.(21–23) Thus, presenting unprotected former harbor areas by emphasizing either the (flood) risks or the benefits (amenities) likely influences how people perceive these areas as attractive living locations. Emphasizing the flood risks evokes negative emotions, due to which people devote greater attention to arguments and require more information to make judgments. Conversely, emphasizing the benefits may evoke positive emotions, which distracts attention from the threat and results in a lower need for detailed information. Finally, trust in governmental institutions has been named an important determinant of people’s information need.(2,15) Typically, citizens lack the expert knowledge to judge risks. Trusting in the expertise of public institutions to control risks helps people to tolerate risks in their environment. Trust is therefore negatively related to risk perception.(24) Risk communication that contains reassuring information is also expected to cause lower interest in detailed information about the risk.(25) Gleicher and Petty(26) studied this principle in a campus crime experiment by manipulating fear, reassurance, and argument quality. Results indicated that when fear was moderate (compared to low) and the expert’s opinion advocated that the suggested crime watch program would be effective, subjects relied on weak arguments about the crime watch program (i.e., they felt reassured by the expert). However, when the expert questioned the efficacy of the crime watch program, subjects required strong arguments in order to hold favorable attitudes towards the crime

1509 watch program. Thus, an expert’s opinion advocating the effectiveness of collective flood control measures (e.g., dikes) could reassure people that governmental institutions have taken adequate measures to control the risk. When such risk communication messages increase institutional trust, people’s need for detailed information about flood risks is likely to decrease. In sum, empirical studies indicate that perceived risk, affective responses, and trust are correlated with information need. In addition, persuasive risk communication messages may be used to influence these antecedents of information need. However, questions remain about the temporal ordering of these variables and the extent to which they contribute uniquely to information need. Although RISP suggests that perceived risk and trust predict affective responses, other studies suggest otherwise. For instance, Ruiter et al.(4) provide an extensive review of the literature on the cognitive and emotional responses to risk information. These authors proposed that attention to risk information arouses fear before or in parallel to threat perception. Moreover, fear and threat perception may independently motivate coping behavior. In this study, we will not make explicit assumptions about the temporal ordering of risk perception, trust, and affective responses. Rather, we focus on the extent to which risk perception, trust, and affective responses mediate the effects of risk communication on information need. In doing so, we treat each of these variables as direct predictors of information need.

3. RESEARCH DESIGN AND HYPOTHESES 3.1. Research Design The very high degree of Internet penetration in the Netherlands (more than 90% of the population) enabled a survey among citizens with Internet access. The survey was performed in the city of Rotterdam and surrounding towns, located in the lower (downstream) area of the Meuse and Rhine rivers. The area is protected against floods from the rivers and the sea by a system of sizeable flood defenses, consisting of dikes, dams, and a storm surge barrier. Manipulations were performed in two steps. First, all participants were introduced to the topic of the study and they received general, textual information describing the housing situation, and the level of risk. That is, they were asked to imagine that

1510 they would live outside the embanked area, which reflected a centrally located, former harbor area not protected by any flood defenses. The level of risk was elucidated by explaining flood likelihood (i.e., “Such high water levels occur on average once in 10 years”) and flood consequences (i.e., “Streets can be covered with water during high water levels”). In the second step the message was refined using realistic textual and pictorial information that emphasized certain aspects of flood risk because both have strong effects on people’s affective responses and judgments of risk.(27) Participants were randomly assigned to one of three frames that explained either the increasing risks due to climate change (frame 1, “climate change”), the high quality of flood risk management in the Netherlands (frame 2, “safety”), or the amenities of living near water such as leisure opportunities and nice views outside the embanked area (frame 3, “amenities”). Participants in the control frame received no information, but instead answered the same questions in relation to their current housing situation in the river area. Appendix A presents an overview of the texts and pictures that were used in the two-step manipulation procedure. 3.2. Research Model and Hypotheses Fig. 1 presents the research model. The unmediated model evaluates the effects of different communication frames on participants’ information needs. The mediated model evaluates the extent to which the hypothesized differences in information need between communication frames are mediated by risk perception, trust in flood risk management, and negative and positive affective responses. H1 contrasts risk communication frames against the control frame. We expect that: H1a: The risk communication frames cause higher information need compared to the control frame. H1b: The H1a contrast is mediated by higher levels of perceived risk and negative affect, as well as lower levels of trust and positive affect. The rationale behind H1 is that all three communication frames identically describe a housing location in a centrally located former harbor area outside the embanked area with related flood risk (step 1 of the manipulation procedure). Thus, participants in all three frames are exposed to fear appeal, independent of the information that refines this message

Terpstra et al. in the second manipulation step. The inherent fear appeal in each of the risk communication frames will contribute to higher risk perception and negative affect, but at the same time to lower trust and positive affect. Therefore, the three frames on average will cause higher information need compared to the control frame, which received no information at all. H2 and H3 contrast different communication frames in order to test the effects of the second manipulation step that refined the message presented in the first manipulation step. We expect that: H2a: The flood safety frame causes lower information need compared to the climate change frame. H2b: The H2a contrast is mediated by lower levels of perceived risk and negative affect, as well as higher levels of trust and positive affect. H3a: The amenities frame causes lower information need compared to the climate change frame. H3b: The H3a contrast is mediated by lower levels of perceived risk and negative affect, as well as higher levels of trust and positive affect. The rationale behind H2 and H3 is that emphasizing consequences of climate change on top of the description of flood risk (cf. manipulation step 1) will aggravate fear appeal. In contrast, information that emphasizes high-quality flood risk management will reassure people and information about amenities will distract their attention from the threat. Therefore, compared to the threatening information about climate change, the reassuring information about flood safety or distracting information about amenities will both lead to lower need for detailed information about the flood risks. The lower levels of information need will be mediated by lower levels of risk perception and negative affect, as well as higher levels of trust and positive affect. 4. METHOD 4.1. Sample Collection Procedures In June 2011, a region-wide sample was drawn from a large panel4 of citizens who are willing to participate in web-based research for a small reward, 4 Sample

collection was contracted out to TNS-NIPO (Dutch Institute for Public Opinion and Market Research), which is a large market research agency in the Netherlands.

Framing Flood Risk which they can keep for themselves or donate to charity. The sample collection procedure was set up to collect at least 400 responses per communication frame and 200 responses for the control frame. In total, 2,302 citizens (response rate 69%) participated in the study. For the purpose of this study, we selected a subset of 1,457 of the 2,302 responses.5 More precisely, 423, 415, 414, and 205 respondents participated in the climate frame, safety frame, amenities frame, and in the control frame, respectively. The four frames were similar with respect to the distribution of gender (χ32 = 3.24, ns), mean age (F3,1453 = 2 0.72, ns), education (χ18 = 19.08, ns), and income 2 = 74.69, ns). Overall, 51% were males, mean (χ78 age was 50.1 years (SD = 13.2), median education was “MBO” (i.e., a medium level of vocational education), and median income was between €38,800 and €51,300 per year, before taxes. Eleven respondents failed to report their education and 367 respondents failed to report their income. 4.2. Measurements and Scale Construction The questionnaire measured risk perception, trust in flood protection, positive and negative affective responses, and information need.6 Each construct was measured using multiple items on seven-point Likert-type scales. We allowed small differences in item wording between frames so that 5 The

complete survey consisted of five frames (experimental conditions) and a control frame. The frames 1–3 contained stories about living in a redeveloped harbor area that, although they are often elevated, are not protected by dikes against river flooding. These frames highlighted climate change (1), safety (2), and amenities (3). The frames 4 and 5 also highlighted climate change (4) and safety (5), but their story line was applied to living in a deep polder near the river protected by dikes. 6 There are various ways to measure and analyze information need/insufficiency. Griffin et al.(13) proposes to measure information insufficiency by measuring the so-called sufficiency threshold and levels of current knowledge, both on a scale from 0 to 100. To obtain a measure of information insufficiency, the sufficiency threshold is controlled for levels of current knowledge in a regression analysis. Ter Huurne and Gutteling(12) followed the same analysis procedure but developed two multi-item scales to measure the sufficiency threshold and current knowledge. Since the results of the latter study showed that current knowledge was not significantly correlated with the sufficiency threshold, the authors suggested that future studies develop “a single sufficiency measure, rather than differentiating the amount of knowledge held and that needed” (p. 858). Since information need is the dependent variable in this study, we prefer to employ a multi-item measure. Following Ter Huurne and Gutteling´s suggestion, we did not include the respondents’ perception of their current knowledge.

1511 items would better fit with the framed situations and respondents’ understanding of the frames would be enhanced. The questionnaire was pretested by professional interviewers using eight in-depth faceto-face interviews. A second quantitative pilot included 102 completed surveys. The purpose of these pilots was to check whether respondents understood the survey questions, the framing texts, how they perceived the pictures, and to trial the survey online. To validate the constructs, we performed factor analysis (principal axis factor analysis) with a Promax rotation. The Promax rotation allows for nonzero correlations between extracted factors, which matches the idea behind our theoretical model that constructs are correlated. The analysis was first performed on a random 50% selection of the sample, and then repeated on the remaining 50% (split-half method). As shown in Table I, results validated the intended factor structure in both subsamples. Internal consistency measured by Cronbach α was satisfactory for all scales. 4.3. Analysis To evaluate the effects of the different communication frames on the scales we perform a series of multiple mediation analyses. First, we contrast code the frames according to the expected differences in information need under H1a, H2a, and H3a. For instance, X1 contrasts the risk communication frames against the control frame. Table II presents these contrasts. The multiple mediation model indicates how well the entire bundle of mediators explains the contrast effects on information need, thereby accounting for intercorrelation among mediators. For instance, previous studies(28) have shown that higher risk perceptions are correlated with higher negative affect, and lower trust. The intercorrelations between variables are presented in Appendix B. As expected, all mediators are intercorrelated in the expected directions; that is, higher levels of risk perception and negative affect are correlated with lower levels of trust and positive affect. In addition, higher information need is correlated with higher risk perceptions and negative affect, and lower trust and positive affect. As shown in Fig. 1, full mediation implies that a significant c-path in the unmediated model decreases to a nonsignificant c -path in the mediated model due to significant indirect effects through multiple

1512

Terpstra et al. Table I. Factor Loadings and Reliabilities Component 1

Information need. Suppose you would consider moving to a neighborhood outside the embanked area in the future. To what extent would you need information about How serious flood damage could be? The measures you could take to prevent flood damage to your possessions? The possibilities for evacuation in case of a flood threat? Who is responsible for flood damage to your possessions? The chance of flooding in the years ahead? Trust in flood management. To what extent do you trust that authorities such as municipalities and water boards will be able to Give timely warnings to citizens in neighborhoods outside the embanked area, so they can move their car or take other protective measures? Make accurate predictions about flood levels in neighborhoods outside the embanked area? Manage the flood risks in neighborhoods outside the embanked area in the next 20 years? Design and plan neighborhoods outside the embanked area so that there will little are no risk of flood damage to homes and possessions of dwellers? Ensure that it will be safe for you (and your family) to live in a neighborhood outside the embanked area? Positive affect. The idea of living in a neighborhood outside the embanked area Gives me a happy feeling. Gives me a good feeling. Gives me a cheerful feeling. Gives me a pleasant feeling. Negative affect. The idea of living in a neighborhood outside the embanked area Gives me a restless feeling. Gives me an unsafe feeling. Gives me a worried feeling. Gives me an anxious feeling. Climate change risk perceptions Due to climate change and flood risks housing prices in neighborhoods outside the embanked area will decrease in the future. Due to climate change flood frequency and flood levels in city harbors outside the embanked area will increase. Rotation sum of squared loadings Cronbach α

2

3

4

5

0.907 (0.877) 0.902 (0.886) 0.902 (0.898) 0.890 (0.850) 0.855 (0.803)

0.856 (0.870)

0.849 (0.841) 0.848 (0.843) 0.842 (0.877)

0.841 (0.858)

0.960 (0.945) 0.929 (0.910) 0.922 (0.933) 0.911 (0.924)

0.943 (0.920) 0.895 (0.876) 0.891 (0.864) 0.880 (0.889) 0.756 (0.706)

0.610 (0.703) 4.90 (4.61) 0.95 (0.94)

5.20 (5.41) 0.93 (0.93)

5.06 (5.55) 0.97 (0.96)

5.54 (5.81) 0.95 (0.94)

2.74 (2.95) 0.61 (0.66)

Note: Principal Principle axis factor analysis with Promax rotation and Kaiser normalization. Analysis performed on a random 50% selection (n = 729) and repeated on the remaining 50% of the sample (n = 728). Results of the repeated analysis in parentheses.

mediators. We speak of partial mediation when a significant c-path in the unmediated model becomes a less but still significant c -path in the mediated model, due to significant indirect effects. In general, a potential decrease in the unstandardized regression coefficients from the c-path to the c -path equals the sum

of the four indirect effects. Significant individual mediators or indirect effects are identified by their significant “ab” products. The unstandardized regression coefficients of the unmediated and mediated models are estimated using a bootstrapping procedure developed

3.78* −10.27** 16.44** −7.07**

0.37 0.07 0.17 0.04

b-path t

11.00**

0.23 −0.73 1.34 −0.57

a-path t

0.78 14.51** 1.02

t

c -path

Risk Perception Trust Negative Affect Positive Affect Total

method is an alternative to the widely used Sobel test for testing mediation, as advocated by Baron and Kenny.(39) Bootstrapping is a nonparametric resampling procedure in which indirect effects are repeatedly estimated (e.g., 5,000 times) to create a nonnormal distribution of the indirect effect estimates. This distribution is then used to construct asymmetric confidence intervals around the point estimates of the indirect effects. The bootstrapping procedure was performed using the SPSS-macro “INDIRECT,” which can be found at http://afhayes.com/spss-sasand-mplus-macros-and-code.html.

Total Effect Direct Effect Mediated Effects

7 The

c-path

Table III shows that the unstandardized regression coefficient B (size 1.02) of the c-path is significant at p < 0.001. These results support H1a that compared to no risk communication, communication about flood risks causes higher information need independent of whether communication emphasizes climate change, flood safety, or amenities. In addition, the mediated model indicates that the unstandardized regression coefficient B (size 0.78) of the c -path is significant (p < 0.001). As noted earlier, the difference between the unstandardized regression coefficients of the c and c paths is equal to the total mediating effect. This mediating effect with size 0.24 is significant because zero is not included in the 99%CI (0.13–0.36). Table III also shows that perceived risk and negative affect act as significant

Unmediated Model

5.1. Risk Communication Versus Control

Mediated Model

Results are presented for the three hypotheses. For the evaluation of these hypotheses, we are interested in the significance of the unstandardized regression coefficients related to the c-paths, c paths, and mediating effects (“ab” products).

Table III. Risk Communication Versus Control

t

12.77** 2.55 6.92** 1.59

5. RESULTS

Note: Coefficients are unstandardized regression coefficients. CI = Bias corrected and accelerated 99% confidence interval. Number of bootstrap resamples: 5,000. R2 = 0.27. *p < 0.01, **p < 0.001.

ab-path

SE

by Preacher and Hayes.(29)7 To signal the significance of indirect effects, we report on the 99% biascorrected and accelerated confidence intervals (CIs). That is, mediators are considered statistically significant when zero is not contained in the 99%CI.

Mediation

0 −1 0 1 H3a

0.16 0.00 0.34 0.02 0.36

0 −1 1 0 H2a

0.03 −0.11 0.13 −0.07 0.13

−1 1/3 1/3 1/3 H1a

0.03 0.02 0.04 0.02 0.04

205 423 414 415

0.09 −0.05 0.23 −0.02 0.24

X3

Upper

X2

Lower

X1

(Bootstrap)

Control Climate Change Safety Amenities Related Hypothesis

N

CI of ab-path

Communication Frames

Mediators

Table II. Contrasts Comparing Communication Frames

Mediation

1513 Mediation

Framing Flood Risk

−0.04 0.00 −0.02 0.00 −0.06 11.73** 1.14 10.66** 2.12

t

−3.04* 1.30 −1.71 1.35

0.36 0.03 0.26 0.06

b-path t

0.36

−0.11 0.06 −0.09 0.07

a-path t

0.01 –0.97 Risk Perception Trust Negative Affect Positive Affect Total

–0.04 Total Effect Direct Effect Mediated Effects

Note: Coefficients are unstandardized regression coefficients. CI = Bias corrected and accelerated 99% confidence interval. Number of bootstrap resamples: 5,000. R2 = 0.21. *p < 0.01, **p < 0.001.

−0.01 0.01 0.01 0.02 −0.01 0.01 0.00 0.01 0.00 0.02

−0.08 0.00 −0.06 0.00 −0.11

Lower SE (Bootstrap)

ab-path Mediated Model

c -path t c-path

Table V shows that information need is slightly lower when risk communication emphasizes amenities compared to climate change. Although this difference is in the expected direction (H3a), it is nonsignificant. Again, the refining information provided in manipulation step 2 did not affect participants’ information need. To test H3b, we examine the sum of all the indirect effects in the mediation model and expect negative “ab” products for individual mediators. Some evidence of mediation is found. The sum of all indirect effects is significant with size –0.08 and a 99%CI excluding zero (–0.14 to –0.02). Perceived risk and negative affect act as significant individual

Unmediated Model

5.3. Climate Change Versus Amenities

Table IV. Climate Change Versus Safety

The data show that information need is slightly lower when risk communication emphasizes flood safety compared to climate change (Table IV). Although this difference is in the expected direction (H2a), it is nonsignificant. Thus, the c-path fails to support that the refining information about flood safety provided in manipulation step 2 affected participants’ information need. However, it is still important to evaluate the indirect effects. That is, a nonsignificant c-path does not necessarily imply the absence of mediation, especially when the mediation process consists of multiple processes. The explanation is that different processes may have opposite (i.e., positive and negative) signs. In such a case, individual processes cancel each other (partly) out, resulting in a less or nonsignificant c-path.(30) To test H2b, we examine the sum of all the indirect effects in the mediation model and expect negative “ab” products for individual mediators. That is, exposure to the safety frame is expected to result in a lower fear appeal effect compared to the climate change frame. Some evidence of mediation is found. The sum of all indirect effects is significant with size -–0.06 and a 99%CI excluding zero (–0.11 to –0.01). Results also show that perceived risk acts as a significant individual mediator. Since the “ab” products of trust, negative affect, and positive affect are all nonsignificant, these findings only partially support H2b.

CI of ab-path

5.2. Climate Change Versus Safety

Upper

Mediators

individual mediators because their “ab” products are significant. Since the “ab” products of trust and positive affect are nonsignificant, these findings partially support H1b.

Mediation

Terpstra et al. Mediation

1514

0.36 0.03 0.26 0.05 −2.86* 1.53 −3.96** 5.28** −0.11 0.07 −0.21 0.26 0.73 0.03 −1.04 Risk Perception Trust Negative Affect Positive Affect Total

–0.05 Total Effect Direct Effect Mediated Effects

Note: Coefficients are unstandardized regression coefficients. CI = Bias corrected and accelerated 99% confidence interval. Number of bootstrap resamples: 5,000. R2 = 0.21. *p < 0.01, **p < 0.001.

Mediation

−0.01 0.02 −0.02 0.04 −0.02 −0.08 0.00 −0.10 −0.01 −0.14 0.01 0.00 0.02 0.01 0.02 −0.04 0.00 −0.06 0.01 −0.08 11.76** 1.18 10.68** 2.03

t b-path t a-path t c -path t c-path

Unmediated Model

Mediation

Mediators Upper Lower (Bootstrap)

ab-path Mediated Model

Table V. Climate Change Versus Amenities

Mediation

1515

SE

CI of ab-path

Framing Flood Risk

mediators. Since the “ab” products of trust and positive affect are nonsignificant, these findings only partially support H3b. 6. DISCUSSION This study looked at the effects of risk framing on people’s need for information about flood risk. Previous studies that applied RISP and related theoretical models supported that information need is correlated with constructs such as risk perceptions, affective responses, and (inversely) with trust in risk management. However, none of these studies actually investigated the extent to which these constructs mediate the effects of risk communication messages on information need. Respondents received one of three communication messages with information about former harbor areas not protected by river levees. The risk messages identically explained flood probability and flood level, but differed in additional information that refined the message by emphasizing either the increasing risk due to climate change (“climate change”), the high quality of flood risk management in the Netherlands (“safety”), or the amenities of living near water in former harbor areas such as leisure opportunities and nice views (“amenities”). Respondents were questioned about the extent they would need information about flood risk in case they would consider moving to a former harbor area. Results indicated that the risk communication frames on average resulted in higher levels of information need. However, communication messages that highlighted positive sides of living in a former harbor area (i.e., safety or amenities) only resulted in marginally lower levels of information need compared to messages that highlighted the negative sides (i.e., increasing flood risk due to climate change). Apparently, the manipulations of safety and amenities frames failed to reduce the fear appeal effect that was established in the first manipulation step. This was also supported by the data. The presence or absence of manipulation effects can be evaluated by inspection of a-paths. The a-paths in Table III support that the frames on average induced quite some fear appeal, which was reflected in all mediators. That is, compared to the control frame, levels of perceived risk and negative affect increased while levels of trust and positive affect decreased. In addition, the a-paths of the safety frame (Table IV) and amenities frame (Table V) were relatively small, indicating that the effects of these frames on the

1516 mediators was only modest. The safety frame only resulted in a marginally lower risk perception, compared to the climate frame. The amenities frame was somewhat more effective and resulted in lower levels of perceived risk and negative affect, and higher levels of positive affect. It is not entirely clear why the safety and amenities frames lacked sufficient power to reduce the fear appeal effect. Although the a-paths provide information about whether manipulations were successful or not, the a-paths do not explain why this is the case. Although the questionnaire materials were qualitatively pretested in eight in-depth face-to-face interviews, we did not perform quantitative manipulation checks on the texts and pictures in the questionnaire (e.g., “To what extent do you find these pictures reassuring?”). Therefore, it remains unclear to what extent the different characteristics of the provided texts and pictures exactly contributed to the differences in fear appeal between frames. For instance, Zaalberg and Midden(31) compared 3D flood simulation with effects of film and slides, and found that the immersive quality of risk information presentation enhanced the extent to which participants searched for coping-related information. There is a challenge for future studies to disentangle the effects of various information formats and their underlying mechanisms. Despite the absence of manipulation checks, results can be viewed in the context of a phenomenon known as the “negativity bias.”(32) According to the negativity bias hypothesis, good news has less impact than bad news.(33) Similarly, negative attributes in communication messages that suggest the presence of risk have greater impact on risk perceptions than positive attributes that suggest the absence of risk.(34,35) So possibly the high level of fear appeal, once it was aroused in the first manipulation step, could not be reduced easily by subsequent positive elements in the message. The negativity bias hypothesis does not explain why framing effects on information need were mediated only by risk perception and negative affect, and not by trust and positive affect. The absence of mediation via trust and positive affect cannot be attributed to potential weak manipulation effects because the frames on average resulted in substantially lower trust and positive affect compared to the control frame (Table III, significant a-paths). It seems that trust and positive affect failed to mediate the effects of the frames on information need because

Terpstra et al. none of their b-paths were significant. Especially the finding that trust did not act as a mediator contradicts with RISP because institutional trust is explicitly incorporated in RISP as part of perceived hazard characteristics.(2) So far, several RISP studies incorporated trust as an (indirect) predictor of information need.(11–16) Only two of these studies(15,16) reported that higher trust is related to lower levels of information need. Since there are various forms of trust,(36) the trust measures in these two studies captured various dimensions such as perceived trust in the expertise, openness, and credibility of organizations, which were combined into a single trust scale. Our study only measured perceived trust in the expertise of flood risk management authorities. So possibly other dimensions than perceived expertise have greater influence on people’s information needs. There are three challenges for further study. First, the role of different trust dimensions within RISP should be better defined. Despite the absence of mediation by trust, the lowered level of trust in expertise of risk management that was caused by the risk communication messages is still an important finding that should not be overlooked. The fear appeal effect on trust is consistent with previous findings that higher levels of negative affect and risk perception are correlated with lower levels of trust.(11,12,28) Lack of trust in expertise could be damaging for the risk management process. According to Poortinga and Pidgeon,(36) the trust dimensions can be defined along two axes—that is, general trust or reliance on an institution’s expertise and skepticism regarding the correctness of the received information. Some level of public skepticism is regarded a healthy ingredient for the risk communication process because it results in critical attitudes towards risk management. However, skepticism may also become a destructive element when it is accompanied by low levels of general trust, leading to cynical attitudes (rejection). Terpstra and Gutteling(37) found that Dutch citizens highly trust the expertise of water management institutions and also perceive relative high credibility of information received from risk managers. In line with these findings, Heems and Kothuis(38) recently argued that Dutch citizens’ trust regarding flood risks could be typified as “uncritical acceptance” or “blind trust.” There is an avenue for further research to study the extent to which risk communication enhances critical attitudes without causing people to distrust or reject information from risk management authorities. Such research is also

Framing Flood Risk needed to better define how trust is connected with information need and information seeking in RISP. Second, this study found that the mediators only partially mediated the effects of the communication frames on information need. This means there are likely other variables that (co-)explain the changes in information need. Although RISP contains a number of other variables (e.g., subjective norms, knowledge, and personal efficacy),(13) none of these were measured. So, from a theoretical perspective it remains unclear which additional variables may have mediated the communication effects on information need. Third, participants responded to the questions directly after their exposure to the frames. It is unclear to what extent the observed differences between the frames remain over a longer time period (e.g., days, weeks) and how message repetition influences this process. This is especially important because the frames aroused fear. Negative affective feelings are associated with effortful, detailoriented cognitive processing of information (systematic processing).(22) However, there are two mechanisms that may hinder the effects of fear appeals. First, affective responses occur rapidly and automatically in response to a stimulus.(5) As we have seen, the messages on average induced quite some fear which was not easily reduced by subsequent positive elements in the message. However, affective responses are generally not enduring but tend to fade after the stimulus has been removed. Second, when the threat is not immediate, people may cope with fear arousal by denying the perceived threat or avoiding further risk information.(4) The question, therefore, is whether information need will fade or stay when fear fades. In conclusion, this study showed that fear appeal increases people’s need for further information about risk because it stimulates people’s risk perceptions and arouses negative feelings. This finding confirms what was predicted by RISP but had not been investigated before. Adding positive elements to a flood risk message hardly reduced these fear appeal effects, a phenomenon that is known as a “negativity bias.” Fear appeal also suppressed levels of public trust and positive feelings. However, these processes seem unimportant for the transmission of fear appeal. Thus, effects of risk communication messages on people’s information needs seem to be mediated only by negative attributes and not by positive attributes. This is a new finding that has both the-

1517 oretical and practical relevance. Theoretically, this calls for a refinement of (information seeking) theories that include positive attributes such as public trust and positive affect. Especially the finding that trust failed to mediate effects of communication messages needs further inquiry. Practically, if authorities aim to stimulate citizens’ information needs and thereby their information seeking behavior, the communication message will be more effective when it increases their risk perceptions and evokes at least some negative affect. This can be achieved, for instance, by using “public friendly” risk maps indicating potential flood depths on risky locations and realistic pictures or videos of recent flood events. At the same time, these findings suggest that fear appeal should be used with caution. Once aroused, it cannot be controlled easily. As a result, risk communication may have unwanted side effects on citizens’ trust in risk management. Regular monitoring of people’s attitudes could be an important activity in preventing extreme levels of distrust or cynicism. ACKNOWLEDGMENTS This article is based on work funded by the Knowledge for Climate (KfC) Research Program for the development of knowledge and services that make it possible to climate proof the Netherlands. None of the conclusions expressed here necessarily reflects views other than those of the authors. APPENDIX A: COMMUNICATION FRAMES Introduction (all respondents) The Netherlands is a water-rich country, where people have been living near major rivers (like the Rhine and Meuse) for a long time, both inside and outside the embanked area. Before presenting the questions, respondents in the control group were reminded of the purpose of the questionnaire as follows: In this questionnaire we would like to know how you think about living in the river area. Before presenting the questions, respondents in the three communication frames were first introduced to one of the three frames that were assigned to them randomly: Many old harbor areas that are located outside the embanked area are no longer in use. This is why local governments would like to redevelop these areas in new residential areas. In the future these

1518 residential areas will remain unprotected by dikes. Although these areas outside the embanked area are often elevated, future residents should take flooding into account, which potentially cause damage. In this questionnaire we are interested in your opinion about living in such a residential area outside the embanked area. If you would live in such a residential area outside the embanked area in the future, then it is important to know that you may experience floods. Especially a combination of a lot of water in rivers and a strong surge of seawater (during a Northwestern wind) can cause high water levels. This hazard has decreased because of the Maeslantkering (the storm surge barrier Nieuwe Waterweg), but has not been completely eliminated. Streets can be flooded during high water levels. Such high water levels occur on average once in 10 years.

Terpstra et al. how the Netherlands can be protected against the water, also in the far future (until the year 2100).

Frame 3: Text and photos of the “amenities frame”: Frame 1: Text and photos of the “climate change frame” Floods are of all times and usually cause serious damage. Moreover, the climate is changing, which increases the water in rivers and causes sea level rise. According to experts, the Netherlands is insufficiently protected against the consequences of climate change. As a result, the flood risks in residential areas outside the embanked area may increase in the future.

Frame 2: Text and photos of the “safety frame”: In the Netherlands we already know for a very long time that water entails risks. The first dikes have been constructed more than 1,000 years ago. Moreover, the government continuously works on water safety. Recently, a new Delta Committee advised

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 climate change, construction companies have explored new ideas to make residential areas more beautiful and attractive. These future water-rich areas outside the embanked area are already under construction.

Finally, respondents in the three frames were reminded of the purpose of the questionnaire: In this questionnaire we would like to know how you think about living in such a neighborhood outside the embanked area. Imagine that you would live in such a neighborhood outside the embanked area, close to the river.

Framing Flood Risk

1519

APPENDIX B: CORRELATIONS

Table B1. Correlations Contrast X1 (Risk Communication Versus Control) Info Need Info Need Risk Perception Trust Negative Affect Positive Affect

Risk Perception

1 0.38*

Trust

Negative Affect

Positive Affect

1

–0.13* 0.36*

–0.21* 0.34*

1 –0.43*

–0.13*

–0.18*

0.47*

1 –0.49*

1

Note: N = 1,457; *p < 0.01.

Table B2. Correlations Contrast X2 (Climate Change Versus Safety) Info Need Info Need Risk Perception Trust Negative Affect Positive Affect

Risk Perception

1 0.33*

Trust

Negative Affect

Positive Affect

1

–0.07 0.27*

–0.26* 0.36*

1 –0.40*

–0.07

–0.19*

0.48*

1 –0.51*

1

Note: N = 838; *p < 0.01.

Table B3. Correlations Contrast X3 (Climate Change Versus Amenities) Info Need Info Need Risk Perception Trust Negative Affect Positive Affect

Risk Perception

1 0.42*

Trust

Negative Affect

Positive Affect

1

–0.05 0.32*

–0.22* 0.37*

1 –0.34*

–0.14*

−0.21*

0.47*

1 –0.48*

1

Note: N = 837; *p < 0.01.

REFERENCES 1. Kellens W, Terpstra T, De Maeyer P. Perception and communication of flood risks: A systematic review of empirical research. Risk Analysis, 2013; 33(1):24–49.

2. Griffin RJ, Dunwoody S, Neuwirth K. Proposed model of the relationship of risk information seeking and processing to the development of preventive behaviors. Environmental Research, 1999; 80(2):S230–S245. 3. McComas KA. Defining moments in risk communication research: 1996–2005. Journal of Health Communication, 2006; 11(1):75–91. 4. Ruiter RAC, Abraham C, Kok G. Scary warnings and rational precautions: A review of the psychology of fear appeals. Psychology & Health, 2001; 16(6):613–630. 5. Slovic P, Finucane ML, Peters E, MacGregor DG. The affect heuristic. European Journal of Operational Research, 2007; 177(3):1333–1352. 6. Witte K, Allen M. A meta-analysis of fear appeals: Implications for effective public health campaigns. Health Education & Behavior, 2000; 27(5):591–615. 7. Lindell MK, Perry RW. The protective action decision model: Theoretical modifications and additional evidence. Risk Analysis, 2012; 32(4):616–632. 8. Ter Huurne EFJ. Information seeking in a risky world. The theoretical and empirical development of FRIS: A framework of risk information seeking. Thesis (dissertation), University of Twente, 2008. 9. Kahlor L. Prism: A planned risk information seeking model. Health Communication, 2010; 25(4):345–356. 10. Kahlor LA. An augmented risk information seeking model: The case of global warming. Media Psychology, 2007; 10(3):414–435. 11. Ter Huurne EFJ, Griffin RJ, Gutteling JM. Risk information seeking among US and Dutch residents: An application of the model of risk information seeking and processing. Science Communication, 2009; 31(2):215–237. 12. Ter Huurne E, Gutteling, J. Information needs and risk perception as predictors of risk information seeking. Journal of Risk Research, 2008; 11(7):847–862. 13. Griffin RJ, Yang Z, Ter Huume E., Boerner F., Ortiz S., Dunwoody S. After the flood. Anger, attribution, and the seeking of information. Science Communication, 2008; 29(3):285–315. 14. Kellens W, Zaalberg R, De Maeyer P. The informed society: An analysis of the public’s information-seeking behavior regarding coastal flood risks. Risk Analysis, 2012; 32(8): 1369–1381. 15. Ter Huurne EFJ, Gutteling JM. How to trust? The importance of self-efficacy and social trust in public responses to industrial risks. Journal of Risk Research, 2009; 12(6):809–824. 16. Kuttschreuter M. Psychological determinants of reactions to food risk messages. Risk Analysis, 2006; 26(4): 1045–1057. 17. Mulilis JP, Duval TS. Negative threat appeals and earthquake preparedness —A person-relative-to-event (pre) model of coping with threat. Journal of Applied Social Psychology, 1995; 25(15):1319–1339. 18. Mulilis JP, Duval TS. The PRE model of coping and tornado preparedness: Moderating effects of responsibility. Journal of Applied Social Psychology, 1997; 27(19): 1750–1766. 19. Kievik M, Gutteling JM. Yes, we can: Motivate Dutch citizens to engage in self-protective behavior with regard to flood risks. Natural Hazards, 2011; 59(3):1475–1490. 20. Finucane ML, Alhakami A, Slovic P, Johnson SM. The affect heuristic in judgments of risks and benefits. Journal of Behavioral Decision Making, 2000; 13(1):1–17. 21. Bohner G, Chaiken S, Hunyadi P. The role of mood and message ambiguity in the interplay of heuristic and systematic processing. European Journal of Social Psychology, 1994; 24(1):207–221. 22. Griffin RJ, Neuwirth K, Dunwoody S, Giese J. Information sufficiency and risk communication. Media Psychology, 2004; 6(1):23–61.

1520 23. Petty RE, Schumann DW, Richman SA, Strathman AJ. Positive mood and persuasion. Different roles for affect under high-elaboration and low-elaboration conditions. Journal of Personality and Social Psychology, 1993; 64(1): 5–20. 24. Siegrist M, Gutscher H, Earle TC. Perception of risk: The influence of general trust, and general confidence. Journal of Risk Research, 2005; 8(2):145–156. 25. Nabi RL. A cognitive-functional model for the effects of discrete negative emotions on information processing, attitude change, and recall. Communication Theory, 1999; 9(3): 292–320. 26. Gleicher F, Petty RE. Expectations of reassurance influence the nature of fear-stimulated attitude-change. Journal of Experimental Social Psychology, 1992; 28(1):86–100. 27. Slovic P, Finucane ML, Peters E, MacGregor DG. Risk as analysis and risk as feelings: Some thoughts about affect, reason, risk, and rationality. Risk Analysis, 2004; 24(2):311–322. 28. Terpstra T. Emotions, trust, and perceived risk: Affective and cognitive routes to flood preparedness behavior. Risk Analysis, 2011; 31(10):1658–1675. 29. Preacher KJ, Hayes AF. Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. Behavior Research Methods, 2008; 40: 879–891. 30. Hayes AF. Beyond Baron and Kenny: Statistical mediation analysis in the new millennium. Communication Monographs, 2009; 76(4):408–420.

Terpstra et al. 31. Zaalberg R, Midden CJH. Living behind dikes: Mimicking flooding experiences. Risk Analysis, 2013; 33(5):866–876. 32. Baumeister RF, Bratslavsky E, Finkenauer C, Vohs KD. Bad is stronger than good. Review of General Psychology, 2001; 5:323–370. 33. Slovic P. Perceived risk, trust, and democracy. Risk Analysis, 1993; 13(6):675–682. 34. Siegrist M, Cvetkovich G. Better negative than positive? Evidence of a bias for negative information about possible health dangers. Risk Analysis, 2001; 21(1):199–206. 35. White MP, Pahl S, Buehner, M, Haye, A. Trust in risky messages: The role of prior attitudes. Risk Analysis, 2003; 23(4):717–726. 36. Poortinga W, Pidgeon NF. Exploring the dimensionality of trust in risk regulation. Risk Analysis, 2003; 23(5): 961–972. 37. Terpstra T, Gutteling JM. Households’ perceived responsibilities in flood risk management in the Netherlands. International Journal of Water Resources Development, 2008; 24(4): 555–565. 38. Heems GC, Kothuis BLM. Flood safety: Managing vulnerability beyond the myth of dry feet. A socio-cultural perspective on how the Dutch cope with the threat of flooding (dissertation). Amsterdam: Waterworks, 2012. 39. Baron RM, Kenny DA. The moderator mediator variable distinction in social psychological-research — Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 1986; 51(6):1173–1182.

You have been framed! How antecedents of information need mediate the effects of risk communication messages.

This study investigates the processes that mediate the effects of framing flood risks on people's information needs. Insight into the effects of risk ...
438KB Sizes 0 Downloads 3 Views