How Does the General Public Evaluate Risk Information? The Impact of Associations with Other Risks
Abstract
There is a considerable body of knowledge about the way people perceive risks using heuristics and qualitative characteristics, and about how risk information should be communicated to the public. However, little is known about the way people use the perception of known risks (associated risks) to judge an unknown risk. In a first, qualitative study, six different risks were discussed in in-depth interviews and focus group interviews. The interviews showed that risk associations played a prominent role in forming risk perceptions. Associated risks were often mentioned spontaneously. Second, a survey study was conducted to confirm the importance of risk associations quantitatively. This study investigated whether people related unknown risks to known risks. This was indeed confirmed. Furthermore, some insight was gained into how and why people form risk associations. Results showed that the semantic category of the unknown risks was more important in forming associations than the perceived level of risk or specific risk characteristics. These findings were in line with the semantic network theory. Based on these two studies, we recommend using the mental models approach in developing new risk communications.
1. INTRODUCTION
People are informed about unknown risks through risk communications, which form a bridge between the results of risk assessments by experts and risk perceptions among the general public. However, the general public seem to interpret risk information differently from the experts.(1,2) So far, several studies have looked at the way the public perceive risks,(2,3) what information they use to judge a risk,(4,5) what kind of information they want about a risk,(6) and how risk messages can influence risk perception (e.g., References 7–9). Nevertheless, it is not clear how this information is used to judge an unknown risk, for instance, how the public process the results of risk analyses and what other knowledge they mobilize. The two studies reported on in this article were designed to examine the process of evaluating risk information and focused on the effects of associated risks.
Previous research has shown that people process risk information by using cognitive mechanisms such as inductive reasoning(10) and heuristics.(2) In addition, several studies have explored which characteristics of a risk and its context influence risk perception,(4,11) and some guidelines are available about how risks can be appropriately communicated.(12,13) A few theories and frameworks of risk perception, such as the mental models approach, assume that people's prior knowledge affects their risk perception (e.g., References 14 and 15). However, it has not yet been specified how people use prior knowledge, and particularly associated risks, when they evaluate an unknown risk.
The first study reported here was a qualitative study. One of its goals was to find out how known risks are associated with unknown risks and how the former are used to interpret the latter. In the second study, a survey, the role of prior knowledge in risk perception was explored by investigating similarities in judgment about unknown risks and associated risks. We conclude by presenting recommendations for the communication of risks to the general public.
1.1. Theoretical Background
Surprisingly, the effect of prior knowledge on the understanding of risk communications has not received much attention in this field of research. Nevertheless, several theories and concepts implicitly assume that prior knowledge is important for risk perception and comprehension. The first approach is the mental models approach, which uses respondents' prior knowledge to list existing ideas and possible misconceptions about a specific risk.(16,17) The respondents' mental model is then compared with an expert model, and the results are used to design a risk communication that fills the gaps in knowledge and changes the respondents' incorrect beliefs. However, the mental models approach uses knowledge schemes to identify the differences between experts and the general public concerning a specific risk. This approach acknowledges that prior knowledge (and thus risk associations) is an important factor that influences risk perception. It is unknown how people use prior knowledge and whether associated risks play a part in risk evaluations. Therefore, we wanted to investigate this in these two studies.
The second approach that assumes an influence of prior knowledge is the construction-integration theory.(18) This approach holds that text information (in general, not only information on risks) is understood in two phases, construction and integration. In the construction phase, a rough scheme is created from the text, the receiver's goals, and prior knowledge. In the integration phase, this scheme is adjusted to the current context by the process of spreading activation until it forms a well-structured representation of the information. Prior knowledge is important because a text is easier to understand when readers can fit new information to what they already know. This is also facilitated when the tenor of the text is in accordance with this knowledge and the readers' personal goals.(19) Thus, associated risks that correspond to these characteristics may improve the understanding of risk communication.
Lastly, the representativeness heuristic is discussed here because when people apply it, their judgments are based on prior knowledge (about the reference group). The probability of an event is estimated by whether this event is representative of similar events stored in memory.(20) The effect of this heuristic on judgments has already been demonstrated. For example, in a study the description of an intelligent but not very creative man led most respondents to categorize him as a computer science or engineering student because this description seemed to be representative of these two categories.(20) The respondents ignored the low base rates of computer science students and engineering students in the general population.(21) To our knowledge, no study has investigated the effect of this heuristic on the way people perceive risks in a more spontaneous or realistic setting.
In the first study, we were particularly interested in finding out how respondents used prior knowledge about known risks to evaluate unknown risks in real-life situations. We used unknown risks so that people could not yet have specific attitudes or beliefs about them.
2. QUALITATIVE STUDY
2.1. Goals
The overall aim of this study was to examine how people handle risk information when they judge an unknown risk. We wanted to explore what information is used to evaluate a risk and how risk characteristics can affect this. We also had two specific goals. The first was to assess whether people associate an unknown risk with known risks to interpret the unknown risk. The second aim was to examine how and why they use associations with known risks to interpret unknown risks.
2.2. Methods
2.2.1. Design
Focus group interviews and in-depth interviews were chosen as complementary research methods. Focus group interviews are group discussions that intend to examine feelings and opinions about a given problem, experience, or specific situation.(22) We also used in-depth interviews with individuals to investigate a particular subject in greater depth than was possible in the focus group interviews.
2.2.2. Respondents
The participants of all in-depth interviews and of two focus group interviews had answered a call for respondents in a free local paper. The other focus group respondents were recruited from a drama group, two choirs, and from among the students at two schools for higher vocational training (taking courses on social work and information management). The interviews took place at Maastricht University and at different locations in the town of Maastricht (e.g., a community center or the respondent's home) between October 2002 and January 2003. The study was presented as an interview about the comprehensibility of newspaper articles, to prevent response biases. Each respondent received a €7 reward for a one-hour interview.
In total, 11 in-depth interviews and seven focus group interviews were carried out. Participants in the in-depth interviews included 10 women and one man. Their mean age was around 42 years, ranging from 23 to 67 years. The number of participants in the focus group interviews ranged between three and 11 respondents per group. In total, nine men and 19 women participated in the focus groups; their mean age was around 40 years, ranging from 17 to 71 years.
2.2.3. Materials
The following materials were used in both types of interview: six different purpose-made newspaper articles, a tape recorder, and a question route. The six newspaper articles concerned different risks. They were written in an objective style and were made more realistic by using a newspaper font and by mentioning the name of the journalist or press agency. We chose these risks because we believed that they were relatively unknown to the general public and were suitable for newspaper articles. For example, one article considered the risk of acrylamide in food, which can be carcinogenic when large amounts are consumed.(23,24) Another article discussed the risk of living in a town near a chemical weapons depot, where the army is going to incinerate these weapons.(25,26) The other newspaper articles were about nuisance caused by artificial light,(27) deoxynivalenol (DON) in wheat products,(28,29) phthalates in cosmetics,(30,31) and residues of medicines in the environment.(32)
The participants of all interviews read and discussed at least four newspaper articles during the interview; in two in-depth interviews, there was enough time to discuss five newspaper articles. The order of the articles was varied so that each article was discussed 10 to 13 times.
2.2.4. Procedure
The interviewer started each session of the focus group and in-depth interviews by welcoming the respondents and explaining how the interview would work. The interviewer emphasized that it was important for the respondents to give their own opinion. This was particularly emphasized in the focus group interviews. After the respondents had read the first article, the interviewer started the tape recorder and asked questions about the article, for example, about the respondents' first impression of the article, the perceived probability of the risk, the number of people involved, and associations with other risks. The respondents in the in-depth interviews were stimulated to give more attention to the process of risk perception, whereas in the focus group interviews, more time was spent on discussion about the risk itself. When all questions had been answered, respondents proceeded to the next newspaper article, until at least four newspaper articles had been discussed. Each respondent was then thanked and paid.
2.2.5. Analyses
The taped interviews were transcribed verbatim and divided according to the topics of the newspaper articles. This resulted in 74 interview sections, 13 sections about phthalates, 10 about DON, 12 about acrylamide, 14 about the incineration of chemical weapons, 13 about residues of medicines in the environment, and 12 about light nuisance.
The interview sections were coded using the QSR NVivo 2.0 qualitative data program.(33) The codes were based on the main issues discussed in the interview, such as credibility, familiarity, consequences, risk evaluation, and risk associations (Table I). A second coder also coded all interviews. The agreement between the two coders was 81%. In cases of disagreement, the codes of the first coder were accepted.
| Issue | Mean Number of Interview Sections |
|---|---|
| Risk evaluation | 3.85 |
| Method of risk evaluation | 2.69 |
| Risk association | 2.65 |
| Familiarity | 1.96 |
| Future behavior | 1.84 |
| Consequences for humans and animals | 1.77 |
| Controllability | 1.49 |
| Information need | 1.43 |
| Credibility | 1.38 |
| Balance of advantages and disadvantages | 1.07 |
| Frequency of event | 0.82 |
| Consequences for the environment | 0.15 |
We first examined the issues that had been mentioned during the interviews. This background information relates to the issues respondents discussed when they were confronted with an unknown risk. Second, we analyzed risk associations, in terms of whether, how, and for what reasons they were used.
2.3. Results
2.3.1. Background Issues
When confronted with one of the newspaper articles about an unknown risk, the respondents often tried to make a risk evaluation (see Table I for the mean number of interview sections devoted to each issue by the respondents). The respondents frequently talked about the probability of the risk occurring and the severity of its consequences. For example, respondents evaluated the probability of the environment being polluted by medicine residues as high but the consequences of this risk were judged to be only slightly harmful. On the other hand, the probability that an accident would happen during the incineration of chemical weapons was estimated to be low, whereas its consequences were judged to be very serious.
The participants also frequently discussed the method of risk evaluation in both types of interviews (Table I). This aspect related to how they evaluated the unknown risk, for example, by relating it to the number of people affected by the risk. Many of the interview sections that were coded as method of risk evaluation were connected to the risk characteristics described by Slovic.(4) These risk characteristics, however, are beyond the scope of this article. Only the risk characteristic of familiarity is discussed here, since it relates to risk associations. Respondents frequently talked about the familiarity of the risk. For example, the respondents often said that they were unfamiliar with the risks of light nuisance and medicine residues in the environment, but they could imagine that these risks existed.
2.3.2. Risk Associations
As can be seen in Table I, respondents frequently associated the unknown risk with known risks. These risk associations were mentioned when the interviewer explicitly asked for them (mean number of sections per interview: 0.72), but more often spontaneously (mean number of sections per interview: 1.93).
Associated risks were used as a stepping-stone to relate the unknown risk to what respondents already knew. We constructed a mental model about the risk of phthalates in cosmetics to obtain an impression of the kind of associated risks respondents used in the interviews.(16,17) As can be seen in Fig. 1, the risk associations took various forms, ranging from associations with similar risks to events that had no clear connection to the present risk.
Mental model of risk of phthalates.
Next, we were interested in how and why risk associations were made. As regards the way these associations were formed, the interviews showed that risk associations were based on personal experience with known risks, information provided by the media, and general knowledge that had been obtained from other risks in the past.
Respondents often appeared to relate the information about the unknown risk presented in the article to a cognitive scheme for a known risk, based on similarity of characteristics. The unknown risk seemed to inherit the characteristics of the associated risk, so that the scheme of the associated risk determined the perception of the unknown risk. For example, after reading the acrylamide article, one respondent assumed that foods such as French fries and chips were risky because she already knew that they contain fat, which is unhealthy. She therefore concluded that acrylamide was also a risk to health. By contrast, acrylamide reminded another respondent of fruit that is unhealthy because it is sprayed with insecticides, which made her conclude that it was impossible to avoid food with acrylamide and that she would also accept this risk.
Sometimes, respondents appeared to use schemes of known risks that were only vaguely related or were not related at all to the unknown risk. In these cases, the respondents only focused on one characteristic of the unknown risk and ignored others. In the above example about acrylamide, the first respondent only focused on fatty foods that may contain acrylamide and that she already knew to be unhealthy, thereby ignoring the fact that acrylamide can also be present in healthy foods such as bread.
In short, the respondents used various risk associations in these examples, leading to different views about accepting or avoiding the unknown risk. In addition, unknown risks could inherit the characteristics of the associated risks. In some cases, the associations were based on one common characteristic so that other features of the unknown risk were ignored.
We discovered four different reasons why respondents used risk associations. First, they were used to characterize the possible severity and consequences of the unknown risk. For example, one respondent compared the effects of phthalates on unborn children with the effect of the drug Softenon (Thalidomide), by commenting: “What shouldn't happen is that this comes onto the market and later it appears that it is totally wrong, and something like the Softenon babies occurs. We just give them those pills and later, ‘oh, damn'…” This example also illustrates an assimilation effect; the properties of the associated risk were adopted by the unknown risk.
Second, risk associations were made to illustrate that people had already tolerated many similar risks although these were known to cause harm. One respondent compared DON to the artificial nutrients mentioned on the labels of food products: “I think that all the other ingredients in food products are not very healthy for people either [just like DON], and those are mentioned on the product. Only, people do not know what they are, so they don't get alarmed by them.”
Third, respondents used associations to clarify that some risky activities also had benefits. For instance, “I think that there's a bigger problem of not getting enough light … perhaps not enough production of melatonin. And I think of people with depressions and so on. Then I think of light therapy; then I think, light is more positive than negative.” This type of risk association can also be described as a contrast effect, in that the consequences of the unknown risk were contrasted with those of an associated risk.
The fourth reason for using risk associations was to illustrate that risks could also be resolved. One respondent associated phthalates with CFCs: “Just like the refrigerator thing, with the CFC, the gas that was in them. And that they no longer have this gas, but something totally different. They will be able to find some solution for this [i.e., phthalates].”
In short, the results suggested that associated risks and their cognitive schemes were important in perceiving unknown risks. Associated risks influenced the perception of an unknown risk via similar characteristics of the two risks' schemes and by narrowing the focus to a limited amount of information about the unknown risk.
2.4. Discussion
Respondents in the in-depth interviews and focus group interviews often talked about issues that were related to risk associations. These risk associations seemed to be based on the respondents' personal experiences, media coverage, and general knowledge. They were used for several reasons: to illustrate the severity and consequences of the unknown risk, to indicate that similar risks had been tolerated in the past, to highlight the benefits related to these risky activities, and to illustrate the possibility that risks can be resolved. Alternatively, the use of associated risks may be described as an assimilation effect or contrast effect.
The cognitive scheme of the associated risk may determine the perception of the unknown risk. This is in line with the construction-integration theory, in that respondents referred to their existing schemes to understand the unknown risks and linked the content of these schemes to the new information. The cognitive scheme and salient qualities of the associated risk seemed to affect the perception of the unknown risk.
The example of the risk of phthalates showed that a mental model offers a good representation of the respondents' knowledge propositions about that risk. This may therefore be in agreement with the construction-integration theory. At the same time, the mental models approach can help risk communicators to understand the public's risk perception and adjust risk communications accordingly. This may facilitate the construction and integration phases in the public's comprehension of the risk message.
3. SURVEY STUDY
This survey study had two principal goals. The first was to explore in a quantitative study the main finding of the qualitative study in greater depth, namely, that the respondents used associated risks to judge an unknown risk. Hence, we empirically investigated whether associated risks are related to unknown risks.
As we mentioned above, the mental models approach assumes that prior knowledge is important in evaluating a risk, for example, through associations with known risks. Because this approach does not explain how associations are formed, our second goal was to find out how people relate unknown risks to known risks. More particularly, we wanted to know what information (risk perception and risk characteristics) people use to choose an associated risk.
We asked respondents to judge four types of risk in terms of several risk perception aspects and qualitative characteristics. The four types were an unknown risk, a risk that they themselves associated with the unknown risk (self-associated risk), a risk that others associated with the unknown risk (risk associated by others), and a risk not related to the unknown risk (unrelated risk). We then calculated the correlations between the risk perception aspects of these four types of risk. If the perception of self-associated risks affected the perception of unknown risks, these perceptions were assumed to correlate (r12). Fig. 2 presents an overview of the correlations and comparisons of the correlations in this study.
Overview of the central analyses.
However, a high correlation between the perceptions of unknown risks and those of self-associated risks could also be caused by a third variable, for example, in that people who are risk averse may judge all hazards as risky. We therefore compared, for each unknown risk, the correlation between the risk perceptions of the unknown risk and those of the self-associated risk (r12) with the correlation between the risk perceptions of the unknown risk and those of an unrelated risk (r14, Comparison I in Fig. 2). If the correlation between the former pair was higher than between the latter pair, this would mean that a third variable like risk aversion cannot be the cause of the high correlation between the risk perceptions of an unknown risk and those of a self-associated risk.
Next, we wanted to know whether the choice of an associated risk is to some extent personal. If that is the case, it would be difficult to use associated risks in risk communication because everyone at risk would need to receive a risk communication that included his or her own personal risk associations. On the other hand, if respondents also accept a risk associated by others, it would be possible to illustrate unknown risks using associated known risks. Therefore, we also compared, for each unknown risk, the correlation between the risk perceptions of the unknown risk and those of the risk associated by others (r13) with the correlation between the risk perceptions of the unknown risk and those of the self-associated risk (r12, Comparison II in Fig. 2). If the correlation between the former pair does not differ substantially from the correlation between the latter pair, it might be concluded that perceptions of risks associated by others do not differ greatly from self-associated risks. This would allow risks associated by others to be used in risk communication.
3.1. Method
3.1.1. Respondents
The respondents were members of an Internet research panel. Seven hundred adult members who formed a representative sample of the Dutch population received an e-mail announcing the study, with a link to an Internet site that presented the first questionnaire. They received four questionnaires at one-week intervals. Respondents were paid in points; they received 100 points for every questionnaire they filled in. If they completed all four questionnaires, they received a 50-point bonus. The maximum number of points (450) could be exchanged for a gift voucher equivalent to €5. The survey was conducted between May and June 2004.
3.1.2. Material
In the first questionnaire, respondents judged three risks that were, at the time, relatively unknown: food irradiation, avian influenza, and the transport of chlorine by train in the Netherlands, resulting in three questionnaire parts. In each part, respondents had to read a short description of the risk, consisting of approximately three sentences. They then filled in the questionnaire, which consisted of risk perception aspects and qualitative characteristics. The risk perception aspects were perceived health damage, severity of the consequences, probability estimation, risk estimation, and a ranking of risks in which the unknown risk should be placed (ranging from the risk of Creutzfeldt-Jakob disease to the risk of coronary heart disease). The internal reliabilities of these aspects were moderate to high in all four questionnaires and for all three risks: Cronbach's α > 0.71. Hence, we used the average sum score of the risk perception aspects in the analyses as a risk perception scale.
The qualitative characteristics were familiarity with the risk, vulnerability, acceptance, fear, voluntariness, controllability, controllability by other people, benefits, known by science, and the time interval at which its consequences would be noticed.(34) Most items were measured on seven-point Likert scales, with higher scores indicating higher levels of the variables measured. Two items, the probability estimate and the ranking of risks, consisted of multiple-choice questions with seven and six options, respectively, to choose from. Each questionnaire consisted of 45 items, 15 items in each part.
At the end of each part of the first questionnaire, respondents were also asked to indicate a risk that they associated with the unknown risk. This resulted in three self-associated risks (e.g., one participant associated X-rays with food irradiation). One week later, the respondents received the second questionnaire, in which they judged the three self-associated risks they had mentioned in the first questionnaire, using the same items. Thus, the second questionnaire could be unique to each respondent. The third questionnaire consisted of three risks that other respondents had mentioned in the first questionnaire and that the respondents had not mentioned themselves (risks associated by others). For each unknown risk, we selected two or three risks that were frequently associated by the respondents. Each respondent was presented with one of these selected risks. For food irradiation, the respondents evaluated either genetically modified food or pesticides on fruit and vegetables. For avian influenza, they evaluated Severe Acute Respiratory Syndrome (SARS) or Bovine Spongiform Encephalitis (BSE). For the transportation of chlorine by train, they evaluated either the transportation of inflammables, an accident at a nuclear power station, or an airplane crash. We made sure that the risk associated by others that was presented to a respondent was not similar to the self-associated risk. Hence, the respondents (or at least some of them) received different questionnaires. In the fourth and final questionnaire, respondents judged risks that were unrelated to the unknown risks and had not been mentioned by any of the respondents, namely, phthalates in cosmetics, an anaphylactic shock caused by an allergy, or flooding of the Netherlands in the next century.
3.2. Results
3.2.1. Respondents
The first questionnaire (about the unknown risks) was completely filled in by 349 respondents. In the end, 256 respondents had completed at least two parts of all four questionnaires. Thus, 73.35% of the respondents of the first questionnaire were included in the analyses after the final questionnaire. The demographics of the final sample were as follows: 55.9% of the respondents were female; the mean age was 41.95 years (SD= 13.43); 27.3% of the respondents had a low education level (primary school or lower professional training), 43.4% a medium level (secondary education and lower secondary professional training), and 29.3% a high level (higher professional training or university).4 The education level of our sample was representative of that of the population in the Netherlands.(35) The respondents were evenly distributed across the Netherlands.
3.2.2. Risk Associations
The first goal of this study was to find out whether the risk perceptions of the unknown risk and those of the self-associated risk were related. This was tested by calculating the correlations between the unknown risks and the self-associated risks, that is, between the scores on the risk perception scales of questionnaires 1 and 2 (r12), for each part. As can be seen in Table II, the risk perception scales of the three unknown risks correlated significantly with the corresponding risk perception scales of the three self-associated risks.
| Self-Associated Risk | ||||
|---|---|---|---|---|
| Part 1 | Part 2 | Part 3 | ||
| Unknown risk | Part 1: Food irradiation | 0.48* | ||
| Part 2: Avian influenza | 0.47* | |||
| Part 3: Chlorine transport | 0.40* | |||
- Note:*p < 0.01.
Thus, this result supported our hypothesis that the risk perception of an unknown risk would be related to the risk perception of a self-associated risk. It may imply that people use the perception of the self-associated risk to judge the unknown risk.
3.2.3. Self-Associated Versus Unrelated Risks
Since the above high correlations could also be the result of a third variable, for example, risk aversion, we compared the correlations between the risk perception scales of the unknown risks and those of the self-associated risks (r12) with the correlations between the risk perception scales of the unknown risks and those of the unrelated risks (r14, Comparison I in Fig. 2). For this purpose, we used Steiger's test for equality of two dependent correlations, Z1bar*, since the correlations originated from a single sample.(36)
If people associate an unknown risk with a risk that they consider to be related in terms of risk perception, rather than with an unrelated risk, the correlations r12 should be significantly larger than the correlations r14. The risks of avian influenza and of transportation of chlorine by train (Parts 2 and 3) had an r12 that was significantly larger than r14, Z1bar*s > 2.33, ps < 0.01. As regards food irradiation (Part 1), no significant difference was observed between r12 and r14, Z1bar*=−0.47, p= 0.32. On the basis of these results, we may conclude that the risk perception of an unknown risk is more strongly related to the risk perception of a self-associated risk than to the risk perception of an unrelated risk. This was in accordance with our hypothesis.
3.2.4. Self-Associated Risks Versus Risks Associated by Others
Next, we wanted to find out whether the self-associated risk was personal or whether it could also be replaced by a risk associated by others. To this end, we compared the correlations between the risk perception scales of the unknown risk and those of the self-associated risk (r12) with the correlations between the risk perception scales of the unknown risk and those of the risk associated by others (r13, Comparison II in Fig. 2). Only in the third part (chlorine transport) was r12 marginally larger than r13, Z1bar*= 1.53, p= 0.06. In the other two parts, r12 did not differ from r13, Z1bar*s > −0.79, ps > 0.21.
In sum, the correlations between the unknown and self-associated risks appeared to be similar to the correlations between the unknown risk and the risk associated by others. People's evaluations of unknown risks, self-associated risks, and risks associated by others were related and sometimes even strongly related, whereas the evaluations of unknown risks were not related to unrelated risks. Thus, we conclude that the association between two risks may not be purely personal. People seem to share, to some extent, a risk association network that a certain risk evokes. This might be based on semantics or representativeness, which leads to the second goal of this study: How are risks associated?
3.2.5. How are Risks Associated?
The second goal of this study was to find out how people relate unknown risks to known risks: Why do they associate these specific risks and not the unrelated ones? We tested three possible modes of association. First, the self-associated risks might be chosen because they had a semantic connection with the unknown risks, with people using the meaning of the words that describe the unknown risk to find a related risk. A second possibility was that the unknown risks and the self-associated risks evoked similar risk perceptions. The third option we studied was that people chose the self-associated risks because these had similar qualitative characteristics as the unknown risks.
These three different association modes were investigated in different ways. First, the semantic connection between unknown and self-associated risks was examined by studying the content of the self-associated risks. We did indeed find that the self-associated risks and the unknown risks often had a semantic connection. As a result, the respondents regularly related the unknown risk to a known risk of the same semantic category. The risk of food irradiation reminded most respondents of one of the two words that characterize this technology, namely, food risks, such as genetically modified food and food poisoning; and radiation risks, such as radioactivity and cell phones. The second risk, that of avian influenza, reminded respondents of other infectious diseases of humans or animals, such as Severe Acute Respiratory Syndrome (SARS) and Bovine Spongiform Encephalitis (BSE). The third unknown risk, the transportation of chlorine by train, reminded most of the respondents of other dangerous transports, for example, those of nuclear waste and petrol. In addition, the respondents mentioned other catastrophic accidents, such as an airplane crash or the explosion of the nuclear power station in Chernobyl. Thus, the semantics of the unknown risks appeared to be important for the choice of self-associated risks.
Second, we examined whether the self-associated risks were chosen because the risk perceptions of the unknown risks were similar to those of the self-associated risks. To this end, the absolute difference in scores between the risk perception scales of the unknown risks and those of the self-associated risks (ΔRP12) were compared with the absolute difference in scores between the risk perception scales of the unknown risks and those of the unrelated risks (ΔRP14). If the respondents associated unknown risks with risks that had similar risk perceptions, ΔRP12 should be smaller than ΔRP14. We conducted paired t-tests between ΔRP12 and ΔRP14 for each part of the questionnaire.
Only in Part 1 was ΔRP12 (ΔM12=−0.77, SD= 1.31) significantly different from ΔRP14 (ΔM12= 0.36, SD= 1.12), t(220) =−10.89, p= 0.0001. However, the difference was contrary to our expectations, in that ΔRP12 was larger than ΔRP14. In the other two parts, ΔRP12 did not differ from ΔRP14, t(221) =−0.37 and t(214) = 1.30.
In sum, the analysis of the second association mode yielded mixed results. The mean differences between the risk perceptions of the unknown risks and those of the self-associated risks were in most cases not significantly different from the mean differences between the risk perceptions of the unknown risks and those of the unrelated risks. Thus, people do not appear to form associations with risks based on the level of perceived risk.
Third, we wanted to find out whether the respondents used certain qualitative characteristics to choose the self-associated risks, such as the fact that both the unknown risk and the self-associated risk might not be controllable by the victims. The qualitative characteristics that were assessed in this part of the study were familiarity, vulnerability, acceptability, fear, voluntariness, controllability, controllability by others, benefits, known by science, and time interval. A qualitative characteristic was considered to be an important determinant of the risk association if the absolute difference between the unknown risk and the self-associated risk for a particular item (ΔQC12) was significantly smaller than the absolute difference between the unknown risk and the unrelated risk (ΔQC14). In such cases, the unknown risk and the self-associated risk would be evaluated as more similar in terms of this characteristic than the unknown risk and the unrelated risk. The paired t-tests for ΔQC12 and ΔQC14 for all items showed significant differences for 18 of the 30 items, ts < −2.28 and ts > 2.27, ps < 0.02. The qualitative characteristics controllability, controllability by others, known by science (Part 1), benefits, time interval (Part 2), familiarity, voluntariness, and controllability (Part 3) yielded a ΔQC12 that was significantly smaller than ΔQC14. The characteristics familiarity, vulnerability, acceptability, benefits, time interval (Part 1), familiarity, vulnerability, benefits (Part 2), vulnerability, and time interval (Part 3) yielded a ΔQC12 that was significantly larger than ΔQC14. There were no other significant effects of the type of risk association on the scores for the qualitative characteristics, ts > −1.85 and ts < 1.70, ps > 0.07.
In short, the qualitative characteristics had mixed results, just like the risk perception scales. Therefore, we cannot conclude that the associations between the unknown risks and the self-associated risks were based on similarity in terms of these qualitative characteristics. Some qualitative characteristics yielded smaller absolute differences between the unknown risks and the self-associated risks than between the unknown risks and the unrelated risks in all three parts, whereas other characteristics yielded larger absolute differences between the unknown and self-associated risks than between the unknown and unrelated risks. Thus, it is unlikely that respondents related risks based on similarity of qualitative characteristics.
3.3. Discussion
This survey study had two goals. The first was to explore, quantitatively, whether people relate unknown risks to other risks. The use of prior knowledge (and associated risks) was assumed to be important in risk perception. However, how risks are related had not yet been studied. Therefore, our second goal was to study how people relate an unknown risk to a known risk.
The findings of this study showed that respondents did indeed link unknown risks to known risks, and that the perceptions of the unknown risk and the self-associated risks were related. The association between two risks was not the result of a personality variable such as risk aversion, nor was it due to a similar level of risk perception, or to similar qualitative characteristics of the two risks. Instead, it appeared that the self-associated risks were semantically connected to the unknown risks. Furthermore, our findings indicated that other risks, associated by other people, could also be acknowledged as associated risks. This may imply that associated risks can influence the perception of unknown risks due to a semantic connection between these risks. However, this remains to be tested empirically.
The choice of a risk association can be explained by a semantic network model.(37,38) In this type of model, concepts that are related to each other are connected by means of nodes and links. The strength of the link between two nodes is based on the frequency of simultaneous activation of the nodes. Concepts are strongly linked when they are often activated together. If one concept is activated, the activation spreads through the network of nodes and links, so that other related concepts are activated.(39) The spreading of activation is determined by the activation of the initial concept and the strength of the links between this concept and connected concepts. A link degenerates when two concepts are no longer activated together. The semantic network model represents an economical storage method because concepts can “inherit” the features of connected concepts.
The semantic network model can be applied to the association between unknown risks and known risks as follows. Learning about an unknown risk (e.g., “food irradiation”) activates the general concept of the unknown risk based on the semantic category it belongs to (e.g., “food risks”). The activation then spreads further to nodes of known risks and instances (e.g., “contaminated food”) that are also connected to the general concept. Personal experience determines which familiar risks and instances are included in the network, which of them are already strongly activated, and which are related to the general concept. Therefore, associated risks could at first sight be expected to be personal, but other familiar risks (e.g., “pesticides on fruit”) may also be acknowledged as associated risks because they are present in the network, only not as strongly as the self-associated risks. For example, extensive media attention to a recent incident may activate some nodes strongly, so that the recent incident is temporarily linked to many other nodes. This might not have occurred without the media coverage.
Other theories and concepts may also be able to explain how people associate risks. For example, people may associate unknown risks with known risks because of certain risk attributes, like Renn's “pending danger, slow killers, cost-benefit ratio, and avocational thrill,”(40) and Sjöberg's “tampering with nature.”(41) In addition, risks may be associated with each other because they are both stigmatized.(42,43) It would be interesting to explore the influence of these attributes in further research.
This study used online questionnaires completed by an Internet research panel, which has several advantages. We found that by using an Internet research panel, we were able to reach a reasonably large sample of respondents with relatively little effort. Response rates for online questionnaires are known to be high, since the questionnaires are easy to fill in and return.(44) However, Internet research has been criticized for its limited opportunities to monitor the respondents; respondents may not fill in the survey honestly or seriously.(45) It has also been argued that the respondent samples of Internet studies are self-selected. Nevertheless, it turned out to be easier to find respondents with various demographic characteristics through web surveys than by paper-and-pencil surveys because people with a wide range of backgrounds now have Internet access.(45) Moreover, several studies have shown that the validity of Internet surveys is similar to that of lab surveys.(44–46)
4. RECOMMENDATIONS AND CONCLUSIONS
How can we improve risk communication by using the findings of these two studies? According to the construction-integration theory, a text is easy to comprehend if the readers' schemes are in accordance with the text and vice versa. It is therefore important to identify the public's knowledge schemes that are related to the risk, so that the content of the risk communication can be tailored to this knowledge. The mental model approach is useful to attune the public's knowledge to new risk information.(16,17) This could be achieved by first separately interviewing experts and a sample of the public on the specific risk. The two resulting mental models could then be mapped to form a frame to which the content of the risk communication should be tailored. Associated risks that are present in the public's mental models can be used in risk communication to guide people toward a particular perception of the unknown risk. For example, according to the findings of our survey study, the known SARS risk can be used in a risk communication about avian influenza (as it was mentioned by 16% of the respondents in the first questionnaire). The known risk activates the semantic category of infectious diseases, so that avian influenza is linked to the characteristics of this category and other accompanying risks, e.g., that it spreads easily and that it needs to be prevented immediately. We recommend pretesting the presumed associated risks to make sure that the associated and unknown risks have semantic similarities, since studies about risk comparisons showed that not all comparisons are accepted by the public (e.g., Reference 47). However, it may be even more important to include certain associated risks in risk communication for the purpose of explaining in what respect(s) these are different from the risk presented, so that these associated risks do not affect the presented risk in undesirable ways.
The interviews revealed that known risks were associated with unknown risks for four different reasons. First, they were used to characterize the possible severity and consequences of the unknown risk. Second, they were applied to show that similar risks had been tolerated before. Third, the associations served to indicate that risky activities could also have benefits. Fourth, they illustrated that previous risks had also been resolved. Thus, depending on the goal, risk communication can use risk associations to promote different effects. For example, if risk managers want to emphasize the severity of the unknown risk, they should associate it with risky activities that are known for their severe consequences. The mental model approach is useful to identify which risk associations result in which effects.
Associated risks may be applied in risk communication in two different ways, in both of which the mental models approach can be important. The first possible form of risk communication focuses on individuals or small groups of people at risk. These individuals and groups can be identified by field workers (e.g., physicians, policymakers, and risk managers) as “risk populations.” The members of the group at risk fill in a detailed questionnaire about their knowledge, attitudes, beliefs, and behavior concerning the risk, which are then used to construct a personal mental model (e.g., Reference 48). Next, all individuals or small groups receive a risk communication that is tailored to their specific mental model. According to construction-integration theory, this would optimize comprehension of the message. A study about tailored (computer-based) risk communication to prevent colorectal cancer showed that respondents who had received tailored messages had a more accurate risk perception and were more likely to correct their misperceptions about their personal risk than the respondents who had received messages without personalized information.(49)
However, this method is expensive and labor-intensive. In addition, the results of our second study showed that people might not need a personal risk association; risk associations made by others can also be related to unknown risks. Thus, tailored interventions should only be used when the target subgroup is expected to be difficult to reach by mass risk communication or differs considerably from the general population.
The second possible method of risk communication uses the mental models approach on a sample of the population and, based on its results, develops a communication with associated risks for the entire population. This type of risk communication takes the form of mass communication and might be less effective in informing the public than an individually tailored communication, since it is uncertain whether every member of the population understands it and takes it seriously. Nevertheless, this method is much cheaper and less labor-intensive. We recommend the mass communication approach when the risk is easy to understand and when no problems are expected in reaching the public.
To conclude, these studies showed that associated risks can assume a prominent, spontaneous role when people respond to an unknown risk or interpret a risk communication: people often associate unknown risks with known risks. The qualitative study also revealed four different reasons why risk associations are made. Moreover, the semantic category of the unknown risk was found to determine people's choice of associated risks, which can be explained by a semantic network theory. In short, our findings support the implicit assumption of the construction-integration theory, the mental models approach, and the representativeness heuristic that prior knowledge (and thus associated risks) is important in the way the general public forms risk perceptions.
