Are perceived prevalences of infection also biased and how? Lessons from large epidemics of mosquito-borne diseases in tropical regions Authors and affiliations: Jocelyn Raude1,2,3* Jean-Baptiste Combes1,4 Patrick Peretti-Watel5,6 Jeremy Ward5,6 Claude Flamand7 Pierre Verger5,6,8 1) EHESP Rennes, Université Sorbonne Paris Cité, France; 2) Aix Marseille University, IRD French Institute of Research for Development, EHESP French School of Public Health, UMR_D 190 “Emergence des Pathologies Virales”, Marseille, France; 3) UMR PIMIT, INSERM 1187, CNRS 9192, IRD 249. Plateforme Technologique CYROI, Université de La Réunion, Réunion, France 4) CNRS, UMR 6051 Arènes-CRAPE Centre de Recherches sur l'Action Politique en Europe, France 5) INSERM, UMR912 "Economics and Social Sciences Applied to Health & Analysis of Medical Information" (SESSTIM), 13006, Marseille, France; 6) Aix Marseille University, UMR_S912, IRD, 13006, Marseille, France; 7) Institut Pasteur de Guyane, Unité d’Epidémiologie, Cayenne, France; 8) ORS PACA, Southeastern Health Regional Observatory, 13006, Marseille, France. 1 *Corresponding author: Jocelyn Raude, PhD, Department of Social and Behavioral Sciences, EHESP French School of Public Health, 15 Avenue du Professeur Leon-Bernard, CS 74312, 35043 Rennes Cedex, France. Tel.: +33 029 902 2615 Email: [email protected] Financial support: Financial support for this study was provided in part by grants from the Health Regional Agencies of Martinique and Guyane, and the French Institute of Microbiology and Infectious Diseases (IMMI). The funding agreement ensured the authors’ independence in designing the study, interpreting the data, writing, and publishing the report. 2 ABSTRACT Objectives: Although people have been repeatedly found to underestimate the frequency of risks to health from common diseases, we still do not know much about reasons for this systematic bias, which is also referred to as “primary bias” in the literature. In this study, we take advantage of a series of large epidemics of mosquito-borne diseases to examine the accuracy of judgments of risk frequencies. In this aim, we assessed the perceived versus the observed prevalence of infection by zika, chikungunya or dengue fever during these outbreaks, as well as their variations among different subpopulations and epidemiological settings. Design: We used data drawn from 4 telephone surveys, conducted between 2006 and 2016, among representative samples of the adult population in tropical regions (Reunion, Martinique, and French Guiana). The participants were asked to estimate the prevalence of these infections by using a natural frequency scale. Results: The surveys showed that (1) most people greatly overestimated the prevalence of infection by arbovirus, (2) these risk overestimations fell considerably as the actual prevalence of these diseases increased, (3) the better-educated and male participants consistently yielded less inaccurate risk estimates across epidemics, and (4) that these biases in the perception of prevalence of these infectious diseases are relatively well predicted by probability weighting function. Conclusions: These findings suggest that the cognitive biases that affect perception of prevalence of acute infectious diseases are not fundamentally different from those that characterize other types of probabilistic judgments observed in the field of behavioral decision-making. They also indicate that numeracy may play a considerable role in people’s ability to transform epidemiological observations from their social environment to more accurate risk estimates. 3 INTRODUCTION The ability for human beings to adequately detect and estimate a variety of risks in their social and natural environments can be undoubtedly considered as an essential cognitive function (1). Indeed, inferences about what occurs in the world are critical to decision making, enabling us to adopt adaptive behaviors in order to avoid harm and preserve health and life. Over the last decades, an extensive body of research has been devoted to the study of risk perception among humans (2). Several competing conceptualizations and measures of perceived risk have been proposed in this the literature–which are undoubtedly due to the intrinsic polysemy of the concept of risk within and across disciplines (3-4). Nevertheless, a trend can be identified within health psychology towards the employment of a somewhat narrower definition, which encompasses perceived likelihood, vulnerability, or susceptibility to harm, such as acquiring a specific disease (5-6). Using this definition, perceived risk refers mainly to the judgments that people make about their personal risk of experiencing harm. Such judgments have consistently been found to motivate people to undertake health protective measures against a vast array of diseases, such as cancer screening or influenza vaccinations (7-9). However, given the difficulty of translating population-based estimates of health threats drawn from large epidemiological or clinical studies into precise and reliable assessments of personal risk or probability of developing a disease, it remains highly problematic to determine whether such risk judgments are actually accurate or even unrealistic. As noted by Leventhal concerning cancer risk (10:81): “Neither the epidemiological nor the biological data alone or in combination are sufficient to answer the clinician’s or patient’s question: Precisely what is the probability that (this specific patient/I) will contract a particular type of cancer at a given point in 4 time.” Moreover, many statisticians and philosophers assume that probabilistic rules do not apply to single events (11). In response to these legitimate criticisms, several authors adopted from the late 70ies a frequentist perspective to approach and assess health-related risk perception (12-13). Basically, the frequentist approach consists of asking people to estimate the probability or frequency of harmful events in specific populations within a given time frame. From this perspective, risk judgments will be considered accurate (or unbiased) if mean perceived frequencies correspond to the actual or estimated objective frequencies of harmful events. In this article, we took advantage of a series of naturally occurring risk events – major outbreaks of mosquito-borne epidemics in tropical regions – to examine the conditions under which perceived prevalence of some acute illnesses is more or less biased. Since the early stages of research on risk perception, it has been shown that people seem to display a range of recurrent and substantial biases when asked to think about the probability or frequency of adverse events. It should be noted however that there remains a strong argument as to whether such biases in judgment of risk frequencies should be attributed to methodological/experimental artefacts or poor cognitive abilities, including limited memory and computational capacities (14). The main research questions we want to address here, through the examination of the data collected within large samples of tropical population in a variety of epidemic settings are the following: Firstly, do cognitive biases that have been observed in a series of laboratory experiments on judgments about the relative frequency of a class of events also exist in a more ecologically realistic situation, in which people are asked to estimate the prevalence of highly visible and widespread illnesses? Secondly, are cognitive biases in judgments about the frequency of some events, which have been documented in the experimental psychology literature, specific to WEIRD (Western, Educated, Industrialized, Rich and Democratic) people or are they potentially 5 universal across sociocultural groups? Given the extensive body of literature that exists on the socioeconomic and sociocultural variations in non-numerical judgments about a range of health risks (15-17), it appears reasonable to assume that there may be significant differences among individuals and groups regarding the nature and magnitude of cognitive biases that affect numerical judgments about health risks, such as people’s estimates of the prevalence of some common conditions in a population. Finally, how reliable is the probability weighting model developed in the field of psychology of judgment and decision-making in predicting the perceived prevalence of these infectious diseases? Theoretical background In a series of famous laboratory experiments aiming to test the basic assumptions of the standard theory of decision making under risk, cognitive psychologists and behavioral economists have shown that probabilistic judgments are subject to numerous biases (e.g., unrealistic optimism or base rate neglect). These are attributable to a range of cognitive heuristics we are likely to use to estimate the number or probability of occurrences, such as the availability, representativeness or affect heuristics (18). In particular, it has been repeatedly found that people are likely to underestimate more frequent events and conversely, to overestimate less frequent events. As applied to epidemiological events, this phenomenon of miscalibration has been called “primary bias” in the psychological literature (14, 19). During the late eighties, a more fine-grained analysis of this cognitive bias has shown that human perception of probability or relative frequency is generally characterized by a series of remarkable properties (20). Primarily, probabilistic judgments are found to be regressive, i.e., there exists an 6 inflexion point above which the perceived frequencies are generally lower than the actual frequencies of adverse events (and the reverse, in which perceived frequencies are generally higher than actual frequencies). Secondly, probabilistic judgments are consistently found to be asymmetrical, i.e., there exists a fixed point at which the perceived frequencies equal the actual frequencies of adverse events. Thirdly, the probabilistic judgments tend to be s-shaped, i.e., the perceived risk frequencies are concave until the above-mentioned inflexion point, after which it is then convex. These empirical findings have led to a model for the perception of risk probability in behavioral economic literature using non-linear functions of transformation of objective probabilities, which show that the smaller frequencies are overweighted and the larger frequencies are underweighted by individuals. According to Prelec (21: 497), these subproportional probability weighting functions are generally “regressive, s-shaped and with a fixed point and invariant inflection point at 1/e = .37.” These research results have nevertheless raised a number of criticisms among psychologists and behavioral scientists. Most of the underlying data has been collected from small, convenient, and non-representative samples of undergraduate students from WEIRD populations so that we can hardly generalize these results to the population as a whole. In a recent meta-analysis, Heinrich et al. (22) have notably shown that about 80% of study participants in the top psychology journals were recruited among undergraduate students, and that about 96% of all psychology samples came from countries that account for only 12% of the world’s population. Furthermore, most of the cognitive biases have been found to disappear in more ecologically realistic conditions by using relative frequencies rather than statistical probability in laboratory experiments (11, 23). Finally, early extensions of research into risk perception to the area of health and illness has led to the collection of contradictory evidence 7 (24-25). In a famous experimental study conducted in the late seventies, Lichtenstein et al (12) found that people tend to misjudge the absolute frequencies of deaths from various biomedical causes. Their subjects were likely to overestimate the small frequencies and underestimate large ones, such as those related to cancer or cardiovascular diseases. It should be noted that this study was successfully replicated by Hakes and Viscusi (13) among a non-representative sample of U.S. citizens, then by Hertwig et al. (26) among a convenience sample of German students. However, when other researchers came to examine the perceived risks related to leading causes of morbidity and mortality by asking questions about their relative frequencies among larger samples of WEIRD individuals, it turned out that subjects were likely to substantially overestimate some of the more relatively common health-related risks, such as those associated with cigarette smoking (27-28), diabetes or hypertension (29). In other words, introducing questions about the relative frequencies of harmful events, rather than their absolute frequencies, seems to lead to inconsistent or even contradictory results in risk perception research. To the best of our knowledge, the capacity of probability weighting models, such as those developed from experimental works conducted by Tversky and Kahneman in the field of behavioral decision-making, to represent and predict adequately such cognitive biases in more ecologically realistic conditions has not to date been seriously investigated (31-32). Furthermore, we still do not know whether the cognitive biases observed in the judgments of probability or frequency are either potentially universal, or specific to WEIRD people. However, risk perception research has shown consistent differences between white/richer/better educated men and non-white/less-rich/less well-educated women in their numerical and non-numerical judgments on a variety of health-related risks (13, 16, 33, 34). Nevertheless, one of the main practical difficulties we have identified in achieving this objective in the health domain is that 8 both communicable and chronic diseases overtly affecting large segments of population (>30%) are rather rare. For instance, cancers and cardiovascular diseases are estimated to affect about 5% and 28% of the French adult population, respectively (35). Setting To further examine whether and how the perceived prevalence of common diseases may be biased, we took advantage of four major epidemics of mosquito-borne diseases which recently occurred in three tropical areas. The first is a large-scaled outbreak of chikungunya fever that occurred on Reunion Island in 2005/06. Reunion Island is a French-speaking tropical territory located in the Indian Ocean, where about 800,000 people of various regional origins live. From March 2005 to February 2006, thousands of cases of chikungunya infection were reported by public health authorities from countries located in the Southwest Indian Ocean. At the end of the epidemic, it was estimated, based upon a large sero-prevalence survey conducted in Reunion Island, that the disease would have infected between 35.9% and 40.5% of the population, with a sudden peak in the number of cases in the winter of 2005/06 (36). The second epidemic was a major outbreak of dengue fever that mostly occurred in 2010 on La Martinique Island, which is a French-speaking territories located in the West Indies. According to the last census, about 380,000 inhabitants permanently reside on La Martinique Island. In October 2010, the clinical incidence of disease that occurred during this epidemic was estimated at about 8.6%–10.4% (with a 95% CI) on the basis of a population-based telephone survey that we performed among 1,006 households. It should be noted that this estimate was largely congruent with the 10% attack rate calculated from the epidemiological data collected by the regional surveillance system (37). The third and fourth epidemics were major outbreaks of chikungunya, then zika disease that 9 hit the population of French Guiana (about 250,000 inhabitants), as well as those of the other Latin American and Caribbean countries in 2014 and 2015/16, respectively. At the time of our surveys, there was no official estimation of the chikungunya and zika attack rate in the region. However, the number of symptomatic cases reported by participants within their households allows us to estimate the clinical prevalence of the diseases at about 12.2%–14.6% of the population, with a 95% CI for chikungunya, and at about 7.1%–8.9% of the population for zika infection. Like zika, chikungunya, and dengue fever are infectious diseases transmitted by bites of infected Aedes mosquitoes – also known as Asian tiger mosquitoes because of their distinctive black and white stripes on their body – and are a prominent species of mosquito in tropical zones. Typical symptoms of these mosquito-borne diseases are similar, since they include sudden fever, headache, fatigue, muscle and joint pains, and skin rash. There are, however, some considerable clinical differences between the diseases, as most people infected by the chikungunya virus have been shown to develop symptoms, whereas dengue fever and zika infections are generally asymptomatic (38-41). Given the substantial proportion of the populations that was affected by these mosquito-borne disease outbreaks, it is expected that most people in our surveys would have experienced the disease, either directly through the observation and discussion of symptomatic cases of these illnesses which occurred in their social circles, or indirectly through the intensive media coverage by regional newspapers, and television. Thus, these three massive epidemics of tropical mosquito-borne diseases may be used as a series of natural field experiments that enable us to examine further how accurately people judge the frequency of harmful events. 10 MATERIAL AND METHODS Participants and Procedure Participants in these studies were adult individuals who were recruited by telephone during major outbreaks of mosquito-borne diseases in the above-mentioned tropical French-speaking regions between 2006 and 2016. A random digit dialling procedure was used to contact the potential informants. Next, a proportional quota sampling method based on gender, age, and occupation was implemented to ensure that the samples were representative of the general population. The interviews were conducted in French or Creole by professional interviewers of a regional survey company (IPSOS). The methodological details of these four studies are displayed in Table 1, including research designs, sample sizes, and maximum margins of error (with a confidence level of 95%) for the participants’ frequency estimates. The surveys were carried out in accordance with the recommendations of the National Data Protection Authority (CNIL), which is responsible for the ethical issue and protection of individual data collected in France. All subjects gave oral informed consent in accordance with the Declaration of Helsinki. [Insert Table 1 here] Measures Sociodemographic and illness-related variables Sociodemographic and illness-related characteristics of the respondents were systematically collected. These included gender, sex, age, education, occupation, household income, location, housing conditions, personal history of mosquito-borne diseases (dates, type of mosquito-borne diseases), and details about the diagnosis and treatment. The participants were 11 also asked about the number of cases of chikungunya (or dengue fever) which occurred within their household during the previous epidemic. Perceived risk of contracting zika, dengue or chikungunya To date, there is still no agreement among scientists on how perceived health risk should be measured in empirical research (8). As noted by Viscusi (27: 1256): “Obtaining meaningful survey responses regarding individuals' risk perceptions is not a straightforward task”. This methodological problem can also be attributed to the fact that elicitation methods are more or less natural and often require considerable cognitive efforts from individuals. As mentioned above, there also exist substantial variations in the definition and measurement of health risks (4). For epidemiologists, a common method to appraise health risk is to assess the incidence (frequency estimate of harmful events within a given time frame) or prevalence (cumulative frequency estimate at a certain time point) of an illness or disability in a given population. Epidemiological information about the relative frequency of an illness can be interchangeably expressed in terms of fractions, odds or percentages. Nevertheless, people’s understanding of these numerical expressions has been found to pose a variety of difficulties in research on risk perception (42-44). To address this issue, participants can be asked to estimate the prevalence of mosquito-borne diseases by using a natural frequency scale based on a reference sample of 100 individuals, which allows easy transformation of their judgements to a percentage scale. Even though subjects in laboratory experiments were shown to have some difficulties using and understanding numerical information, there is sound evidence that the utilisation of relative frequencies to elicit probabilistic judgments is a better method than directly dealing with probabilities or fractions (11). Furthermore, ordinal or discrete rating scales (e.g., 1-5 or 0-10 scales), which are widely used in the risk perception literature, were found to be inappropriate 12 for measuring probabilities or frequencies (45). In our surveys, the format and phrasing of the questions were adapted from the item developed by Viscusi (27): “Among 100 people living in (Region), how many do you think have been infected with (Disease) since the beginning of the epidemic? Please give a value between 0 and 100”. Thus, the percentage obtained through this method provides a yardstick to evaluate the accuracy of judgments that people express about the relative frequency of particular illnesses (the perceived risk) by comparing them with the estimates of “objective” risk based on epidemiological or survey data (the observed risk). Statistical Analysis Arithmetic means and 95% confidence intervals of the estimates of relative frequencies of chikungunya or dengue fever were calculated for each sociodemographic and illness-related variable. To compare statistical differences in mean risk estimations among various categories of respondents, t-tests or analyses of variance were calculated across epidemiological contexts. Generalized linear models (GLM) were then used to determine the multivariate associations between sociodemographic characteristics, health status and risk estimation in the four studies. The sociodemographic and illness-related variables were dichotomized, when appropriate, to reduce the potential instability of these models. All the data collected in the surveys were treated and analyzed using IBM SPSS 19.0 (Statistical Package for the Social Sciences) and R 3.3.1 (with R Studio 0.99). To examine the relationship between the perceived (subjective) and observed (objective) prevalence of the diseases, we used one of the most popular probability weighting models developed in the field of behavioural decision-making (21), when compared with more conventional predictive models. As noted in the introduction, behavioral studies of judgment and 13 decision-making under risk have demonstrated through a series of laboratory experiments that the relationship between subjective and objective probabilities of events can be adequately described by a nonlinear transformation of the relative frequencies function: w(p) = exp(–β(–ln p)α, where (0, 0) ≤ (α, β) ≤ (1, 1) ) This subproportional, regressive and s-shaped function implies that there exists an inflection point under which the deviation between the perceived and observed prevalence of the mosquitoborne diseases gets smaller as actual prevalence increases (and vice versa). The parameters of the equation for the judgements of relative frequency of infection across epidemiological settings can be estimated with double log coordinate using simple ordinary least squares (21). We estimate the following Equation 1: w(p) = exp(β(log(p)α) (Equation 1) We change the variable log(p) = x and get the Equation 2 w(p) = exp(βxᾱ) (Equation 2) of which we take the ln and multiply by minus 1, log(w(p)) = βxᾱ, and change the variable log(w(p)) = y, we take the log again and get Equation 3: log(y) = log(β) + αlog(x) (Equation 3) Equation 3 can easily be estimated by Ordinary Least Square (OLS). The estimation is simple enough, as long as values are strictly positive before taking the logarithm. We first estimate Equation 3 by using standard GLM function in R. Then we perform an estimation of equation 1 using a Nelder and Mead (46) optimization which is available through the optim function in R. Results differ, and based on Mean Squared Prediction Error (MSPE), we prefer the Nelder-Mead estimation compared with OLS. MSPE compares the true values for 14 w(p), with the estimated values for the latter based on the estimation of α and β . MSPE gives a standard statistic to assess the capability of the model to provide correct estimations (47). Comparing the MSPE with the OLS and the Nelder-Mead estimation enables a comparison of models. These models also allow determining whether different subgroups of a population follow the same perceptive curve. RESULTS Are there differences in perceived prevalence among subpopulations? All the populations were found to have greatly overestimated the prevalence of diseases regardless of the epidemiological setting. Indeed, the magnitude of the epidemics was significantly overestimated by an average of 16 percentage points in La Reunion (t = - 22.1; p < 0.001), 26 percentage points in La Martinique (t = - 32.1; p < 0.001), 26 percentage points (chikungunya epidemic) (t = - 24.4; p < 0.001), and 23 percentage points (zika epidemic) (t = - 31.9; p < 0.001) in French Guiana. However, as shown in table 2, there were considerable statistical differences in the average overestimations, depending on the sociodemographic and illness-related characteristics of the respondents. With the notable exception of age in study 1, all the sociodemographic variables under investigation – such as household income, level of education, gender, age, and history of infection – were found to influence significantly the way in which individuals estimated the proportion of people who have been infected at the time of the survey (p < 0.01) across regions. These results suggest that social and cultural conditions may play an important role in the magnitude of the biases that have been repeatedly observed when people are asked to estimate the prevalence of a health disorder. [Insert Table 2 here] 15 As the sociodemographic and health-related variables are known to be strongly intercorrelated, a multivariate analysis using generalized linear models was performed in order to assess the robustness of the influence of each of these personal factors on the perceived prevalence, independent of other factors. As shown in table 3, the results of the multivariate analyses were relatively consistent with those from earlier bivariate analyses. However, it should be noted that only gender and the level of education of the participants were found to affect directly judgments about the relative frequency of the diseases across regions and epidemiological contexts (p < 0.001). Therefore, after adjusting for potential confounding variables, female participants were found to overestimate the prevalence of the diseases by an average of 6 to 11 percentage points more than their male counterparts, while the most educated participants tended to overestimate them on average by 7 to 15 percentage points less than the other participants. [Insert Table 3 here] What is the relationship between perceived and observed prevalence of the diseases? The mean percentage deviation – which measures biases in perceived prevalence relative to the actual prevalence of the disease – was statistically significant for each of the 4 studies (p < 0.001). However, the average overestimations calculated for each group show large differences, depending on the magnitude of the epidemic. As shown in figure 1, the prevalence estimations provided by the participants are closer to the line of identity, which represents the correct prevalence estimation, as the magnitude of the epidemic increases. Thus, the overestimation bias among participants was significantly higher for the “smaller” epidemics (from +284% to +182% in studies 2 and 4, respectively) than for the largest epidemic (+43% in study 1). In other words, the perceived prevalence tended to increase more slowly than the observed prevalence. [Insert Figure 1 here] 16 To examine the relationship between estimated and actual prevalence, we used the probability weighting function developed by Prelec (1998). The best fitting model (using Error! Reference source not found.) is w(p) = exp(–0.641(–ln(p))0.616. These parameters enable to estimate the inflexion point, i.e., the point at which perceived and actual prevalences are equal, at 73.0%. That means that below a prevalence threshold of 73%, individuals tend to overestimate the actual prevalence (and vice-versa). For the male participants, the best fitting model is w(p) = exp(–0.732(– ln(p))0.619, with an inflexion point at 64.4%, whereas that of female participants is w(p) = exp(– 0.562(–ln(p))0.664, with an inflexion point at 83.5%. For the best educated participants, the best fitting model is w(p) = exp(–0.837(–ln(p))0.468, with an inflexion point at 48.9%, whereas that of less educated participants is w(p) = exp(–0.586(–ln(p))0.642, with an inflexion point at 79.9%. Graphically, this relationship is shown in Figure 1 by the asymmetric s-shaped curves representing the perceived prevalence predicted by the probability weighting model. The results are displayed separately for the male and female participants, and for the most educated and the less educated participants in Figure 1. After the prevalence estimations were averaged across different subgroups of the population, arithmetic means for each epidemiological setting fell fairly close to the predicted curves. To test further the hypothesis that perceived prevalences can be adequately represented by a subproportional, regressive, and s-shaped function of observed prevalence, we compared the fit of the probability weighting model with the fit of a linear model (w(p) = a + n × p) and a power model (ln(w(p)) = ln(a’) + n’ × ln(p)). For the sake of comparison, all models were estimated by using ordinary least squares (OLS). The probability weighting model was found to fit the data drawn from the surveys better than the other models (R² = 13.0%, t = 19.4, P < 0.001, versus R² = 12.6%, t = 22.0, P < 0.001 for the linear model and R² = 11.8%, t = 22.2, P < 0.001 for the power model). 17 DISCUSSION During the last decade, we took advantage of a series of large epidemics of mosquitoborne diseases as a natural experiment to test empirically the hypothesis that there exists a primary bias in the way people perceive the risk of infection. Indeed, since the pioneering studies of Lichtenstein et al. (12), the accuracy of people’s estimates of health-related risk frequencies has been the subject of intense and stimulating discussions among psychological and behavioural scientists (14). Today, there is sound evidence, based on a series of laboratory experiments, that people from WEIRD societies tend to overestimate the absolute frequency of rare epidemiological events such as deaths caused by botulism or measles. In contrast, they are likely to underestimate the incidence of frequent epidemiological events such as mortality attributable to lung cancer or cardiovascular diseases. This evidence is convergent with the well-established psychophysical law according to which individuals have the tendency to underestimate large magnitudes in volume, height, weight or even calories (48). Nevertheless, subsequent studies involving the general population have shown that a number of common health risks (e.g. those related to cigarette or alcohol consumption) were substantially overassessed by the public when they were asked about their relative frequency (27-28, 49). Consistent with these previous empirical analyses, we found in our surveys that (1) people from a variety of French tropical regions substantially overestimated the prevalence of infection by arbovirus, and (2) these risk overestimations fell considerably as the actual prevalence of these diseases increased. The strong convergence of the results of our studies with a range of investigation conducted in the domain of chronic diseases indicates that this deviation may be robust, regardless of the type of health threat and/or population. Furthermore, we found that 18 probability weighting functions better predicted the magnitude of biases observed in perceptions of prevalence of these infectious diseases than did more ordinary predictive models, such as a linear regression model. This is a striking result since this cognitive bias in probabilistic judgments was experimentally elicited through indirect methods, using gambles and lotteries as a model to investigate judgments or decisions made under uncertainty. Overall, this empirical evidence as well as that from the previous afore-mentioned studies, demonstrates that there is no systematic underestimation in the perception of prevalence of the most common health threats in both WEIRD and non-WEIRD population samples, since neither arboviral infection nor lung cancer could be considered to be an infrequent epidemiological event. Therefore, these inconsistent findings regarding risk magnitude judgments should lead us to question whether the primary bias is an artefact triggered by particular experimental conditions, as its existence seems to depend to a large extent on the method used to elicit and measure it. To date, many researchers in the field of risk perception have ignored some of the methodological and conceptual distinctions – for instance, absolute frequency versus relative frequency – which are essential in determining the accuracy of risk beliefs, as well as in characterizing the magnitude and pattern of the cognitive bias observed in their studies. However, there is now a crucial need for discussion and clarification of the effects induced by the concepts and methods employed to investigate risk frequency judgements. As discussed by Gigerenzer (11:83), by using certain frequency formats rather than others to elicit probabilistic judgments, “we can make apparently stable errors disappear, reappear, or even invert”. Where risk perception is concerned, it seems highly plausible that people underestimate relatively common health risks when they are questioned about their absolute frequencies, while they overestimate them when they are asked about their relative frequencies, based on a reference sample of 100 or 1,000 individuals. 19 Furthermore, little is known about the possible social and cultural variations in estimating the frequency or prevalence of diseases, especially in the context of an outbreak (13). Our surveys show that all the groups of participants have a tendency to overestimate the prevalence of infection by arbovirus. Nevertheless, and consistent with previous studies dedicated to risk magnitude judgments, we found systematic differences among subgroups of the population in the way people perceive these health risks. In particular, when other sociodemographic factors were held constant, better-educated and male subjects consistently exhibited far fewer inaccurate risk frequency judgments across epidemics than did their less-educated and female counterparts. There is considerable evidence that male and better-educated individuals are less subject to measurement errors when they have to estimate risk frequency (13). However, we do not know much about what drives these systematically larger disproportionate biases in health risk perception among females and less-educated individuals. In our opinion, formal education and gender could be a proxy for group differences in numeracy skills, which refers to person’s capacity to correctly understand, process and use numeric information to form judgments and make decisions (44). Indeed, it has been consistently shown that highly-numerate individuals are generally less subject to a range of errors when asked to translate some basic mathematic or probabilistic expressions, such as proportions or percentages, into meaningful information about risks (50). Therefore, it might be reasonable to assume that the ability to transform epidemiological information (e.g., observations of cases of chikungunya, zika or dengue fever) arising from the social environment into accurate judgments about their relative frequencies relies to a large extent on the numeracy skills that an individual might have developed over his or her life course. CONCLUSION 20 To conclude, it should be noted that the pattern of perceived prevalence found across our surveys is relatively compatible with the probability weighting models developed in the field of psychology of judgment and decision-making (20), which assumes a nonlinear transformation of the probability scale among decision-makers (with an overweighting of probabilities lower than about 40% and conversely). This is particularly true if one considers that the potential inflexion point between overestimation and underestimation of the perceived prevalence decreases as the level of formal education of the participants increases. Nevertheless, given that a vast majority of western, educated, and rich individuals were enrolled in laboratory experiments devoted to heuristics and cognitive bias (22), these empirical findings provide the opportunity to refine the models of cognitive processes that are proposed in the research literature to explain and predict the emergence of these biases. Last but by no means least, even though scientists have proposed a range of possible explanations for gender and education gaps in health-related risk perceptions, the cause of these sociodemographic and cultural variations in the magnitude of this bias remains unclear and, therefore, there is still a need for a better understanding of why some subgroups of the population make less inaccurate risk frequency estimates than others. 21 REFERENCES 1. Sharot T. The optimism bias. Current Biology. 2011; 21(23): 941-45. R941-R945. doi: 10.1111/j.1539-6924.1993.tb00741.x 2. Bodemer N, Gaissmaier W. Risk perception. In: Cho H, Reimer TO, McComas KA, eds. The SAGE handbook of risk communication. Thousand Oaks, CA: Sage; 2015. p 10-23. 3. Slovic P. Are trivial risks the greatest risks of all? Journal of Risk Research. 1999; 2(4): 28188. Available from: URL: http://dx.doi.org/10.1080/136698799376727 . Accessed 10 March 2017. 4. Althaus CE. A disciplinary perspective on the epistemological status of risk. Risk Analysis. 2005;25(3):567-88. doi:10.1111/j.1539-6924.2005.00625.x 5. Weinstein ND. Perceived probability, perceived severity, and health-protective behavior. Health Psychology. 2000;19(1):65. doi: 10.1037/0278-6133.19.1.65 6. Brewer NT, Weinstein ND, Cuite CL, Herrington JE. Risk perceptions and their relation to risk behaviour. Annals of Behavioral Medicine. 2004;27(2):125-30. 7. Peters E, McCaul KD, Stefanek M, Nelson W. A heuristics approach to understanding cancer risk perception: Contributions from judgment and decision-making research. Annals of Behavioral Medicine. 2006;31(1):45-52. doi: 10.1207/s15324796abm3101_8 8. Weinstein ND, Kwitel A, McCaul KD, Magnan RE, Gerrard M, Gibbons FX. Risk perceptions: Assessment and relationship to influenza vaccination. Health Psychology. 2007;26(2):146. doi: 10.1037/0278-6133.26.2.146 22 9. Sheeran P, Harris PR, Epton T. Does heightening risk appraisals change people’s intentions and behavior? A meta-analysis of experimental studies. Psychological Bulletin. 2014;140(2):511. doi: 10.1037/a0033065 10. Leventhal H, Kelly K, Leventhal EA. Population risk, actual risk, perceived risk, and cancer control: A discussion. Journal of the National Cancer Institute. Monographs. 1999;25:81-5. 11. Gigerenzer G. How to make cognitive illusions disappear: Beyond “heuristics and biases”. European Review of Social Psychology. 1991;2(1):83-115. Available from: URL: http://dx.doi.org/10.1080/14792779143000033 Accessed 10 March 2017. 12. Lichtenstein S, Slovic P, Fischhoff B, Layman M, Combs B. Judged frequency of lethal events. Journal of Experimental Psychology: Human Learning and Memory. 1978;4:551–78. Available from: URL: http://dx.doi.org/10.1037/0278-7393.4.6.551 Accessed 10 March 2017. 13. Hakes JK, Viscusi WK. Dead reckoning: Demographic determinants of the accuracy of mortality risk perceptions. Risk Analysis. 2004;24(3):651-64. doi: 10.1111/j.02724332.2004.00465.x 14. Gigerenzer G, Fiedler K, Olsson H. Rethinking cognitive biases as environmental consequences. In: Todd PM, Gigerenzer G, eds. Ecological rationality: Intelligence in the world. Oxford, England: Oxford University Press; 2012. p 80-110. 15. Savage I. Demographic influences on risk perceptions. Risk analysis. 1993;13(4):413-20. 16. Finucane ML, Slovic P, Mertz CK, Flynn J, Satterfield TA. Gender, race, and perceived risk: The 'white male' effect. Health, Risk & Society. 2000;2(2):159-72. Available from: URL: http://dx.doi.org/10.1080/713670162 Accessed 10 March 2017. 17. Kpanake L, Chauvin B, Mullet E. Societal risk perception among African villagers without access to the media. Risk Analysis. 2008;28(1):193-202. doi: 10.1111/j.1539-6924.2008.01008.x 23 18. Gilovich, T, Griffin D, Kahneman D. Heuristics and biases: The psychology of intuitive judgment. Cambridge, England: Cambridge University Press; 2002. 19. Johnson EJ, Tversky A. Affect, generalization, and the perception of risk. Journal of Personality and Social Psychology. 1983;45(1):20. doi: 10.1037/0022-3514.45.1.20 20. Tversky A, Kahneman D. Advances in prospect theory: Cumulative representation of uncertainty. Journal of Risk and Uncertainty. 1992;5(4):297-323. doi: 10.1007/BF00122574 21. Prelec D. The Probability Weighting Function. Econometrica. 1998;66(3):497-527. doi: 10.2307/2998573 22. Henrich J, Heine SJ, Norenzayan A. The weirdest people in the world? Behavioral and Brain Sciences. 2010;33(2-3):61-83. doi: 10.1017/S0140525X0999152X 23. Cosmides L, Tooby J. Are humans good intuitive statisticians after all? Rethinking some conclusions from the literature on judgment under uncertainty. Cognition. 1996;58(1): 1-73. Available from: URL: http://dx.doi.org/10.1016/0010-0277(95)00664-8 Accessed 10 March 2017. 24. Weinstein ND. Accuracy of smokers' risk perceptions. Annals of Behavioral Medicine. 1998;20(2):135-40. 25. Sutton SR. How accurate are smokers’ perceptions of risk? Health, Risk & Society. 1999;1(2):223-30. doi: 10.1080/13698579908407020 26. Hertwig R, Pachur T, Kurzenhäuser S. Judgments of risk frequencies: Tests of possible cognitive mechanisms. Journal of Experimental Psychology: Learning, Memory, and Cognition. 2005;31(4):621. doi:10.1037/0278-7393.31.4.621 27. Viscusi WK. Do smokers underestimate risks? Journal of Political Economy. 1990;98(6):1253-69. doi: 10.1086/261733 24 28. Lundborg P, Lindgren B. Do they know what they are doing? Risk perceptions and smoking behaviour among Swedish teenagers. Journal of Risk and Uncertainty. 2004;28(3):261-86. doi: 10.1023/b:risk.0000026098.84109.62 29. Frijling BD, Lobo CM, Keus IM, Jenks KM, Akkermans RP, Hulscher ME, Grol RP. Perceptions of cardiovascular risk among patients with hypertension or diabetes. Patient Education and Counseling. 2004;52(1):47-53. Available from: URL: http://dx.doi.org/10.1016/S0738-3991(02)00248-3 Accessed 10 March 2017. 30. Weinstein ND. Exploring the Links Between Risk Perceptions and Preventive Health Behavior. In: Suls J. ed. Social Psychological Foundations of Health and Illness. Oxford, UK: Blackwell; 2003. p 22-53. 31. Pashler H, Harris CR. Is the replicability crisis overblown? Three arguments examined. Perspectives on Psychological Science. 2012;7:531–36. doi: 10.1177/1745691612463401 32. Ioannidis J, Doucouliagos C. What's to know about the credibility of empirical economics? Journal of Economic Surveys. 2013;27(5):997-1004. doi: 10.1111/joes.12032 33. Raude J, Fischler C, Setbon M, Flahault A. Scientist and public responses to BSE‐related risk: A comparative study. Journal of Risk Research. 2005;8(7-8):663-78. Available from: URL: http://dx.doi.org/10.1080/13669870500194825 Accessed 10 March 2017. 34. Peters E. Numeracy and the Perception and Communication of Risk. Annals of the New York Academy of Sciences. 2008;1128: 1–7. doi:10.1196/annals.1399.001 35. [DREES] Direction de la recherche, des études, de l'évaluation et des statistiques. L’état de santé de la population en France. Rapport 2015. 2015. Available from: URL : drees.socialsante.gouv.fr/IMG/pdf/rappeds_v11_16032015.pdf. Accessed 10 March 2017. 25 36. Gérardin P, Guernier V, Perrau J, Fianu A, LeRoux K, Grivard P, Michault A, de Lamballerie X, Flahault A, Favier F. Estimating Chikungunya prevalence in La Reunion Island outbreak by serosurveys: Two methods for two critical times of the epidemic. BMC Infectious Diseases. 2008;8(1):99. doi: 10.1186/1471-2334-8-99 37. Rosine J, Adélaïde Y, Anglio J, Blateau A, Bousser V, Davidas M, Ledrans M, Romagne MJ, Suivant C, Quénel P. Bilan de l'épidémie de dengue en Martinique, 2010. Bulletin de veille sanitaire. 2011;9/10:2-5. 38. Burke DS, Nisalak A, Johnson DE, Scott RM. A prospective study of dengue infections in Bangkok. The American journal of tropical medicine and hygiene. 1988;38(1):172-80. 39. Gallian P, Cabié A, Richard P, Paturel L, Charrel RN, Pastorino B, DeLamballerie X. Zika virus in asymptomatic blood donors in Martinique. Blood. 2017;129(2):263-66. Available from: URL: https://doi.org/10.1182/blood-2016-09-737981 . Accessed 10 March 2017. 40. Burt FJ, Rolph MS, Rulli NE, Mahalingam S, Heise MT. Chikungunya: a re-emerging virus. The Lancet. 2012;379(9816):662-71. Available from: URL: http://dx.doi.org/10.1016/S01406736(11)60281-X . Accessed 10 March 2017. 41. Sissoko D, Moendandze A, Malvy D, Giry C, Ezzedine K, Solet JL, Pierre V. Seroprevalence and risk factors of chikungunya virus infection in Mayotte, Indian Ocean, 2005-2006: A population-based survey. PLoS ONE. 2008;3(8):3066. doi:10.1371/journal.pone.0003066 42. Diefenbach M, Weinstein ND, O’Reilly J. Scales for assessing perceptions of personal risk susceptibility. Health Education Research. 1993;8:181–92. Available from: URL: https://doi.org/10.1093/her/8.2.181 . Accessed 10 March 2017. 26 43. Rothman AJ, Kiviniemi MT. Treating people with information: An analysis and review of approaches to communicating health risk information. Journal of the National Cancer Institute. Monographs. 1999;25:44-51. doi: 10.1093/oxfordjournals.jncimonographs.a024207 44. Peters E. Numeracy and the perception and communication of risk. Annals of the New York Academy of Sciences. 2008;1128(1):1-7. doi: 10.1196/annals.1399.001 45. Viscusi WK, Hakes J. Risk ratings that do not measure probabilities. Journal of Risk Research. 2003;6(1):23-43. Available from: URL: http://dx.doi.org/10.1080/1366987032000047789 . Accessed 10 March 2017. 46. Nelder JA, Mead R. A simplex method for function minimization. The computer journal. 1965;7(4):308-13. 47. Abadie A, Diamond A, Hainmueller J. Synthetic control methods for comparative case studies: Estimating the effect of California’s tobacco control program. Journal of the American statistical Association, 2010;105(490):493-505. 48. Chandon P, Wansink B. Is obesity caused by calorie underestimation? A psychophysical model of meal size estimation. Journal of Marketing Research. 2007;44(1):84-99. Available from: URL: http://dx.doi.org/10.1509/jmkr.44.1.84 . Accessed 10 March 2017. 49. Lundborg P, Lindgren B. Risk perceptions and alcohol consumption among young people. Journal of Risk and Uncertainty. 2002;25:165–83. doi: 10.1023/A:1020695730192 50. Reyna VF, Nelson WL, Han PK, Dieckmann NF. How numeracy influences risk comprehension and medical decision making. Psychological Bulletin. 2009;135(6):943. doi: 10.1037/a0017327 27 Table 1. Characteristics of the studies Geographical area Date Type of disease Actual prevalence (95% CI) Research design Maximum margin of error Observations (n) Population (N) Study 1 La Reunion Island April 2006 Chikungunya 37.2% (33.9%-40.5%) Cross-sectional study 3.0% 1,035 794,000 Study 2 Study 3 La Martinique Island French Guiana September 2010 April 2015 Dengue fever Chikungunya 9.5% (8.6%-10.4%) 13.9% (12.2%-14.6%) Cross-sectional study Cross-sectional study 3.1% 4.2% 1,006 553 380,000 250,000 28 Study 4 French Guiana June 2016 Zika 8.0% (7.1%-8.9%) Cross-sectional study 2.9% 1,129 250,000 Table 2. Differences between the perceived and observed prevalence of chikungunya, zika or dengue fever by sociodemographic and health characteristics in studies: arithmetic means (95% CI). Total Gender Male Female Education Some high school and lower High School graduate Some college College graduate and higher Household income Lowest Lower-middle Upper-middle Highest Age group 18 to 29 30 to 44 45 to 59 60 or older Occupation Employed or self-employed Unemployed (looking for work) Unemployed (not looking for work) Student Retired Personal history of infection Present None Number of cases in the household 0 1 2 >2 Study 1 (n=1,035) 16.0 (14.6; 17.4) *** 11.3 (9.4; 13.3) 20.2 (18.2; 22.2) *** 26.3 (23.0; 29.5) 19.2 (16.7; 21.6) 11.2 (8.4; 14.0) 6.9 (4.5; 9.3) *** 22.4 (16.8; 27.9) 21.3 (18.9; 23.6) 11.7 (9.7; 13.9) 10.1 (6.5; 13.7) NS 17.1 (14.7; 19.4) 16.1 (13.7; 18.5) 14.0 (10.6; 17.4) 16.2 (11.9; 20.5) *** 11.0 (8.8; 13.2) 20.3 (17.1; 23.5) 24.9 (21.2; 28.6) 16.9 (13.8; 19.9) 8.6 (4.0; 13.3) *** 20.7 (18.4; 22.9) 12.7 (10.8; 14.5) *** 11.7 (9.4; 14.0) 15.4 (11.3; 19.5) 16.4 (13.1; 19.7) 21.6 (19.9; 25.3) Study 2 (n=1,006) 26.2 (24.6; 27.8) *** 22.5 (20.2; 24.9) 29.2 (27.0; 31.4) * 28.8 (24.9; 32.7) 28.2 (25.0; 31.4) 26.9 (23.7; 30.1) 22.4 (19.8; 25.1) ** 29.0 (23.6; 34.4) 27.6 (24.5; 30.6) 26.4 (23.7; 29.2) 20.1 (16.8; 23.4) ** 31.2 (28.2; 34.1) 25.1 (22.3; 27.9) 26.4 (23.1; 29.7) 20.9 (17.0; 24.9) ** 25.4 (22.8; 28.1) 28.3 (24.9; 31.7) 27.2 (22.4; 32.0) 32.1 (27.6; 36.7) 21.2 (17.4; 25.0) ** 32.8 (27.8; 37.9) 25.4 (23.8; 27.1) *** 24.9 (23.0; 26.9) 31.9 (25.7; 38.2) 33.4 (27.9; 38.9) 35.7 (28.3; 43.2) *significant at 5% level, **significant at 1% level, ***significant at 1‰ level. 29 Study 3 (n=553) 25.3 (22.8; 27.8) *** 19.6 (16.3; 22.8) 31.1 (27.4; 34.8) *** 30.3 (23.7; 36.8) 31.0 (26.3; 35.9) 30.5 (25.6; 35.3) 13.5 (9.6; 17.3) *** 35.9 (31.8; 40.0) 23.9 (19.9; 27.9) 17.2 (11.2; 23.2) 9.0 (3.5; 14.6) ** 32.2 (28.0; 36.3) 25.5 (20.9; 30.0) 19.6 (14.6; 24.6) 19.6 (11.9; 27.3) *** 18.9 (15.5; 22.2) 34.6 (28.5; 40.7) 36.7 (28.9; 44.5) 31.7 (25.8; 37.7) 17.7 (9.0; 26.5) ** 39.4 (30.2; 48.6) 24.0 (21.4; 26.6) *** 22.4 (19.8; 25.0) 37.2 (28.1; 46.4) 34.1 (22.4; 45.9) 58.4 (48.0; 68.7) Study 4 (n=1,127) 22,7 (21.3; 24.1) *** 16,4 (14.2; 18.6) 26,1 (24.4; 27.9) *** 23,9 (18.6; 29.0) 26,6 (23.8; 29.4) 23,4 (20.6; 26.1) 19,2 (17.1; 21.3) *** 27.7 (24.9; 30.4) 23.7 (21.4; 25.9) 17.6 (14.9; 20.3) 17.9 (13.7; 22.0) *** 19.9 (18.1; 21.6) 29.8 (26.0; 33.6) 25,5 (21.0; 30.0) 26,9 (22.6; 31.3) 19,6 (14.5; 24.7) *** 32,4 (28.0; 36.8) 21,6 (20.1; 23.0) *** 21.5 (20.0; 23.0) 27.3 (23.6; 31.1) 32.1 (-92.0; 108) 44,4 (33.9; 55.0) Table 3. Association between the sociodemographic characteristics, health status and estimated prevalence of chikungunya, zika or dengue fever across epidemics: adjusted coefficients (95% confidence intervals) calculated from generalized linear models. Intercept Gender Male Female Education Some college and lower College graduate and higher Household income Lower-middle and lower Upper-middle and higher Age group 15 to 29 30 to 44 45 to 59 60 or older Occupation Employed or self-employed Unemployed (looking for work) Unemployed (not looking for work) Student Retired Personal history of infection None Present Number of cases in the household None One or more MLE Study 1 (n=1,035) 48.9 (44.2; 53.7)*** Study 2 (n=1,006) 35.7 (25.2; 46.3)*** Study 3 (n=553) 31.9 (26.7; 37.2)*** Study 4 (n=1,127) 22.4 (17.7; 27.1)*** Reference 6.3 (3.6; 9.1)*** Reference 8.9 (5.7; 12.2)*** Reference 10.9 (7.4; 14.4)*** Reference 8.9 (6.1; 11.8)*** Reference -8.7 (-12.1; -5.4)*** Reference -6.9 (-10.6; -3.3)*** Reference -9.6 (-14.3; -4.9)*** Reference -7.0 (-9.9; -4.0)*** Reference -4.5 (-8.4; -0.6)* Reference -6.0 (-10.2; -1.8)** Reference -4.7 (-9.5; 4.0) Reference -2.2 (-6.5; 1.9) -4.8 (-9.6; 0.1) -0.6 (-7.1; 5.9) Reference -6.9 (-11.9; -1.8)** -8.2 (-13.4; -2.9)** -11.7 (-22.3-1.2)* Reference -4.8 (-9.7; 0.2) -6.0 (-11.4; -0.6)* -3.8 (-14.4; 6.7) Reference -4.5 (-8.4; -0.6)* -10.2 (-14.5; -5.9)*** -12.0 (-19.5; -4.5)*** Reference 4.4 (0.5-8.4)* 7.1 (2.7; 11.5)** 1.3 (-3.8; 6.3) -5.6 (-12.1; 0.8) Reference 4.0 (-7.2; 15.3) 2.2 (-7.9; 12.2) 3.4 (-6.5; 13.3) 4.9 (-4.9; 14.7) Reference 10.3 (4.9; 15.6)*** 5.8 (-0.0; 11.6) 6.1 (-0.8; 12.9) -4.4 (-16.7; 7.8) Reference 4.5 (-0.7; 13.3)* 1.6 (-3.2; 6.3) 1.5 (-4.1; 7.1) 3.2 (-5.2; 11.7) Reference 5.2 (1.7-8.6)** Reference 4.1 (-1.5; 9.8) Reference 15.4 (9.3; 21.5)*** Reference 10.4 (3.0; 17.8)** Reference 2.4 (-1.2; 5.9) 448.5 Reference 3.1 (-0.5; 6.7) 502.1 Reference 3.5 (-2.4; 9.4) 514.7 Reference 1.0 (-5.4; 7.4) 448.4 *significant at 5% level, **significant at 1% level, ***significant at 1‰ level. 30 Figure 1. Predicted estimation curves, mean perceived prevalence and 95% confidence intervals by gender (top) and level of formal education (bottom). 31
© Copyright 2025 Paperzz