Rare disaster information can increase risk-taking

The recent increase in the frequency and impact of natural disasters1 highlights the need to provide the public with accurate information concerning disaster prevalence. Most approaches to this problem assume that providing summaries of the nature and scale of disasters will lead people to reduce their exposure to risk2. Herewe present experimental evidence that such ex post ‘news reports’ of disaster occurrences can increase the tolerance for risk-taking (which implies that rare events are underweighted3). This result is robust across several hundred rounds of choices in a simulated microworld, persists even when the long-run expected value of risky choices is substantially lower than safe choices, and is contingent on providing risk information about disasters that have been (personally) experienced and those that have been avoided (‘forgone’ outcomes). The results suggest that augmenting personal experience with information summaries of the number of adverse events (for example, storms, floods) in di erent regions may, paradoxically, increase the appeal of a disaster-prone region. This finding implies a need to communicate long-term trends in severe climatic events, thereby reinforcing the accumulation of events, and the increase in their associated risks, across time4. For the past 20 yearsMunichRe have surveyed the previous year’s natural catastrophes. Their most recent report1 states: ‘It is not just that the number of natural catastrophes studied over the decades has increased . . . as a result of climate change, but that the impact of these events (as anticipated) has also become much greater and more costly’. Thus, although climate change is gradual, its impact on communities is not only incremental and chronic but can also be sudden and acute because climate change alters the prevalence and severity of discrete climate-related negative events (for example, storms, floods, crop failures)5. This fact underscores the importance of understanding how people react to information about the risk of natural catastrophes. A common response to this communication problem is to assume that more information is better, and that providing descriptive summaries of risk levels will lead people to reduce their exposure to relevant risks. This approach is taken in many fields; examples include information about vehicle accidents in a given area6–8, the risk of forest fire9, flood risks10, and terrorist attacks (for example, the US Traveler Enrollment Program). Although evaluations of the response to these systems are scarce8,9, the hoped-for positive effect of summarized information does not always materialize. Several studies suggest that publicly available information summaries concerning catastrophic events sometimes have the paradoxical effect of decreasing overall risk estimates11–13. For example, in a study of residents living in an area close to, but unaffected by the 2011 Tohoku tsunami13, participants were presented with scenarios involving waves of varying heights and asked whether each requires an evacuation. Comparing responses made before and after the tsunami, these unaffected residents’ estimates of wave heights warranting an evacuation were higher after the disaster. This suggests an increased tolerance for risk following the provision of information about a disaster that one avoided. In two experiments, we investigated the (causal) effect of providing ex post information about each individual rare negative event that occurred in a simulated microworld (Fig. 1). Our participants made a choice about where to ‘live’ for each one of 400 rounds. The ‘microworld’ contained three regions, each having a village with multiple dwellings: participants could earn points for choosing to live in a particular dwelling in one of the villages; but lost many points when a catastrophe hit their dwelling. Points won represent an experimental analogue of the ‘utility’ garnered from a profitable and peaceful life in one’s chosen ‘home’, whereas points lost represent the ‘disutility’ of a life impacted by a major disaster. Throughout the experiment, all participants were given an accurate description of the ex ante risks of catastrophe in each region (Fig. 1). However by varying the type of ex post round-by-round feedback about when and where catastrophes occurred (described below), we examined how the reporting of negative events affected participants’ choice between regions that carried different levels of risk (Table 1). One region of the microworld was safe: catastrophes never occurred, but the available points on each round were modest. A second region offered more points if no disaster occurred but rare catastrophes occurred (10 in 100 rounds) which affected 9% of dwellings in the region (incurring a loss of points). A third region had very rare catastrophes (1 in 100 rounds) that were of higher impact (affecting 90% of dwellings in the region), although offering the same benefits and overall risks as the second region (that is, an individual home owner faced the same probability of disasters in both regions; Table 1). In the OwnHouse condition, participants learnt of a catastrophe only when their current dwelling was damaged. In the Local Village condition participants learnt when a catastrophe hit ‘their’ village, and saw how many dwellings were damaged in that village. In the All Villages condition, participants learnt when a catastrophe hit any village, and saw how many dwellings were damaged there. These three feedback conditions were designed to mimic the different amounts of information that people might have access to: if they relied solely on their personal experience (Own House), if (additionally) they could access local information sources (Local Village), or they could also source information from afar via the media or authorities (All Villages). There were two environments (each encountered for 200 rounds): a balanced or ‘moderate’ environment where all three villages offered the same expected value (that is, average number of points), and a ‘severe’ environment in which the two risky regions had 25% lower expected value than the safe region