Click here to login if you're an NAE Member
Recover Your Account Information
Author: Jeffrey Czajkowski
A significant aim of natural disaster research is to improve the science, or the hazard assessment, of risk associated with such disasters. This goal could be achieved by, for example, enhancing the accuracy of short-term extreme weather or long-term climate forecasts or by increasing the validity of the hazard component of natural disaster catastrophe models.
It is assumed either that users of the information will fully understand the scientific data and incorporate that understanding in rational decisions based on a systematic analysis of tradeoffs between benefits and costs, or that losses will be better predicted and managed based on the enhanced scientific aspects of a catastrophe model. Yet, although hazard assessments have improved, many forms of losses from natural disasters have increased over time, associated with innumerable instances of inadequate investments in loss reduction measures and poor decision making before and after events.
As reported by the United Nations International Strategy for Disaster Reduction, “Experience has shown that a purely technical assessment of risk, however sophisticated and cutting-edge, is by itself unlikely to trigger actions that reduce risk. Successful risk assessments produce information that is targeted, authoritative, understandable, and usable” (UNISDR 2015, p. 148). Research provides empirical evidence of individuals exhibiting systematic behavioral biases and using simplified decision rules when making choices with respect to low-probability/high-impact events such as natural disasters. The findings show that such choices and the resulting behavior are significantly influenced by individual interpretation, which is dependent on how the scientific information is framed and presented, so it is essential to incorporate understanding of decision biases in the assessment and subsequent communication of natural hazard risks.
Unfortunately, little of this behavior-based knowledge has been incorporated into natural disaster risk assessment and catastrophe modelling. The use of appropriate risk management strategies based on such knowledge could reduce natural disaster losses.
Progress in Forecasting and Modelling
Recent decades have seen significant progress in the ability not only to observe and understand the weather but also to provide more accurate forecasts (Hirschberg et al. 2011; NRC 2010). This is congruently true for extreme weather events such as hurricanes, as evidenced by the National Hurricane Center’s reduced annual average track forecast errors from 1970 to 2014 (figure 1). For example, the 72-hour track forecast error improved from nearly 450 nautical miles on average in 1970 down to less than 100 nautical miles in 2014, as shown by the yellow least squares trend line. These forecast improvements have been credited with a number of associated benefits, such as a substantial reduction in the number of direct fatalities thanks to more timely evacuation (Gladwin et al. 2007; Rappaport 2000, 2014).
Major advances in the science and modelling of extreme weather hazards (see Lin et al. 2012 for a discussion of storm surge modelling) have led to the widespread use of catastrophe models for natural hazard risk assessment since the early 1990s by the insurance industry. This usage has in turn led to the further implementation of natural hazard risk transfer mechanisms such as reinsurance and capital markets (Grossi and Kunreuther 2005), allowing for the relatively uneventful absorption of natural hazard economic losses by the insurance industry in recent years.
There remain serious concerns, however. First, the evidence suggests an upward trend in economic losses from natural disasters worldwide (figure 2), to an estimated annual average of about $250 billion (UNISDR 2015). This rise is correlated with population and exposure growth in high hazard areas (UNISDR 2013, 2015), leading to more people affected by natural disasters, interdependencies in economic and social systems that increase vulnerability to disruptions, and potentially exacerbated hazard risks from climate change impacts (UNISDR 2015).
Second, reduced mortality benefits have been limited to select developed countries, largely because of a lack of capacity to forecast disasters and provide early warnings in developing countries (UNISDR 2015). Moreover, in developing countries noninsured and nondirect property and other losses, and the costs associated with recovery, are difficult to quantify and hence thought to be substantially underestimated (UNISDR 2015).
Finally, even in a relatively sophisticated natural disaster risk management landscape like the United States there have been both inadequate investments in hazard-related loss reduction measures (e.g., with Hurricane Katrina in 2005 and Hurricane Sandy in 2012) and poor decision making. As an example of the latter, during the 2013 Oklahoma City tornado residents should have sheltered in place but were advised by a local meteorologist to evacuate south in their cars.
Illustrative examples show how a traditional natural hazard forecast risk context (i.e., risk “space”) may be placed in a broader overview of event risk in time, importantly including behavioral implications of intertemporal decision making. The paper further describes an economic (i.e., benefit-cost) model of decision making in this risk space, highlighting potential sources of bias demonstrated in recent research.
Defining the Natural Hazard Forecast Risk Space
Natural hazard risk is defined as the probability of a natural hazard event occurrence and its expected impact (Kunreuther and Useem 2010). Thus, the concept of natural hazard risk has two key components, hazard probability and impact, each of which has an element of uncertainty associated with it. Geoff Love and Michel Jarraud of the World Meteorological Organization provide a schematic of this natural hazard risk space (figure 3), with the probability of the hazard on the y-axis and the impact on the x-axis (Love and Jarraud 2010); uncertainty is represented by the shaded shapes surrounding the three illustrative risks shown (e.g., C would have more uncertainty in impact vs. likelihood given the size, shape, and position of the circle).
Physical scientists working in the area of natural hazard forecast risk often concentrate on the likelihood of occurrence (y-axis), such as the return period for a flood event. Thus, as an illustrative example of this predisposition, only 0.6 percent of the National Oceanic and Atmospheric Administration’s 2008 budget of $4 billion was directed to social science activities, which are more likely to focus on the impact side of a risk (NRC 2010).
Likewise, in a catastrophe modelling framework of combined hazard, exposure, and vulnerability components leading to loss (Grossi and Kunreuther 2005), emphasis is typically on hazard, although the other components may significantly affect losses and hence overall risk. For example, a Risk Management Solutions study found that loss estimates could change by a factor of 4 when property exposure data gaps were filled or inaccurate information was corrected (RMS 2008).
There is a clear need to better understand the impact side of the natural hazard risk equation—including perceptions of expected impact—if overall risk reduction is the goal (Kunreuther and Useem 2010). For example, Botzen and colleagues (2015) find that in New York City flood risk perception is influenced by underestimation of hazard impact. Significantly, since the US tornado tragedies in 2011, impact-based warnings for tornadoes (figure 4) have been implemented by the National Weather Service (NWS).1
Integrated loss modelling between the physical sciences and other disciplines such as engineering and the social sciences is critical to enhance understanding of the numerous factors behind natural hazard risk (Kunreuther and Useem 2010; Morss et al. 2011; Tye et al. 2014). Integrated disciplines (e.g., physical scientists and economists) are evident in natural hazard impact assessment research on hurricane risks in coastal locations (Czajkowski and Done 2014) and on inland flooding from tropical cyclones (Czajkowski et al. 2013).2 And a 2010 overview of the literature on integration of socioeconomic considerations in weather research cites six successful programs (NRC 2010, pp. 34–35).
Natural Hazard Forecast Risk in Time
Pre- and Postevent Planning
Forecasts of natural hazard risks are directly tied to an event, whereas the extent of overall impacts is a function of risk over time in the affected areas.
While most activity surrounding a natural hazard event is focused on the crisis management stages of preparation and response (during and immediately afterward), the socioeconomic impacts are tightly linked with the pre-event prevention, mitigation, and recovery planning activities and the postevent long-term recovery process. Narrowly focusing risk reduction efforts on the event itself will likely not support optimal total risk reduction efforts. Herman Leonard and Arnold Howitt of the Kennedy School of Government at Harvard provide a time-oriented view (figure 5) that extends to include the oft-underappreciated stages of pre-event preparation and postevent recovery (Leonard and Howitt 2010).
The ability to expand the timescale of the natural hazard risk event space to earlier and later stages is critical. For example, how would warning messages of a potential natural hazard event risk in the relatively distant future affect current pre-event preparation activities (NOAA 2015)?
Accounting for Behavioral Biases
Although expanding the timescale of the risk space is essential, interjecting the notion of time is potentially problematic given temporal behavioral biases such as underweighting the future through hyperbolic discounting3 (Kunreuther et al. 2012). In this way of thinking, although the costs of pre-event preparation and mitigation are immediate and certain, the benefits associated with action are in the distant future and therefore uncertain in both time and return. Even if properly discounted benefits that accrued over time (i.e., at a constant and appropriate discount rate) outweighed the upfront costs, individuals would tend to disproportionately discount the future given their aversion to delayed gratification (Kunreuther et al. 2012).
Other intertemporal behavioral biases (Kunreuther et al. 2012) that could hamper optimal pre-event mitigation are myopic planning (a limited time horizon of only the next few years), underestimation of the risk (the probability or impact of a hazard), and affective forecasting errors (poor predictions of emotional states based on feelings today).4 These biases make clear the importance of behavioral tendencies in decision making in the natural hazard risk context.
Decision Making in Natural Hazard Risk Context
Intuitive vs. Deliberative Decision Making
From a rational economic perspective in the natural hazard risk space, individual decisions at a point in time are based on expected utility theory.5 According to this theory, an individual confronted with the need for a decision with uncertain outcomes will decide based on the outcome with the greatest expected utility.
Table 1 shows the application of expected utility theory in the context of a natural disaster. When one is deciding to evacuate from a forecasted hurricane, utility (or disutility) is assigned to each possible future state (landfall hit or miss) given the possible action (stay or evacuate), and each future state is assigned a probability (p) with all probabilities summing to one. The choice of staying or evacuating is determined by selecting the action with the highest expected outcome across all possible states; in table 1, the choice is to evacuate.
In reality, however, the decision-making process is often quite complex (especially over multiple forecast periods)6 and rarely do the people under the warning act rationally. Rather, the combination of systematic behavioral biases coupled with simplified decision rules leads to choices that differ from those predicted by expected utility theory (Kunreuther and Useem 2010; Kunreuther et al. 2012).
Kahneman (2011) highlights the difference between intuitive and deliberative thinking, documenting extensive research on intuitive biases that operate in lieu of ideal deliberative decision making and result in suboptimal choices for low-probability/high-consequence events such as natural disasters. For example, the availability bias estimates the likelihood of disaster occurrence based on the saliency of the event as opposed to objective hazard probabilities. Or protective action is not taken because the subjective probability of expected impact is below some threshold level of concern. Kunreuther and Useem (2010) discuss a host of other behavioral biases revealed by the research, many of them associated with group behavior, risk culture, fear and other emotions, and trust.7
Factors that Drive Positive Behavior
A considerable amount of research has sought to identify what factors drive positive behavior in this context, controlling for behavioral biases. Meyer and colleagues (2013) used a realistic simulated storm environment to better understand risk perception and decision making, and in a subsequent study they interviewed more than 2000 respondents in real time under the threat of hurricane strikes during the 2010–2012 hurricane seasons (Meyer et al. 2014). Beatty and colleagues (2015) used a big data approach to analyze water bottle sales before and after a hurricane.
In a recent review and assessment of risk communication and behavior, NOAA (2015) points to work by Mileti and colleagues (2006) that identified a number of factors and categories consistently found to matter in the context of warning response: sociodemographic (female, white, more education, and children present); personal (experience, knowledge of hazard and actions, self-efficacy, fear, risk and vulnerability perception, more resources available, large and strong social network); source/channel (environmental or social cues present, official source, in person, familiar source, multiple sources); information (specific, credible, certain, frequent, consistent, and with guidance on actions); and threat (less lead time available, greater severity, close, confirmed).
Importantly, however, little of this behavior-based knowledge has been incorporated in natural disaster risk assessment or mitigation planning. Possible approaches are briefly articulated in the next section.
Despite significant advances in recent decades in observing, understanding, and forecasting extreme weather, the impacts and threats from natural disasters remain extensive. This paper has provided a definition and context for decision making in the natural disaster risk space where behavioral biases play a significant role. Efforts to reduce natural disaster risk will have to incorporate appropriate risk management strategies based on this behavior-based knowledge. The following measures are recommended based on this overview:
Alliance Development Works. 2012. WorldRiskReport 2012: Environmental Degradation and Disasters. Berlin: Bündnis Entwicklung Hilft. Available at www.nature.org/ourinitiatives/habitats/oceanscoasts/ howwewor k/world-risk-report-2012-pdf.
Beatty TK, Shimshack JP, Volpe RJ. 2015. Disaster preparedness and disaster response: Evidence from bottled water sales before and after tropical cyclones. Working paper, University of Virginia, Charlottesville.
Botzen W, Kunreuther H, Michel-Kerjan E. 2015. Divergence between individual perceptions and objective indicators of tail risks: Evidence from floodplain residents in New York City. Working paper, Wharton Risk Management and Decision Processes Center, Philadelphia.
Chavas DR, Yonekura E, Karamperidou C, Cavanaugh N, Serafin K. 2012. US hurricanes and economic damage: An extreme value perspective. Natural Hazards Review 14(4):237–246.
Czajkowski J. 2011. Is it time to go yet? Understanding household hurricane evacuation decisions from a dynamic perspective. Natural Hazards Review 12(2):72–84.
Czajkowski J, Done J. 2014. As the wind blows? Understanding hurricane damages at the local level through a case study analysis. Weather, Climate, and Society 6(2):202–217.
Czajkowski J, Villarini G, Michel-Kerjan E, Smith JA. 2013. Determining tropical cyclone inland flooding loss on a large scale through a new flood peak ratio-based methodology. Environmental Research Letters 8(4):044056.
Gladwin H, Lazo JK, Morrow BH, Peacock WG, Willoughby HE. 2007. Social science research needs for the hurricane forecast and warning system. Natural Hazards Review 8(3):87–95.
Grossi P, Kunreuther H. 2005. Catastrophe Modeling: A New Approach to Managing Risk, Vol. 25. Boston: Springer Science and Business Media.
Harrison J, McCoy C, Bunting-Howarth K, Sorensen H, Williams K, Ellis C. 2014. Evaluation of the National Weather Service Impact-based Warning Tool. WISCU-T-14-001 Report. Available at www.seagrant.sunysb.edu/Images/Uploads/PDFs/Hurricanes- NWS-IBW_finalreport.pdf.
Hirschberg PA, Abrams E, Bleistein A, Bua W, Delle Monache L, Dulong TW, Gaynor JE, Glahn B, Hamill TM, Hansen JA, and six others. 2011. Weather and climate enterprise strategic implementation plan for generating and communicating forecast uncertainty information. Bulletin of the American Meteorological Society 92:1651–1666.
Kahneman D. 2011. Thinking, Fast and Slow. New York: Farrar, Straus and Giroux.
Kunreuther H, Useem M. 2010. Learning from Catastrophes: Strategies for Reaction and Response. Upper Saddle River, NJ: Prentice Hall.
Kunreuther H, Meyer R, Michel-Kerjan E. 2012. Overcoming decision biases to reduce losses from natural catastrophes. In: The Behavioral Foundations of Public Policy, ed. Shafir E. Princeton, NJ: Princeton University Press.
Leonard HB, Howitt AM. 2010. Acting in time against disasters: A comprehensive risk management framework. In: Learning from Catastrophes: Strategies for Reaction and Response, ed. Kunreuther H, Useem M. Upper Saddle River, NJ: Prentice Hall.
Lin N, Emanuel K, Oppenheimer M, Vanmarcke E. 2012. Physically based assessment of hurricane surge threat under climate change. Nature Climate Change 2(6):462–467.
Love G, Jarraud M. 2010. Forecasting and communicating the risk of extreme weather events. In: Learning from Catastrophes: Strategies for Reaction and Response, ed. Kunreuther H, Useem M. Upper Saddle River, NJ: Prentice Hall.
Malmstadt J, Scheitlin K, Elsner J. 2009. Florida hurricanes and damage costs. Southeastern Geographer 49:108–131. Available at http://fsu.academia.edu/JillMalmstadt/Papers.
Mendelsohn R, Emanuel K, Chonabayashi S, Bakkensen L. 2012. The impact of climate change on global tropical storm damages. Nature Climate Change 2:205–209.
Meyer R, Broad K, Orlove B, Petrovic N. 2013. Dynamic simulation as an approach to understanding hurricane risk response: Insights from the Stormview lab. Risk Analysis 33(8):1532–1552.
Meyer RJ, Baker EJ, Broad KF, Czajkowski J, Orlove B. 2014. The dynamics of hurricane risk perception: Real-time evidence from the 2012 Atlantic hurricane season. Bulletin of the American Meteorological Society 95:1389–1404.
Mileti DS, Bandy R, Bourque LB, Johnson A, Kano M, Peek L, Sutton J, Wood M. 2006. Annotated Bibliography for Public Risk Communication on Warnings for Public Protective Action Response and Public Education (rev. 4). Available at www.colorado.edu/hazards/publications/informer/infrmr2/ pubhazbibann.pdf.
Morss RE, Wilhelmi O, Meehl G, Dilling L. 2011. Improving societal outcomes of extreme weather in a changing climate: An integrated perspective. Annual Review of Environment and Resources 36(1):1–25.
Murnane RJ, Elsner JB. 2012. Maximum wind speeds and US hurricane losses. Geophysical Research Letters 39(16):L16707.
Murphy A, Strobl E. 2010. The impact of hurricanes on housing prices: Evidence from US coastal cities. Federal Reserve Bank of Dallas Research Department Working Paper 1009.
NRC [National Research Council]. 2010. When Weather Matters: Science and Service to Meet Critical Societal Needs. Washington: National Academies Press.
NOAA [National Oceanic and Atmospheric Administration]. 2015. Risk Communication and Behavior Assessment: Findings and Recommendations. Internal Report. Silver Spring, MD.
Nordhaus W. 2006. The economics of hurricanes in the United States. NBER Working Paper 12813. Cambridge, MA: National Bureau of Economic Research. Available at www.nber.org/papers/w12813.
Nordhaus W. 2010. The economics of hurricanes and implications of global warming. Climate Change Economics 1(1):1–20.
Rappaport EN. 2000. Loss of life in the United States associated with recent Atlantic tropical cyclones. Bulletin of the American Meteorological Society 81(9):2065–2073.
Rappaport EN. 2014. Fatalities in the United States from Atlantic tropical cyclones: New data and interpretation. Bulletin of the American Meteorological Society 95:341–346.
RMS [Risk Management Solutions]. 2008. A Guide to Catastrophe Modelling. The Review: Worldwide Insurance, London. Available at http://forms2.rms.com/rs/729-DJX-565/images/rms_guide_ catastrophe_modeling_2008.pdf.
Schmidt S, Kemfert C, Hoppe P. 2009. Simulation of economic losses from tropical cyclones in the years 2015 and 2050: The effects of anthropogenic climate change and growing wealth. DIW Discussion Paper 914. Berlin: Deutsches Institut für Wirtschaftsforschung e.V.
Schmidt S, Kemfert C, Hoppe P. 2010. The impact of socio-economics and climate change on tropical cyclone losses in the USA. Regional Environmental Change 10:13–26.
Strobl E. 2011. The economic growth impact of hurricanes: Evidence from US coastal counties. Review of Economics and Statistics 93(2):575–589.
Tye MR, Holland GJ, Done JM. 2014. Rethinking failure: Time for closer engineer-scientist collaborations on design. Proceedings of the Institution of Civil Engineers 168(2):49–57.
UNISDR [United Nations International Strategy for Disaster Reduction]. 2013. From shared risk to shared value: The business case for disaster risk reduction. Global Assessment Report on Disaster Risk Reduction. Geneva.
UNISDR. 2015. Making development sustainable: The future of disaster risk management. Global Assessment Report on Disaster Risk Reduction. Geneva.
Zhai AR, Jiang JH. 2014. Dependence of US hurricane economic loss on maximum wind speed and storm size. Environmental Research Letters 9(6):064019.
1 Information about NWS impact-based warnings is available at www.weather.gov/impacts/. For an assessment of the impact-based tool see Harrison et al. (2014).
2 A number of recent impact-focused assessments for extreme events are available; e.g., Chavas et al. (2012), Malmstadt et al. (2009), Mendelsohn et al. (2012), Murnane and Elsner (2012), Murphy and Strobl (2010), Nordhaus (2006, 2010), Schmidt et al. (2009, 2010), Strobl (2011), and Zhai and Jiang (2014).
3 Hyperbolic discounting rapidly discounts valuations for small time periods and slowly discounts valuations for longer periods. Exponential discounting, on the other hand, discounts by a constant factor per unit delay, regardless of the length of the delay.
4 Intertemporal bias of duration neglect (Kunreuther et al. 2012) may also exist in the postevent recovery phase, when there is a tendency to overestimate the time to recover and hence future protection would be overvalued.
5 Other social science theories of decision making in a natural hazard context include the psychometric paradigm of psychology (perception of hazards taking into account qualitative information [e.g., dread] rather than just statistical [i.e., probability]); the cultural theory of risk in anthropology (social and cultural influences on risk perception); the mental models approach of psychology and risk (individuals have a “mental model” of reality, influenced by social interactions and experiences, that they use as a lens to view risky situations); the protection motivation theory of psychology (people protect themselves based on their perception of severity, probability, effectiveness of protective action, and self-efficacy); and the social amplification of risk framework of geography (risks are amplified or attenuated due to individual, social, and cultural factors) (NOAA 2015).
6 For an illustration of this decision over time from a dynamic perspective, where for each forecast period the individual may choose to evacuate or wait for an additional forecast, see Czajkowski (2011).
7 Chapter 4, “Cognitive Constraints and Behavioral Biases,” discusses these in more detail as does chapter 5, “The Five Neglects: Risks Gone Amiss,” from an expected utility perspective.