Printer Friendly

The dynamics of hurricane risk perception: real-time evidence from the 2012 Atlantic hurricane season: surveys of coastal residents conducted in 2012 as hurricanes were approaching reveal widespread misunderstanding of the extent and nature of threats posed by tropical cyclones.

Over the past century, hurricanes have been the single largest source of property damage from natural hazards in the United States. In the last decade alone, losses from hurricanes have been estimated at $290 billion (2012 U.S. dollars), with two storms--Katrina in 2005 and Sandy in 2012-collectively inflicting over $120 billion in damage (Blake et al. 2011, 2013; Pielke et al. 2008; Pielke 2012). What makes the scale of these losses particularly troublesome is that hurricanes are now among the best understood of all natural hazards, and in recent years there have been dramatic increases in track forecasting abilities and warning times (e.g., Cangialosi and Franklin 2013; Gall et al. 2013). These scientific advances, however, have seemingly not been matched by commensurate increases in preventive adaptation. To illustrate, 36 hours in advance of Hurricane Sandy residents were warned that the storm would likely bring "life-threatening storm surge flooding" to the Northeast (NWS 2012). Yet, 230,000 cars were still lost in the storm from floods (Taylor 2013)--a loss that, at least in hindsight, would seem to have been avoidable.

This article reports the findings of a unique program of research designed to shed light on potential reasons for this adaptation paradox. We report data from field surveys that measured the evolution of coastal residents' risk perceptions and preparation plans as two hurricanes--Isaac and Sandy--approached the United States during the 2012 hurricane season. In these studies, perceptions and preparation decisions were measured in real time as they were being made by residents threatened by the storms. These data thus provide the first longitudinal look at how hurricane risk perceptions and responses evolve over time during storm threats and how these perceptions compared to the objective risks residents were facing.

The data yield a surprising--and potentially disturbing--view of hurricane threat response. Despite the ubiquity of information available about Isaac and Sandy, residents misperceived the actual risks that they faced in terms of the intensity, nature, and duration of impacts. Surveyed residents, for example, overestimated the probability that their homes would be affected by hurricane-force winds, but then they displayed limited degrees of concern over this prospect. These residents also underestimated the threat posed by flooding--including people living adjacent to water areas. The consequence was a systematic pattern of miscalibrated preparation decisions, with residents taking actions that were suitable for a short-lasting wind event but not for a significant wind or flood catastrophe with a long-term recovery period for which evacuation would be required. In addition, these misperceptions also appeared to be manifested in longer-term investments in protection; only 54% of residents living within a half mile of water areas threatened by Sandy, for example, reported holding separate flood insurance policies. These issues point to the importance of adjusting hurricane warnings and information campaigns, and of evaluating policy options in the light of these misunderstandings of hurricane hazards.

BACKGROUND AND METHOD. Over the years, a large survey-based literature has been developed that describes the kinds of beliefs coastal residents have about the long-term risk posed by hurricanes (e.g., Peacock et al. 2005; Trumbo et al. 2011), as well as the basis of shorter-term preparation decisions, particularly those involving evacuation (e.g., Baker 1991; Dash and Gladwin 2007; Dow and Cutter 1998; Huang et al. 2012; Lindell et al. 2005; Lindell and Prater 2008; Morss and Hayden 2010; Zhang et al. 2007). While this work has been useful in providing insights into such issues as the intrahousehold drivers of decisions to evacuate (e.g., Baker 1991; Lindell and Prater 2008) and media utilization during storms (e.g., Zhang et al. 2007; Broad et al. 2007), it has been less informative about how residents perceive hurricane threats when they are arising and about the accuracy of decisions to take protective action. One primary reason for this gap is that past findings have been based on surveys conducted weeks or even years after storms have past, when memories of what risk perceptions were before the storm and the process by which preparation decisions were made may have faded and were possibly distorted by hindsight bias (e.g., Brown et al. 1994; Fischhoff and Beyth 1975). (1) As a result, we know little about how risk perceptions evolve over time as storms move toward a coast when the outcome of a storm is still in doubt and, most critically, about the suitability of preparation actions.

In an attempt to obtain this knowledge, we conducted a program of survey research during the 2012 hurricane season that measured risk perceptions and preparation decisions as they were being made by threatened residents. The storm season offered two opportunities for study: Hurricane Isaac, which made landfall on the coast of the Louisiana just west of the Mississippi River in late August (Berg 2012), and Hurricane Sandy, which made landfall on the coast of New Jersey near Atlantic City in late October (Blake et al. 2012). The surveys were conducted by phone and were initiated 72 hours (for Sandy) or 48 hours (for Isaac) before each storm's predicted landfall and then repeated with different random samples three shifts a day until 6 h before predicted landfall (see Fig. 1). The surveys were timed to allow measures of subjective storm beliefs to be paired with objective storm information carried in the 0500,1100, and 1700 EDT National Hurricane Center advisories.

Each survey instrument contained between 60 and 80 questions (depending on screens) that focused on five domains: 1) current beliefs about the objective characteristics of the storm and warnings, 2) perceptions of the threat posed by the storm, 3) sources of information about the storm, 4) preparation actions, and 5) personal background characteristics. The nature and wording of the specific items evolved from the experience gained designing two prior real-time survey instruments that were developed for use in 2010 (Hurricane Earl) and 2011 (Hurricane Irene). The 2012 surveys contained several new items not contained in previous versions (e.g., probability assessments for different kinds of threats) and were tested for comprehension by the field survey firm (Kerr and Downs Research) prior to administration.

For the Isaac study, respondents were drawn from a random sample of households in coastal ZIP codes with land telephone lines along the middle Gulf Coast from southeastern Louisiana to Alabama, as well as the two westernmost counties in the Florida Panhandle (see Fig. 1). For the Sandy study, respondents were drawn from coastal ZIP codes along the mid-Atlantic region from Virginia to northeastern New Jersey. Within each survey shift, approximately 50-60 surveys could be completed, producing a total of 893 completed surveys across both storms. In Table 1 we provide the basic demographic profile of each sample for each storm along with, for comparison, the corresponding 2010 population demographics of the associated counties from which the sample was pooled. While there was some storm-to-storm variation in samples, most participants were homeowners between the ages of 30 and 80 with at least some college education, and approximately three-quarters of participants reported total household incomes over $40,000. As such, the sample tended to be somewhat older, more educated, and more likely to own homes than the mean of the general population in the surveyed areas.

The absolute response rate for the Isaac and Sandy surveys (percentage of phones dialed that yielded a completed survey) was 7.1% for Isaac and 10% for Sandy, a number consistent with recently published norms for telephone surveys in public opinion polls (Pew Research Center 2012, table on p. 5). The realized cooperation rates (the percentage contacted who participated), however, was much higher than the Pew norms, being 39% for Isaac and 49.3% for Sandy. As a point of reference, these cooperation rates are on a par with response rates reported in recent mail-based posthurricane surveys (e.g., Huang et al. 2012).

The location of respondents' vis-a-vis evacuation zones was reasonably well known for Isaac but less so for Sandy, where evacuation orders varied by municipality. As we detail in appendix B [the appendixes are in an online supplement ( -12-00218.2)], most respondents to the Isaac surveys lived in mandatory evacuation areas (zone A or category 1) with the exception of some in Harrison County, Mississippi, who lived inland from evacuation zones. In most areas in Isaac's track, evacuation was advised both for those in beachfront areas as well as those in low-lying areas and adjacent to streams prone to flooding from rain. Because respondents likely varied in their awareness of such orders, our subsequent analysis will focus on stated awareness rather than actual orders (for which we have incomplete measures).

A natural source of concern when contacting respondents via landlines was the possibility of nonresponse bias due to an increasing tendency for those who were most at risk from the storm (or were most concerned about risk) to leave the study area as the storms approached. Our data offered two means for testing this possibility: by measuring temporal changes in home-contact and survey completion rates over time and by seeing if there were temporal changes in sample demographics that might be associated with decisions to evacuate in past studies (e.g., Lindell et al. 2005; Huang et al. 2012). Because homes were contacted on a randomized basis, nonresponse bias due to evacuation would be evidenced by a decrease in the rate of successful telephone contacts over time as the storm approached. As shown in Fig. 2, however, such a decrease was not observed; indeed, in the case of Sandy, successful contact rates actually increased over time--possibly because of more residents being at home due to work cancellations. In the case of Sandy, a logistic trend analysis of the number of homes successfully contacted supported a significant positive trend as the time to landfall wore on ([chi square] (1, N = 9)= 13.77; p =<0.01), whereas there was no significant trend in the case of Isaac ([chi square] (1, 6) = 2.28; p = 0.13).

Further reassurance that nonresponse bias was likely not a major factor in our surveys was that the demographic profile of the respondent pool remained largely stable over time. For both storms we regressed five different profile variables that have been found in the past to be correlated with propensities to evacuate--age [question (Q) 72; appendix A], education (Q75), past storm damage experience (Q63), gender (Q86), and distance from water (Q80; available only in the Sandy study; e.g., Baker 1991; Lindell and Prater 2008)--with time until landfall. For the Sandy study, none of these univariate analyses could reject a null hypothesis of temporal stationarity. In the Isaac study, there was a significant tendency for the sample to be slightly younger later in the survey period (f = -2.44; p = .015). Age, however, was separately found not to be a significant predictor of two risk-perception variables that might be associated with decisions to evacuate in the absence of an order: ratings of perceived safety in home and probability of wind damage. (2)

FINDINGS. The survey data provided a rich array of cross-sectional and spatial-temporal data about storm knowledge, perceptions, and preparation actions. Below we report the most salient features of these data, focusing on three categories: awareness of the storms and warnings, the accuracy of the mental models that residents held about storm threats, and the suitability of short- and long-term preparations. In appendix C the sample sizes underlying each figure are reported.

Storm and warning awareness. Both Isaac and Sandy were major local, regional, and national media events. Local news stations provided continuous coverage of the storm from the time warnings were first issued until after landfall, and the Weather Channel set an all-time viewership record during Sandy, when 39 million U.S. households tuned in to watch the television network's coverage on 29 October (Bibel 2012). This impact was matched by high levels of web viewing, with the Weather Channel web-based platforms (, mobile apps) receiving over 450 million page views that same day (Bibel 2012).

Reflecting this ubiquity of media attention, survey respondents displayed universal (100%) awareness of each storm (Q1), with respondents indicating that they were keeping regularly abreast of storm information. Across all time periods, 88% of Isaac respondents indicated having received their latest information (from any source) within the previous 2 hours (Q23), as did 79% in Sandy. The primary source of this information (Q24) was television for 90% of respondents in Isaac and 87% in Sandy. In contrast, Internet websites and social media were less commonly utilized; 21% of respondents in Isaac and 15% in Sandy reported that their last information came from any Internet source (either alone or in conjunction with TV) and, of the 43% of respondents across both storms who had a social media account (e.g., Facebook, Twitter), only 4.5% indicated that it was used as a source of storm information. (3)

Despite the high awareness of the storm threats and frequent monitoring of information, respondents' knowledge about the warnings that had been issued for their locations was surprisingly imperfect. For example, during our Isaac surveys, hurricane warnings were continuously in effect in Louisiana and Mississippi, with watches in Alabama and Florida. Nevertheless, when asked 11% of respondents were unaware or unsure whether watches or warnings of any kind had been issued (Q13), and among those who were aware only 66% who were under a hurricane warning correctly reported this (Q14). (4) Likewise, in the 36 hours just before Sandy made landfall, 20% of respondents in coastal New Jersey, Delaware, and Maryland (the main threat areas that were sampled) were still unaware or uncertain whether warnings had been issued for their areas, and of those aware 40% thought it was something other than a hurricane warning.

What makes this result somewhat surprising is that residents in both studies had recent experience with tropical cyclones and, as noted, media coverage of both storms was extensive. The surveyed area of the Gulf Coast, for example, had been under some kind of tropical cyclone warning five times since 2008, and the mid-Atlantic region had been affected by hurricane Irene just the year before (Avila and Cangialosi 2011). One possible explanation for particularly high rates of confusion in Sandy, however, is that as the storm approached the coast the National Hurricane Center decided to switch from issuing traditional hurricane watches and warnings to "hurricane wind warnings" in anticipation of an extratropical transition prior to landfall (Blake et al. 2012).

Accuracy of mental models: Misperceiving intensity and impact. In addition to misconstruing warnings, residents also displayed relatively poor mental models of the meteorological threats each storm posed. [Mental models are the cognitive representations of real-world objects, events, and processes that people form in their minds (Jones et al. 2011).] The data suggest that perceptions were marked by two prominent biases: an overestimation of wind intensity--believing hurricane wind conditions winds were far more likely to occur than was actually the case--and an underestimation of impact--being relatively unconcerned about the prospect of such winds and a tendency to underestimate the threat posed by storm surge and flooding.

To illustrate these biases, in Fig. 3 we plot the time course of respondents' subjective beliefs about the probability that their homes would experience hurricane-force winds of 75 mph (33.5 m [s.sup.-1]) or greater (Q16; red line) along with the corresponding objective probabilities derived from the National Hurricane Center wind forecasts, pooled across states. The objective benchmarks were constructed using the published cumulative hurricane-force wind probability in a given advisory for the city closest to residents' location. For example, in Sandy the benchmark forecasts were Norfolk for southeastern Virginia (VA), Ocean City for Maryland (MD) and Delaware (DE), and either Atlantic City or Newark for New Jersey (NJ).

The data show that residents systematically overestimated their likelihood of experiencing hurricaneforce winds, with estimates, at times, averaging 5 times those of the scientific estimates for both storms. For example, as Sandy was approaching coastal New Jersey, the National Hurricane Center cumulative hurricane wind probability at Atlantic City remained below 30%, yet the New Jersey sample consistently reported subjective estimates between 70% and 80%.

But while residents fully expected the arrival of a hurricane, paradoxically few of them expressed high degrees of worry over this prospect. Respondents were asked a number of questions designed to elicit expected personal storm impacts, including rating (on a 100-point scale) how safe they felt riding out the storm in their homes (Q30), the probability that the winds would be such to risk property damage (Q17), the probability that property damage might be such as to threaten personal safety (Q18), and whether they believed that the storm would hit and be a danger to them (Q31). In Fig. 4 we plot the time course of these multiple measures, which shows that residents' high expectations of experiencing hurricane-force winds shown earlier (Fig. 3) were not manifested in high levels of concern about these winds.

For example, across all time periods only 13% of respondents threatened by Isaac and only 17% in the case of Sandy thought that the storm posed personal danger. Likewise, the judged probability that the winds would be strong enough to inflict some kind of property damage was consistently lower than the probability that the winds would be of hurricane force. For example, in the survey period before expected landfall in Isaac, the average judged probability of hurricane force winds was 40% (Fig. 3), but the average judged probability of property damage was 22% for any personal danger (with the likelihood of severe damage being predictably lower). Likewise, in the last survey period before landfall in Sandy, the average judged probability of hurricane force winds was 58%, but the average judged probability of any property damage was 30%.

A potentially more worrisome aspect of the findings, however, is that the data also show evidence that residents consistently misperceived the likely source of the danger posed by both storms, with residents for whom the greatest objective threat was from water believing it was from wind. While we lack specific objective information about the actual primary threat faced by each respondent, we can form tentative inferences about the accuracy of risk beliefs by examining how they covaried with factors inherently associated with water and wind risk, such as the respondent's proximity to water (Q80), whether the person lived in an evacuation zone (either objective or perceived; Q44), and building type (Q55).

The survey provided two measures of respondents' beliefs about the relative threat posed by wind versus water: a question that asked which of six impacts posed the greatest threat from each storm (wind, flooding from storm surge, a combination of wind and surge, flooding from rain, tornadoes, or some other impact; Q32), and their subjective probabilities that they would experience damage to home or safety from wind or flood (Q17-22). As noted above, for each hazard we asked respondents to assess the probability of property damage alone as well as the probability that the damage would be severe enough to threaten personal safety (e.g., Q17 vs Q18). To construct a composite index, we first took the average of these probability assessments for flood, and then we subtracted it from the corresponding average for wind. In the Isaac surveys, the flood threat was asked only in terms of storm surge; however, in the Sandy surveys, we solicited separate probabilities for the risk of flood from storm surge and that from rain. In this latter case, we defined the subjective water threat as the larger of these two mean stated probabilities.

As we noted earlier, almost all of the Isaac surveys were conducted among individuals living in either zone A or category 1 surge zones, where evacuations had been ordered in advance of the storm (see appendix B). Despite this location, 56% of respondents identified the greatest threat they faced was that of wind, while only 33% identified either surge flooding or combined wind and surge flooding as the major threat. Likewise, the mean stated probability of damage from wind was 7% higher than that for water across Isaac respondents.

Of course, one explanation for this result is that respondents were unaware that they were living in flood-prone areas (a common finding in past evacuation studies; e.g., Baker 2005a,b; Arlikatti et al. 2006). To test for this, in Fig. 5a we plot the distribution of primary threat beliefs by two indicators of water threat that would have been salient to residents: whether they had heard they were living in an area where evacuations had been ordered (Q44) and whether they were living within 500 feet of the Gulf Coast (or, in the Louisiana sample, Lake Ponchartrain), as computed from geocoded linear distance [N = 40 within 500 ft (152 m) and N = 314 beyond 500 ft].

The figures show that while living adjacent to the water indeed heightened concerns about flooding, a plurality--40% within 500 ft of the water--still believed that the primary threat was from wind, and beliefs about the larger threat of wind were actually higher among those who had heard that they were living in an area where evacuations had been ordered.

Because we gathered a richer array of measures about location and beliefs in the Sandy study, it allowed us to undertake a more detailed analysis of wind-versus-water misperceptions. In Fig. 5b we plot two measures of the relative degree to which residents believed the greatest threat was from wind over water as a function of the distance of a residents' home from a water body (Q80): the wind-bias index described above (the difference between the subjective probability of damage from wind vs water), and the difference between the proportion of respondents who identified wind as the greatest source of threat versus any water threat (Q32; surge, wind and surge, rain). While here we see that, indeed, awareness of the threat of water grew as the respondent's proximity to water grew, both the wind-bias indices are always strictly positive; even those living on the water believed that the greater threat they faced was from wind rather than water.

Because the probability-based wind-bias index appeared to be the measure of relative belief that was most responsive to variation in objective risk, as a final analysis on the Sandy data we regressed this bias measure against a battery of indicator-coded variables that capture the normative drivers of the relative risk of wind versus water [distance to water (Q80) and building structure type (Q55)] and individual difference factors that might drive perceptions, including gender (Q86), education (Q75), age (Q72), ownership of a flood policy (Q60), and whether the respondent had previous experience living through a hurricane (Q63, and Q64). In this analysis the effect of distance was captured by four binary (indicator) variables that contrasted beliefs at each successive distance with those held by waterfront residents. The results of this analysis, reported in Table 2, supports only one marginally significant moderator of the tendency to believe that wind is the main threat posed by the storm--the respondent's age. Controlling for other factors, older respondents were more inclined to see wind as the greater risk over water. In contrast, there was no significant effect of increasing distance from a waterfront location, storm experience, housing type, or education or ownership of a flood policy. The absence of an effect of ownership of a flood policy would seem particularly surprising; even those who are sufficiently concerned about the threat of floods that they paid to insure against it believed that the greatest threat Sandy posed was from wind, not water.

Finally, there was also suggestive evidence that residents underestimated the likely duration of the impact of each storm. In our Sandy survey (though not for Isaac), we asked residents how long they expected to be without power during and after the hurricane (Q42). In Fig. 6 we plot the distribution of answers broken down by states where the path of the storm suggested the greatest impact would lie (New Jersey and Delaware) and where less of an impact was anticipated (Maryland and Virginia). The data suggest that residents were relatively optimistic about the duration of impact; the majority of residents thought either that, if they lost power, then it would be for less than 2 days (with 20% in all four states believing they would never lose power) or they held no belief about duration. In contrast, only 28% of coastal respondents in New Jersey and Delaware expected that they might be without power for more than 2 days--only slightly more than the expectations of residents in Maryland and Virginia (22%), where there would have been objective reasons to expect a smaller impact. What is notable about this optimism is that as Sandy approached, residents were widely warned to prepare for outages that could last 7-10 days, or the longest that had been experienced during Hurricane Irene the year before (see Lupkin 2012). (5) The optimistic beliefs, however, imply that many respondents either failed to hear such warnings or believed that if there were long outages, they were going to be experienced by people other than themselves.

PROTECTIVE ACTIONS. Short-term preparation. Although respondents' beliefs about the threats posed by Isaac and Sandy differed in important ways from actual risks, an overwhelming proportion of respondents undertook at least some short-term preparatory action in advance of both storms, and almost all felt well prepared for the storms by the time that they arrived. The evolution of preparedness levels is depicted in Fig. 7, which plots the percentage of respondents for Isaac and Sandy (pooled) who indicated taking at least some preparatory action (Q37) and those who felt they were ready for the storm over time (Q40).

The data tell a clear and seemingly reassuring story: despite misperceptions that may have existed about how strong the winds would be at their homes and sources of danger, virtually all respondents took the storm seriously enough to undertake preparations--and to carry out these steps early. For example, when Sandy surveys began on the evening of 26 October--72 hours before the storm made landfall--over 75% of respondents had already taken some preparatory action, and by the time the storm arrived on 29 October, over 94% felt sufficiently well prepared to endure whatever Sandy had to offer.

On the other hand, an analysis of the kind of preparations that were being taken provides a less encouraging view of readiness. In Fig. 8 we plot the time course of undertaking four major protective actions (Q37), pooling over storms: buying household supplies (e.g., groceries), putting up removable storm shutters (if owned), purchasing or readying a power generator, and developing an evacuation plan if needed (e.g., securing a hotel reservation).

The data show a disconcerting pattern of preparation: while the vast majority of respondents sought basic supplies in advance of each storm (6 h before landfall 88% reported doing so), more effortful actions were comparatively limited. For example, across time periods only 25% of respondents had made plans for where they would go'if and an evacuation were ordered or needed, and in the last survey period when each storm was within 6 hours of predicted landfall less than 55% of residents who owned removable window protections (such as shutters) had put them up and 11% had secured or prepared electric generators.

Perhaps even more alarming was the observed limited compliance with evacuation advice. Though the survey methodology precluded us from directly measuring the percentage of respondents who actually complied with evacuation advice, it nevertheless provided two implicit measures: the change in the percentage of respondents who believed they were living in communities where evacuation had been advised yet who were still home to answer the survey as the time of landfall approached, and the change in the successful home contact rate (from Fig. 2). Because home telephones were randomly dialed, increasing actual evacuation rates over time should be mirrored by a decrease over time in the percentage of the sample of respondents who indicated that they were living in evacuation areas (or were home at all).

In Fig. 9 we plot the evolution of these implied compliance measures as well as stated intentions to leave among those respondents living in advised evacuation areas (from Q44, 49). Hurricane evacuation advisories are typically issued at least 36 hours before a storm's anticipated landfall, and, consistent with this practice, we see a sharp increase in awareness of evacuation warnings 30 hours before landfall. What is potentially disturbing, however, is that the data suggest that there was limited apparent compliance with this advice. Among those respondents who believed that they were living in communities where evacuations were ordered, the percentage who stated they intended to leave was, ironically, highest (55%) before a significant proportion indicated that they were aware that orders had been given (at 36 hours prior to landfall). Moreover, the percentage indicting intentions to leave decreased over time as awareness of orders grew. While we do not have direct measures of the percentage of actual compliance, further indications that the actual rate of evacuation was quite low is reflected by the absence of a decrease in the percentage of respondents living in evacuation areas who were home to answer the survey as the time of landfall approached. Specifically, 30 hours before landfall 54% of the respondents who were contacted said they were living in communities where evacuations had been advised. At 6 hours before landfall, however, this percentage remained almost the same--49%, a statistically negligible decrease that would be consistent with little, if any evacuation.

One possible explanation for the lack of intentions to take effortful actions is that residents believed that their particular homes were at limited risk of damage from either wind (for those who owned shutters) or flooding. To explore this, we analyzed the bivariate relationships that existed between the conditional likelihoods of installing shutters (given ownership; Q34, Q35, Q37) and evacuation intention (given living in an advised evacuation area; Q44, Q49) by respondents' beliefs about the probability that their homes would suffer damage from either winds (for shutters) or flooding (for evacuation). The data show only a weak association between the two constructs. For shutters, there was no significant relationship between shutter usage and subjective damage likelihood ([chi square] = .48 (1, N = 184);p > 0.1). For example, 59% of those who believed that there was a greater than a 50-50 chance of experiencing wind damage to their homes (N = 32) put up their shutters, which was only nominally higher than that observed among people who believed that there was less than a 50-50 chance (N = 152; 52%). For evacuation there was a significant positive effect of risk beliefs on evacuation intentions ([chi square] = 8.44 (1, N = 284); p = 0.014), but it was small in absolute terms; across all times periods and storms, 38% who thought that there was greater than a 50-50 chance of experiencing damage from rain or surge flooding (N = 50) expressed an intention to evacuate, compared to 20% among those who thought that there was less than a 50-50 chance (N = 234).

Long-term protection. The storm surveys also explored the degree to which residents had invested in long-term protection prior to Isaac and Sandy. This was either in the form of making improvements to their homes that would make them more resilient to damage from storms or owning flood insurance policies (Q65, Q59, Q60, respectively). The data suggest a troubling absence of such long-term investments in protection. Among respondents threatened by Sandy who had lived in their homes more than 11 years, only 17% reported having invested in storm-safety improvements in their homes (19% for all tenures). The percentages for Isaac were somewhat higher (38% for those in their homes more than 11 years, 35% overall), but they were still low considering frequent incidence of hurricanes along the central Gulf Coast.

Ownership of federal flood policies was also limited. For example, in areas threatened by Sandy, only 53% of those living within a half mile (0.8 km) of water (bay or ocean) indicated that they owned flood policies (54%, including those who were uncertain whether the coverage was separate from the regular homeowners' policy), with this percentage only slightly higher (57%; 59% adjusted for uncertainty) among those who indicated living within one block of water. In areas threatened by Isaac, ownership of flood policies among those living in proximity of water was higher but still far from complete; among those living within a half mile of open water, only 70% indicated that they had purchased a federal flood policy. Although this incidence of flood insurance purchase might seem to be acceptable, it is certainly lower than desirable.

What explains the low ownership of flood policies among those at high risk from flood? One contributing mechanism may have been a mistaken belief among residents that their regular homeowners' policies covered them for flood losses. Specifically, across our whole sample, among the 42% who expressed the belief that they were insured against flood losses, only 51% indicated that they own a separate federal flood policy, with 3% being unsure. This implies that potentially half of the respondents who thought that they were covered in the event of a flood loss were, in fact, not.

DISCUSSION. One of the greatest challenges facing forecasters and emergency management officials worldwide is to design natural-hazard communication strategies that successfully encourage individuals in threatened areas to take appropriate protective actions--both in their responses to immediate threats as well as their long-term decisions about housing and personal risk management (Morss et al. 2010; Demuth et al. 2012). The enormous property losses that have occurred as a result of tropical cyclones in recent years, however, suggest that communication efforts have not been as effective as they might be. Individuals living in areas prone to flood risk have been found to chronically underinsure (e.g., Kousky and Michel-Kerjan 2012), and individuals fail to evacuate in the face of explicit warnings when faced with hurricane risks (Baker 1991; Huang et al. 2012).

What makes this problem particularly vexing in the case of tropical cyclone threats is that in recent years researchers have witnessed large gains in public awareness of these storms. When hurricanes approach coastlines in the United States, they are major media events; in our work, not a single respondent was unaware that his location was threatened either by Hurricane Isaac or Sandy, and the vast majority of respondents reported keeping regularly abreast of the latest storm news as each storm approached, with over 80% of respondents indicating their latest information was less than 2 hours old. Yet somehow this ubiquitous awareness did not translate into uniformly appropriate protective actions; only 55% of the respondents that we sampled whose homes were equipped with removable window protection installed it as the storms approached, and only a small proportion of those who believed that they were living in areas where evacuations had been advised expressed an intention to leave; we had no problem finding residents in evacuation areas at home to answer their phones as each storm approached.

The goal of this research was to complement earlier attempts to better understand the factors that underlie decisions to undertake protective action in the face of hurricane threats by reporting the findings of two "real time" surveys of coastal residents as hurricanes Isaac and Sandy approached the United States in 2012. The data provide the first look at how hurricane threat perceptions evolve over time in response to warnings as storms approach the coast, and how protective decisions are being made when the storm's outcome is still in doubt.

The findings provide what might be seen as a disquieting--and in some cases paradoxical--view of hurricane threat perceptions and response. As noted above there was universally high awareness about the threat posed by Isaac and Sandy as each approached the coast, but there also was evidence that residents held poor mental models of both the nature and duration of the personal impacts that the storms could have. One of the surprising results was that individuals overestimated the probability that their locations would be impacted by winds of hurricane force (75 mph or more) compared to scientific estimates provided by the National Hurricane Center, yet this pessimism did not translate to correspondingly high degrees of concern about the damage that such winds might cause or induce residents to take the kind of protective actions that such beliefs would seem to warrant. Only a fraction of those owning removable storm shutters put them up, few secured backup generators in anticipation of long power outages, and roughly only 20% made evacuation plans should they be needed.

There was also little evidence in the data that preparation was inhibited by social pressures, by beliefs that certain measures would be ineffective, or by barriers to undertaking them (Lindell and Pratter 2012). For example, when respondents who were aware they were living in evacuation areas were asked why they did not intend to leave (Q53), only 1% cited physical limitations, 1% cited that they were advised to stay by friends or relatives, and 7% cited that they desired to protect their homes. The most common reason was a belief that there was simply no need to; 75% indicated that they felt safe staying put.

Were these feelings of safety misplaced? One of the major findings of our work was that many residents misconstrued the primary locus of the threat posed by hurricanes as coming from wind rather than water. This is a bias, we should note, that has been observed in other contexts. For example, in surveys among Texas residents after Hurricanes Lili, Bret, and Rita, Lindell and Prater (2008), found that coastal residents similarly underestimated the risks posed by storm surge relative to wind, and concern about wind damage was more strongly associated with intentions to evacuate from future storms. Likewise, an excessive focus on wind rather than flooding risk was been cited as a major cause of lives lost in France during Cyclone Xynthia in February 2010 (Vinet et al. 2012). What was particularly notable was that we observed the tendency to underestimate the relative threat posed by water in Isaac and Sandy even among those for whom the threat should have been most salient; for example, in our Sandy survey, even people having waterfront properties and who held flood insurance policies felt that there was a higher probability that their homes would suffer damage from wind than flooding.

While the forces that gave rise to these poor mental models are uncertain, we can offer some speculations. First, some of the findings might be explained by endemic biases in how people perceive and respond to risk that have been observed in other contexts. For example, it has long been observed that when responding to hazards--be they natural, health, or man made--people are prone to believe that they will be less likely to suffer harm than others--an effect termed the optimistic bias (Shepperd et al. 2013; Trumbo et al. 2011; Weinstein 1980). The optimistic bias provides a natural explanation for why residents might display upwardly biased beliefs that the storms would bring hurricane-force winds to their locations but then express limited concern that such winds would cause personal harm.

But while inherent optimism might explain some aspects of the data, we suggest that other observed biases may have their root in how the risks of hurricanes are often communicated to residents. For example, one factor that would seem likely to contribute to an overweighing of wind over water risk is that that storm intensity is currently exclusively conveyed by National Oceanic and Atmospheric Administration (NOAA) by the Saffir-Simpson scale, which describes the maximum sustained winds that a storm possesses, not its maximum storm surge or flood threat. While the National Hurricane Center made clear efforts to warn residents of flood risk of each storm, our surveys revealed that residents nevertheless had a higher awareness of a storm's maximum winds rather than flood potential. Specifically, when respondents were asked to report what they believed Isaac's and Sandy's maximum winds and predicted maximum storm surges to be, respondents were much better at the former than the latter; whereas 88% of respondents in Isaac and 79% in Sandy could recall the wind forecast (Q7), only 67% in Isaac and 63% in Sandy could recall the storm surge forecast (Q9). Simple greater mental availability of the wind threat could explain at least some of the bias.

Another likely contributing factor is that in many cases wind damage is inherently easier to mentally simulate than flooding damage (Meyer 2006). Whereas we experience (modestly) high winds and see its consequences on a regular basis, flood events are rare. Mental simulation of flood losses would be particularly difficult for individuals whose homes are not in beachfront locations, where surge risks might easily be imagined. A New York resident living in a high-rise building in lower Manhattan during Sandy might thus be forgiven for overlooking "storm-surge risk" as a major personal threat, when, in fact, it was the greatest threat faced during the storm due to flooding, which could prohibit escape from the building and make the building uninhabitable for long periods.

What might be done to improve residents' mental models of tropical cyclone threats? As a starting point, the findings of this work strongly support recent calls for hurricane communication to focus less on a hurricane's maximum wind strength (which is typically found in small areas near the center) and more on the impacts that residents living in different areas are likely to experience, particularly with respect to flood (e.g., Demuth et al. 2012; Huang et al. 2012), or other attributes of a hurricane's wind field, such as size, duration, or directional uniformity (Czajkowski and Done 2013). Achieving this goal, however, is unlikely to be easy, as it will almost certainly require more than emphasizing flood risks in advisories and disseminating flood-risk maps to residents. As Hurricane Isaac approached the Louisiana coast, for example, the National Hurricane Center's advisories emphasized flooding (from surge and rain) as the primary threat posed by the storm (e.g., advisory 28, 27 August), and in Sandy the advisory headlines similarly emphasized surge risks. Likewise, residents cannot be assumed to develop better intuitions simply by providing better maps and evacuation-education programs before storms; prior research suggests that many residents do not know their evacuation zones, even when aided by a map (e.g., Baker 2005a,b; Arlikatti et al. 2006; Zhang et al. 2004).

Hence, if there is to be a solution, it will likely require an orchestrated suite of communication activities that characterize the strength of a storm in terms of both its size and nature of impacts, rather than just wind strength. For example, the Met Office has recently experimented with the use of color-coded "risk grids" that simultaneously convey the probability and severity of storm impacts (Demeritt 2012), and Morss et al. (2010) provide further support for the ability of individuals to utilize probabilistic forecasts.

Likewise, officials could consider exploring tools that would allow residents to more easily mentally simulate how storms could induce damage. To illustrate, in Sandy one of the greatest sources of personal property losses was from private automobiles--a loss that could easily have been avoided had residents simply known the damage that flood waters can do to a car and move them out of harm's way as the storm was approaching.

Of course, there are likely strong limits to what better education and more targeted communication might hope to achieve. In many cases the greatest source of decision errors in the face of hazards is that individuals are uncertain about the correct course of action and end up choosing familiar default options that are decidedly suboptimal for a given situation-such as choosing to stay when one is unsure whether to evacuate or, in the tragic case of Hurricane Sandy, choosing to evacuate by taking a familiar road that goes through an unmarked surge zone (Koplowitz 2012).

In this case we follow Thaler and Sunstein (2008) and others by suggesting that communities work to develop stronger sets of "decision defaults" that reduce the uncertainty that typically accompany individual decisions about when and how much to prepare. For example, Kunreuther and Michel-Kerjan (2009) have argued in support of long-term flood insurance contracts that have automatic annual decisions about renewal. Similar mechanisms could be extended to short-term preparedness, such as communities developing a program that annually distributes hurricane kits to all residents from which households can opt out--shifting the focus of decision making from that of whether one should prepare to whether one should not prepare.

Finally, our hope is that this research will spawn additional attempts to conduct real-time measurement of responses to natural hazards. The technical challenges of doing such work, however, are formidable. One of the limitations of relying on landlines as used here that we noted at the outset is the risk of sample-selection bias as storms approach; those who are more concerned with risk will be more likely to evacuate their homes, possibly resulting in a biased view of actual intended evacuation and storm preparation levels. While we offered evidence that in the case of Isaac and Sandy there was little sample attrition (e.g., home contact rates on the last day were not significantly different than the first), this cannot be expected to be the case in general given more severe storms. Likewise, another source of bias is the fact that wireless phones are increasingly replacing landlines as the major telecommunication channel used by households, particularly those who are younger [in 2012 the Centers for Disease Control and Prevention (CDC) estimated that 34% of U.S. households have only wireless phone service; Blumberg and Luke 2012],

As such, consideration needs to be given to alternative contact methods, such as brief surveys delivered to smart phones. Those methods, however, will have their own challenges, at least at this point in time. Aside from the pragmatic difficulties of implementing surveys on mobile phones [which are partially restricted under the Telemarketing Consumer Protection Act (TCPA) of 1991 (6)], there would be a loss of precise locational information, which is critical if one hopes to map risk perceptions to objective risk. One possibility might be to integrate real-time surveys into weather- and protection-related smart-phone apps where respondents give prior consent to responding to brief surveys and surrendering GPS location information. Such an approach might allow future research not just to replicate the work reported here but also investigate spatial dynamics, such as movement after warnings have been issued.

ACKNOWLEDGMENTS. This research was supported by research funds provided by the Wharton Risk Management and Decision Processes Center, the Florida Catastrophic Storm Risk Management Center, and Grants NSF-SES 0838650 and NSF-SES 0951516 from the National Science Foundation.


Arlikatti, S., M. K. Lindell, C. S. Prater, and Y. Zhang, 2006: Risk area accuracy and hurricane evacuation expectations of coastal residents. Environ. Behav., 38, 226-247, doi:10.1177/0013916505277603.

Avila, L. A., and J. Cangialosi, cited 2011: Tropical cyclone report: Hurricane Irene. National Hurricane Center, 45 pp. [Available online at www.nhc.noaa .gov/data/tcr/AL092011_Irene.pdf.]

Baker, E. J., 1991: Hurricane evacuation behavior. Int. J. Mass Emerg. Disasters, 9, 287-310.

--, 2005a: Nassau County hurricane evacuation behavioral survey. Report prepared for the U.S. Corps of Engineers.

--, 2005b: Suffolk County hurricane evacuation behavioral survey. Report prepared for the U.S. Corps of Engineers.

Berg, R., 2012: Tropical cyclone report: Hurricane Isaac (AL092012). National Hurricane Center, 78 pp. [Available online at /AL092012_Isaac.pdf.]

Bibel, S., cited 2012: The Weather Channel's Hurricane Sandy coverage garners the network's highest-ever ratings. [Available online at http://tvbythenumbers -hurricane-sandy-coverage-garners-the-networks -highest-ever-ratings/155312/.]

Blake, E. S., C. W. Landsea, and E. J. Gibney, 2011: The deadliest, costliest, and most intense tropical cyclones from 1851 to 2010 (and other frequently requested hurricane facts). NOAA Tech. Memo. NWS NHC-6, 47 pp. [Available online at www.nhc]

--, T. B. Kimberlain, R. J. Berg, J. P. Cangialosi, and J. L. Beven II, 2013: Tropical cyclone report: Hurricane Sandy (AL182012). NOAA Rep., 157 pp. [Available online at /AL182012_Sandy.pdf.[

Blumberg, S. J., and J. V. Luke, 2012: Wireless substitution: Early release of estimates from the National Health Interview Survey, January-June 2012. Division of Health Interview Statistics, National Center for Health Statistics, Centers for Disease Control and Prevention, 19 pp. [Available online at /nchs/data/nhis/earlyrelease/wireless201212.pdf.]

Broad, K., A. Leiserowitz, J. Weinkle, and M. Steketee, 2007: Misinterpretations of the "cone of uncertainty" in Florida during the 2004 hurricane season. Bull. Amer. Meteor. Soc., 88,651-667, doi:10.1175/BAMS-88-5-651.

Brown, R. S., C. W. Williams, and P. R. Lees-Haley, 1994: The effects of hindsight bias on causal attribution on human response to natural events. /. Appl. Soc. Psychol., 24,661-674, doi: 10.1111/j.l559-1816.1994.tb00606.x.

Cangialosi, J. P., and J. L. Franklin, 2013: 2012 National Hurricane Center forecast verification report. NOAA National Hurricane Center Tech. Rep., 79 pp. [Available online at /pdfs/Verification_2012.pdf.]

Czajkowski, J., and J. Done, 2013: As the wind blows? Understanding hurricane damages at the local level through a case study analysis. Wea. Climate Soc., 6, 202-217, doi:10.1175/WCAS-D-13-00024.1.

Dash, N., and H. Gladwin, 2007: Evacuation decision making and behavioral responses: Individual and household. Nat. Hazards Rev., 8, 69-77, doi:10.1061 /(ASCE)1527-6988(2007)8:3(69).

Demeritt, D., 2012: The perception and use of public weather services by emergency and resiliency professionals in the UK. Report for the Met Office Public Weather Service Customer Group, 38 pp. [Available online at /departments/geography/people/academic/demeritt /DemerittPWSCGreport02032012FINAL.pdf.[

Demuth, J. L., R. E. Morss, B. H. Morrow, and J. K. Lazo, 2012: Creation and communication of hurricane risk information. Bull. Amer. Meteor. Soc., 93,1133-1145, doi: 10.1175/BAMS-D-l 1-00150.1.

Dow, K., and S. L. Cutter, 1998: Crying wolf: Repeat responses to hurricane evacuation orders. Coastal Manage., 26, 237-252, doi:10.1080/08920759809362356.

Fischhoff, B., and R. Beyth, 1975: I knew it would happen: Remembered probabilities of once--Future things. Organ. Behav. Hum. Perform., 13, 1-16, doi:10.1016/0030-5073(75)90002-1.

Gall, R., J. Franklin, F. Marks, E. N. Rappaport, and F. Toepfer, 2013: The Hurricane Forecast Improvement Project. Bull. Amer. Meteor. Soc., 94, 329-343, doi:10.1175/BAMS-D-12-00071.1.

Huang, S.-K., M. J. Lindell, C. Prater, H.-C. Wu, and L. K. Siebeneck, 2012: Household evacuation decision making in response to Hurricane Ike. Nat. Hazards Rev., 13, 283-296, doi:10.1061/(ASCE)NH.1527 -6996.0000074.

Jones, N. A., H. Ross, T. Lynam, P. Perez, and A. Leitch, 2011: Mental models: An interdisciplinary synthesis of theory and methods. Ecol. Soc., 16, (46). [Available online at /art46/.]

Koplowitz, H., 2012: Glenda Moore, Staten Island mother who lost 2 sons in Hurricane Sandy, returns to site of tragedy. [Available online at www.ibtimes .com/glenda-moore-staten-island-mother-who -lost-2-sons-hurricane-sandy-returns-site-tragedy -858289.]

Kousky, C., and E. Michel-Kerjan, 2012: Hurricane Sandy's storm surge and the NFIP: Flood insurance coverage in New York and New Jersey. Wharton Center for Risk Management and Decision Processes Issue Brief, 4 pp. [Available online at /WRCib2012c_Sandy-and-the-NFIP.pdf.]

Kunreuther, H. C., and E. 0. Michel-Kerjan, 2009: At War with the Weather: Managing Large-Scale Risks in a New Era of Catastrophes. MIT Press, 464 pp.

Lindell, M. K., and C. S. Prater, 2008: Behavioral analysis: Texas evacuation study. Texas A&M University Hazard Reduction and Recovery Center Tech. Rep.

--and--, 2012: The Protective Action Decision Model: Theoretical modifications and additional evidence. Risk Anal., 32,616-632, doi: 10. Ill l/j.1539 -6924.2011.01647.x.

--, J.-C. Lu, and C. S. Prater, 2005: Household decision making and evacuation in response to Hurricane Lili. Nat. Hazards Rev., 6, 171-179, doi:10.1061 /(ASCE)1527-6988(2005)6:4(171).

Lupkin, S., 2012: Hurricane Sandy: New Jersey warned of 7- to 10-day power outages. [Available online at -prepares-10-day-power-outages/story?id=17579827# .UZVdQbWR-O0.]

Meyer, R. J., 2006: Why we underprepare for hazards. On Risk and Disaster: Lessons from Hurricane Katrina, R. J. Daniels, D. F. Kettl, and H. Kunreuther, Eds., University of Pennsylvania Press, 153-174.

Morss, R. E., and M. H. Hayden, 2010: Storm surge and "certain death": Interviews with Texas coastal residents following Hurricane Ike. Wea. Climate Soc., 2, 174-189, doi:10.1175/2010WCAS1041.1.

--, J. K. Lazo, and J. L. Demuth, 2010: Examining the use of weather forecasts in decision scenarios: Results from a US survey with implications for uncertainty communication. Meteor. Appl., 17, 149-162, doi:10.1002/met.196.

Neisser, U., E. Winograd, E. T. Bergman, C. A. Schreiber, S. E. Palmer, and W. S. Weldon, 1996: Remembering the earthquake: Direct experience vs. hearing the news. Memory, 4, 337-358, doi: 10.1080/096582196388898.

Norris, F. H., and K. Kaniasty, 1992: Reliability of delayed self-reports in disaster research. /. Trauma. Stress, 5, 575-588, doi:10.1002/jts.2490050407.

National Weather Service, cited 2012: Hurricane Sandy advisory number 26. [Available online at www.nhc]

Peacock, W. G., S. D. Brody, and Wes Highfield, 2005: Hurricane risk perceptions among Florida's single family homeowners. Landscape Urban Plann., 73, 120-135, doi:10.1016/j.landurbplan.2004.11.004.

Pew Research Center, 2012: Assessing the representativeness of public opinion surveys. Pew Research Center Rep., 51 pp. [Available online at www the%20Representativeness%20of%20Public%20 Opinion%20Surveys.pdf.]

Pielke, R. A., Jr., cited 2012: Updated: Normalized hurricane damage 1900-2012. [Available online at -normalized-hurricane-losses.html.]

--, J. Gratz, C. W. Landsea, D. Collins, M. A. Saunders, and R. Musulin, 2008: Normalized hurricane damage in the United States: 1900-2005. Nat. Hazards Rev., 9,29-42, doi:10.1061/(ASCE)1527-6988(2008)9:l(29).

Shepperd, J. A., W. M. P. Klein, E. A. Waters, and N. D. Weinstein, 2013: Taking stock of unrealistic optimism. Perspect. Psychol. Sci., 8, 395-411, doi:10.1177 /1745691613485247.

Taylor, A., 2013: Hurricane Sandy: 80 days later. [Available online at /hurricane-sandy-80-days-later/100440.]

Thaler, R. H., and C. R. Sunstein, 2008: Nudge: Improving Decisions about Health, Wealth, and Happiness. Yale University Press, 320 pp.

Trumbo, C., M. Lueck, H. Marlatt, and L. Peek, 2011: The effect of proximity to Hurricanes Katrina and Rita on subsequent hurricane outlook and optimistic bias. Risk Anal., 31, 1907-1918, doi:10.1111/j.1539-6924.2011.01633.x.

Vinet, F., D. Lumbroso, S. Defossez, and L. Boissier, 2012: A comparative analysis of the loss of life during two recent floods in France: The sea surge caused by the storm Xynthia and the flash flood in Var. Nat. Hazards, 61, 1179-1201, doi:10.1007/sll069 -011-9975-5.

Weinstein, N., 1980: Unrealistic optimism about future life events. J. Pers. Soc. Psychol., 39, 806-820, doi:10.1037/0022-3514.39.5.806.

Zhang, F., and Coauthors, 2007: An in-person survey investigating public perceptions of and responses to Hurricane Rita forecasts along the Texas coast. Wea. Forecasting, 22, 1177-1190, doi:10.1175/2007WAF2006118.1.

Zhang, Y., C. S. Prater, and M. K. Lindell, 2004: Risk area accuracy and evacuation from Hurricane Bret. Nat. Hazards Rev., 5, 115-120, doi:10.1061/(ASCE)1527 -6988(2004)5:3(115).

(1) Studies that have directly measured recall accuracy for natural hazards have shown reasonably high test-retest reliability in stated reports, something that would seem to assuage this concern (e.g., Neisser et al. 1996; Norris and Kaniasty 1992). The limitation of this work, however, is that, because there were no measures taken before events, little is known about whether postevent reports are influenced by hindsight bias and the degree to which test-retest reliability was inflated by temporal nonindependence of the measures.

(2) Details of these analyses are available upon request.

(3) It is possible, of course, that respondents were using social media for functions other than as a source of factual storm information.

(4) The most common error was to believe that they were still under a hurricane watch (22% in Isaac and 27% in Sandy).

(5) The optimistic assessments of likely durations of power outages are consistent with those uncovered by Baker (2005a,b) in prestorm surveys among residents in Nassau and Suffolk Counties in Long Island, where only 4% of respondents in Nassau and 15% in Suffolk believed they would lose power for more than 2 days in the advent of a category 1 hurricane--the strength of Sandy.

(6) The TCPA allows surveys to be conducted on cell phone numbers as long as they are manually dialed. Even in such cases, rates of compliance may be affected because it may be the responsibility of the recipient to pay for incoming call time.

AFFILIATIONS: Meyer and Czajkowski--University of Pennsylvania, Philadelphia, Pennsylvania; Baker--Florida State University, Tallahassee, Florida; Broad--University of Miami, Miami, Florida; Orlove--Columbia University, New York, New York CORRESPONDING AUTHOR: Ben Orlove, Columbia University, 833 International Affairs Building, New York, NY 10027 E-mail:

The abstract for this article can be found in this issue, following the table of contents.

DOI: 10.1175/BAMS-D-12-00218.1

A supplement to this article is available online (10.1175/BAMS-D-12-00218.2)

In final form 12 January 2014

TABLE 1. Respondent socioeconomic/demographic profile summary

  Characteristic                     Isaac survey     Isaac survey
                                    respondents (%)   counties (%)

Homeowner status *
  Homeowner                               93               59
  Rent                                     6               27
  Other/refused/vacant (counties)          1               14
Age *
  Under 30                                 3               40
  30-60 (respondents)/30-59               52               41
  61-80 (respondents)/60-79               34               16
  Over 80 (respondents)/80+                6               3
  Other/refused                            5               --
Race *
  African American or black               13               22
  Caucasian or white                      83               71
  Other/refused                            4               7
Education level **
  Some high school/high school            25               42
  Some college/college graduate           53               46
  Postgraduate                            14               7
  Other/refused/less than high             8               5
    school (counties)
2011 total household income **
  Less than $15,000                        5               13
  $ 15,000-$39,999 (respondents)/         16               23
    $15,000-$34,999 (counties)
  $40,000-$79,999 (respondents)/          25               33
    $35,000 to $74,999 (counties)
  Over $80,000 (respondents)/             21               31
    over $75,000 (counties)
  Other/refused                           33               --
Resident type
  Live here year-round                    98               --
  Vacationing                              2               --
  Other/refused                            0               --

  Characteristic                     Sandy survey     Sandy survey
                                    respondents (%)   counties (%)

Homeowner status *
  Homeowner                               89               57
  Rent                                     9               30
  Other/refused/vacant (counties)          1               13
Age *
  Under 30                                 4               39
  30-60 (respondents)/30-59               41               42
  61-80 (respondents)/60-79               40               15
  Over 80 (respondents)/80+                8               4
  Other/refused                            7               --
Race *
  African American or black                8               13
  Caucasian or white                      83               69
  Other/refused                            9               18
Education level **
  Some high school/high school            26               41
  Some college/college graduate           48               43
  Postgraduate                            20               10
  Other/refused/less than high             6               6
    school (counties)
2011 total household income **
  Less than $15,000                        2               11
   $15,000-$39,999 (respondents)/         10               19
    $15,000-$34,999 (counties)
  $40,000-$79,999 (respondents)/          14               31
    $35,000 to $74,999 (counties)
  Over $80,000 (respondents)/             23               39
    over $75,000 (counties)
  Other/refused                           51               --
Resident type
  Live here year-round                    97               --
  Vacationing                              1               --
  Other/refused                            2               --

* County data from 2010 U.S. census.

** County data from 2007-2011 American Community Survey. Education
level based on total population of residents that are 25 years
old and older (approximately 88% of the total adult population).
2011 total household income based on occupied (owner and renter)
housing units income data in 2011 inflation-adjusted dollars.

TABLE 2. Regression of wind-water belief bias: Hurricane Sandy.

No. of observations       385
F(8, 376)                 2.07
Prob > F                  0.0382
R-squared                 0.0404
RMSE                      26.85

Predictor *               Estimate     Standard     t value   Pr > t

Single-family home         5.163176    3.256452     1.59      0.114
Within 1 block of water   -5.710127    3.828923    -1.49      0.137
Within 1 mile of water    -5.827326    3.153236    -1.85      0.065
Have a separate flood     -4.323446    3.157173    -1.37      0.172
Education level           -1.492738    1.303257    -1.15      0.253
Age                        2.219887    1.03686      2.14      0.033
Male                       0.0377965   3.149295     0.01      0.990
Experienced hurricane     -2.603693    3.200698    -0.81      0.416
  in past
Constant                   9.720106    7.77089      1.25      0.212

* Single family home = 1 for Q55 "detached single-family home," 0
otherwise (81% of the home type observations for Sandy are
single-family detached homes); within I block of water = 1 for Q80
"directly on the water" and "within 1 block of water," 0 otherwise;
within 1 mile of water = 1 for Q80 "within 1/2 mile of the water"
and "within 1 mile of the water," 0 otherwise; over 1 mile of water
= 1 for Q80 "more than 1 mile of the water," 0 otherwise and is the
omitted dummy category. Have a separate flood policy = 1 for Q60
"yes," 0 otherwise; education level = Q75 discrete values 1-5; age
= Q72 discrete values 1-6; male = 1 for Q86 "male," 0 otherwise.
Experienced hurricane in past = 1 Q63 "yes" or Q64 "yes," 0
COPYRIGHT 2014 American Meteorological Society
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2014 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Meyer, Robert J.; Baker, Jay; Broad, Kenneth; Czajkowski, Jeff; Orlove, Ben
Publication:Bulletin of the American Meteorological Society
Geographic Code:1USA
Date:Sep 1, 2014
Previous Article:North Atlantic tropical cyclones and U.S. flooding: North Atlantic tropical cyclones are responsible for major flooding over large areas of the...
Next Article:Impact of typhoons on the ocean in the Pacific: the interactions between typhoons and the ocean vary greatly depending on the properties of the storm...

Terms of use | Privacy policy | Copyright © 2021 Farlex, Inc. | Feedback | For webmasters