Wind Insurance and Mitigation in the Coastal Zone

Daniel R. Petrolia, Joonghyun Hwang, Craig E. Landry and Keith H. Coble

Abstract

This paper presents the only study analyzing the decision to purchase wind coverage for individuals whose standard homeowner’s policy excludes wind, and one of very few analyses of the decision to undertake wind mitigation measures. Because these two decisions are closely related, a simultaneous mixed-process approach is used that allows for correlated disturbances across probit (insurance) and tobit (mitigation) equations. Results indicate a positive correlation between errors of the insurance and mitigation models; conditioning on covariates, households that hold wind insurance tend to engage in greater levels of wind mitigation. (JEL C35, Q54)

I. INTRODUCTION

Coastal properties along the Atlantic and Gulf Coasts are at risk both of wind and flood damages due to tropical storms and hurricanes. Although the source of risk is the same (hurricanes), mitigation measures and insurance products to address wind and flood peril differ. Flood mitigation generally focuses on elevation of the property, along with other waterproofing measures. Wind mitigation involves steps to increase the structural soundness of the property, such as installation of storm shutters, reinforcing doors and windows, and the use of particular roof designs and attachments that reduce the risk of the roof being blown off.

Regarding insurance in the United States, coverage for flood peril is almost universally excluded from standard residential and commercial property insurance policies offered by private insurers, and offered instead through the National Flood Insurance Program. Although coverage for wind peril is generally included in standard property policies, in many coastal areas at high risk of wind damage due to tropical storms and hurricanes, wind coverage is excluded. Property owners who still wish to have such coverage must purchase either a separate wind-only policy from a private insurer or, as is more often the case, obtain coverage through state-run insurance programs (Kousky 2011; USGAO 2008).

The objective of this paper is to address the question What factors influence homeowners’ decisions in the face of wind risk? Specifically, What factors affect wind insurance purchase and wind mitigation decisions, and how are these two decisions related to one another? We find that although there are many studies in the literature focused on insurance and mitigation under hurricane risk, they are by and large focused on flood risk, with relatively little written with regard to wind risk. The only exceptions that appear to focus specifically on wind damage mitigation are studies by Carson, McCullough, and Pooser (2013), Peacock (2003), and to some extent, Meyer et al. (2014).1The present study is similar to that of Peacock (2003) and Meyer et al. (2014) in that it collects household-level survey data, including data on risk perceptions, whereas Carson, McCullough, and Pooser’s (2013) is not at the household level and thus contains no subjective risk information. The present study differs from all three of the previous studies in that we also collect risk preference information, and we sample households from all five of the Gulf states, including Florida’s Atlantic coast, whereas Peacock (2003) and Carson, McCullough, and Pooser (2013) focus on Florida only, and Meyer et al. (2014) focus on the middle Atlantic (from Virginia to northeastern New Jersey) and the central Gulf Coast (from the easternmost parishes in Louisiana to the two western-most counties in Florida). Also, Meyer et al. (2014) do not conduct any behavioral analysis but simply report summary statistics of data collected. Although Nyce and Maroney (2011) analyze factors that influence wind coverage rates in Florida, there are no studies, to our knowledge, that examine the decision to buy wind coverage when it is excluded from one’s regular homeowner’s policy. Thus, we offer the only known analysis focused on the decision to purchase wind coverage.

The analysis utilizes data from a 2010 survey of 790 coastal homeowners in Alabama, Florida, Louisiana, Mississippi, and Texas. The survey focuses on mitigation and insurance decisions, perceptions, and preferences regarding both flood and wind risk stemming from hurricanes. Results indicate that respondents living in areas where wind peril is excluded from regular homeowner policies, that is, where wind risk is highest, actually tend to undertake fewer mitigation activities. Also, those who buy wind-only policies also tend to undertake more mitigation activities. Respondents who have experienced storm damage in the past as well as those living in the coastal zone are both more likely to buy a wind-only policy and to undertake more mitigation.

II. OVERVIEW OF STATE-OPERATED WIND INSURANCE PROGRAMS

Originally, most private insurance plans provided protection for wind peril (Insurance Information Institute 2013). Insurers, however, have canceled such coverage in areas where a high risk of hurricanes exists, and for this reason, several state governments have been pressured to intervene in the wind insurance market. State-run insurance programs take a variety of forms including Fair Access to Insurance Requirements (FAIR) plans,2 wind pools, hybrid programs that write both dwelling and hazard-specific policies, and reinsurance funds that provide secondary insurance for primary insurers.

State-run windstorm underwriters associations, also known as “wind pools” or “beach plans,” offer wind coverage where private insurance is not available. As of 2008, wind pools covered more than $17 billion worth of property (Pompe and Rinehart 2008). Today, wind pools exist in Alabama, Florida, Georgia, Hawaii, Louisiana, Mississippi, North Carolina, South Carolina, and Texas (American Insurance Association 2013; Kousky 2011). State programs have different pricing goals. For example, Alabama Insurance Underwriting Association and Louisiana Citizens Property Insurance Corporation are legally bound to set prices above those in the private market (Alabama Insurance Underwriting Association 2013a; Kousky 2011; Louisiana Citizens Property Insurance Corporation 2013). The language adopted by the Texas Windstorm Insurance Association is that “Rates must be reasonable, adequate, not unfairly discriminatory, and noconfiscatory as to any class of insurer.”3 Florida Citizens originally required higher prices than the private market, but the requirement was abandoned in 2007; Florida Citizens is now more actively competing in the private market (Kousky 2011).

According to a survey of Gulf Coast states conducted by the Texas Department of Insurance (2012), average residential wind-only premiums are highest in Alabama and Mississippi (approximately $11 per $1,000 of insured value); Texas and Louisiana have average prices slightly above $5 per $1,000 of insured value, and Florida has the lowest average price (approximately $4 per $1,000 insured value). These programs offer a variety of premium discounts for homeowners that meet certain building codes and/or adopt additional mitigation measures (Alabama Insurance Underwriting Association 2013b; Citizens Property Insurance Corporation 2014; Louisiana Citizens Property Insurance Corporation 2013; Mississippi Windstorm Underwriting Association 2013; Texas Windstorm Insurance Association 2013). For more information on this topic, see Kousky (2011) for a detailed summary of state-run insurance programs, including their origins, how they are funded and operated, pricing strategies, and means for dealing with claims during catastrophic event years, and Dixon, Macdonald, and Zissimopoulos (2007) for a summary of changes in market conditions specific to wind insurance in the Gulf states since the 2005 hurricane season.

III. CONCEPTUAL MODEL

We assume that individual i makes a marginal benefit/cost calculation based on the subjective expectation of utility Ui = U(Hi,Xi,Ci,Mi), where Hi is consumption of housing, Xi is consumption of a numeraire good, Ci = {0,1} represents the decision to purchase or not purchase wind coverage when it is excluded from the homeowner’s policy, and Mi ≥ 0 represents the number of mitigation measures undertaken on the structure. Modifying the flood insurance model presented by Petrolia, Landry, and Coble (2013), we assume that individuals choose insurance and mitigation optimally, giving rise to reduced-form Marshallian-type demand equations given by Embedded Image and Embedded Image, where Embedded Image represents the perceived likelihood of a major storm event; Embedded Image, perceived conditional expected loss (which depends upon the level of mitigation); λi, perceived likelihood of insurance payoff (insurer credibility); γi, perceived likelihood of disaster assistance; αi, risk aversion; δi, prior experience with wind damage; πc(pi,Mi), price of insurance (which depends upon objective measures of risk for property i, pi, and may also be a function of some subset of mitigation measures Mi taken); πM(pii), price of mitigation (which depends upon objective measures of risk for property i, pi, the level of wind mitigation assistance available, τi, and likely to be nonlinear in M); and wi, household wealth.

Thus, conceptually, these two decisions are influenced by many of the same independent factors, but also jointly affected by prices of insurance and mitigation. Mitigation also has indirect effects on insurance coverage by influencing the level of the conditional expected loss, Li(Mi), and insurance price, πC(pi,Mi) (to the extent that prices are adjusted for mitigation). Generally speaking, wind insurance and mitigation can be jointly determined but also may be chosen sequentially. Also, if households do not recognize the linkage, it is also possible that choices regarding insurance and mitigation are independent. We posit that the effect of the mitigation decision on the insurance decision is indirect, via the arguments Li(Mi) and πC(pi,Mi) (and possibly πM(pii), if it too, is a function of M), with the expectation that ∂Li/∂Mi ≤ 0 and ∂πC/∂Mi ≤ 0. Consequently, the overall effect of the mitigation decision on the insurance decision is indeterminate. Demand for wind insurance and mitigation measures can also be affected by their relative prices, which may change given the possible nonlinearity in mitigation pricing. Prices of insurance will vary spatially, while mitigation prices will also exhibit individual-level, idiosyncratic variations. In general, basic theory provides little guidance in modeling joint insurance and mitigation purchase decisions, and econometric identification can be problematic.

Below we discuss the remaining explanatory factors, specifically, why each is included, what the theoretical expectations are, and what has been found empirically in previous literature. Because relatively few studies focus on wind mitigation, and none, to our knowledge, focus on wind insurance, we rely heavily on the literature for flood insurance and mitigation; the few studies that focus on wind mitigation; literature on evacuation and other decisions under hurricane, earthquake, and wildfire risk; as well as choice under risk in general, to guide our analysis. See Table 7 for a summary of hypothesized signs for each variable.

Rees and Wambach (2008) show that demand for insurance depends on the perceived likelihood and conditional magnitude of loss. We hypothesize that both the likelihood of purchasing wind coverage and the extent of mitigation are increasing in perceived likelihood of loss, which we proxy with perceived likelihood of a major storm event, and increasing in the conditional magnitude, which is measured as the percentage of home value lost conditional on a major storm event occurring. Empirically, Petrolia, Landry, and Coble (2013) find that flood insurance purchase is increasing in terms of magnitude of conditional iw0b loss, but not for perceived storm frequency. Regarding mitigation, Botzen, Aerts, and van den Bergh (2009) and Peacock (2003) find a significant effect of perceived risk on flood mitigation. Whitehead (2005) and Stein, Dueñas-Osorio, and Subramanian (2010) find that increased perceptions of hurricane risk increase the likelihood of evacuation. Martin, Martin, and Kent (2009) and Talberth et al. (2006) find that perceptions of wildfire risk significantly affect risk reduction behavior.

Dixon, Macdonald, and Zissimopoulos (2007) discuss the potential negative impacts of perceived insurance contract uncertainty resulting from extended claims litigation related to the 2005 hurricane season. The concept of insurer credibility is very closely related to the concept of probabilistic insurance (Wakker, Thaler, and Tversky 1997), where it is found that individuals tend to demand disproportionate premium discounts to compensate for insurer default risk. We hypothesize that confidence in the insurer to honor the terms of a policy should increase the probability of purchasing a wind policy, but have the opposite effect for mitigation, ceteris paribus. If individuals perceive insurers to be noncredible, then they will likely substitute toward mitigation. Empirically, Petrolia, Landry, and Coble (2013) find a positive effect of insurer credibility on the flood insurance purchase decision.

The charity hazard hypothesis (Browne and Hoyt 2000; Kaplow 1991; Raschky and Weck-Hannemann 2007) posits that eligibility for postdisaster aid should decrease the likelihood of individual-level protection. We adopt the same expected effect here, that is, that heightened perceptions of eligibility for postdisaster aid will have a negative effect on both wind insurance purchase and wind mitigation investments. Empirical findings, however, are mixed: Botzen and van den Bergh (2012a, 2012b) and Raschky et al. (2013) find empirical evidence in support of the charity hazard hypothesis. Petrolia, Bhattacharjee, and Hanson (2011) find that those who are confident in being rescued after a hurricane are less likely to evacuate. On the other hand, Browne and Hoyt (2000) and Petrolia, Landry, and Coble (2013) find an opposite effect, namely, a significant positive effect on insurance purchase.

Models of insurance demand in the expected utility framework generally assume some degree of risk aversion, in other words, that the individual’s utility increases at a decreasing rate with respect to wealth (Varian 1992). Accordingly, we hypothesize that both decisions to purchase wind coverage and to mitigate are increasing in the degree of risk aversion. Empirically, measures of risk aversion have been found to be significant explanatory factors in a variety of settings (Holt and Laury 2002; Kachelmeier and Shehata 1992; Lusk and Coble 2005). Petrolia, Landry, and Coble (2013) find this to be true specifically for the case of purchasing flood insurance, in other words, the probability of purchasing a flood policy is increasing in the degree of risk aversion. Regarding mitigation, Briys and Schlesinger (1990) and Dionne and Eeckhoudt (1985) find that the choice to mitigate is positively related to risk aversion.

It is hypothesized that personal experience with storm damage will heighten one’s awareness of the risks associated with hurricanes, and consequently increase the probability of both purchasing wind insurance and undertaking mitigation. Empirically, this has been found to be the case: Baumann and Sims (1978), Browne and Hoyt (2000), Carbone, Hallstrom, and Smith (2006), Kriesel and Landry (2004), Kunreuther (1978), and Petrolia, Landry, and Coble (2013) find that past damage positively affects flood insurance choice, and Botzen, Aerts, and van den Bergh (2009) and Brody et al. (2009) find the same relationship for mitigation choice. Peacock (2003) finds a slightly different relationship: that past damage is not significant, but a more general measure of past experience with hurricanes is significant in explaining mitigation decisions. In related settings, Naoi, Seko, and Ishino (2012) find that past earthquake damage positively affects seismic retrofitting in the future; Petrolia, Bhattacharjee, and Hanson (2011) find mixed evidence as to whether the decision to evacuate under hurricane risk is influenced by past evacuation decisions, and Martin, Martin, and Kent (2009) find no significant effect of past wildfire experience on risk reduction behavior.

Regarding characteristics of the property, it is hypothesized that because damages associated with hurricane winds generally increase with greater proximity to the coastline, the probability of both purchasing wind coverage and mitigating is also increasing in proximity to the coastline. Kriesel and Landry (2004), Landry and Jahan-Parvar (2011), and Petrolia, Landry, and Coble (2013) have found this to be the case empirically for flood insurance purchases, and Botzen, Aerts, and van den Bergh (2009) and Peacock (2003) find the same for mitigation decisions. Similarly, the effect of living in areas of increased risk distinct from proximity to the coast, such as being in a noncoastal flood zone, can affect insurance and mitigation decisions. Landry and Jahan-Parvar (2011) and Petrolia, Landry, and Coble (2013) find that being located in areas of increased flood and/or erosion risk increases the probability of holding a flood insurance policy. In the case of coastal wind risk, this risk is generally reflected in proximity to the coastline, but specific to the mitigation decision, we have an additional indicator of increased risk: whether a homeowner’s policy excludes wind peril. Lohse, Robledo, and Schmidt (2012) find that when insurance is not available, higher levels of self-insurance are optimal, and Carson, McCullough, and Pooser (2013) extend this finding to cases in which insurance is “very expensive or scarce.” Thus, we hypothesize that individuals are more likely to mitigate when they live in the areas where wind coverage is excluded from regular homeowner’s policy.

Assuming greater vulnerability in older homes, the benefits of wind insurance could increase with home age. At the same time, however, older homes tend to require higher premiums to be insured. Empirically, Grace, Klein, and Kleindorfer (2004) find that the age of the home is negatively correlated with the demand for noncatastrophic homeowner’s insurance but is positively correlated with catastrophic homeowner’s insurance. Similarly, in the case of mitigation, the benefits from undertaking mitigation for older homes are not necessarily higher, because the effectiveness of mitigation features on a poorer structure may be questionable. Empirically, Carson, McCullough, and Pooser (2013) find that the age of the home negatively affects mitigation, and Dumm, Sirmans, and Smersh (2011) find that homes built under newer building codes that require mitigation activities sell more than otherwise similar homes. Peacock (2003) finds a positive effect on mitigation for a closely related variable, years in residence in the home, but also finds a positive effect of stronger building codes, indicating that his findings are not necessarily inconsistent with the former findings. Given the possible interpretations, we hypothesize that the effect of age of the home is sign indeterminate.

The effect of living in a condominium relative to a single-family structure on the decision to mitigate is not clear. It may be that such residents are less likely to mitigate given the higher probability of restrictions on what activities residents may undertake. On the other hand, they may be required to undertake certain mitigation activities that they would not undertake otherwise. It is also not clear whether such residents are more or less likely to purchase wind insurance, and we are aware of no prior study that addresses this issue. Petrolia, Landry, and Coble (2013) find that mobile home residents are less likely to purchase flood insurance, and Dixon, Macdonald, and Zissimopoulos (2007) find evidence that following the 2004 and 2005 hurricane seasons, changes in insurance price and availability varied by structure type. For example, premiums increased more rapidly for light metal and light wood-frame structures built before 1995 than for other types of structures. Thus, we hypothesize the same price effect for wind coverage. Its effect on mitigation, however, is not clear; given that these residents are less likely to insure, they may be more likely to mitigate. On the other hand, there may be fewer mitigation options available to mobile home residents. In the evacuation literature, Petrolia, Bhattacharjee, and Hanson (2011) and Whitehead (2005) find that mobile home residents are more likely to evacuate for hurricanes, though it is unclear how much we should expect this relationship to transfer over to mitigation.

Although we are aware of no legal requirements for wind coverage attached to a mortgage, Dixon, Macdonald, and Zissimopoulos (2007) find evidence of some lenders declaring existing loans in default, or refusing new loans, if insurance was inadequate. Additionally, Grace, Klein, and Kleindorfer (2004) find that the mortgage contract has a negative effect on the demand for noncatastrophic insurance but has a positive effect on catastrophic insurance. Based on these findings, we hypothesize that the decision to hold a wind policy is positively related to the presence of a mortgage. It is less clear how the presence of a mortgage influences the decision to mitigate. It is possible that residents currently paying on a mortgage will have less disposable income to spend on mitigation; on the other hand, it is possible that some lenders will encourage, if not require, homeowners to undertake certain mitigation activities as a condition for securing a loan.

We hypothesize that the probability of holding wind coverage and undertaking mitigation is increasing in wealth, which we proxy with a measure of household income. Browne and Hoyt (2000) and Petrolia, Landry, and Coble (2013) find this to be the case for flood insurance, and Brody et al. (2009), Carson, McCullough, and Pooser (2013), and Peacock (2003) find this to be true for mitigation. Botzen, Aerts, and van den Bergh (2009), however, find no such effect. Naoi, Seko, and Ishino (2012) find that household income has a positive effect on both purchasing earthquake insurance and seismic retrofitting, though they find that insurance purchase is negatively affected by overall wealth.

Individual mitigation efforts and demand for insurance can be influenced by mitigation projects and resources available at the community level. Community-level mitigation actions, such as maintaining green space to control flood waters or tree buffers to inhibit wind, may encourage individual risk-management actions by drawing attention to risk factors or it may crowd out (substitute for) individual actions. Petrolia, Landry, and Coble (2013) find that as Community Rating System scores decrease (indicating increased community-level flood preparedness), the likelihood of holding flood coverage increases, but they also find that among Florida respondents, the opposite is true. In other words, they find evidence that within Florida, increased community-level flood-mitigation activity has a negative effect on individual-level insurance purchases. Prante et al. (2011) find evidence that public spending on wildfire risk reduction can decrease the level of private mitigation. With regard to wind risk, however, community-level options for mitigation are very limited. Nonetheless, the U.S. federal government does make financial resources available for hazard mitigation, disbursing funds at the county level. These funds can be used for community projects or allocated to individuals for household mitigation projects, such as elevation of structures, retrofitting of buildings for wind resistance, or acquisition of homes in high-hazard areas. Thus, mitigation grant dollars can be used to implement community projects or subsidize individual mitigation efforts. Given these divergent uses and mixed results in the literature, we adopt no sign-determinate hypothesis regarding available mitigation grant dollars.4

Kunreuther and Kleffner (1992) argue that individuals may undertake mitigation even if they have full insurance coverage in order to prevent injury or death, and Carson, McCullough, and Pooser (2013) find that the presence of children in the home increases the likelihood of mitigation. Petrolia, Landry, and Coble (2013), however, find no significant effect of the presence of children on the insurance decision. Based on these findings, we adopt the hypothesis that the presence of children increases the number of mitigation activities. As for insurance, we have no expectation regarding the direction of the effect of children.

It is also possible that the insurance and mitigation decisions are a function of demographic differences, such as race and gender, but there is no clear theoretical basis. Petrolia, Landry, and Coble (2013) find that Hispanic households are significantly more likely to hold a flood policy, although Peacock (2003) finds that mitigation behavior of Hispanics is not statistically different from the behavior of whites, but is significantly lower for blacks. Given these mixed findings, we adopt no sign-determinate hypotheses on race. Evidence on the effect of gender on insurance or mitigation choice appears to be very limited. Botzen, Aerts, and van den Bergh (2009) and Petrolia, Landry, and Coble (2013) test for but find no significant effect of gender on mitigation or insurance, respectively. Although Cohen and Einav (2007) find that females are more likely to choose a low deductible on automobile insurance, Kruse and Thompson (2003) find no significant differences between males and females in their willingness to pay for a hypothetical risk-mitigation investment. Based on this scant evidence, we adopt no sign-determinate hypotheses for gender. We separate the sample into three categories based on both gender and marital status—married households, single males, and single females—to capture real gender effects rather than just differences in reporting.

It is reasonable to hypothesize that because of geographic, legal, and other administrative differences, behavior varies by state. Empirically, Brody, Kang, and Bernhardt (2010) find that Florida communities undertake more nonstructural mitigation activities relative to communities in Texas, but find no significant difference between these two states for structural mitigation activities. We have no directional expectations but hypothesize that there are indeed differences across states that are otherwise not captured by other variables, and so we include state indicator variables to capture any such effects.

IV. SURVEY INSTRUMENT AND DATA

An online survey was administered in August and September 2010 by Knowledge Networks to their Knowledge Panel®, to obtain household-level information regarding risk preferences, risk perceptions, and risk management decisions. The target population was property owners aged 18 or over within 96 coastal counties in Alabama, Mississippi, Texas, Louisiana, and Florida. Out of 1,536 sampled, 1,070 (69.6%) responded, and 859 consented access to their street address. Table 1 reports a demographic comparison of selected population and our sample. It shows that our sample reasonably represents the population within the 96 coastal counties.

TABLE 1

Comparison of Population and Sample Demographics, All Shown as Percentage of Total (N = 790)

Figure 1 presents a map of the targeted sample. Four hundred eighty respondents (61%) were from Florida, 182 (23%) were from Texas, 97 (12%) were from Louisiana, and 31 (4%) were from Alabama and Mississippi. The survey contained 41 questions and took 20 minutes on average to complete. Based on street address provided by respondents, we identified distance of the property from the nearest shoreline using geographic information system techniques, and age of the home from publically available county property records.5

FIGURE 1

Map of Sample Respondents

Source: Courtesy of John Cartwright, Geosystems Research Institute, Mississippi State University

Figure 2 contains the survey questions used to measure wind insurance and mitigation decisions. The wind insurance question asked respondents to indicate whether they have wind coverage included on their regular property insurance, have separate wind insurance, or do not have any wind insurance. The mitigation question asked respondents to indicate all the mitigation features they installed on their properties, and based on their responses, we aggregated the number of mitigation activities.6

FIGURE 2

Survey Questions Used to Measure Wind Insurance and Mitigation

Tables 2 and 3 report the frequencies of responses to the wind insurance and wind mitigation questions by state. The lion’s share of respondents reported having wind coverage included on their regular policy. Of those for which it is excluded, however, Alabama/Mississippi has the highest proportion of those buying separate wind coverage (71%, although this proportion is based on very few observations), followed by Texas (48%), Florida (28%), then Louisiana (18%). Regarding mitigation activity, Florida has the highest proportion of mitigators (i.e., those reporting at least one mitigation activity) at 79%, followed by Alabama/Mississippi (71%), Louisiana (53%), and Texas (50%). Overall, the results imply a mean of just over two activities per household for Florida, just under two for Alabama/Mississippi, and just over 1 for Louisiana and Texas.

TABLE 2

Frequency of Respondents by Wind Insurance Type and State

TABLE 3

Frequency of Respondents by Number of Wind Mitigation Activities and State

Table 4 presents the data in a different way, directly comparing insurance coverage and mitigation activity. Reported are the frequencies of respondents by number of wind mitigation activities undertaken by wind insurance type. Over 40% of those with no wind coverage whatsoever, this is, those whose regular homeowner’s policy excludes wind and who chose not to buy a separate wind policy, also have undertaken no mitigation activities, whereas only 26% of those with a separate wind-only policy (again, because their regular homeowner’s policy excludes wind peril) have undertaken no mitigation activities. As Table 4 makes clear, these latter respondents are also more likely to undertake multiple mitigation activities. For example, half of these respondents have undertaken three or more mitigation activities. By contrast, over 70% of those who hold no wind coverage whatsoever have undertaken at most one mitigation activity, and almost 90% have undertaken two or fewer. Finally, as Table 4 illustrates, those who have wind coverage included in their regular policy fall in between these two categories. These preliminary results indicate that the decision to mitigate and the decision to insure may be positively correlated. The survey also asked respondents to indicate the main reason for not installing additional (or any) mitigation features (see Table 5). The two leading reasons for not mitigating were high up-front installation costs and the perception that further mitigation was not necessary.

TABLE 4

Frequency of Respondents by Number of Wind Mitigation Activities and Wind Policy Type

TABLE 5

Survey Question: For Storm-Resistant Features That Your Home Does NOT Have, What Is the Main Reason That You Have Not Installed Them?

Figure 3 contains the survey questions used to measure risk preferences and risk perceptions. Risk preferences were measured using a real-money Holt and Laury (2002) experiment included at the end of the survey. Respondents were asked to make five choices each between low-variance and high-variance risks of loss and gain. We included measures over both the loss and gain domain because, consistent with prospect theory (Kahneman and Tversky 1979), and as a departure from traditional expected utility theory, it is possible that individuals make decisions under risk of loss differently than under risk of gain. It is also important to recognize the potential limitations of such a measure, using relatively low stakes and relatively high probabilities as we do here, to capture the preferences associated with high-stakes low-probability risks, such as those inherent in insuring and mitigating against hurricane risk (Kachelmeier and Shehata 1992). The following is an example question over the loss domain: Throughout the choice sets, the probabilities were set at 0.1/0.9, 0.3/0.7, 0.5/0.5, 0.7/0.3, and 0.9/0.1, respectively, and dollar values were fixed. Respondents were told that one of their choices from each of gain and loss sets would be selected randomly, and it would affect their actual payoffs so that incentive compatibility was ensured. Respondents received $10 before making the choices so that the outcome of the loss gambles would not affect money that was already earned for taking the survey. We use the total number of low-variance choices over the loss and gain domains, respectively, as measures of increasing risk aversion.

FIGURE 3

Example of Experimental Question Used to Elicit Risk Preferences and Survey Questions to Measure Risk Perceptions

Recognizing that there are a variety of ways to measure perceptions of risk associated with hurricanes (Baker et al. 2009; Botzen, Aerts, and van den Bergh 2009; Kellens et al. 2011; Meyer et al. 2014; Peacock 2003; Sims and Bauman 1987), we measured perceived conditional expected damage and risk of hurricane landfall. We also measured perceptions about credibility of insurer and the likelihood of being eligible for governmental postdisaster assistance (survey questions presented in Figure 4).

FIGURE 4

Survey Questions Used to Measure Insurer Credibility and Disaster Assistance Perceptions

We also include data on county-level federal mitigation grants obtained from the Federal Emergency Management Agency, including the Flood Mitigation Assistance program since 1996, and the Repetitive Flood Claims and Severe Repetitive Loss programs since 2008. Because our survey was conducted in 2010, grants awarded through 2010 were aggregated into a single measure and then assigned to sample respondents based on county of residence.

Other relevant data collected include housing type (house, condominium/apartment, or mobile home), mortgage contract, past wind damage experience, and other demographic factors including income, presence of children in the household, race, marital status, and gender. These data are detailed further in the model section.

V. ECONOMETRIC MODEL

We model the difference between the utility from holding and not holding a wind insurance policy as an unobserved variable y*C such that Embedded Image [1] where xC is a vector of explanatory factors and βc are parameters to be estimated. We assume that εC has mean zero and a standard normal distribution with variance normalized to one (because the sample data contain no information about the scale). We do not observe the net benefit of the purchase, only whether it is made or not.7 Therefore, our observation is Embedded Image [2]

We model the unobserved utility-maximizing number of mitigation activities as Embedded Image [3] where xM is a vector of explanatory factors and βM are parameters to be estimated. It is possible that in some cases, mitigation activities were undertaken on the home prior to purchase by the current resident, and thus, these were not explicitly chosen. We acknowledge this potential shortcoming of our underlying choice model. One could argue that although not chosen explicitly, these were chosen implicitly as desirable attributes of the house when purchased. This may not necessarily be the case, however. For this reason, although we model this as an explicit decision, the more conservative interpretation of this equation is where the unit of analysis reflects merely the characteristics of the structure rather than the respondent’s actual choice.

The survey instrument asked respondents only about particular mitigation activities, whereas in reality, some respondents may have undertaken additional mitigation activities that we do not observe. Thus, the data are censored from below by zero and from above by the maximum number of mitigation activities queried.8 We assume that εM has mean zero and a normal distribution with variance σ2. However, we do not observe Embedded Image directly, but rather we observe yM, where Embedded Image [4] where Embedded Image is the maximum number of mitigation activities queried.

The above models can be estimated independently. However, as discussed in the preceding conceptual model section, because these choices are so closely related (both responses to risk of wind damage, potentially determined simultaneously, and influenced by some of the same observed and unobserved factors), a joint approach is more appropriate. As noted earlier, however, there is no way to know from our data the particular decision process used by each homeowner. Additionally, given the close relationship, there is a high potential for endogeneity. Normally, endogeneity is remedied using an instrumental variable approach, which requires the identification of one or more instrumental variables that explain one decision but not the other. In our case, however, it is not clear that any such factors exist, that is, any variable that can be argued to explain one can just as easily be argued to explain the other. To circumvent this modeling challenge, we avoid the potential for endogeneity altogether by excluding each from the other’s equation, and instead, treating each decision as “seemingly unrelated” to the other, but allowing for estimation of a correlation coefficient that reflects the unobserved commonalities between the two decisions. In other words, this approach assumes that neither decision affects the other directly, but rather that both decisions are influenced by the same observed and unobserved factors. Given the mixed directional effects noted in the conceptual model section, we have no clear expectation on sign for the correlation coefficient between insurance and mitigation.

Note well that we are not arguing that this is how such decisions are actually made by homeowners, but rather that this is the best way to model these decisions in a way that captures the relevant commonalities but avoids the endogeneity problem.9

Consequently, we assume correlated disturbances, such that Embedded Image [5]

where ρ is the covariance between εC and εM. Under the null hypothesis that ρ equals zero, it reduces to a set of independent models. Under the alternative hypothesis that ρ is nonzero, however, full-information maximum likelihood, in other words, joint estimation, is, in general, more efficient. A generalized linear approach that allows for mixed processes for two seemingly unrelated regressions, in this case, a probit equation (for binary wind coverage) and a tobit equation (for censored wind mitigation count) can be utilized.10 We rely on the user-written command “CMP” (conditional mixed process) in Stata (see Roodman 2011). The CMP routine has been used elsewhere in the literature for similar cases of mixed processes, for example, by Hottenrott and Peters (2012), Dias (2010), and Teisl and Roe (2010). CMP is also advantageous in that models can vary by observation. In the present case, the decision to purchase wind-only coverage is limited to a subset of households in the sample for whom wind is excluded from their standard homeowner’s policy, whereas the decision on mitigation extent is relevant to all households in the sample.

We start with the likelihood for the independent probit model, which takes the form Embedded Image [6] where Φ(·) is the cumulative normal distribution and h −1(0) = (−∞, − xCiβCi], h−1 (1)( − xCiβCi ,∞), and the likelihood for the independent tobit model, which takes the form Embedded Image [7] where φ(·) is the normal probability density and Embedded Image [8]

Supposing that we observe some yi = (yCi,yMi)’ = (0,yMi)’, where 0 < yMi < ȳM, then, with the error structure Σ defined above, the joint likelihood function takes the form Embedded Image [9] (see Roodman 2011 for more details).

Table 6 reports variable descriptions, and Table 7 reports summary statistics and expected coefficient signs for the explanatory variables included in vectors xC and xM. Both models contain measures of risk aversion over loss and gain and other information, including perceived insurer credibility, perceived likelihood of eligibility for postdisaster aid, and perceived storm frequency. The insurance equation also includes a measure of perceived conditional expected damage, which is omitted from the mitigation equation because respondents answered this question after they had already undertaken their respective wind mitigation activities, meaning that expected damage is likely a function of mitigation.

TABLE 6

Description of Variables

TABLE 7

Summary Statistics

Ideally, any model of choice for a market good should contain variables for its own price and those of close substitutes and complements. However, we do not observe what would have been the cost of insurance and mitigation for those respondents who chose not to purchase it, and mitigation costs can vary widely. Furthermore, prices for wind policies are generally set at the state level, with any price differences usually a function of mitigation activities undertaken (for discounts). Thus, insurance price differences generally reflect only differences in risk exposure. This is similar to the National Flood Insurance Program, whose rates vary according to flood zone, whether the structure was built before or after the publication of the local flood map, number of stories, structure type, and elevation above the base flood. Mitigation costs are not directly observed and can be influenced by many idiosyncratic factors. This being the case, rather than including a proximate insurance price variable that may be construed as a single measure of overall risk exposure or a cost estimate for mitigation measures, our empirical models instead contain variables that should reflect objective risk exposure (as opposed to subjective perceptions of risk exposure, which we also include), including measures of proximity to the coastline, housing type, presence of a mortgage, and state of residence (to reflect any state-specific price and/or any other political or administrative differences). This is similar to the arguments used by Petrolia, Landry, and Coble (2013) in their study of federal flood insurance.

Finally, the models contain measures of other potentially relevant factors, including county-level federal mitigation grant dollars. The only available data, however, relate to flood mitigation expenditures (since the National Flood Insurance Program is a federal program). We use the sum of Flood Mitigation Assistance, Repetitive Flood Claims, and Severe Repetitive Loss Program funds from 1996 to 2011 to account for available mitigation dollars. The influence of primarily flood mitigation grants on individual wind insurance and mitigation is unclear; if households see flood and wind mitigation as complementary, flood mitigation may crowd out wind-risk management. The models also include age of property, personal experience with storm damage, income, presence of children, race, gender, and marital status.

VI. RESULTS

Table 8 reports the results of the simultaneous wind coverage and wind mitigation regressions, including marginal effects for the probit regression. Note that the wind coverage model applies only to those households for whom wind coverage is excluded from the standard homeowner’s policy; these respondents face the explicit choice of purchasing a separate wind policy or living with no wind coverage. The number of observations for this model is 225. The mitigation model applies to the entire sample of 790. The correlation coefficient for the error distribution is statistically significant and positive, indicating that the two decisions—to purchase wind insurance and to mitigate wind risk—are related via their respective disturbance terms, confirming that the simultaneous approach is superior to independent regressions. Our results indicate that wind insurance and wind mitigation measures are positively correlated, after conditioning on the covariates that have been included in our model, for those households for whom wind coverage is not included in the standard homeowners’ policy. This finding is consistent with the results of Talberth et al. (2006) who found that in the context of wildfire risk, insurance purchase and mitigation behavior are complementary.

TABLE 8

Results of Simultaneous Probit (Insurance) and Tobit (Mitigation) Regressions

Wind Insurance

Consistent with the findings of Holt and Laury (2002); Kachelmeier and Shehata (1992); Lusk and Coble (2005); and Petrolia, Landry, and Coble (2013), risk aversion over the loss domain has a positive and significant effect on the insurance decision: a one-unit increase in the number of low-variance gambles chosen in the Holt and Laury exercise increases the probability of holding a wind insurance policy by 3.4%. Also consistent with the findings of Petrolia, Landry, and Coble (2013), risk aversion over the gain domain has no significant effect, indicating that a measure of risk preference measured over the gain domain has no statistical significance explaining decisions of insuring the risk of loss. Although we find that perceived insurer credibility and perceived eligibility for postdisaster assistance have the hypothesized signs, they are not statistically significant. In addition, we find no significant effect of perceptions of expected storm frequency, conditional expected damage, or past wind damage experience on the probability of purchasing separate wind coverage, although the sign is as hypothesized on two of the three. We find a significant and negative effect of county-level federal mitigation grant dollars, indicating that perhaps flood mitigation undertaken at the community or individual level may be seen as complementary to wind-risk management, thus acting as a substitute for private wind coverage.

Consistent with our hypothesis, we find that respondents who live in the coastal zone (here defined as within three kilometers of the shoreline) are significantly more likely to hold a wind policy (specifically, 32% more likely). This result is consistent with that of Kriesel and Landry (2004), Landry and Jahan-Parvar (2011), and Petrolia, Landry, and Coble (2013), who find a negative correlation between the likelihood of holding a flood insurance policy and the distance from the shoreline.

Consistent with the findings of Petrolia, Landry, and Coble (2013) and Grace, Klein, and Kleindorfer (2004), the coefficient on mortgage is significant and positive, indicating that those with a mortgage are more likely to carry wind coverage. Although we are unaware of any explicit state-level requirements for holding wind coverage as part of a mortgage, as mentioned earlier, this may be evidence of lenders encouraging or insisting upon coverage as a prerequisite for obtaining a loan. Given that many of these respondents likely live in areas also subject to increased flood risk, it may also be a side-effect of the flood insurance program’s requirement for flood coverage on mortgaged properties.

We find that individuals who live in a condominium are less likely to purchase a wind policy relative to individuals who live in a single-family home, but find no significant differences for mobile-home respondents. Contrary to the findings of Grace, Klein, and Kleindorfer (2004), we find no significant effect of age of the home on the insurance decision.

Based on both the reported coefficients and postregression Wald tests of parameter equivalence on individual state indicators, we find that respondents living in Alabama/Mississippi and Texas are significantly more likely to purchase wind coverage relative to those in both Florida (the omitted base) and Louisiana. Results also indicate that respondents in Louisiana are not significantly different from those in Florida.

Consistent with the findings from other studies (Baumann and Sims 1978; Browne and Hoyt 2000; Kunreuther 2006; Landry and Jahan-Parvar 2011; Petrolia, Landry, and Coble 2013), we confirm a significant positive impact of household income on the probability of purchasing wind insurance. Our results indicate that 7% increase in income results in 28.3% increase in the probability of purchasing a wind policy. We find that the presence of children in the home significantly reduces the likelihood of holding a wind policy. We also find that single females are less likely to purchase a wind policy relative to married households.

Wind Mitigation

The wind mitigation model includes the same explanatory variables as the insurance model with exceptions of the expected damage variable being omitted and a variable indicating whether wind peril is excluded from the homeowner’s regular policy being included. We wish to reiterate the caveat that it is likely that in some cases mitigation activities were undertaken on the home prior to purchase by the current resident, and thus, these were not explicitly chosen. So although we model this as an explicit decision and interpret it as such, the more conservative interpretation of these results is that they reflect the characteristics of the structure rather than the respondent’s actual behavior.

The variable indicating if wind peril is excluded from the homeowner’s regular policy is significant in the mitigation equation but, contrary to our original hypothesis, is negative. This indicates that after controlling for other effects, those for whom wind peril is excluded from the regular homeowners’ policy (implying areas of high wind risk) actually undertake 0.18 fewer mitigation activities than those in lower risk areas. This result runs somewhat counter to the findings of Carson, McCullough, and Pooser (2013) and Lohse, Robledo, and Schmidt (2012) that higher levels of mitigation activity are optimal when insurance is unavailable.

Contrary to the findings of Briys and Schlesinger (1990) and Dionne and Eeckhoudt (1985) our results show that neither risk aversion over loss nor gain domains has a significant effect on the decision to mitigate. We also do not find a significant effect of insurer credibility on mitigation decision. Consistent with the finding from the insurance model, perceived eligibility of postdisaster assistance is not significant, but county-level federal mitigation grant dollars has a negative effect on the number of mitigation activities undertaken. Again, this could imply that flood mitigation activities are seen as complementary to wind-risk management.

Contrary to previous literature and our own hypotheses, we find no significant effect of risk perceptions on the decision to mitigate. A possible explanation for this result may be found in the work of Bubeck, Botzen, and Aerts (2012), who argue that such relationships are often insignificant or even negative because individual risk perceptions may decrease after mitigation measures have been installed. Consistent with the findings of Botzen, Aerts, and van den Bergh (2009), Naoi, Seko, and Ishino (2012), and Brody et al. (2009), we find a positive correlation between past damage experience and mitigation activities, indicating that respondents who experienced wind damage in the past undertake 0.29 additional mitigation activities.

As hypothesized, and also consistent with the findings of Botzen, Aerts, and van den Bergh (2009) and Peacock (2003), we find that individuals who live in the coastal zone undertake 0.38 additional mitigation activities. Consistent with the findings of Carson, McCullough, and Pooser (2013) and Dumm, Sirmans, and Smersh (2011), age of the home is found to have a significant and negative effect on mitigation activities. This could be because homes that were built under newer building codes were required to install more wind mitigation features. Another possible explanation is that home value is generally negatively correlated with its age, so home owners may be less willing to spend money on mitigation on older homes.

Consistent with the finding from the wind insurance model, we find that individuals who live in a condominium are less likely to mitigate relative to individuals who live in a single-family home. We find no significant effect of mobile homes, however, indicating that residents of mobile homes undertake neither more nor less mitigation to single-family homes. Although we find that a mortgage contract is positively correlated with purchasing a wind policy, we find that it is negatively correlated with mitigation, indicating that those who have a mortgage contract tend to undertake 0.19 fewer mitigation activities. Although the reason for this effect is not clear, it is possible that cash flow could be an issue: residents who are making payments on their home have less disposable income available to spend on mitigation relative to residents whose homes are paid off.

Also, we find that Louisiana residents undertake 0.93 fewer mitigation activities, and Texas residents undertake 0.83 fewer mitigation activities relative to Florida residents. The Florida-Texas relationship is consistent with the findings of Brody, Kang, and Bernhardt (2010), although their analysis is at the community level. Results indicate that respondents in Alabama/Mississippi are not significantly different from those in Florida.

Although the sign on income is positive in the mitigation equation, and thus consistent with the findings of Carson, McCullough, and Pooser (2013) and Peacock (2003), it is not statistically significant. Consistent with the finding from the insurance model, we find a significant and negative effect of the presence of children in the home on mitigation. This finding contradicts the findings of Carson, McCullough, and Pooser (2013), who find a greater probability of the decision to mitigate in the presence of children but find no significant effect on extent of mitigation. The latter measure is more akin to the present one. However, their measure was at the community level: the proportion of the population in the home’s zip code under the age of 18. Given that ours is a direct measure of the presence of children in each home, ours may represent a more accurate account of this effect. Consistent with the finding from the insurance model, we find that single female respondents tend to mitigate less than married households. Finally, we find that white respondents tend to mitigate less.

VII. SUMMARY AND CONCLUSIONS

We present the results of what we believe to be the only household-level analysis of the decision to purchase a wind-only insurance policy when wind peril is excluded from the homeowner’s regular policy, and one of very few analyses of the extent of mitigation activities specific to wind risk. We believe this to be the only study that models these two closely related behaviors jointly, so as to explore the conditional correlation of these activities and achieve more efficient estimation. One of the major findings is that the decision to buy a wind-only policy and the decision to mitigate are positively correlated; that is, those who choose to insure against wind damage are also more likely to mitigate wind damage (after conditioning on covariates). We also find that among those respondents who live in areas where wind peril is excluded from the regular policy, which is generally an indication of very high wind risk, there is actually less mitigation activity, not more. Within this group, however, we find that those who choose to buy a wind-only policy also undertake a larger number of mitigation activities, whereas those who carry no wind coverage whatsoever undertake the smallest number of mitigation activities, even fewer than those who live in lower-risk areas where wind peril is included in their regular homeowner’s policy.

These findings may be explained by premium discounts offered to those who have undertaken wind mitigation activities. These discounts offer an incentive to those who already have wind coverage to mitigate and for those who do not already have wind coverage but have undertaken mitigation to purchase coverage. However, it needs to be noted that premium discounts vary by state and locality, and that there is not necessarily a direct mapping of mitigation activity to discount. In some instances, a home must adopt a particular building code to qualify for discounts, rather than undertake individual mitigation measures. We find that the leading reasons for not undertaking further (or any) mitigation are affordability and the belief that the homeowner’s property does not need any additional mitigation. These results lend credence to the recommendation of Kunreuther, Pauly, and McMorrow (2013) that mitigation can be encouraged by addressing the affordability issue; specifically, by tying mitigation and insurance purchase to the home’s mortgage and allowing for payment over time to avoid the cash-flow problem associated with high up-front costs for mitigation. As for the perception of benefits, states may wish to better inform homeowners of the mitigation options available and how these activities can reduce the probability and extent of damage in the event of a storm. Additional research on cost-effective mitigation measures that can be easily related to homeowners appears warranted.11

Neither perceived insurer credibility nor disaster assistance appears to significantly affect the likelihood of holding a separate wind policy or the extent of mitigation. On the other hand, we find that county-level federal flood mitigation grant dollars have a negative effect on both purchasing wind coverage and undertaking mitigation activities, indicating that there may be some degree of perceived substitutability between flood mitigation and wind-risk management.

We find that wind-only policy uptake rates are substantially higher in Texas and Alabama/Mississippi, compared to Florida and Louisiana. On the other hand, we find that extent of mitigation activities is higher in Florida relative to Alabama/Mississippi, Louisiana, and Texas. Thus, our results suggest that Alabamans, Mississippians, and Texans tend to favor insurance over mitigation as a means to protect against wind, Floridians tend to favor mitigation over insurance, and Louisianans tend to do less of both, relative to their neighbors. That Floridians tend to purchase less insurance relative to their neighbors is somewhat surprising, given that Florida’s wind insurance program is the only one of the five Gulf states whose state program directly competes in the private market, implying that it should have relatively lower premiums relative to the other states whose programs are explicitly markets of “last resort” (Kousky 2011). In fact, a recent survey found that the average premium per $1,000 of coverage was less than $4 in Florida, over $5 in Louisiana and Texas, and over $10 in Alabama and Mississippi (Texas Department of Insurance 2012). Given that we find the relative take-up rate in Florida to be the lowest, there must be other factors besides price that explain differences across states and that our model does not capture.

Consistent with the findings of Petrolia, Landry, and Coble (2013), we find that respondents who are risk-averse with respect to monetary loss are significantly more likely to buy wind insurance; however, risk aversion has no effect in the mitigation model, suggesting, perhaps, that risk preferences could be more relevant for insurance decisions, but that mitigation decisions are dominated by other factors. Also we do not find a significant impact of conditional expected damage or expected storm frequency on wind coverage, nor do we find any significant effect of expected storm frequency on mitigation. Also consistent with previous literature (Botzen, Aerts, and van den Bergh 2009; Brody et al. 2009), we find that actual past experience with wind damage is significant in explaining the extent of mitigation. Regarding the insurance decision, however, our results run contrary to the previous literature (Baumann and Sims 1978; Carbone, Hallstrom, and Smith 2006; Kunreuther 1978; Kriesel and Landry 2004; Petrolia, Landry, and Coble 2013), showing that there is no significant effect of past damage experience on the decision to purchase a wind policy. We also find, consistent with the literature, a positive effect of household income on insurance (Baumann and Sims 1978; Browne and Hoyt 2000; Kunreuther 2006; Landry and Jahan-Parvar 2011; Petrolia, Landry, and Coble 2013) but contrary to previous findings, find no such effect on mitigation (Carson, McCullough, and Pooser 2013; Peacock 2003).

Finally, we wish to make a few comments regarding the way forward. Given the divergence in types of coverage, types and levels of government efforts to encourage mitigation and coverage, and the types of incentives to undertake them, future research should focus on better understanding how individuals receive and process the myriad of information and options available to address coastal wind as well as flood risk, and how to reduce the transaction costs associated with obtaining coverage and undertaking mitigation for both. There have been several proposals made in favor of merging flood and wind peril into a single (or all-perils) policy (see Brown 2010; Derrig et al. 2008; Dixon, Macdonald, and Zissimopoulos 2007; Pidot 2007; Pompe and Rinehart 2008; and USGAO 2008), as well as proposals to couple insurance contracts with multiyear home-improvement loans to encourage mitigation (Michel-Kerjan and Kunreuther 2011). Future research should focus on understanding how and the extent to which consumers would respond to such alternatives, and whether such approaches would simplify the process for consumers and encourage mitigation as well as participation in insurance programs. Future research should also explore alternative ways to measure insurance uptake and mitigation efforts, given differences across states in operation of wind pools and idiosyncratic differences in possible mitigation activities.

Acknowledgments

Special thanks to John Cartwright, Geosystems Research Institute, Mississippi State University, for providing maps and much of the GIS-based data. This research was conducted under award NA06OAR432026406111039 to the Northern Gulf Institute by the NOAA Office of Ocean and Atmospheric Research, U.S. Department of Commerce, and supported by the National Institute of Food and Agriculture, under project #MIS-033140 “Benefits and Costs of Natural Resources Policies Affecting Ecosystem Services on Public and Private Lands.”

Footnotes

  • The authors are, respectively, associate professor, Department of Agricultural Economics, Mississippi State University, Mississippi State; chief economist, Florida Fish and Wildlife Commission, Tallahassee; associate professor, Department of Agricultural and Applied Economics, University of Georgia, Athens; and Giles Distinguished Professor, Department of Agricultural Economics, Mississippi State University, Mississippi State.

  • 1 Simmons, Kruse, and Smith (2002) also focus on wind damage mitigation, but they focus on the impact of mitigation on the resale value of the property rather than the decision to mitigate.

  • 2 Fair Access to Insurance Requirements (FAIR) plans were established under the Housing and Urban Development Act of 1968, with the intent to provide insurance to individuals who cannot obtain it in the voluntary market. Currently, 32 states and Washington, D.C., have FAIR plans (Kousky 2011).

  • 3 Insurance Code, Title 10, Subtitle G, Chapter 2210, Subchapter A, Section 2210.355, Paragraph C, State of Texas 2009 (available at www.statutes.legis.state.tx.us/Docs/IN/htm/IN.2210.htm; accessed December 1, 2014).

  • 4 Hatori, Matsushima, and Kobayashi (2004) claim that households may partly base their mitigation decisions on social norms, in particular, observing the actions of their neighbors. Carson, McCullough, and Pooser (2013) test this empirically using participation in the My Safe Florida Home program. This program, however, encourages mitigation at the household level, not the community level, which is what we are particularly interested in testing. Additionally, our dataset contains no measure of social norms, that is, of neighbor’s decisions, thus we are not able to control for such an effect.

  • 5 Unlike the other states, Louisiana parish (county) tax assessors do not make year built publically available, nor do any of the major property data firms own such data. Thus, for Louisiana respondents, we use year of purchase as a proxy for age of the home. To test for sensitivity to this proxy, we also estimated the effect of this variable excluding Louisiana properties but found no significant differences.

  • 6 We treated “don’t know” responses as “no” responses. We excluded elevation of the property from the count of mitigation activities because this applies to flood risk rather than wind. We also reviewed other mitigation activities written in by respondents and, when legitimate, included these in the total count. The additional wind mitigation activities added to the total count include metal roof, hip roof, polypropylene screening, reinforced garage door, concrete blocks, and boarded windows.

  • 7 Unlike the mitigation question, the wind insurance question did not include a “don’t know” option. It is possible that some respondents did not actually know whether or how they were covered for wind damages.

  • 8 The survey allowed for respondents to write in additional mitigation features not listed. Nevertheless, this format is still susceptible to limiting respondents’ responses to those features explicitly listed.

  • 9 One may be inclined to think a selection model would be appropriate in this case, given the decision to buy a wind policy is observed only if wind peril is excluded. But wind peril is excluded based on where the resident lives, and so the selection stage would be one of where a respondent chooses to live. But the decision of where to live is independent of the choice of whether to buy a wind-only policy. Following the logic of Agresti (2002), in this case the probability that an observation is missing, that is, the probability that a respondent lives in a location where wind peril is excluded from the regular policy, is independent of that observation’s value, in other words, that it is independent of whether the respondent chooses to buy a separate policy or not. Agresti concludes that under such circumstances, with a likelihood-based analysis, it is not necessary to model the missingness mechanism.

  • 10 All else equal, it would be better to use a count model for the mitigation data, but our circumstances demanded the consideration of a simultaneous estimation approach, which required that we explore the possibility of using a tobit instead (because we are aware of no means of estimating a binary equation and a count equation simultaneously). The use of a tobit for count data is not unheard of (see Campa 1993; Fair 1978; Romer and Snyder 1994), and Greene (2000) discusses the use of the tobit in this kind of situation as a logical choice (though perhaps not ideal in all cases). To test the sensitivity of adopting the tobit instead of a count model, we compared the mitigation model results from the simultaneously estimated tobit to those of a Poisson and negative binomial. We find only small differences, in terms of relative magnitude, in the predicted marginal effects. We find no reversals of sign or significance. These results are available from the authors upon request.

  • 11 Bubeck, Botzen, and Aerts (2012) argue that mitigation behavior is only weakly related to risk perceptions and, instead, advocate for a greater focus on “threat and coping appraisals.” “Threat appraisal” describes how an individual evaluates how threatened he or she feels by a certain risk. “Coping appraisal” refers to how an individual thinks about the benefits of possible actions and begins to evaluate his or her own competence to carry them out.

References