COASTAL BUILDING CODES AND HURRICANE DAMAGE

Carolyn A. Dehring* University of Georgia

and

Martin Halek University of Wisconsin - Madison

May 11, 2012

* Terry College of Business, Department of Insurance, Legal Studies, and Real Estate, University of Georgia, 206 Brooks Hall, Athens, GA 30602, phone (706) 542-3809, FAX 706- 542-4295, [email protected]

The authors are grateful to the Honorable Kenneth M. Wilkinson, CFA, Lee County Property Appraiser, Lee County, , for data, and especially to Jim Sherron. We thank Chris Jones for his coastal engineering insights. We also thank Peter Colwell, Ed Coulson, Chris Dehring, Jim Kau, Olivia Mitchell, Harold Mulherin, Henry Munneke, and Tony Yezer for helpful comments. We acknowledge financial support provided by the Terry-Sanford Research Award and the National Science Foundation DEB-0823293 and DEB-0218001 as part of the ongoing research at Coweeta LTER.

JEL Classification: H59, H76, Q54

Abstract

We explore whether federal and state level changes to coastal building standards are effective in mitigating losses to coastal property following . We find properties built seaward of and after the reestablishment of the Coastal Construction Control Line, and those built under the National Flood Insurance Program and located in an A-Zone, had more damage relative to similarly located structures built before these regulatory changes. We show the NFIP regulations allowed for weaker foundation requirements and lower elevations relative to the earlier county code. This likely led to greater flood damage, as supported by analysis of individual structural components.

2 1. INTRODICTION

In the United States, government limits the bundle of rights associated with real property

ownership through building codes, also referred to as construction codes. These codes set

standards for various aspects of construction, including fire prevention, structural integrity, and

general health and safety. Unlike in Canada, where a national building code is used, local

building codes in the U.S. may be based on a model code or a state code. Building codes are

enforced by local governments as an exercise of police power.

Building construction in areas prone to flood hazard is often regulated at the federal or

state level. In the State of Florida, coastal setback lines, known as coastal construction control

lines (CCCLs), designate areas having the potential for extreme fluctuation in the event of one-

hundred year storm. Special siting and design criteria apply to construction seaward of this line.

In addition, most Florida communities participate in the National Flood Insurance Program

(NFIP), where residents are eligible for subsidized federal flood insurance.1 As a condition of

participation in this program, building codes must incorporate minimum building standards set

forth by the Federal Emergency Management Agency (FEMA).2 In Special Flood Hazard Areas

(SFHAs), these building standards involve minimum building elevations and foundation

requirements.

There are two competing hypotheses regarding the effects of building codes on safety and

property risk. Building code proponents argue that home buyers are unlikely to possess the

technical expertise needed to accurately assess the structural integrity of a housing unit, and that

such codes are necessary to prevent market participants from developing land in a manner that

1 NFIP insurance on over 20% of the policies presently in force is priced at 35 to 40% of the actuarially fair rate. A recent study by the Property Casualty Insurers Association of America estimates that on average full-risk premium properties would be charged rates 23.3% higher if coverage were provided through the private market. 2 Local building ordinances may be stricter than FEMA standards.

3 endangers adjacent property (see Oster and Quigley 1977, Seidel 1980, Chivers and Flores

2002). Kunreuther (1974) suggests land use restrictions and building codes as efficient ex-ante disaster loss reduction mechanisms. In support of this view, Fronstin and Holtmann (1994) attribute their finding that newer properties in Florida’s Miami-Dade county sustained greater damage than older properties in Hurricane Andrew to a general erosion of the South Florida

Building Code, although they do not explicitly control for building code changes in their

empirical analysis.

The alternative hypothesis suggests an adverse effect of building codes on safety.

Lindsey (1976) assumes a less omniscient government agent where government regulation focuses purely on visible factors, possibly at the sacrifice of overall quality. For example, a focus on elevation with regard to flood peril may compromise the effectiveness of other building technologies designed to minimize the effects of concomitant perils. Further, building codes associated with the provision of subsidized insurance may create moral hazard by inducing risk taking. Both Shavell (1979) and Stiglitz (1983) have rationalized that the acquisition of insurance against some contingency is associated with a decreased incentive to avoid or prevent the insured loss, because the insured does not bear the full consequences of their actions.3 Finally,

independent of any insurance provision, moral hazard can also result from a false perception of

safety.4

In this paper, we investigate the effectiveness of coastal construction code changes associated with Florida’s Coastal Construction Control Line and the National Flood Insurance

Program in reducing property risk from hurricane exposure. Using a sample of residential

3 For example, compulsory insurance regulations coupled with no-fault laws similarly create moral hazard costs which result in increased traffic fatalities (see Cohen and Dehejia 2004). 4 Peltzman’s (1975) results suggest that drivers exhibit greater risk taking in their driving in response to increased auto safety regulation.

4 properties from the Florida county in which Hurricane Charley made landfall in 2004, we investigate if the likelihood of property damage is related to the construction code regime in place at the time of construction, while controlling for other physical and locational attributes thought to influence hurricane risk exposure. Then, conditional on some damage being sustained, we determine whether the extent of damage is related to the codes in place at the time of construction. We find post-NFIP construction located in certain Special Flood Hazard Areas had a higher incidence of damage and, conditional on damage, sustained greater damage relative to similarly located property from Hurricane Charley. We also find properties built both seaward of and after the reestablishment of the 1991 CCCL sustained greater damage relative to similarly located property. In the case of the CCCL, a “permit run” prior to the reestablishment of the line suggests an adverse market perception associated with these regulations. We attribute our NFIP results to lower required elevations and less stringent foundation requirements relative to what was in place before the NFIP. Our findings raise questions about the optimal scale of code design.

The remainder of this paper is organized as follows. In Section 2, we provide some background into the coastal construction code changes in Lee County, Florida, that are the subject of our study. In Section 3 we present our basic econometric models. In Section 4, we discuss our data and the sampling issues associated with hurricane damage reporting. Our empirical results are reported and discussed in Section 5. In Section 6 we extend this analysis by incorporating data from individual building components and pre-NFIP building codes. The paper concludes with a discussion of the policy implications from our findings.

2. BACKGROUND

5 The state of Florida experienced significant population gains during the 20th century.

Since 1920, the state’s population growth rate per decade was consistently in the top 7 among

states. Between 1900 and 1980, average growth rates per decade exceeded 50% for some regions

in the state.5 It was important for both the state and local communities to manage the development that accompanied this population growth. This was especially true in communities having environmentally sensitive coastal areas subject to coastal flooding and beach erosion.

1970s to 1984 ’s Lee County first adopted coastal construction codes in the mid-

1970s. These coastal building codes established two or three zones on each of the county’s multiple barrier islands. Zones were delineated by distance from the shoreline. Building requirements were strictest in the most seaward zone, which included those properties that front the . In this zone, buildings had to be elevated above the National Geodetic

Vertical Datum (NGVD) of 1929, which is often referred to as the mean sea level, and anchored to pile foundations. A pile is a column, typically made of wood, steel, or concrete, that is driven deep into the ground to provide support for a structure. The size of the piles, as well as their spacing, depth of embedment, and force bearing capacity were specified in the code. The code also specified horizontal wave and uplift pressures. Finally, buildings were required to be elevated between 12 and 13.5 feet above the mean sea level, measured from the first finished floor, or from the underside of the building in the most seaward zone(s).

In the largest and most landward zone, which included properties on the eastern sides of

the islands, columns on footings (pier foundations) were permitted as an alternative to pile foundations. The footing is a widened section at the base of a column (pier) on which a column

5 These growth rates were 90% in the Southeast region, 66% in the Southwest region, and 52% in the Central region. (Smith 2005)

6 rests. In general, pile foundations are perceived to be the superior technology for resisting

hurricane induced damage. In more landward zones, enclosed rooms with non-load bearing,

breakaway walls were permitted under the first finished floor. Buildings in all zones were

required to withstand wind forces of up to 110 miles per hour. The top half of Table 1

summarizes the major components of the building code for zones 1, 2 and 3 for Captiva Island,

one of the islands in the sample. 6

1984: Initial Participation in the National Flood Insurance Program

Since its inception in 1972, the National Flood Insurance Program has made flood insurance coverage available for purchase to property owners living in qualifying communities.

To qualify, a community must submit an application to the NFIP documenting flood plain management regulations that have been adopted in the community. These regulations are designed to reduce damage from storms in flood-prone areas. Flood Insurance Rate Maps

(FIRMs) depict Base Flood Elevations (BFEs) throughout the county, above which new structures and substantial improvements to old structures must be raised.7 NFIP building

standards also require that materials and construction methods minimize flood damage. The

regulations further prohibit the location of any electrical or heating equipment below the BFE,

and more generally any human habitation below the BFE.

6 For example, on Captiva Island, Zone 1 is 50 feet wide and the most seaward zone in which construction is permitted. Zone 2 is a 90 foot wide zone immediately landward of Zone 1. Zone 3 is the remaining portion of the island (everything landward of Zone 2). Some zones were further divided into sub-zones. Codes are nearly identical on Estero and Gasparilla Islands. Pre-NFIP code is not available for Sanibel Island. Lee County Ordinances 76-7, 76-10, 76-15 and 77-1. 7 The base flood, or 100-year flood, is the flood having a 1% chance of being equaled or exceeded in any given year. The BFE is the height of the base flood, in relation to mean sea level, and is the minimum building elevation standard under the National Flood Insurance Program. Base flood elevations are derived from detailed analyses at selected spatial intervals.

7 Lee County joined the National Flood Insurance Program in 1984.8 All inhabitable

barrier island land in the county received either an “A-Zone” or “V-Zone” designation, both of

which indicate Special Flood Hazard Areas (SFHA).9 A-Zone land is subject to rising water from

coastal flooding. Elevation requirements in the A-Zone stipulate that the top of the proposed

lowest floor (including basements) must be elevated to or above the base flood elevation. V-

Zone (velocity) land is subject to storm wave action in addition to the rising water from coastal

flooding. V-Zone elevation requirements warrant the lowest supporting horizontal member to be

located at or above the base flood elevation level. The lowest supporting member is the lowest beam or joist that supports the elevated building. In essence, for the same base flood elevation,

V-Zone buildings will be higher by the difference between the underside of the building and the

top of the lowest floor of the building, or 1 – 2 feet on average. In the V-Zone, all construction

must be securely anchored on piles or columns, and designed and anchored to withstand all

anticipated weight or force to be borne by the base flood.10 There are no requirements regarding

the foundation system for A-Zone properties. Piles, columns on footings, or monolithic slab

foundations are all permissible in the A-Zone. A-Zone and V-Zone building requirements on

Captiva Island are presented in the bottom half of Table 1.11

1991: The Reestablishment of the 1978 Coastal Construction Control Line

Coastal setback lines were introduced in 1971 as part of the Beach and Shore Protection

Act. These coastal construction setback lines, now referred to as Coastal Construction Control

8 Flood plain management regulations were passed on July 11, 1984 after the June completion of the FIRM, and made effective August 31, 1984.8 9 A-Zone areas are estimated to have a 1% annual chance of flooding and a 26% chance of flooding over the life of a 30-year mortgage. V-Zone areas are estimated to have a 1% or greater annual chance of flooding, a 26% chance of flooding over the life of a 30-year mortgage, and an additional hazard associated with storm waves. 10 It is worth noting the change in foundation requirements in the V-Zone are from a pre-NFIP prescriptive code to a more performance-based code which merely states the technical objectives to be achieved. 11 Codes are identical on Estero and Gasparilla Islands.

8 Lines (CCCL), apply to beach or dune areas having the potential for extreme fluctuation in the

event of a one-hundred year storm. Construction in the area must adhere to special siting and

design criteria, as structures must be able to withstand physical forces and waves from storms,

water pressure from flooding, and the effect of soil loss from storm-induced erosion.

Improvements on structures constructed partially or totally seaward of a CCCL must have the

lowest horizontal structural beam located above the predicted breaking wave crest.12 Importantly,

the CCCL elevation is engineered to incorporate additional risks such as scour, wind, and long-

term beach erosion, whereas the base flood elevation accounts for rainstorm and coastal flood

risks.

The placement of the Coastal Construction Control Line is determined by engineers. The

Department of Environmental Protection (DEP) uses the 100-year storm surge line to indicate

the upland limits of the effect of a one hundred year coastal storm. Placement is determined through the application of engineering predictive models that factor in tidal cycles, erosion,

water depth, and topography.13

In 1991, Lee County underwent a reestablishment of its existing 1978 Coastal

Construction Control Line.14 The 1991 CCCL was generally landward of the original 1978

CCCL. To develop in this area requires the builder to obtain a CCCL elevation certificate, prepared by or under the direct supervision of a registered land surveyor, professional engineer,

12 The design elevations for the lowest horizontal structural member of a structure are based on estimates of a wave height superimposed on a storm tide consistent with a 100-year storm, which has a 1% chance of occurrence. 13 In 1986, Lee County established a Coastal Building Zone (CBZ) in accordance with Florida’s Coastal Zone Protection Act of 1985. Also, in 2001, Florida modified its state building code. This event is not significant when controlled for in our later analysis. We cannot test for the effects of these changes because there is no spatial variation in our study area. 14 The effective date of the CCCL reestablishment was May 30, 1991. We note here that there are no properties in the sample that are landward of the original 1978 CCCL.

9 or architect licensed by the State of Florida. Similarly, a permit is required to modify, repair or

rebuild when the proposed changes involve changes to the structure’s foundation.15

3. EMPIRICAL MODELS We develop two models designed to test whether the likelihood of hurricane damage of

residential homes is related to the building codes in place when construction on the structure

commences. We then examine whether the magnitude of hurricane damage of residential homes

is related to the building codes in place at the time of construction, conditional on some damage

being sustained. Model specifications control for location, building age, structural attributes of

the property thought to influence storm exposure and construction technology, and

implementation of the two aforementioned coastal construction regimes: Lee County’s

participation in the National Flood Insurance Program, and the reestablishment of Lee County’s

Coastal Construction Control Line.

The first model considers the likelihood of hurricane damage, controlling for

participation in the NFIP and reestablishment of the CCCL. Equation (1) shows the binary logistic regression model:

DAMAGEi  0  1GULFDISTi  2BLDGAREAi  3LOTAREAi  4 PRICESFi 3  5 AGEi  i ISLANDi 1 AZONEi   2 AZONE 3VZONEi POSTNFIPi (1) i1

  4 5POSTCL91i CL91i   i

15 Structures damaged after a storm may be rebuilt unless the storm has altered the shoreline such that there is no longer a viable building site landward of the beach.

10 The dependent variable to be explained is the binary event of a residential property either

incurring or not incurring structural damage in a hurricane. For our determination of the

dependent variable, DAMAGE takes the value one if there is any reported damage, and zero

otherwise.16

The independent variables of particular interest relate to the various coastal building code

regimes. The indictor variable POSTNFIP takes the value 1 if the property’s construction

commenced after the August 31, 1984 effective date of the county’s flood plain management

regulations. The coefficient 1 reflects whether properties in an A-Zone location are more or less

likely to incur damage than properties in a V-Zone location before NFIP program participation.

If A-Zone land is generally less prone to hurricane induced damage than V-Zone land, then we

might expect the coefficient 1 to be negative. The coefficients  2 and 3 reveal any additional

likelihood of damage in the A-Zone and V-Zone, respectively, following county participation in

the NFIP. A negative (positive) sign on either of these coefficients would suggest a reduced

(increased) likelihood of damage in that zone from the NFIP code changes.

As discussed earlier, the effect of the various code changes on damage is a priori

ambiguous. Consider the NFIP building regulations. It is reasonable to assume that the higher

and more flood-proof the building is, the lower the risk of flood inundation. However, NFIP

regulations (a condition for subsidized flood damage insurance) may create a perception of

safety or transfer responsibility away from the homeowner, which may lead to moral hazard by

inducing risk taking behavior. For example, a household may use less expensive or inferior

material in the construction of the property, and may be less likely to protect its property in ways

not explicitly mandated by regulations (e.g. taking last minute precautions to shutter doors and

16 We discuss property damage estimation procedures in Section 4.

11 windows). Along these lines, the highly subsidized flood insurance provided by the NFIP may

encourage a property owner to simply “build to the limit” of the insurance policy.17 Also, elevation and flood-proofing are not costless, and consumers may substitute less expensive technologies or inferior workmanship where available to compensate for the increased costs of

NFIP compliance. Finally, because county coastal construction codes were already in place prior to NFIP participation, any effect of code changes on damage must be interpreted relative to the existing code. Thus, the cumulative effect of NFIP code changes on property damage is best resolved empirically.

A similar specification is employed to control for changes in the location of the CCCL.

The indicator variable CL91 indicates properties designated seaward of the 1991 reestablishment

of the CCCL. This variable is interacted with the POSTCL91 indicator variable, which denotes

whether construction on the property commenced after the CCCL reestablishment date of May

30, 1991. The coefficient  4 indicates whether properties seaward of this line are more or less

likely to incur hurricane damage in general. The coefficient 5 reveals any additional likelihood

of damage to properties seaward of this line that are built following the 1991 reestablishment,

relative to other similarly located structures built before this reestablishment. Here again the

effect of the CCCL reestablishment on damage is a priori ambiguous because of the existing

county codes are already in place.

Other variables in the logistic regression model that may explain the occurrence of

damage can be classified as locational or structural in nature. The continuous variable

GULFDIST reflects the distance in feet from the building site to the Gulf of Mexico. We expect

17 The maximum insurance available through the NFIP is $250,000 for residential properties. Although this limit is small relative to the overall value of most coastal dwellings, this insurance does provide an inexpensive first layer of flood protection.

12 its coefficient 1 to be negative as we expect greater distance from the ocean is associated with

less likelihood of damage, other things equal. The coefficients 1 ,  2 , and  3 indicate whether properties located on Gasparilla Island, Estero Island, and Sanibel Island, respectively, have a greater likelihood of damage relative to those on Captiva Island. Because Charley’s eye struck the southern end of North Captiva Island (just above Captiva Island), we expect less damage was

sustained on the other islands implying that 1 ,  2 , and  3 are all less than zero.

Structural variables in our regression model include structure size, lot size, real price per square foot, and structure age. The variable BLDGAREA is the size of the residential structure in

the year prior to Hurricane Charley. The coefficient 2 reveals if larger buildings, based on square feet, are more or less likely to incur damage relative to smaller buildings, all else equal.

The variable LOTAREA is the lot size in square feet. We use this variable as a proxy for the safety externality argument. After controlling for house size, in a larger lot structure house is generally farther away from other structures. Damage to properties on larger lots will be less likely if greater distance between houses means the structure is less exposed to damage by moving debris. On the other hand, a structure on a larger lot may receive less protection from other structures against wind damage. The variable PRICESF is the 2003 assessed value of the improvements (not the land) divided by the appropriate GDP deflator, and divided by building area. This variable is used as a proxy for construction quality. While smaller houses tend to have higher price per square foot, holding square footage constant, we might expect damage to be less likely with a higher price per square foot.18 Finally, the variable AGE denotes the age of the

18 We do not know the number of building stories at the time damage was incurred. Thus, building area cannot be used as a clean proxy for the height of the dwelling.

13 house in years as of 2003. Controlling for code changes, we expect its coefficient 5 to be

positive, as older houses are more likely to suffer physical and functional depreciation.

A second model specification investigates whether the amount of hurricane damage to

residential properties is associated with the building codes in place when construction on the

dwelling commenced, conditional on the property sustaining damage. Specifically, the binary

dependent variable DAMAGE in Equation (1) is replaced with the continuous dependent variable, DAMAGE % .

The dependent variable to be explained in this model is the extent of structural damage

incurred, which we represent by the reported damage amount of the property in percentage

terms. The natural log of DAMAGE % is used in the regression analysis because of the skewness

of its distribution.19 All the independent variables in this second model are the same as in

Equation (1), however their corresponding coefficients now have different interpretations. For

example, the coefficients  2 and 3 enable estimation of the additional percentage change in

damage to those properties built post-NFIP in the A-Zone or V-Zone, respectively, relative to

20 pre-NFIP construction. Similarly, the coefficient 5 leads to estimation of the additional

percentage change in damage for properties designated seaward of the 1991 CCCL, and built

after this reestablishment, relative to other similarly located structures built before the 1991

reestablishment. Our expectations on the signs of the coefficients are unchanged from Model 1.

That is, a change in a given variable thought to increase (decrease) the likelihood of damage is

similarly expected to increase (decrease) the percent of damage sustained, conditional on

19 The skewness coefficient for the damage amount implies a long tail distribution in the positive direction. Transforming this variable allows for a better linear fitting multivariate regression model. 20 Unlike the coefficient of a continuous independent variable in a full-log regression, the interpretation of an indicator variable’s parameter is not directly obvious. It is the exponential of the indicator variable’s parameter that yields its marginal effect on the dependent variable.

14 damage. Since the natural logs of GULFDIST , BLDGAREA , LOTAREA , and PRICESF are used in the regression analysis, along with the natural log of the dependent variable DAMAGE % ,

the coefficients 1 , 2 , 3 , and 4 reflect a percentage change in DAMAGE % based on a

percentage change in the corresponding independent variable.

4. DATA

Hurricane Charley was the second major hurricane of the 2004 Atlantic hurricane season

and the most powerful storm to strike Southwest Florida since 1960. The Category Four

hurricane made landfall on North Captiva Island, just north of Captiva Island, with maximum

winds near 150 miles per hour. North Captiva Island was severed into two parts when the right

eye-wall of Charley passed over the island. The most pronounced damage pattern was in a small

area of Captiva Island where damage resulted from a mix of both tornado and hurricane wind

damage. All barrier islands in our sample sustained predominant erosion.21

Data for our research, provided by the Lee County Property Appraiser's Office, includes

barrier island (city), lot size, structure size, assessed value of structure per square foot, and NFIP

flood zone categorization. Visual inspection of each property through the Lee County Property

Appraiser website was used to determine the distance of the building site to the Gulf of Mexico,

and whether the building site is seaward or landward of the 1991 CCCL.22

Of critical importance is the correct modeling of the relevant code regimes. Code requirements do not apply to structures under construction and for which a valid and unexpired building permit was issued prior to the effective date of the code change. Only permit data

21 Today, Charley stands behind Hurricane Katrina in 2005 and Hurricane Andrew in 1992 as the third costliest hurricane in United States history, with property damage estimated by the National Hurricane Center (NHC) at $15 billion. 22 We note there that no property in the sample was seaward of the original 1978 CCCL.

15 beginning after 1980 was available to us. Accordingly, our sample includes only those properties for which building permits were granted after January 1, 1980 and for which we were able to verify a valid building permit date. However, information on when construction of a building was completed allows us to also accurately model the age of the building.23

Property damage estimates resulting from Hurricane Charley were assigned by the Lee

County Property Appraiser's Office, whether the original source was a Lee County Emergency

Operations Center (EOC) damage report, a homeowner damage report, or a Property Appraiser’s

field inspection.24 The damage estimates were in some cases categorical (e.g. minor, major, etc.),

and in some cases a percentage estimate of overall damage was provided. Our final sample

includes percentage estimates derived from professional appraisers with the county.25 Because

non-reported structures (damaged or undamaged) were included in the mass field inspections, sample selection issues related to reporting were negated. All post-1980 construction in the study area is included in our final sample.

[Insert Table 2]

Descriptions for the variables used in the analysis can be found in Table 2. Table 3 displays descriptive statistics for our cross sectional sample of 264 residential properties in Lee

County’s Coastal Building Zone. Of these, 233 properties incurred some damage while 31

properties incurred no damage. Total damage DAMAGE% is measured as the percent of structural damage caused by Hurricane Charley. The percent of damage ranges between 0% and

23 In some cases, the difference between the permit date and the date of completed construction was several years. 24 Lee County residents received two separate mass mailings shortly after Hurricane Charley instructing them to provide categorical information regarding the extent of damage to their property. The county also accepted insurance company documentation regarding property damage to homeowners. 25 This would not have been the case had categorical assessments been utilized, unless only Property Assessor’s inspected categorical assessments were used. However, these contain less information on the extent of damage than a percentage. Moreover, inspected categorical assessments present the same problem of not having a comprehensive sample of damaged properties.

16 80%, with mean and median damage at 15% and 10%, respectively. On average properties are

1,544 feet, about one-third of a mile, from the Gulf of Mexico. The closest property is 84 feet from the Gulf, and the farthest property from the Gulf is 2.3 miles away. The size of the actual residential dwellings in our sample ranges from 669 square feet to 10,725 square feet. The mean and median of building area are similar at 2,697 square feet and 2,365 square feet, respectively.

The average lot area of the parcels in our sample is approximately 18,457 square feet (0.42 acres). Overall, the mean and median real price per square foot for buildings are $105.47 and

$98.33, with a minimum of $33.64 and a maximum of $277.31. The age of each residence is estimated by the difference between year 2003 and the year construction of the residence was completed. On average, residential buildings are just over 11 years old with the oldest being 24 years.

Table 4 provides damage incidence by code regime and location. Most of the 264 properties are built post-NFIP (229) and post-CCCL reestablishment (170). The majority of the sample is located in the A-Zone. There are 160 properties in the A-Zone, and 104 in the V-Zone, respectively. Similarly, the majority of observations in the sample are landward of the 1991

CCCL, 166, as compared to 98 properties seaward of this line. In general, 74% (26/35) of pre-

NFIP structures are damaged from Charley, compared to 90% (207/229) of post-NFIP construction. Similarly, 84% (59/70) of properties seaward of the CCCL and built prior to re- establishment sustained damage, as compared to 93% (26/28) of those properties built after the reestablishment and seaward of the CCCL.

If building regulations affect the perceived trade-off between safety provision and cost of compliance, we would expect both land prices and the timing of development to be related to such regulation (see Dehring 2006). We note here that an apparent permit run on properties

17 seaward of the 1991 CCCL right before its establishment suggests market participants viewed

the regulations as having consequence. Moreover, this reestablishment appears to have spurred

preemptive development. Figure 1 and Figure 2 show the frequency distribution of building

permits by year for the sample. Specifically, Figure 1 shows the distribution for the 98 properties

seaward of the 1991 CCCL. A spike in the distribution can be seen in 1991, the year of the

CCCL reestablishment. In 1991, 23 permits were issued for property seaward of this line. All of

these were issued before the May 30, 1991 effective date of the CCCL. Figure 2 shows no such

spike in permits landward of this line. The next section discusses the empirical results of our

logistic and linear regression models.

5. RESULTS To assess the effects of coastal construction code changes on the likelihood of residential

properties sustaining damage while controlling for location and structural attributes, we

performed a binary logistic regression using the dependent variable, DAMAGE . We find

evidence that coastal building code regime explains the likelihood of damage to property. These

findings are presented in Table 5.

[Insert Table 5]

The likelihood of sustaining damage appears to be effected by some locational and

structural characteristics of residential properties in our sample. The coefficient on SANIBEL is negative and significant, suggesting properties located on Sanibel Island were less likely to sustain damage relative to properties located on Captiva Island. The positive coefficient on

BLDGAREA implies that larger dwellings, based on square feet, were significantly more likely to incur damage in Hurricane Charley relative to smaller buildings. Finally, the positive coefficient

18 on AGE suggests a greater likelihood of damage for every additional year the structure has aged, after controlling for building code regime.

We find regulatory variables to be significant in this model. Prior to participation in the

NFIP, we find no difference in the likelihood of damage for an A-Zone location as compared to a

V-Zone location. However, the positive coefficient of POSTNFIPAZONE indicates that residences built after the implementation of NFIP regulation and located in the A-Zone are significantly more likely to sustain damage than similarly located structures built prior to NFIP regulation. We also find a lower likelihood of damage for structures located seaward of the 1991

CCCL, but no change in this likelihood following the CCCL reestablishment. This finding likely reflects existing codes in place. Thus, the main finding concerning the regulatory regime under which a property is constructed suggests that in the A-Zone, properties built in accordance with the NFIP’s coastal building standards are more likely to have been damaged compared to similarly located properties built before the county joined the NFIP.

Out of our sample of 264 residential properties, 31 did not incur any damages in

Hurricane Charley. For the other 233 properties that did incur damages, Table 6 presents multivariate linear regression results of the models used in our analysis of total damages sustained by these properties.

[Insert Table 6]

Similar to the logistic regressions, we examine the impact of NFIP and CCCL regulation in reducing (or increasing) total property damages relative to pre-NFIP construction and pre-

CCCL construction, conditional on some damage being sustained. We again find structural and location variables to be significant. Distance from the Gulf of Mexico, structural age and Island location are all characteristics associated with the amount of damage. The extent of damage is

19 greater for older structures and for properties closer to the Gulf of Mexico. The coefficient of

LN(GULFDIST) is -0.2367, implying that a 10% increase in the distance from the gulf is

associated with a decrease in total damages of 2.37%. The coefficient on AGE is 0.040

suggesting that a one-year increase in the age of a building increases total damages by 4.0%. The

geographic location of the properties relative to Captiva Island also tends to influence the extent

of damage. The coefficients on GASPARILLA and SANIBEL are significant and negative,

suggesting residences located on these islands sustain approximately 51% and 49% less damage,

respectively, than those located on Captiva Island. Similarly, the coefficient on ESTERO is

positive and significant, suggesting Estero Island residences sustain 67% more damage than

those on Captiva.

In our regressions on the percent of damage sustained in Charley, we do not find more or

less damage associated with either an A-Zone location or a location seaward of the CCCL.

However, we again find a significant result related to A-Zone properties built after the implementation of NFIP regulation. The positive coefficient on POSTNFIPAZONE of 0.57 suggests that relative to properties built prior to NFIP regulation, those built after incur almost

57% more total damages, holding other things constant. We also find that properties seaward of

the 1991 CCCL and built following the reestablishment incur 47.5% more total damages than

similarly located residences built prior to the re-establishment.

Our findings suggest a greater likelihood of damage associated with NFIP A-Zone

property, and, conditional on damage, a greater extent of damage to be associated with the

Coastal Construction Control line and the NFIP A-Zone. While consistent with the alternative

hypothesis of building codes discussed earlier, our results are troubling given that both programs

are designed to protect structures against damage from flood and storm events. However, any

20 conjectures regarding the cause of our results is not satisfying without further empirical support.

To better explain our results, at least with regard to the NFIP A-Zone, we compare the code requirements under the NFIP against the previous code regime for each individual property. 26

6. PROPERTY-LEVEL CODE REQUIREMENT COMPARISON

For each property in the sample we determine the location of the building on each lot to determine the appropriate pre-NFIP zone classification (1, 2 or 3). From this we obtain the pre-

NFIP required elevation and foundation requirement for each property (the requirements under the old county code). Because pre-NFIP codes for Sanibel Island were not made available, our examination includes the 200 properties on Captiva, Estero, and Gasparilla Island.

First, we compare building elevation requirements under the old county code to those under NFIP regulation. Changes in required elevations by code regime, flood zone and damage frequency are presented in Table 7. For the full sample we see that 71 properties had a reduction in minimum required elevation of between 1 and 4 feet when the county joined the NFIP. Of these properties, 63 were in the A-Zone, and 8 were in the V-Zone. Of the rest of the sample, 120 properties had no change in elevation, and 17 had increases in elevation. For the damaged post-

NFIP sample (excluding Sanibel), the regression relationship between the percent of damage and elevation change (in feet) is significant at 5%, and suggests a 1 foot decrease in elevation increases damage by 1.267%.27

[Insert Table 7]

Taking a closer look at A-Zone land, we find the majority of A-Zone land, 63 of 98

properties, had lower minimum elevations under the NFIP. By contrast, only 8 of 102 V-Zone

26 We thank an anonymous referee for this suggestion. 27 DAMAGE% = 18.194 -1.267ELEVCHG

21 properties had decreased elevation. Of the 63 A-Zone properties with reduced minimum

elevations, 58 were post-NFIP construction. Of these, 54 properties, or 93% of the post-NFIP A-

Zone construction with decreased elevation, were damaged. However, we also see a high

percentage of post-NFIP, A-Zone construction with no change or increased elevation with

damage, 95% and 60%, respectively. With V-Zone land, a higher percentage of post-NFIP

construction with increased elevation incurred damage (100%) as compared to those with a

decrease elevation (88%). Thus, while it is notable that so many A-Zone properties had a

reduction in minimum elevation with the NFIP, we find no overwhelming descriptive evidence that it is these elevation changes in the A-Zone are associated with our earlier A-Zone findings.

Moreover, the regression relationship between the percent of damage and elevation change (in

feet) is not significant when applied only to the A-Zone land. This suggests that some other facet

of the code change is driving our earlier A-Zone results.

Next, we look at damaged A-Zone properties to see whether those that would have been

required to have pile foundations under the old code (Zone 1 and 2) sustained more damage than

those that would have been allowed weaker foundations (Zone 3).28

[Insert Table 8]

As can be seen in Table 8, most of these A-Zone properties would have been classified as

Zone 3 under the old code regime. However 21 properties in the A-Zone would have been

classified as Zone 1 or 2 (the more seaward zones) under the old code. The mean damage

percentage for these 21 properties is 26.1%, which is 43% higher than the mean damage

percentage on the 57 Zone 3 properties. Median damage for these Zone 1 or 2 properties is 112%

28

22 more compared to the Zone 3 properties (26.5% versus 12.5%). For the damaged post-NFIP A-

Zone sample, the regression relationship between the percent of damage and a Zone 1 or 2 designation under the old code is significant at 5%, and suggests a Zone 1 or 2 designation under the old code increases damage by 7.729%.29

Reduced elevation and weaker foundations in the A-Zone would seem to be associated

with a greater flood risk. To further analyze our findings concerning elevation and foundation requirements in the A-Zone, we apply logistic regression models where an indicator variable denoting damage to various building components acts as the dependent variable. Specifically, the four dependent variables used in these models are ROOF, EXTWALL, INTWALL, and FLOOR.30

These variables indicate whether there was any damage to the roof, exterior wall of the structure, interior wall of the structure or floor system, respectively.31 Coefficients of the independent

variables are interpreted similarly to our first logistic regression model. Here, we expect damage

to the floor and the interior wall in a flood or storm surge to be related to elevation or building

foundation, while roof and exterior wall damage should be associated more with wind damage.

Using each of these structural components as the dependent variable, we find a slightly

significant difference in the likelihood of roof damage or exterior wall damage for those A-Zone

properties built post-NFIP implementation relative to those built pre-NFIP. There is stronger

significant evidence that the likelihood of interior wall damage and floor damage are both higher

for those A-Zone properties built post-NFIP. The results of this analysis are presented in Table 9.

[Insert Table 9]

7. CONCLUSION

29 DAMAGE% = 18.342 +7.729ZONE1OR2 30 We are able to include Sanibel Island A-Zone property in this regression, as we do not control for old codes. 31 Reports on damage to building components only indicate presence or absence, so we cannot use Model 2 with these variables.

23

Losses from future hurricanes will likely increase as coastal populations continue to grow. One estimate of residential and commercial coastal property exposures in Florida stands at $1.94 trillion, however this number will certainly increase as growth continues. Hence, the ex- ante mitigation of losses to coastal properties from natural disasters has become a vital concern.

Moreover, the economic significance of any proposed effective loss control device prior to implementation is of significant importance.

Our study investigates whether state and federal mandated changes in coastal building standards mitigate hurricane damage to residential properties in high hazard coastal areas. We find that code changes associated with National Flood Insurance Program are associated with more structural damage to coastal property. After controlling for locational and physical attributes of the structure, we find coastal zone properties located in an NFIP designated A-Zone that were built under compliance with requirements set forth by the Federal Emergency

Management Agency have an increased likelihood of property damage, and greater extent of property damage, conditional on some damage being sustained, relative to similarly located pre-

NFIP built structures. A comparison of the building requirements under the NFIP relative to what was in place at the county level before the NFIP provisions reveals that most A-Zone land in the sample was actually subject to a “decrease” in coastal building standards in terms of both lower required elevation and less stringent foundation requirements. Additional analysis on structural component damage in the A-Zone reveals a higher incidence of floor and interior wall damage, consistent with flood damage.

We also find a greater extent of property damage, conditional on some damage being sustained, to property seaward of the Coastal Construction Control Line that was built after the re-establishment of this line. Because of the nature of the siting and design requirements for

24 properties seaward of the CCCL, we cannot undertake the type of comparison we did for A-Zone

land for properties seaward of this line. However, a permit run in the months prior to the

redrawing of the CCCL for those properties that would have been affected by the CCCL re- establishment suggests two things. First, that the nature of the CCCL code compliance mattered to land market participants. Second, the redrawing had the unintended consequence of preemptive development, an outcome that runs counter to the goal of the very regulations that establish the CCCL.

The empirical findings concerning overall structural damage are consistent with the

literature in terms of land prices and coastal construction code changes. Dehring (2006)

examines the impact of these same coastal code changes on vacant land prices on Lee County’s

barrier islands. Land prices are found to decrease by up to 30% following code changes

associated with the NFIP and the CCCL reestablishment. Our current study implies that market

participants may have indeed priced the effect of the code changes accurately.

Our findings raise questions concerning the optimal scale of code design and

enforcement. There has been a shift in recent decades away from local codes to state or model

codes. Currently 41 states mandate a model code or state code, compared to 15 in 1976 (Colwell and Kau, 1982). While model codes introduce efficiencies in design, they may be subject to

political interference by manufacturers and trade associations. Further, these codes may be based

purely on visible factors, possibly at the sacrifice of overall construction quality. On the other

hand our findings may suggest that the NFIP’s A-Zone standards are not well suited for barrier

islands.

More quantitative post-hurricane event damage assessments are clearly warranted.

Currently, accessing comprehensive and accurate data has restricted opportunities for further

25 research on the effectiveness of regulation in mitigating structural risk to property in high hazard areas. If such data were made available (e.g. NFIP participation by parcel), this would facilitate a more accurate assessment of moral hazard in this context. Cooperation from FEMA, insurance companies, professional appraisers, coastal engineers and social scientists would improve both the quantity and quality of research in this critical area. Our investigation is but one instance where empirical evidence suggests that regulation modifications produce unintended effects in terms of structural damages caused by hurricanes.

26 References

Blake, E.S., Brown, D.P., and R.J. Pasch, 2005, Hurricane Charley, National Hurricane Center.

Chivers, J. and N. Flores, 2002, Market Failure in Information: The National Flood Insurance Program, Land Economics, 78: 515-21.

Cohen, A., and R. Dehejia, 2004, The Effect of Automobile Insurance and Accident Liability Laws on Traffic Fatalities, The Journal of Law and Economics, 47: 357-393. Colwell, P. F., and J. B. Kau. 1982. The Economics of Building Codes and Standards. In M. Bruce Johnson (editor) Resolving the Housing Crisis: Government Policy, Decontrol and the Public Interest. San Francisco: Pacific Institute for Public Policy Research, Ballinger Publishing Company.

Dehring, C., 2006, Building Codes and Land Values in High Hazard Areas, Land Economics, 4: 513-29.

Ehrlich, I., and G. Becker, 1972, Market Insurance, Self-Insurance, and Self-Protection, The Journal of Political Economy, 80: 623-648.

Federal Emergency Management Association, 2004, Hurricane Charley Rapid Response Coastal High Water Mark Collection,. Hazard Mitigation Technical Assistance Program, Task Order 326.

Florida Department of Environmental Protection, 2004, Hurricane Charley. Post-storm Beach Conditions and Coastal Impact Report with Recommendations for Recovery and Modifications of Beach Management Strategies, Bureau of Beaches and Coastal Systems.

Florida Department of Environmental Protection, 2005, Hurricane Charley Characteristics and Storm Tide Evaluation, Bureau of Beaches and Coastal Systems.

Fronstin, P., and A. Holtmann, 1994, The Determinants of Residential Property Damage Caused by Hurricane Andrew, Southern Economics Journal, 61: 387-397.

Hunter, J., February 2, 2006, Testimony in National Flood Insurance Program hearing before the Committee on Senate Banking, Housing and Urban Affairs, United States Senate.

Kriesel, W., and C. Landry, 2004, Participation in the National Flood Insurance Program: An Empirical Analysis for Coastal Properties, The Journal of Risk and Insurance 71: 405-420.

Kunreuther, H., 1974, Disaster Insurance: A Tool for Hazard Mitigation, The Journal of Risk and Insurance, 41: 287-303.

Lindsay, C., 1976, A Theory of Government Enterprise, The Journal of Political Economy, 84: 1061-77.

27 Oster, S., and J. Quigley, 1977, Regulatory Barriers to the diffusion of Innovation: Some Evidence from Building Codes, Bell Journal of Economics, 8: 360-76.

PCI White Paper. June 2011. “True Market-Risk Rates for Flood Insurance.” Property Casualty Insurers Association of America.

Peltzman, S., 1975, The Effects of Automobile Safety Regulation, The Journal of Political Economy, 83: 677-726.

Shavell, S., 1979, On Moral Hazard and Insurance, Quarterly Journal of Economics, 93: 541- 562.

Seidel, S. R., 1980, The Effect of Building Codes on Housing Costs, In Housing Costs and Government Regulations: Confronting the Regulatory Maze, New Brunswick: Center for Urban Policy Research, Rutgers University.

Smith, S. K. 2005. “Florida Population Growth: Past, Present and Future.” University of Florida Bureau of Business and Economic Research. Unpublished working paper.

Stiglitz, J., 1983, Risk, Incentives and Insurance: The Pure Theory of Moral Hazard, The Geneva Papers on Risk and Insurance, 8: 4-33.

28 TABLE 1 Comparison of Construction Standards on Captiva Island Before and After the NFIP Zone 1 and 2 (pre-NFIP) Zone 3 (pre-NFIP)

Excavation / Grading No lowering of existing natural ground No restriction elevations

Foundation Pile Pile or Column on footing

Dimension 8 to 10 inches per side Spacing 6 feet apart Embedment -8 to -5 feet NGVD

Pile Clusters and 4 to 5 feet max for pile cap and bracing No restriction Bracing elevation

Understructures No substantial walls or partitions below Wall and partitions below the first first finished floor in Zone 1a floor elevation in area IIa shall be considered expendable during severer storm activity

Minimum elevation 12 ft. NGVD 12.0 ft. NGVD

Applies to Underside of building support structure* Minimum 1st finished floor

Horizontal wave 200-600 pounds/(feet)2. Varies by wave n/a pressures elevation (ranging from 13.0-15.5 ft.)

Uplift pressures 35-80 pounds/feet. varies by elevation of n/a deck underside

Pile forces Horizontal loads of 2 kips to 3.8 kips x 0.5 kip x max pile dimension max pile dimension

Design considerations Stable soil conditions shall not be n/a assumed above 2.6 feet NGVD

A-Zone (NFIP) V-Zone (NFIP)

Specific Pile or Continuous foundation Pile or Columns on footing that can withstand all applied loads of the base flood flow Minimum elevation Base Flood Elevation (BFE)

Applies to: Lowest floor Lowest supporting beam

General New construction must be

- anchored to prevent flotation, collapse, or lateral movement of structure - constructed with materials and utility equipment resistant to flood damage - constructed by methods and practices that minimize flood damage

new and replacement water supply, sanitary sewage and on-site waste disposal systems shall be designed to minimize or eliminate infiltration of flood waters, or located to avoid impairment of contamination

Lowest horizontal structural member Lee county ordinances 76-15 and 84-17 * Wall and partitions below the first floor elevation in area IIa shall be considered expendable during severer storm activity

29 TABLE 2 Variable Descriptions Variable Name Description DAMAGE = 1 if any reported damage to structure DAMAGE% Percentage of overall damage to structure ROOF = 1 if any reported damage to roof of structure EXTWALL = 1 if any reported damage to exterior walls of structure INTWALL = 1 if any reported damage to interior walls of structure FLOOR = 1 if any reported damage to floor system of structure GULFDIST Distance in feet from structure site to Gulf of Mexico BLDGAREA Size of the residential structure in square feet LOTAREA Size of the structure lot in square feet PRICESF Real price per square foot of the structure ($2003) AGE Age of structure in 2003 CAPTIVA = 1 if structure located on Captiva or North Captiva Island GASPARILLA = 1 if structure located on Gasparillo Island ESTERO = 1 if structure located on Estero Island SANIBEL = 1 if structure located on Sanibel Island AZONE = 1 if structure located in A-Zone VZONE = 1 if structure located in V-Zone POSTNFIP = 1 if commenced structure construction after NFIP regulation POSTNFIPAZONE = 1 if commenced structure construction in the A-Zone after NFIP regulation POSTNFIPVZONE = 1 if commenced structure construction in the V-Zone after NFIP regulation CL91 = 1 if structure located seaward of the 1991 CCCL POSTCL91 = 1 if commenced structure construction after 1991 CCCL regulation POSTCL91CL91 = 1 if commenced structure construction seaward of the 1991 CCCL after 1991 CCCL regulation

30

TABLE 3 Descriptive Statistics Variable Mean Median Std Dev Min Max DAMAGE% 14.74 10.00 14.18 0 80.00 GULFDIST 1,543.81 932.16 2,082.04 83.83 12,399.84 BLDGAREA 2,697.53 2,365.00 1,334.25 669.00 10,725.00 LOTAREA 18,457.10 13,256.06 18,213.11 4,118.01 184,720.01 PRICESF 105.47 98.33 44.02 33.64 277.31 AGE 11.30 11.00 6.33 1.00 24.00 CAPTIVA 0.67 1 0.48 0 1 GASPARILLA 0.05 0 0.21 0 1 ESTERO 0.04 0 0.19 0 1 SANIBEL 0.24 0 0.43 0 1 POSTNFIP 0.87 1 0.34 0 1 POSTNFIPAZONE 0.53 1 0.50 0 1 POSTNFIPVZONE 0.34 0 0.47 0 1 CL91 0.37 0 0.48 0 1 POSTCL91 0.47 0 0.50 0 1 POSTCL91CL91 0.10 0 0.30 0 1 Notes: Obs.=264 for the full sample. The A-Zone contains 160 observations and the V-Zone contains 104 observations. 31 observations had estimates of zero damage percent.

31

TABLE 4 Damage Frequency by Regulatory Regime in Effect and Structure Location Pre Regulatory Change Post Regulatory Change Total Damage No Damage Damage No Damage A-Zone (NFIP) 13 8 125 14 160 V-Zone (NFIP) 13 1 82 8 104 Totals 26 9 207 22 264

Seaward of CCCL 59 11 26 2 98 Landward of CCCL 17 7 131 11 166 Totals 76 18 157 13 264 Notes: Obs.=264 for the full sample. The determination of which regulatory regime is based on the date the building permit was granted compared to August 31, 1984 for NFIP regulatory change and May 30, 1991 for CCCL regulatory change.

32 TABLE 5 Logistic Regression on Damage Occurrence Wald Independent Variables Coefficient Standard Error Chi-Square Pr > ChiSq Intercept -3.5243 6.1659 0.3267 0.5676 LNGULFDIST 0.0502 0.3247 0.0239 0.8771 LNBLDGAREA 1.2219 0.6654 3.3723 0.0663* LNLOTAREA -0.5107 0.4599 1.2331 0.2668 PRICESF 0.00308 0.00685 0.2022 0.6530 AGE 0.0937 0.0558 2.8179 0.0932* GASPARILLA -1.7464 1.0852 2.5899 0.1075 ESTERO -1.3750 1.0325 1.7734 0.1830 SANIBEL -1.4412 0.6653 4.6924 0.0303** AZONE -1.6002 1.2645 1.6014 0.2057 POSTNFIPAZONE 2.1547 0.7876 7.4852 0.0062*** POSTNFIPVZONE 0.2744 1.2395 0.0490 0.8248 CL91 -1.1258 0.6187 3.3105 0.0688* POSTCL91CL91 0.9850 0.9375 1.1039 0.2934 Notes: Obs. =264. The model includes indicator variables regarding construction of structures after NFIP regulation and after CCCL regulation. The dependent variable indicates if damage occurred. The parameter estimates are maximum likelihood estimates. The p-values of the coefficients are based on the Wald chi-squared test statistic, which is the square of the coefficient divided by its estimated standard error. *, **, *** denote statistical significance at 10 percent, 5 percent and 1 percent levels, respectively, based on a two tailed test that the true coefficient is zero.

33

TABLE 6 Multivariate Regressions on the Natural Log of Damage Percentage

Independent Variables Coefficient Standard Error t-Value Pr > |t| Intercept 3.99031 1.45869 2.74 0.0067*** LNGULFDIST -0.23671 0.08658 -2.73 0.0068*** LNBLDGAREA -0.07269 0.15921 -0.46 0.6484 LNLOTAREA -0.03072 0.11675 -0.26 0.7927 PRICESF 0.000063 0.00164 0.38 0.7026 AGE 0.04035 0.01559 3.28 0.0012*** GASPARILLA -0.51303 0.29469 -1.74 0.0831* ESTERO 0.67052 0.34080 -1.97 0.0504* SANIBEL -0.49317 0.16198 -3.04 0.0026*** AZONE 0.21442 0.32328 0.66 0.5079 POSTNFIPAZONE 0.57048 0.25748 2.22 0.0277** POSTNFIPVZONE 0.21849 0.25122 0.87 0.3854 CL91 0.14386 0.14644 0.98 0.3270 POSTCL91CL91 0.47560 0.20146 2.36 0.0191** Notes: Obs. =233. The model includes indicator variables regarding construction of structures after NFIP regulation and after CCCL regulation. The dependent variable is the natural log of damage percentage. *, **, *** denote statistical significance at 10 percent, 5 percent and 1 percent levels, respectively, based on a two tailed test that the true coefficient is zero. Variance Inflation Factors (VIF) showed no evidence of collinearity among the independent variables. Brausch-Pagan test statistics for heteroscedasticity were not significant.

34 TABLE 7 Change in Minimum Required Elevation (in feet): Post-NFIP - Pre-NFIP elevation change A-Zone V-Zone Full Pre- Post- Post- Pre- Post- Post- (in feet) Sample all NFIP NFIP damaged all NFIP NFIP damaged -4 1 1 0 1 1 0 0 0 0 -3 18 18 1 17 16 0 0 0 0 -2 20 19 2 17 15 1 0 1 1 -1 32 25 2 23 22 7 0 7 6 total reductions 71 63 5 58 54 8 0 8 7

0 112 25 3 22 21 87 13 74 67

1 8 1 0 1 1 7 1 6 6 4 1 1 1 0 0 0 0 0 0 12 4 4 2 2 1 0 0 0 0 13 2 2 1 1 1 0 0 0 0 14 2 2 1 1 0 0 0 0 0 total increases 17 10 5 5 3 7 1 6 6

total 200 98 13 85 78 102 14 88 80

Notes: Obs. = 200. The 64 Sanibel Island properties are not included.

35

TABLE 8 Damage Frequency for Post-NFIP Regulatory Regime Construction in the A-Zone A-Zone Mean damage% Median damage % Damage Zone 1 or 2 21 26.1% 26.5% Zone 3 57 18.3% 12.5% Total 78 Notes: Obs. =78.

36 TABLE 9 Logistic Regressions on Structural Component Damage Occurrence to A-Zone structures, Conditional on Some Overall Damage Independent Variables Roof Extwall Intwall Floor Intercept -5.067 1.2418 -5.3410 -0.8209 (5.0021) (5.9934) (5.1796) (5.1836) LNGULFDIST 0.2590 -0.1230 -0.4945 -0.5583 (0.3138) (0.3654) (0.3309) (0.3356) LNBLDGAREA 0.4694 .6075 1.5508 ** 0.2502 (0.6060) (0.6906) (0.6653) (0.6057) LNLOTAREA -0.3440 -0.6262 -0.4961 -0.0237 (0.4126) (0.4847) (0.4340) (0.4218) PRICESF 0.00791 0.00766 0.00083 0.00565 (0.00798) (0.00950) (0.00861) (0.00852) AGE 0.1322 *** 0.0147 0.0795 0.0971 * (0.0510) (0.0583) (0.0532) (0.0535) GASPARILLA -1.2721 -2.3962** -1.5691 * -0.4835 (0.8656) (0.9610) (0.9204) (0.9054) ESTERO -0.4733 -2.4783 ** -1.9459 * -1.6957 (0.8808) (1.0342) (0.9504) (1.0037) SANIBEL -0.1671 -3.3242 *** -2.1057 *** -1.6919 *** (0.5425) (0.7083) (0.6139) (0.5894) POSTNFIP 1.2397 * 1.7082 * 1.9152 ** 2.3809 *** (0.6812) (0.8748) (0.7670) (0.8231) CL91 0.6455 -0.2536 -0.3886 -0.6772 (0.5766) (0.6184) (0.5883) (0.5851) POSTCL91CL91 1.4423 14.0343 0.8239 1.5578 (1.1917) (350.1) (0.9627) (0.9639) Notes: Obs. =135. The dependent variable of each model indicates if damage to the structural component occurred. The parameter estimates are maximum likelihood estimates. Standard errors are in parenthesis. The p-values of the coefficients are based on the Wald chi-squared test statistic, which is the square of the coefficient divided by its estimated standard error. Of the 138 damaged A-Zone properties reported in Table 3, 3 were located on Sanibel Island for which we do not have pre-NFIP codes. This results in 135 observations for this table. *, **, *** denote statistical significance at 10 percent, 5 percent and 1 percent levels, respectively, based on a two tailed test that the true coefficient is zero.

FIGURE 1 Permits by Date: All properties seaward of 1991 CCCL. N = 98

25

20

15

10

5

0

1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 FIGURE 2 Permits by Date: All properties landward of 1991 CCCL. N =166

25

20

15

10

5

0

1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003

39