Affordability of National Flood Insurance Program Premiums: Report 2 (2016)

3

Data for Analyses of National Flood Insurance Program Policy Options

Congress, through the Biggert-Waters Flood Insurance Reform Act 2012 (BW 2012) and Homeowner Flood Insurance Affordability Act 2014 (HFIAA 2014), requested the Federal Emergency Management Agency (FEMA) to undertake a suite of complicated and technical tasks to propose a draft affordability framework for the National Flood Insurance Program (NFIP). The analytical calculations needed to do such analysis were described in Chapter 2. Making these calculations requires the construction of one or more microlevel databases with the necessary data for analysis of representative properties in the nation’s floodplains. Ideally, the database(s) will include data on property characteristics, the socioeconomic characteristics of the property owner and occupant (if different from the owner), and the NFIP policy (if there is a policy in force on that property).

At present, FEMA has access to the NFIP policy database that includes some of these data and to flood insurance rate maps (FIRMs) that in some places could be used to characterize the likelihood of floods that reach different stages in different areas of the floodplain. To evaluate affordability policy options, however, additional data on variables not in the NFIP database and existing FIRMs will be needed.¹ This chapter describes the data

__________________

¹ The content of this chapter is consistent with findings of several other reports that document the need for additional data pertinent to the NFIP (i.e., GAO, 2014a; King, 2013; PwC, 1999). For example, GAO (2014a) could not calculate forgone premiums—the difference between subsidized policies and full-risk premium policies—as there was a lack of property elevation data for pre-FIRM subsided policyholders.

in the NFIP policy database and the data that can be derived from FIRMs. With this as background, data gaps needed for conducting the kinds of analyses described in Chapter 2 are identified, and means for filling such gaps discussed.

THE NFIP POLICY DATABASE

FEMA’s NFIP policy database includes about 4.5 million records and 76 data fields. The database was created for tracking NFIP policies insured by the Federal Insurance and Mitigation Administration (FIMA) that resides within FEMA.² The flood insurance policies in the database include those written by direct servicing agents, the Write-Your-Own (WYO) Program agents, and private insurance agents in companies not associated with the WYO program. In May 1998, the NFIP created a manual that insurance companies must abide by when collecting and submitting policyholder information. All of the policy information submitted is compiled into the NFIP policy database, which is updated on a monthly basis (NFIP, 2013).

Depending on the type of structure being insured, the NFIP uses three different forms for flood insurance policy applications:

1. Dwelling forms are used for homeowners, residential renters, or owners of residential buildings that contain one to four units.
2. General property forms are used for owners of residential buildings with five or more units, as well as for owners or lessees of nonresidential buildings or units.
3. Lastly, residential condo building association forms are provided to residential condo associations on behalf of the association and their unit owners (FEMA, 2014).

The review of the NFIP policy database for this report was based on a snapshot of the NFIP policy data from October 2013.³ The October 2013 NFIP policy database includes the following information and attributes as categorized below.

Policies. The NFIP policy database includes general identifying information about the policyholder, including name and address. Since FEMA

__________________

² The NFIP maintains a claims database that was not part of this review.

³ The assessment of the NFIP data reported in this chapter was complicated by difficulties in accessing and understanding the NFIP database. For example, neither a data dictionary nor metadata was available for the database. By following ISO 8000 and 9001 standards, as well as Federal Geographic Data Committee metadata standards, the NFIP could adopt a well-recognized process of data management that will help users access and understand the content of the NFIP database.

tracks policies by the individual, rather than by the property, unique policy numbers are assigned to each policyholder and property holder. When a policyholder moves, a new policy number is assigned. Some aspects of the policy are protected by FEMA under privacy concerns, such as names and addresses, and are not available to the public. Other general policy information within the NFIP policy database includes

• policy status (active, canceled, etc.);
• number of policy terms (number of years the policy is effective: 1 year, 3 years, etc.);
• whether the policy was required for disaster assistance, and if so, by which agency;
• company code of the WYO company responsible for the policy (where applicable);
• whether the policy is for a single-family or multifamily property; and
• whether it is for a residential or nonresidential property.

Location. Several attributes related to the spatial location of the insured property can be found in the NFIP policy database, including the latitude and longitude coordinates of the property; the property address, city, state, and zip code; the FEMA region; and which Census block (and block group) the property falls within. Information on how accurate the horizontal geocoding is for the property is also provided so the user knows how well the policy is located. The NFIP application does not contain the latitude and longitude coordinates of the property. This information is generated by FEMA using outside firms to geocode the property address. In addition, FEMA includes the attributes from the FIRMs as part of the policy database. The NFIP community and county that the insured structure is located within, as well as the map panel number and flood zone, as obtained from the FIRMs, are also provided.

Chosen Coverage. Insurance deductible and coverage amounts for both the property and the contents are included within the policy database, as are premiums. Policy endorsement dates, original effective dates (for rollover policies), current effective dates, and expiration dates are also provided.

Premiums/Policy Type. Several attributes within the NFIP policy database are utilized for the insurance premium calculations. Some of these elements include

• whether it is a new policy or a renewed policy;
• what flood zone was used for rating the policy;

• if the policy has a V zone (which are areas within the special flood hazard area (SFHA) with additional hazards associated with storm-induced waves) risk-factor rating, where a qualified professional assesses the building’s location, its support system, and its ability to withstand wind and wave action. If the professional certifies that the property has a lower risk of flood damage based on these three factors, then the property becomes eligible for a premium discount;
• whether it is a pre- or post-FIRM property;
• type of residence;
• whether the policy falls under any BW 2012 categories and, if so, which BW 2012 category it would fall under. Some examples of the BW 2012 categories include single-family nonprincipal residences, businesses, severe repetitive loss pre-FIRM subsidized properties, and multifamily residences;
• whether the property is in a Community Rating System (CRS) community, and if so, which CRS class that provides premium discounts to all homeowners in the community ranging from 5 percent (Class 9) to 45 percent (Class 1);
• the policy’s NFIP community program type (regular or emergency);
• the location of the contents within the structure; and
• any obstruction types/categories associated with the structure. Some of the factors used to establish the obstruction categories include the size of the structure (less than or greater than 300 square feet), whether the structure has breakaway walls, if the building has an enclosure or crawl space with proper openings, whether there is machinery or equipment and is it above or below the base flood elevation, whether there is an elevator and is it above or below the base flood elevation, and whether the building is elevated.

Building Characteristics. The policy dataset provides several building attributes that can be used to review and assess flood risk at the structure level. Some of these characteristics include when the structure was built, whether the property is in the course of construction, the number of units within the property, the number of floors in the building, the type of basement or enclosure it has (if any), and whether the building is elevated and/ or flood proofed.

Elevation Data. The following fields are provided in the NFIP policy database, but the information within them is not fully populated for all policies:

• Base flood elevation (BFE) from the FIRMs
• Whether there is an elevation certificate for the property and, if so, what the diagram number is

• Elevation of the lowest floor
• Elevation difference between the BFE and the lowest floor
• Lowest adjacent grade

FEMA utilizes many of the policy attributes listed above, in addition to building characteristics and elevation information discussed below, to establish a risk-rating method for each elevation-rated policy. Policies issued for properties outside the SFHA are not risk rated, as is the case for pre-FIRM subsidized policies that are now being phased out. For properties outside the SFHA, elevation data will be missing.

Miscellaneous Attributes. There are a few other fields included within the NFIP policy database that do not necessarily fit within the categories mentioned above, but that may still be of value for an affordability analysis. These include

• whether the property is state owned,
• the federal policy fee,
• the community probation surcharge amount, and
• the insurance to value indicator.

Most of the fields included in the NFIP policy database are well populated. The October 2013 policy database had a 95 percent completion rate or better in regard to general information about the policy and policyholder, the geographic location of the insured structures, data from the community’s FEMA FIRMs, the insurance deductible and coverage amounts for each policy, the policy premiums and risk ratings, and insured building characteristics under each policy. Although this information is needed for constructing a database as inputs for evaluating the affordability of insurance policies, there were some limitations related to the completeness of the policy data as well. Most importantly, about 70 percent of the policy records lack information about structure elevation relative to the BFE.

For assessing current risk, it is necessary to know the current flood zone for the property. The reported flood zone in the database, however, is the one that was used for the initial policy risk rating. To identify the actual current NFIP flood zone that the structure lays within, a geospatial analysis would need to be performed whereby all of the NFIP policies are intersected with FEMA’s National Flood Hazard Layer (NFHL). This can be a time-consuming process at the national level that can take a geographic information system (GIS) specialist 4 to 6 weeks to complete, but it would provide accurate location information for assessing actual policy risk ratings and any premium adjustments that may be needed as a result. Appendix H

reports the data fields found within the October 2013 version of the NFIP policy database.

FLOOD INSURANCE RATE MAPS

FEMA has completed an ambitious program to provide the nation with coverage of digital flood insurance rate maps (DFIRMs) for approximately 1 million of the 3.2 million stream miles in the nation; the 1 million miles are located where the majority of the nation’s population lives. The first phase of this program, called Flood Map Modernization, operated from 2003 to 2008, and a subsequent phase, called Risk Mapping Assessment and Planning (Risk MAP), is now in operation (FEMA, 2009a, 2009b). Currently, following instructions from BW 2012, a technical mapping advisory council is preparing a report on several mapping topics including how to improve, in a cost-effective manner, (a) accuracy, general quality, ease of use, and distribution and dissemination of FIRMs and risk data; and (b) performance metrics and milestones required to effectively and efficiently map flood risk areas in the United States.

Lenders use FIRMS to determine whether flood insurance is required by property owners. Private insurance agents use the FIRM to help quote a policy premium. This determination is made on the basis of a horizontal criterion: Does the building lie within or outside the mapped SFHA? The current FIRMs, by showing the boundary of the SFHA, adequately support this flood insurance purchase requirement determination.⁴ For quoting a premium, an essential component of FIRMs is the BFE, which is the water surface elevation that would result from a flood having a 1 percent chance of being equaled or exceeded in any year at the mapped location.⁵ The BFE is a vertical, rather than a horizontal, criterion used in flood insurance purchase requirement determinations. The NFIP risk-based premium

__________________

⁴ If a property owner whose building is classified as being within the floodplain wishes to protest that determination, a procedure is available, for both the owner and the NFIP, to process a Letter of Map Amendment.

⁵ In addition, local communities regulating land development are expected by the NFIP to require the first-floor elevation of buildings to be at or above the base flood elevation.

(elevation-rated premium) is based on the difference between the lowest habitable floor elevation of the property and the BFE for the zone within the SFHA, as well as a limited number of other property characteristics. Base flood elevations are shown on floodplain maps that have been prepared with high-quality land surface elevation information and detailed flood modeling studies.

The implementation of FEMA’s Risk MAP program in 2009 began an effort to provide communities with flood information and tools to enhance mitigation planning, providing more information than the boundary of the SFHA and the BFE. Risk MAP continued the focus on technological advancements that included elevation data acquisition and more accurate mapping of areas impacted by levees and coastal flood hazards (FEMA, 2009a). Of note for affordability analysis is the inclusion of flood risk assessment products (also known as nonregulatory products) with the maps. Nonregulatory products were an additional feature of the mapping process and FEMA provides a package of nonregulatory products under Risk MAP. Examples of such products include changes since the last FIRM (if digital flood data are available from the previous study), water surface elevation grids and flood depth grids, percent annual chance grids and percent 30-year chance grids (the percent chance that an area will flood over the course of a typical 30-year mortgage), flood risk assessments, and areas of mitigation interest. One of the benefits of these products is that many of these studies include elevation data from the 10 percent, 4 percent, and 2 percent annual chance flood events in addition to the 1 percent (SFHA or 100-year floodplain) and 0.2 percent annual chance flood events (500-year floodplain).

OTHER DATA SOURCES

The questions posed to FEMA will require data for policyholders and the insured properties, as well as uninsured properties and their owners, which cannot be found in the NFIP policy database or derived from the DFIRMs. Particularly important data gaps include the absence of first-floor elevation data for many policies and the absence of any data on uninsured properties. Furthermore, even if all of the data in the NFIP policy database were complete and accurate, the database could not be used to simulate affordability assistance programs that are means tested because the data-

base does not contain income, wealth, or housing cost data. This section discusses other data sources that may be available to FEMA to address these and other data needs.

Decennial Census/American Community Survey Information

The decennial census of population and the continuing American Community Survey (ACS) are sources of information that may help FEMA answer some kinds of policy questions. The census provides complete population counts for the nation and subnational geographic areas down to census tracts (small, relatively stable geographic areas of about 2,500 to 8,000 people), block groups (statistical divisions of census tracts of about 600 to 3,000 people), and individual blocks once every 10 years. The data collected in the census are limited to basic demographic information (age, gender, race/ethnicity, and household relationship) and housing tenure (owner or renter). The data provide the basis for population estimates that are updated each year for states, counties, cities, and towns. These estimates can help FEMA identify population growth in flood-prone areas in a general way.

More helpful to FEMA is likely to be the information provided each year from the ACS, which, beginning in 2005, has collected detailed social and economic characteristics across the nation from a large sample of households. The ACS is conducted monthly, and data products are released every year for small areas down to census tracts and block groups. The content of the ACS questionnaire is roughly the same as what used to be in the once-a-decade decennial census “long-form” sample. There are questions on education, place of birth, citizenship, household relationship, income, employment, housing costs (mortgage/rent, utilities), housing characteristics (number of rooms, number of units in the structure, when the house was built, etc.), and other topics (NRC, 2007, 2015a). See Appendix F for a table of currently available ACS information for census tracts and block groups of potential relevance for FEMA.⁶

All data for census tracts and block groups released each fall are summarized over the preceding 5 calendar years (60 months); there are no 1-year (12-month) products available as there are for larger geographic areas (no data are available at all for blocks from the ACS). The latest census tract and block group data available are for 2009-2013, which covers the Great Recession and some economic recovery. The next round of data

__________________

⁶ Public-use microdata samples (PUMS) from the ACS, which would allow FEMA to specify additional tabulations, are not useful for FEMA’s purposes because no area is identified in the PUMS with fewer than 100,000 people.

for tracts and block groups, for 2010-2014, will be available at the end of 2015.

Although the ACS has a large sample (about 2.3 million responding households each year), even cumulated over 5 years, the ACS sample is only about two-thirds of the census long-form sample. This means that variability due to sampling error is higher than in the long-form sample for small geographic areas, and sampling variability in the long-form sample is itself high for very small areas (see NRC, 2007, Tables 2.7 and 2.8). Moreover, the need to cumulate over 5 years creates challenges when interpreting estimates and, especially, tracking changes over time. The ACS also has different residence rules from the census—the ACS asks people to indicate where they have been living in the past 2 months; the census asks people for their “usual” residence. For areas with seasonal populations, such as beach or lake communities, the ACS estimates may represent the “usual” (year-round) population if the “season” is short, or a combination of year-round and seasonal residents if the “season” is more than 2 months (as is likely to be the case, for example, for “snowbirds” moving between northern states and Florida, Arizona, etc.).

Even with high sampling variability and the issues with 5-year accumulations, the ACS provides a low-cost way for FEMA to overlay characteristics of interest for households in small areas (e.g., home ownership, median rent and mortgage amounts, median house value, median income, and age of housing stock) on maps of current and projected future flood-prone areas. FEMA could also reimburse the Census Bureau to prepare special tabulations by reaggregating the underlying ACS data to conform to geographic areas defined by FEMA to match floodplain boundaries. Such special tabulations could provide a clearer picture of areas in which there may be households at risk of floods and of inability to afford flood insurance premiums. Such areas could not be smaller in population size than block groups, which are the smallest areas currently released from the ACS, but the specially defined areas could have more relevant boundaries for analyzing flood risk and insurance affordability.

Published ACS data represent the aggregate of household characteristics in a block group, census tract, town, township, village, or city (or special tabulation area) and are not at the level of the individual policyholder. Consequently, care must be taken when using ACS small-area data in policy analysis of the likely effects of alternative flood insurance program provisions. For example, median household income could be the same in an area of homogeneous incomes and in an area with both very-high-income and very-low-income households, so that it would not be appropriate to impute the median value to all households without additional information. A more telling indicator of the distribution of income is to examine ratios of household income to the poverty level, such as the percentage of house-

holds with income at or below the poverty level and with income at more than twice the poverty level. Using the poverty level also takes account of the fact that households differ in size and composition and hence in income needs. The ACS currently provides ratios of household income to poverty level for census tracts but not block groups, and might be able to provide them for FEMA-defined flood-prone areas.⁷

Possible uses of ACS data for flood insurance program policy analysis could include the following:

• In small areas containing large numbers of current policyholders, ACS data for those areas could help indicate the likely affordability of premiums for the remaining households by using a combination of what is known about the policyholders (although currently very little information is available on policyholders) and all residents in a model to predict characteristics of interest for nonpolicyholders.
• In small areas that have no or small numbers of current policyholders but are in areas that are likely to face increased flood risk, ACS data could help FEMA identify areas with residents who may be at high risk of not being able to afford flood insurance premiums. FEMA could then decide to invest resources in those areas for additional targeted information from surveys, administrative records, and commercial sources to support an affordability analysis.

Given that many small areas are likely to have a mix of policyholders and nonpolicyholders, the use of ACS data for flood insurance program policy analysis is limited due to this heterogeneity, unless and until FEMA obtains additional data on policyholders for modeling purposes.

Federal Agency Administrative Records

It could be possible for FEMA to make arrangements to obtain information on household income and other characteristics for policyholders and other owners of at-risk properties from another federal agency. Access to federal administrative information would take time to arrange but, once established, could provide an inexpensive timely flow of key information that is regularly updated. For example, adjusted gross income could be

__________________

⁷ For such areas, tables could display, for example, the percentage of families below 100 percent of poverty, the percentage between 100 and 200 percent of poverty, the percentage between 200 and 300 percent of poverty, the percentage between 300 and 400 percent of poverty, and the percentage above 400 percent of poverty. Of course, even in flood-prone areas, probably not all and maybe not even a high percentage of families with incomes that are low relative to the poverty level or other families will face premiums that are high relative to their income.

available from the Internal Revenue Service (IRS) or covered earnings and Social Security benefits from the Social Security Administration (SSA). Such access would require justification of a federal purpose that would accord with IRS or SSA regulations about allowable access, and it would also likely require that FEMA use the data and conduct its policy modeling under the terms of a memorandum of understanding in a secure environment, similar to that of a Federal Statistical Research Data Center (RDC),⁸ to protect the confidentiality of the information.

Commercial Sources

Several commercial enterprises now collect data at the individual property level and perform their own internal analyses to predict home prices using GIS and related statistical modeling software. Some of these companies provide analytical services to the NFIP and also serve as data providers. Making arrangements to obtain information on, say, building elevation from remote sensing technologies and GIS or property values from local property records or realty sites would require a sustained effort, but could have substantial benefits for FEMA. Once experience is gained with such data sources, they could provide an inexpensive way of regularly updating key characteristics for at-risk properties. If these commercial entities have data that FEMA can use in a microsimulation they could sell such data to FEMA, as an alternative to expecting FEMA to gather new data. As one other example, there are private firms that maintain databases on mortgage balances at the individual property level. These data would be used to establish whether the property owner faces a mandatory flood insurance purchase requirement. However, getting access to the data which are proprietary can be expensive or maybe even not possible.

Local Tax Assessment Records and Other Related Sources

Most tax assessor offices maintain information needed for estimating and collecting property taxes. Of potential interest to an affordability study, this includes an assessment of the property’s value, usually an estimate of the improved value (just the structures), as well as the land value. The extent to which such data are well organized, digitized, and easily made available to the public will vary among communities.

__________________

⁸ An RDC is a location where a user (with appropriate clearance) can have access to restricted-use data (e.g., microdata) that would not otherwise be accessible. Analyses would be performed using the center’s computer, and results would still be subject to all of an agency’s disclosure rules. The Census Bureau originally established a network of RDCs around the country; these RDCs now house data from other agencies as well. Available at http://www.census.gov/about/adrm/fsrdc/about/available_data.html (accessed on October 7, 2015).

Municipalities will generally make assessors’ data available (either for free or for a fee), although for most this will require contacting the office and making a formal request. While online tools to look up assessors’ information are increasingly common, most communities do not simply provide their entire database for download just on their website. There is also substantial variation in whether communities maintain historical data on past sales, or only the current assessment. Furthermore, assessment practices can vary and some communities will maintain both the appraised value and the assessed value (when these differ), but others may only provide the assessed value, which may require calculations to convert back to the appraised value.

U.S Army Corps of Engineers

Many coastal and riverine areas of the United States have recently received new FIRMs or have new flood studies planned under the Risk MAP program or both, but not all of these FIRMs will include information of flood likelihoods and stages needed for the estimation of damage claims. There are other sources for securing such information. The U.S. Army Corps of Engineers (USACE) has multiple flood risk assessment tools available from its Hydrologic Engineering Center (HEC) that are used by engineers worldwide. HEC-HMS (Hydrologic Modeling System), which is a computer modeling software package, computes design flood hydrographs for specified return periods, such as 10-year or 100-year floods. HEC-RAS (River Analysis System) takes the highest discharge from the design flood hydrograph and calculates the corresponding flood water surface elevation above geodetic datum, using a map of land surface terrain and channel morphology often derived from light detection and ranging (LiDAR) data.

The USACE also has made available depth-damage curves for estimating property damage from flood events. Such curves give estimates of damages to a structure or its contents as a percent of its value based on the depth of water at the site. These are used for USACE flood damage reduction studies, but are publicly available for other uses as well. The most up-to-date curves available are generic, nationwide functions for residential structures with basements based on damage estimates from major flood events in the United States between 1996 and 2001.⁹

Hazus

Hazus-MH is a national, GIS-based software model developed for FEMA by the National Institute of Building Sciences. The objective of

__________________

⁹ See http://planning.usace.army.mil/toolbox/library/EGMs/egm04-01.pdf.

Hazus is to provide an analytical platform for estimating the effects of natural disasters in a standardized way across the nation for use by public officials in planning and evaluating mitigation measures. It is designed to estimate damages from multiple types of hazards. In the flood model, Hazus couples a flood hazard analysis, which estimates the depth of flooding in the user-defined study area, with an estimate of economic damages from the flood. Hazus has also been used in academic studies (e.g., Dierauer et al., 2012; Kousky and Walls, 2014).

Hazus includes likelihoods of different flood depths for use in estimating flood hazard as well as depth-damage curves that can be used to estimate damages for different kinds of structures. The hazard component of Hazus can be used to fill in hazard information that may be missing from FIRMs for particular locations. Hazus uses depth-damage curves to relate depth of flooding to building and contents damages for each representative property. Hazus has many such curves in its library, which varies by property type (e.g., single-family residential, mobile home, and light industrial). Damage curves may also vary by characteristics of the structure. For example, for single-family properties these are year built, number of stories, and type of basement. These curves can be applied to individual structures. Hazus does not, however, have data on individual structures in its database. Instead, the properties in Hazus are what is reported from the U.S. Census of Population and Housing at the census block level, while nonresidential data come from Dun & Bradstret.¹⁰ Hazus assumes the building stock is evenly distributed throughout a census block. This makes the database ill suited for microsimulaton. Therefore, only the hazard and damage procedures can be used in microsimulation modeling.

LiDAR

Building elevation data are often missing from the NFIP policy database and information available on building characteristics for properties that do not have an NFIP policy can only be acquired from other sources. LiDAR, which is a remote sensing detection system using light from a laser to measure distance, can be used to obtain ground elevations and other property data. One application of this technology uses lasers mounted on a fixed-wing airplane along with other instruments to determine the elevation of the earth surface. LiDAR has become the industry standard for obtaining accurate ground elevations efficiently. For example, in the early 2000s North Carolina acquired statewide, high-resolution LiDAR-derived topography and imagery. Table 3-1 shows the different accuracy levels and the monetary cost depending on how much LiDAR is obtained. In general,

__________________

¹⁰ Dun & Bradstret is an American public company.

TABLE 3-1 Cost of LiDAR for Various Levels of Accuracy

NOTE: USGS, U.S. Geological Survey.
SOURCE: http://nationalmap.gov/3DEP/neea.html (accessed on October 7, 2015).

the minimum LiDAR order is 500 square miles, but significant savings can result if acquisition areas are greater than 5,000 square miles. For flood insurance study–related information, QL2 is the quality standard that is typically used.

LiDAR-based structure elevations can be used along with the BFE to determine the risk the structure has in relation to flooding. LiDAR technology has become one of the basic building blocks to determine ground surface elevations.

FILLING DATA GAPS

Gaps in the NFIP policy database and from DFIRMs are (a) elevation data for policies that are not elevation rated, needed to estimate future premiums, claims, and NFIP revenues; and (b) policyholder socioeconomic characteristics, needed for establishing cost burden and simulating eligibility, benefits, and costs of means tested assistance programs. There are no data—including the data in (a) and (b)—for properties, property owners, and occupants that are located in at-risk areas but are not covered by NFIP policies. These data may be needed for evaluation of policy options that might expand takeup.

These, as well as other less significant data gaps, might be filled using some of the other data sources described above. However, if other data sources are not sufficient then the approaches described in this section may

be needed. It is likely that a combination of the strategies will be needed not only for constructing an initial database, but also for creating future versions of the database.

Proxy Variables

A proxy means using one measure to stand in for another, when one measure is readily available and the other is not. For example, ground elevation at a property might be known from a DFIRM. A local tax assessment database, or a LiDAR report, might include the style of the building or the presence of a basement for that structure. This information about the property might allow for adding height to ground elevation and using that as a proxy for the elevation of the first habitable floor.

As another example, BW 2012 called for an affordability study to focus on “individuals who cannot afford” to pay NFIP risk-based rates. Policy makers may prefer to use household income as the measure of ability to pay. However, such a measure for the policyholder may not be available, but data on assessed property value may be.

Although a proxy variable might be used when the originally intended variable is very difficult and expensive to obtain, the two variables might be weakly correlated, suggesting that the proxy variable is measuring something substantially different from what had been intended. Analysts will be expected to explain the reasons for the use of proxy variables and any cautions about how results of the analysis should be interpreted by policymakers.

Sample Survey

The FEMA website states that the agency only collects the minimum amount of information necessary to administer the NFIP. Compliance with the requirements of BW 2012 and HFIAA 2014 and the need to evaluate NFIP policy option alternatives over time provide a reason for FEMA to collect information beyond what is currently collected.

FEMA could commission a spatial sample of homeowner/at-risk properties selected from the NFIP policy database. Particular data needs are elevation data for non-elevation-rated properties (may have to pay for the elevation certificate) and homeowner and occupant characteristics. For those selected, a survey might be administered to obtain such data. Alternatively, for those selected the needed data might be obtained through changes to the insurance application form (e.g., a supplementary form). In all cases, the survey results would be confidential and would not be entered into the NFIP database. Of course, this only will get information from current policyholders. To obtain data for properties that are not covered,

an additional spatial sample might be drawn for addresses in the nation’s floodplains, with data collected by surveying the owners/residents at the selected addresses.¹¹

A survey has the advantage that it can be professionally designed to obtain the desired information on a consistent basis. A drawback of a survey is its monetary cost. Depending on the interviewing mode (and personal interviews could be desirable for a FEMA survey because of the ability to capture information by observation) and the extent of follow-up needed to bring response rates up to acceptable levels, the cost of a completed survey case could be at least $200 to $300 per interview. Survey response rates have been falling for several decades in the United States and other countries; indeed, public opinion polls conducted by telephone may typically only obtain a 10 percent response. U.S. Office of Management and Budget guidelines require federal surveys to plan sufficient nonresponse follow-up to obtain an 80 percent response rate or, if this rate is viewed as unattainable, to plan for a study of a sample of nonrespondents to permit estimation of any nonresponse biases and their effects on key estimates. In the case of a FEMA survey, the information on current NFIP policies could help in modeling nonresponse adjustments for that portion of the sample, but it could not help for the portion comprising at-risk properties without flood insurance coverage.

How many completed survey cases are needed for a FEMA survey will be a function of the extent of disaggregation of microsimulation model results that is desired (greater disaggregation requires a larger sample to attain adequate precision of estimates) and the budget the agency can allocate to the effort. Arguably, FEMA can justify a one-time investment in a sample survey—even an expensive one—as providing the most accurate basis for an NFIP policy options microsimulation model. But the need for policy modeling is continuing and would not be used only one time. Flood risk will change for currently covered properties, other currently at-risk properties, and properties that at present are minimal risk. Finally, even with a large well-executed survey, there will be missing items for some

__________________

¹¹ Spatial sampling involves selecting a limited number of locations (a sample) in geographic space for faithfully measuring phenomena that are subject to “dependency” and “heterogeneity.” Dependency refers to the phenomenon that observations at neighboring locations are more similar to one another than are observations at locations farther apart. Dependency suggests that a value at one location can predict the value at another location. Spatial heterogeneity refers to attributes of geographical variation. Spatial heterogeneity suggests that dependencies can change across space (also referred to as “nonstationary”) and, therefore, it may be unwise to trust an observed degree of dependency beyond a region that may be small. Spatial sampling techniques are more efficient than conventional sampling when surveying spatially distributed targets, where spatial autocorrelation and heterogeneity are prevalent (Banerjee et al., 2004).

properties. Surveys typically impute such missing values from the information provided by other respondents.

SUMMARY

A task of the committee was to discuss data needs and data gaps—the completeness and quality of data needed for policy analysis. Data needed to evaluate alternative options include data about flood insurance policies, property characteristics, and property owner/resident socioeconomic characteristics. At present, FEMA has ready access to two internal databases: the NFIP policy database and FIRMs. To evaluate affordability policy options, additional data on variables not presently available or that might replace or supplement the FEMA data are needed.

A particularly important gap in the data for many policies is the absence of first-floor elevation data that are necessary for estimating the damage to the structure from floods of different magnitudes. Although some of those data are now being collected for properties inside the SFHA, such data are not available and are not being collected for properties outside the SFHA. Also, even if all of the data in the NFIP policy database were complete and accurate, the database cannot be used to simulate affordability assistance programs that are means tested because the database does not contain income, wealth, or housing cost data. Furthermore, the NFIP database does not contain information for nonpolicyholders located in flood-prone areas and cannot be used to analyze whether an alternative policy option that would reduce premiums or provide assistance might promote takeup among such households. These and other data gaps need to be filled, and the report discusses approaches to filling those data gaps.