Benefits and costs of natural infrastructure: Evidence from the New Jersey coast

Abstract

This paper empirically estimates the economic impacts of a large-scale public investment in natural infrastructure aimed at adapting to climate change and increasing coastal resilience. I utilize temporal and spatial variation in investment in dunes to provide a hedonic property value estimate of the economic benefits. I identify the net effect of treatment utilizing the doubly robust Oaxaca-Blinder estimator and show that coastal housing price increases attributable to constructed dunes are approximately 3.6 percent. A decomposition of the average impact suggests that the policy intervention generates ancillary costs related to impaired ocean views and privacy concerns that partially offset large protection benefits.

Full text

1
Benefits and Ancillary Costs of Natural Infrastructure:
Evidence from the New Jersey Coast
Steven J. Dundas
1
Article Forthcoming in Journal of Environmental Economics and Management
DOI: 10.1016/j.jeem.2017.04.008 (accepted 25 April 2017)
Abstract: This paper empirically estimates the economic impacts of a large-scale public
investment in natural infrastructure aimed at adapting to climate change and increasing coastal
resilience. I utilize temporal and spatial variation in investment in dunes to provide a hedonic
property value estimate of the economic benefits. I identify the net effect of treatment utilizing the
doubly robust Oaxaca-Blinder estimator and show that coastal housing price increases attributable
to constructed dunes are approximately 3.6 percent. A decomposition of the average impact
suggests that the policy intervention generates ancillary costs related to impaired ocean views and
privacy concerns that partially offset large protection benefits.
Keywords: Dunes; Natural infrastructure; Climate adaptation; Oaxaca-Blinder; Hedonics; Policy
evaluation
JEL codes: H54, Q51, Q54, Q58
1
Assistant Professor, Department of Applied Economics and Coastal Oregon Marine Experiment Station, Oregon State University. Email:
steven.dundas@oregonstate.edu. I would like to thank Nicolai Kuminoff, David Lewis, Kerry Smith, Marty Smith, Laura Taylor, and Roger von
Haefen for their valuable comments on earlier drafts of this paper. Additional thanks to Kelly Bishop, Zack Brown, Michael Hanemann, Alvin
Murphy, Walter Thurman, and participants at the 2015 AERE Sessions of the AEA Meetings, the 2014 NBER Summer Institute (EEE) and 2014
Camp Resources and in seminars at Arizona State University, Louisiana State University, Oregon State University, the University of Delaware,
North Carolina State University, and the College of New Jersey for their comments. Special thanks to Rachel Albritton for assistance in the
development of a GIS viewshed tool. I am grateful for funding that supported the development of this research from the Jenkins Fellowship at
North Carolina State University, the Center for Environmental Economics and Sustainability Policy at Arizona State University, and NOAA
National Centers for Coastal Ocean Science/Center for Sponsored Coastal Ocean Science through a NOAA Cooperative Institutes Program award
NA11OAR4320091A to the Cooperative Institute for Marine Resources Studies at Oregon State University. I also thank the editor (Jay
Shimshack) and two anonymous reviewers for their helpful comments and suggestions.

2
1. Introduction
Landfall of Superstorm Sandy in October 2012 produced unprecedented damage along the U.S.
Eastern seaboard with losses totaling approximately $70 billion dollars. The densely developed
coastlines of New York and New Jersey were among the hardest hit areas. In particular, Long
Beach Island (LBI), NJ experienced storm surges upward of 8 feet above mean sea level. In
sections of this 18-mile long barrier island, the surge of water ripped homes off of their
foundations and created sand deposits four feet thick up to three blocks from ocean (Barone et al.
2014). Despite the enormous destructive force of the storm, three communities on the island
(Surf City, Harvey Cedars, and Brant Beach) were spared from the worst of Sandy’s damaging
waves and high water levels. To illustrate, consider post-Sandy disaster expenditures by the
Federal Emergency Management Agency (FEMA) and home damage estimates in two similar
communities on LBI. Ship Bottom received aid totaling approximately $69 million and 46
percent of the homes in the borough were damaged. In contrast, FEMA’s liabilities in the
borough of Harvey Cedars, less than five miles north of Ship Bottom, were about $11 million
and only 6.2 percent of homes sustained damage.
2
Why was there such a striking difference? In
the decade prior to Sandy, the beach and dune system in Ship Bottom had eroded severely and
did little to buffer the force of the storm surge while Harvey Cedars received a 22-foot tall
engineered dune system through a federal shoreline protection project. Similar parallels can be
drawn for Surf City and Brant Beach compared to their neighbors without federally constructed
dunes.
Although the science showing a causal link between storms like Sandy and anthropogenic
climate change is still evolving (e.g. Trenberth, et al. 2015), the magnitude of the damages from
2
For an accounting of Sandy’s damages and FEMA expenditure by community on Long Beach Island, see Table C.1 in the online appendix.

3
Sandy shifted discussion of climate change adaptation to the forefront of coastal public policy. In
Sandy’s wake, the federal government allocated approximately $4.5 billion to the U.S. Army
Corps of Engineers (USACE) to continue and expand dune construction efforts on vulnerable
Mid-Atlantic beaches (Disaster Relief Appropriations Act of 2013, P.L. 113-2). Reconstructing
nature-based dune systems represents a clear shift in federal coastal policy away from more
traditional grey infrastructure (e.g. jetties, seawalls) to stabilize coastlines. These “natural”
infrastructure investments are costly to initially construct, averaging $1 million to $10 million
per mile, and require repeated replenishment to replace eroding sands.
Given the proposed scale of federal involvement and the increasing suite of threats from
climate change in vulnerable coastal communities, it is crucial to quantify the economic benefits
associated with these costly ex-ante natural infrastructure investments.
3
This paper provides
suggestive empirical evidence on the magnitude of these benefits by identifying the capitalized
value of existing constructed dunes in the LBI housing market prior to landfall of Superstorm
Sandy. To my knowledge, the benefits estimated here represent the first empirical values for an
ex-ante public policy aimed at adaptation to climate change. This work adds to a growing
literature on climate change adaptation where previous work largely focused on ex-post private
adaptation in agriculture (e.g. Kelly et al. 2005), forestry (e.g. Guo and Costello 2013), and fuel
choice (e.g Mansur et al. 2008) among others.
4
A key difference with public adaptation is the
scale of the change, which may lead to ancillary effects beyond the intended scope of the
intervention. Here, the ancillary impacts of constructed dunes are found to be negative, partially
3
Recent research by Davlasheridze et al. (2017) suggests ex-ante spending aimed at hurricane risk reduction by FEMA reduces property damages
nearly twice that of ex-post spending on similar investments.
4
Other related ex post studies have focused on estimating the value of risk information provided by storm events, such as Hurricane Andrew in
Florida (Hallstrom and Smith 2005), and Hurricane Floyd in North Carolina (Bin and Polasky 2004).

4
offsetting benefits associated with the primary objective of the policy (i.e. storm protection) and
potentially changing the sign of a benefit-cost analysis.
The empirical challenges to valuing these effects are two-fold. First, there are many
unobserved confounders correlated with storm protection in coastal environments.
5
To address
this concern, I assemble a comprehensive dataset that exploits the spatial and temporal
discontinuities in constructed dune systems to identify the likely impact of the policy
intervention. The characteristics of the LBI housing market and the sequence of events leading to
the policy interventions prior to Sandy are also ideally suited to minimize potential confounds
related to correlated unobservables, sample selection, and multi-scale capitalization. Second,
dune construction represents a substantive change to the local landscape and may impact
ancillary service flows. On LBI, constructed dunes and widened beaches provide storm
protection but may also obstruct ocean views from waterfront homes and increase visitor traffic
on local beaches. These potential impacts jointly determine the overall effect of a policy
intervention as the behavior of economic agents can either enhance or offset the policy’s goal of
providing protection services. To investigate this further, I utilize high resolution spatial data on
dune quality and location, viewshed potential, and beach width to decompose the policy effect
into three service flows affected by the dune.
I estimate the casual effect of this public investment in natural infrastructure on housing
markets by adapting the Oaxaca-Blinder estimator (Oaxaca 1973; Blinder 1973) to a quasi-
experimental hedonic model. This estimation strategy is useful as it allows for flexibility in
estimating treatment effects due to its double robustness properties, utilizing the relative
5
For example, Bin and Kruse (2006) estimate that homes in flood zones subject to wave action have sales prices that are 27 percent higher as
compared to homes not located in a federally designated floodplain, demonstrating the conflation of amenities and risk.

5
strengths of both of regression and matching techniques (Kline 2011; Sloczynski 2015).
Practically speaking, weights are used to generate exact covariate balance between treated and
control groups and produce a consistent, unbiased counterfactual to identify the net effect of
dune construction on housing prices. My results suggest that federal dune construction policy
increases housing prices from 3.0 to 6.3 percent, with my preferred Oaxaca-Blinder estimate
suggesting a 3.6 percent capitalization. This market response is precisely estimated and
economically important, suggesting that federal dune construction effectively transfers an
average of $3,229 per year to owners of protected beachfront properties.
6
The magnitude and
significance of this result is supported by a number of robustness checks. While back-of-the-
envelope benefits of the dune policy appear to be sizable ($170 million), the gains are less than
the engineering and maintenance costs of dune construction activities ($261 million) in this
location.
Given the potential for ancillary impacts of dune construction and the policy implication that
dunes may be a net cost, I estimate additional models to investigate the spatial extent of the
capitalization and the values associated with specific service flows impacted by dune
construction. The spatial models indicate the capitalization effect extends to the second block of
homes and has a non-monotonic pattern (i.e., ocean block homes gain higher benefits than
oceanfront homes). Decomposition of the effect of the dune into amenity streams related to
protection, ocean views, and visitor traffic /privacy suggest the protection, or climate adaptation,
benefits are quite large, but are partially offset due to reduction in ocean views and privacy
concerns associated with the policy implementation. In other words, if the dunes provided only
6
The annualization factor used in all calculations in this paper is: AF = r/[1-(1+r)-n] . The discount rate (r) is assumed to be 5% and average tenure
of single family homes (n) is assumed to be 15 years (2011 American Housing Survey).

6
protection services as intended and did not impact other ancillary service flows, a benefit-cost
test would likely be favorable for the policy. Ancillary costs aside, the sheer magnitude of the
devastation from Sandy provided ex-post information on the protection benefits of the dunes.
Federal liabilities from flood insurance payments and disaster aid were reduced substantially in
the three protected towns on LBI. Given that Sandy has been estimated to be a 700-year storm
event (Hall and Sobel 2013), it raises questions about incorporating these ex-post benefits into a
traditional ex-ante benefit-cost analysis and whether society should be insuring against fat tail
events (e.g., 100 year storm events) when allocating resources to climate adaptation.
In addition to valuing coastal dunes in the context of a public adaptation to climate change
and adapting the Oaxaca-Blinder estimator to hedonic modeling, this paper also provides some
further contributions. In terms of benefit transfer, the estimated hedonic model and methods here
provide a framework to predict the effect on housing prices of different mixes of storm
protection benefits and ancillary impacts in different locales. While I identify the ancillary
effects after the policy occurs, future research should aim to integrate them into ex ante analyses
of proposed policies. More broadly, this work also demonstrates the potential significance of
disentangling the direct and ancillary effects of public policies aimed at climate adaptation to
identify who wins, who loses, and by how much. My results suggest that such adaptation may
generate economically significant ancillary benefits and costs that can significantly impact
implementation of public policy investments subject to federal benefit-cost analysis
requirements.
The remainder of the article proceeds as follows. In section 2, I discuss challenges with
recovering policy effects with hedonic models and implications of a policy impacting multiple
amenity flows. Section 3 describes my data and section 4 outlines a brief policy background and

7
the research design. Section 5 discusses results, robustness checks, and policy implications.
Section 6 explores spatial extent of capitalization and the decomposition of the policy effect.
Section 7 concludes.
2. Recovering Policy Effects with Hedonic Models
Credible estimation of benefits associated with changes in environmental amenities continues to
be an empirical challenge. Fortunately, the hedonic property value model paired with sound
research designs has emerged as a powerful tool to estimate benefits. For example, recent research
has utilized hedonic methods to estimate the distributional effects of the 1990 Clean Air Act
Amendments (Bento et al. 2015), the heterogeneous effects of shale gas development on different
housing types (Muehlenbachs et al. 2015), expectations of shale gas development (Boslett et al.
2016), the property value implications of environmental health risks from polluting firms (Currie
et al. 2015) and the economic impacts of siting new power plants (Davis 2011).
There are a number of challenges associated with identifying and interpreting hedonic
estimates of environmental amenities. While the challenges discussed below do not constitute an
exhaustive list, they highlight primary concerns with using the hedonic framework to value coastal
dunes. First, the extent of the market needs to be defined appropriately due to potential for
imprecise estimates from a restrictive definition or biased estimates from too broad of a definition
(Michaels and Smith 1990). Here, the market is defined as all housing transaction on a single
barrier island, considered by many realtors as its own market due to the island’s physical location
and character.
7
Second, the spatial scale of the impacted amenity should be delineated
7
The barrier islands to the north and south of LBI are state and federally protected land, leaving the densely developed island a center for real
estate along the central coast of New Jersey. The island is self-contained with only a single access point by causeway.

8
appropriately to avoid introducing bias into model estimates (Abbott and Klaiber 2011). In this
work, the effect of the dune is delineated with fixed effects in two spatial dimensions, capturing
the scale of policy implementation by neighborhood and the policy proximity effects within
neighborhoods. The extent of potential bias resulting from these choices is explored with various
robustness checks on my preferred model specification and is discussed in detail in Section 5.
Next, endogeneity caused by unobserved, spatially delineated variables correlated with the
non-market good of interest has the potential to confound estimates. Here, the dune system was
only completed in three sections of LBI due to conflict over partial property easements needed
for dune construction. Thus, I have a repeated cross-section of housing transactions of similar
beach homes – some protected by a new constructed dune system and many that were not. I can
utilize this spatial variation in policy intervention within a single market of similar homes to then
identify the average treatment effect of the dune purged of biases associated with endogeneity
and omitted variables that plague traditional hedonic approaches. Another potential concern is
that sorting may bias estimates due to the implied selection of desirable locations over
undesirable based on unobservables. To alleviate this concern, I use a simple check to determine
if the policy had an effect on the likelihood of a transaction following Muehlenbachs et al.
(2015). I regress log annual transactions in each of the 12 neighborhoods on LBI on a policy
indicator variable along with neighborhood and year fixed effects. The effect of the policy is
small and statistically insignificant, lessening sample selection concerns.
8
Another concern of particular importance is interpreting coefficient estimates as marginal
willingness to pay (MWTP). This interpretation relies on the standard assumption that the hedonic
price function does not shift over time. The issue centers on whether or not the policy being
8
The coefficient estimate on the policy variable is -0.026 with a t-stat of -0.2.

9
evaluated generates a change in the variable of interest that alters the structure of the hedonic
equilibrium. For relatively small changes, it is plausible that the gradient does not change (i.e.
Palmquist 1992). If the change is large however, the hedonic price function may shift to clear the
market and the slope coefficients may represent a capitalization effect, not MWTP. Kuminoff and
Pope (2014) demonstrate that preferences must remain unchanged over time and the supply and
demand curves cannot be altered by the change in the non-market good in order for capitalization
effects to be interpreted as a welfare measure. Yet, recent work by Banzhaf (2015) suggests that
coefficient estimates from a quasi-experimental hedonic approach may be interpreted as lower
bound of Hicksian equivalent surplus even for non-marginal changes. The research here identifies
a capitalization effect and I provide a discussion welfare interpretation with the presentation of the
results in Section 5.
In addition to the policy effect, this work is also interested in understanding the individual
amenity flows impacted by dune construction activities. The primary purpose of this exercise is to
inform policy design of large federal projects. As noted in the introduction, a simple benefit-cost
test calculated from results here suggests the dune policy generates benefits that do not justify the
policy’s costs. Intuition and previous research indicates storm protection, the primary purpose for
this policy, is likely a valuable amenity. Decomposition of the policy effect could then identify
amenity costs that partially offset the expected protection benefits. On LBI, dunes and a wider
beach provide storm protection but the height of the dune also compromises ocean views in homes
in close proximity. The additional beach width increases recreation opportunities for homeowners
but also may increase tourist activities, leaving these same homeowners concerned about visitor
traffic and privacy. Therefore, I view the effect of the construction of the dunes (D) on property
values (p) as arising through three channels – storm protection (s), ocean views (v), and beach

10
width (b) – with H and L representing vectors of housing and locational attributes influencing
housing prices that are unaffected by the dune:
     
, , , ,p f s D v D b D

HL
(1)
dp f ds f dv f db
dD s dD v dD b dD

  

(2)
Estimation of the effect of constructed dunes captures the summation of these three impacts, but
does not offer insight into their relative magnitudes or even sign. The interplay of these impacts
determines the overall effect of a policy intervention as the behavior of economic agents can
either enhance or offset the policy’s goal of providing storm protection services. It should be
noted that there are other potential ancillary effects associated with dune construction (e.g. non-
use values for native shoreline habitat) that are not addressed directly in this research.
An additional contribution related to decomposition of the policy effect warrants discussion here.
It relies on the interpretation of storm protection benefits from constructed dunes as a value for
climate adaptation. In the adaptation literature, Mendelsohn (2000) makes the distinction between
types of adaptation: private versus joint (i.e. public) and anticipatory (ex ante) versus reactive (ex
post). Guo and Costello (2013) provide the additional distinction between small adaptive changes
in continuous choices (intensive margin) versus large, discrete changes with investment in new
capital stock (extensive margin). To that end, federal dune policy anticipates storm events and sea
level rise related to climate change and protects coastlines with large geoengineering projects to
minimize future damages. In other words, the constructed dunes can be considered an ex ante
public adaptation to climate change along the extensive margin. Prior research on climate adaptation
has largely focused on ex post changes in private behavior in the context of agriculture (e.g. Kelly et

11
al. 2005), forestry (e.g. Guo and Costello 2013) and fuel choice (e.g. Mansur et al. 2008). In contrast,
public policies, such as coastline stabilization, will inevitably involve coordinated national
strategies that target vulnerable regions and sectors of the economy.
A key distinction between the model of private adaptation presented in Guo and Costello (2013)
and a model of public adaptation is to define the value of adaptation in terms of a public policy
outcome with ancillary impacts:
Value of Policy = Value of Adaptation +/- Value of Ancillary Impacts (3)
The last term in (3) is a potential consequence of a large, discrete change that may enhance or
reduce the value of a given policy. An analytical example illustrating this concern is provided in
the online appendix. In the following sections, I detail the data and research design leading to an
estimate of the value of the dune policy, or the left-hand side of (3). Then, I explore the right-hand
side of equation (3) by estimating additional models interpreting storm protection benefits as the
value of adaptation and ocean views and visitor access/privacy issues as the ancillary impacts.
3. Data
The empirical analysis to identify the impacts of constructed dunes focuses on LBI, an 18-mile
long barrier island in eastern Ocean County, NJ (Figure 1). LBI is bordered by the Atlantic Ocean
to the east, Manahawkin Bay to the west, and the only vehicular access is via the State Route 72
bridge over the bay. Approximately 20,000 people reside on LBI year round, but with the island’s
close proximity to Atlantic City (25 miles), Philadelphia (55 miles), and New York City (75 miles),
the summer population often swells to over 100,000 people.
9
9
This arises from people using their homes as summer residences only or renting their homes to vacationers. While the rental market is substantial
on the island, it is highly decentralized and records on rental transactions are not maintained by any realty agencies that operate on the island.
Therefore, this analysis focuses on sales transactions only.

12
Housing sales for LBI were compiled utilizing deed records from the Ocean County Tax
Administrator for all recordable transactions from January 1, 2000 until October 28, 2012. The
spatial extent of the market is restricted slightly by the USACE project scope. The northernmost
municipality, Barnegat Light, is not part of the USACE project due to an existing dune system that
resulted from a previous grey infrastructure investment. The timing of the market brackets the
initial release of information about the USACE’s intention to build dunes and the landfall of Sandy.
The storm made landfall October 29th, 2012 and I assume that transactions with a closing date on
or before October 28th, 2012 are not confounded by information related to the storm.
The universe of residential transactions during this time period contains 8,978 sales records with
data on sales price, specific address with block and lot, and age of the structure. Additional data
on housing characteristics (number of bedrooms, number of bathrooms, square footage, lot size,
and indicators for the presence of a garage, hot tub, and fireplace) were obtained from the offices
of both county and municipal tax assessors.
10
Initially, 1,481 transactions are removed due to a
lack of necessary housing characteristics or unrealistic values for certain variables (e.g., bedrooms
= 0). Transactions not deemed arms-length (i.e. $1 sales price) are also removed from the data set,
reducing the number of potentially viable transactions to 5,263. Lastly, the coastal location of the
study area and the fundamentals of the real estate market during this time frame led to a small
number of knockdowns, where developers were razing older beach cottages and replacing them
with larger, more luxurious homes. Suspect transactions (351) were identified by observing the
same property being sold multiple times in a single year or with quick re-sale (i.e. within 12 months
of previous sale) at a substantial higher price. The resulting data set contains 4,912 residential,
arms-length sales that are suitable for the empirical analysis.
10
The hot tub data allow control for the “Jacuzzi effect” discussed in Kahn and Walsh (2014).

13
Table 1 provides summary statistics for key transaction and housing characteristic variables for
the full sample in the first two columns. The high sales mean of $942,344 (2012 dollars) is a
function of the NJ real estate market, the desirable coastal location, and the prevailing market
conditions during the study time frame. The average home has approximately 4 bedrooms and 3
bathrooms with interior living space totaling 1,746 square feet and is about 36 years old. The
average lot size is relatively small (~ 0.13 acres), reflective of the density of development on the
island. Key distance and location variables are calculated using publicly available GIS data from
multiple sources, including the NJDEP, the NJ Geographic Information Network, and the US
Census Bureau. The impact of insurance is controlled for using a flood zone fixed effect as
determined by the National Flood Insurance Program maps.
11
All homes on the island lie in one
of five potential flood zones, with 3 (VE, AO, and AE) located in the Special Flood Hazard Area.
12
Housing data are linked to a geo-coded parcel map obtained from the Ocean County
Department of Planning in order to help define the spatial extent of the capitalization from the
dunes. Identification of the exact spatial location of each transaction allows for the inclusion of
spatial fixed effects that define the scale of capitalization and aid in reducing the potential
confounds of omitted variables. Proximity to the beach is utilized as a spatial proximity fixed
effect under the assumption that homes in each proximity band relative to the dune are impacted
in a similar manner by the policy. Given the geography and spatial pattern of development (i.e.,
north-south roadways intersecting the island at equal intervals parallel to the beach), this is a
reasonable assumption for this location. Each parcel is classified into one of 5 bands: oceanfront,
11
The flood maps are currently undergoing a revision so the existing maps have an effective date of September 29, 2006 - before any dunes were
constructed. The three municipalities receiving dunes did not file a collective letter of map revision with FEMA to alter the 2006 flood maps to
include the flood protection benefits of the dunes.
12
Flood zone designation is the primary driver of cost of premiums in the NFIP program with the median premiums in Ocean County, NJ for
single-family homes are $3,144 for V zones, $806 for A zones, and $376 for X zones (Kousky and Kunreuther 2013).

14
ocean block, second, and third block from the ocean, and bayfront. Note that ocean block homes
are not oceanfront, but immediately adjacent to the oceanfront and residents do not have to cross
a major roadway to access the beach. A neighborhood fixed effect represents the scale at which
the policy intervention occurs. Four of the five towns on LBI constitute individual
neighborhoods while the large, discontinuous municipality, Long Beach Township, is divided
into eight distinct and locally recognizable neighborhoods.
4. Policy Setting and Empirical Strategy
This section begins by providing background for the policy setting in New Jersey that lends support
to assumptions used to justify the research design. I then describe a simple difference-in-
differences analysis to show the plausibility of my identifying assumptions. Lastly, I provide the
intuition and econometric foundation for using Oaxaca-Blinder as my preferred estimator to
identify the net effect of federal dune policy on coastal housing markets.
4.1 Policy Setting
In 1999, the USACE first publicly announced plans for the construction of dunes and beach
renourishment along the oceanfront of a majority of LBI (USACE 1999). At the time, the natural
dunes along the island’s six municipalities were virtually nonexistent with their growth impeded
by dense residential development (Barone et al. 2009). This development and the property rights
system in New Jersey resulted in the need for the state’s Department of Environmental Protection
(NJDEP) to obtain voluntary partial property easements from all oceanfront property owners for
proper dune construction to begin. In every municipality, many of these property owners
vehemently opposed the perpetual easements and the project could not begin as planned.

15
Property owners’ primary concern is their perception that dunes diminish property value
resulting from ancillary costs such as lost ocean views, loss of use of their property, reduced
privacy, and concerns regarding the perpetual easements.
13
This resistance led to overwhelming
public opinion that dune projects would not commence, even after a project cooperation agreement
authorizing federal money for the project was signed in 2005 (Smothers 2006; Urgo 2006).
14
For
instance, an April 2006 New York Times article stated, “Work was scheduled to start this month
but without all the easement agreements signed, that is unlikely. The delay places the federal
money in peril; it will be taken back if none is spent by…Sept. 30. The state allocation…would
disappear as well.” (Smothers 2006, pp B2). In other words, no community self-selected into
receiving the policy treatment and the policy was not expected to be implemented.
Under the threat of funding loss, the NJDEP and USACE initiated construction on the first dune
in 2006 in Surf City despite not having all necessary signed easements from property owners.
15
The NJDEP justified the move by seeking a preliminary injunction against the holdouts, claiming
the properties were being maintained in an unsafe manner and inaction on the easements was
equivalent to failing to abate a nuisance related to severe erosion (Milgram v. Ginaldi 2008).
16
The
threat of the injunction was the impetus that allowed dune construction to commence. However,
the outcome of this case ultimately supported the property owners, enforcing the notion that these
13
Opposition is not limited to Long Beach Island. In 2001, the State of New York withheld permits for the USACE, effectively terminating the
Fire Island dune project citing property owner opposition, among other concerns (Rather 2001). On Absecon Island, NJ, home of Atlantic City,
the anti-dune group D.U.N.E managed to get a referendum on the dune project in front of voters, ultimately delaying the process for a number of
years.
14
Figure C.1 in the online appendix provides graphical evidence supporting this belief, showing that the housing market did not react to the
agreement.
15
Why Surf City was chosen as the first location by the state is not clear. Author communication with NJDEP and USACE officials did not reveal
a specific reason and I did not find any observable characteristic about the municipality or homes that would drive this decision.
16
The legal conflict centers on two well-established doctrines of property rights in NJ. First, the public trust doctrine maintains access to waterways
and shorelines for the general public and allows eminent domain takings and prescriptive easements on private coastal property if deemed in the
public interest. Second, waterfront property owners maintain vested property rights to views, access, and ocean breezes and have a right to challenge
any government project that would infringe on those rights. The inherent tension between these property rights creates legal challenges to
implementing the dune policy where holdouts refuse to sign the voluntary easements.

16
takings must follow eminent domain procedures when property owners decline to voluntarily sign
the easements. The second dune in Harvey Cedars was constructed in 2009-2010 after the mayor
decided to use eminent domain against six holdouts following the Milgram ruling.
17
Lastly, the
third dune was constructed in Brant Beach in 2012 after the USACE and township officials agreed
to simply alter the engineering plans to avoid construction activities on the property of the
remaining holdouts, thus eliminating the barriers to starting construction. A comprehensive
timeline of events surrounding the implementation of the policy is provided in online appendix
(Figure C.2).
The legal and political economy here led to a situation where these three communities received
constructed dunes and neighboring communities on the same barrier island did not. The
communities receiving the policy did not self-select into treatment as the required set of easements
from oceanfront property owners was not finalized. A potential concern that the municipalities
with lower levels of protection from the natural dunes had more incentive to seek these political
and legal solutions can be quickly dismissed by an examination of pre-policy dune size in each
town. Communities receiving the dunes had a pre-policy average cross-sectional sand area (ft2) in
the frontal dune of 34.5 and all other communities averaged 35.6. These values represent 6.4 and
6.6 percent of the protection necessary for a major storm event dictated by FEMA and demonstrate
that the entire island was highly susceptible to dune failure. Figure 2 provides an illustration of
this measurement. The evidence is also inadequate for an argument that income levels in each
community may drive selection. The median household income for the island is $76,212 and the
17
One resident sued Harvey Cedars seeking more than the $300 compensation offered for the taking. The NJ Superior Court originally ruled that
the dunes were a public good that provided general benefits to all Harvey Cedars residents and awarded the resident $375,000 as compensation for
loss of ocean views (Harvey Cedars v. Karan, 2012). On appeal, the New Jersey Supreme Court set the precedent that if private markets shift as a
result of the dune construction, then the dune has potential to produce specific benefits to homeowners closest to the dune (Harvey Cedars v. Karan,
2013). In September 2013, Karan settled out of court for $1.

17
first two dunes went to the communities with the lowest and the highest income levels. Both
communities have adjacent neighbors with similar median incomes that did not receive the
constructed dunes.
4.2 Research Design
The policy context surrounding this complex issue lends credibility to plausibly assuming that
there are not any confounding factors that influence housing prices other than the dune policy. The
implementation of the policy varies spatial and temporally across LBI. That is, three communities
received the constructed dunes in different years. To illustrate, I conduct a simple difference-in-
difference (DID) exercise. I define treated communities as Surf City and Harvey Cedars and
control communities as all others. I drop all transactions from the third treated community (Brant
Beach) because the final dune was completed very late (June 2012) in the timeframe of this
analysis. I then define a pre-policy (2003-2005) and a post-policy (2010-2012) period removing
observations from 2006 – 2010 during dune construction activities across both treated areas. With
the remaining 1,212 observations, I estimate the following standard DID model:
log( ) ( * )
ijt j t j t ijt
price Dune Post Dune Post
    
    
(4)
where Dunej is an indicator for treated community and Postt is an indicator for the post-policy
period. Results are presented in Table 2 and displayed graphically in Figure 3. The estimation
suggests that housing prices in the treated and control communities in the pre-policy period were
very similar and the difference is not statistically significant. In the post-period, the difference in
prices between the groups is highly significant, with treated sales prices approximately 6.9 percent

18
higher than control homes. The difference-in-difference estimate (δ) suggests a 7.1 percent
increase in housing values attributable to constructed dunes in this simple exercise.
18
For the full analysis with spatially and temporally varying policy implementation, an
observation is considered treated if the sales transaction occurs in one of the three neighborhoods
receiving a constructed dune after a distinct treatment timing date. This definition assumes that the
housing market does not respond to the policy until it is fully implemented and transactions during
the construction phase are removed from the data set. This timing is appealing as it removes any
potential uncertainty involving the amenity impacts of the projects and any impacts construction
activities may have on prices. As a result, there are 357 transactions in the treatment group and
4,470 in the control group.
In the full data, comparison of the means of observable characteristics of the homes suggest
relative balance between the treated and control groups (see columns 3-6 of Table 1) To confirm,
I use t-tests to assess the balance on observables and I am unable to reject the null hypothesis that
the means are equal for sales price and nearly all structural characteristics of the homes.
19
There is
also strong evidence for pre-treatment parallel trends in the treated and control communities.
Figure 4 displays two panels showing linear trends in housing prices over time with pre- and post-
policy trends plotted separately for two treated communities: Surf City (top panel) and Harvey
Cedars (bottom panel). Solid lines indicate trends in treated home sales while dotted lines indicate
trends in control communities. As shown in both panels of the Figure 4, the treated homes were
selling for slightly less than control group homes in the pre-policy periods but with very similar
trends. After dune construction activities, sales prices in treated homes were now higher than the
18
All percentage effects of dummy variable coefficients in this paper are calculated following Halvorsen and Palmquist (1980).
19
The lone exception is number of bedrooms, where there is a significant, yet small, difference between the two groups of 0.17.

19
control homes, demonstrating a positive impact of the dunes on housing values in both treated
areas. The two graphs also suggest that the capitalization of the dunes persists after construction
is completed. To test if dune construction had any spillover effects into control communities, I
conduct a falsification test where I remove all treated observations so the data contain only control
group transactions. Next, I falsely assign treatment to control communities that are directly
adjacent to treated communities. The results suggest a net effect of treatment that is not
significantly different from zero, lending support to assumption that the control outcomes are
likely not affected by the treatment intervention.
20
Although the evidence presented above lends strong support to my initial identifying
assumptions, I investigate the assignment of treatment further given the legal and political actions
that resulted in Surf City, Harvey Cedars, and Brant Beach receiving dunes before other
communities on the same barrier island. The primary factor leading to the spatial and temporal
variation in treatment is the easement issue for oceanfront property owners. Therefore, I look to
see if there is anything observable about oceanfront homes in each group that could be driving
selection into treatment. As shown in Table 3, there appears to be observable differences in key
variables that may help explain selection into treatment, specifically distance to public access and
lot size. Oceanfront homes in the control group are further from public access points (i.e. more
private) with an average distance of 184 feet, compared to 78 feet for the treatment group. The
concern among oceanfront property owners is that public access may increase with federal and
state funds being used to construct the dunes, which, in turn, would potentially decrease property
values and reduce their own enjoyment of the beach. This apprehension was discussed extensively
20
The coefficient on the Dune*Post variable (i.e. δ) for the Oaxaca-Blinder (DID) falsification test is -0.013 (-0.025) with a standard error of 0.011
(0.020). These results are presented, along with another test using the DID decomposition model, in Table C.2 of the online appendix.

20
in the local and national media prior to the construction of the first dune (e.g. Smothers 2006; Urgo
2006).
21
Additionally, the control group oceanfront homes sit on slightly larger lots and may
experience a larger loss in first floor ocean views than homes on smaller lots. These differences in
oceanfront home characteristics between the two groups is suggestive of selection on observables
and warrants investigating an estimator that applies the conditional independence assumption
(Rosenbaum and Rubin 1983).
4.3 Oaxaca-Blinder Estimator
Choosing between the treatment assignment assumptions is problematic due to suggestive
evidence that both are plausible. In order to accommodate this set of circumstances, I utilize the
logic of the Oaxaca-Blinder decomposition (Oaxaca 1973; Blinder 1973) to estimate the impact of
the dunes. The Oaxaca-Blinder (OB) estimator was developed and is still widely used by labor
economists to decompose wage differentials. The basic intuition of the estimator is to determine
differences in an outcome of interest between two groups. The decomposition identifies a
component of explained variation attributable to differences between observable characteristics
between each group and a second component that identifies the net effect of belonging to a
particular group. This latter part is commonly referred to as the unexplained component of the
decomposition and can be interpreted as a consistent estimate of the net effect of treatment if group
membership is defined in such a manner (Sloczynski 2015). In a hedonic modeling context, the
explained component can identify differences in housing prices attributable to structural,
locational, and environmental characteristics associated with the homes in each group. The
unexplained component then identifies the net effect of group membership, which in this context
is the treatment effect associated with federal dune construction projects.
21
It should be noted, however, that these fears were unfounded as no new public access, restrooms, or parking were added as a result of the policy.

21
Variants of this estimator have been used previously in housing market analyses to decompose
temporal changes in the distribution of housing and land prices (e.g. Mcmillian 2008; Qin et al.
2016). The OB logic also recently received cursory treatment in the context of sorting models
(Banzhaf and Walsh 2013) and interpreting hedonic capitalization as marginal willingness to pay
(Kuminoff and Pope 2014). While these previous studies use Oaxaca decompositions to look at
changes over time in housing markets, this study makes direct use of the decomposition to examine
differences between two housing groups in a policy evaluation setting to identify the net effect of
treatment.
22
Recent research into the finite sample performance of the OB estimator by Kline (2011) suggests
this technique is ideal for this setting because the estimator consistently identifies the net effect of
treatment under assumptions of either treatment exogeneity or conditional independence. In other
words, the estimator is doubly robust. Kline (2011) also shows that the OB estimator has useful
small sample properties for unbalanced research designs with small treatment groups relative to
the controls. These properties bode well for this research where only 357 observations out of 4,827
are in the treatment group.
To implement this logic to identify the treatment effect of federal dune policy, the model of
potential outcomes can be specified as follows:
'
Price ,
i
d d d
ii

X
(5)
 
E | , 0 for 0,1
d
i i i
Dune d




X
(6)
22
The usefulness of the Oaxaca-Blinder estimator in a policy evaluation setting was recently highlighted by Kline and Moretti (2014) where the
authors identify the long-run economic impacts of the Tennessee Valley Authority on local economies.

22
where
'
i
X
is a vector of all housing attributes,
d

is a vector of coefficients,
i
d

is the error term,
and the superscript d indicates assignment in either the treatment (d=1) or control (d=0) groups.
The differences in expected outcomes between the two groups can be decomposed in three steps:
   
       
 
 
   
 
''
1 1 0
' ' ' '
1 0 * *
' ' '
1 * * 0
1*
| 1 | 0
| 1 | 0 | 1 | 1
| 1 | 1 | 0 ( )
|1
o
i i i i i i
i i i i i i i i
i i i i i i
i i i
E Price Price E Dune E Dune
E Dune E Dune E Dune E Dune
E Dune E Dune E Dune
E Price Price Dune E

   
   

      

           
         

   

XX
X X X X
X X X
 
*| 1 | 0
o
i i i i
Price Dune E Price Dune
   
  
   
(7)
The second line of (7) adds and subtracts the unobserved counterfactual. The reference coefficients
*

estimate the counterfactual price structure and are determined by the weighting matrix (Ω):
* 1 0
ˆˆ
()I
  
  ΩΩ
(8)
The third line consolidates terms and the fourth line is the resulting decomposition. The first term
of the fourth line in (7) is equivalent to the net effect of treatment (i.e. the unexplained component)
and the second term captures the price differential attributable to characteristics of the treatment
and control groups (i.e. the explained component).
Operationally, I run separate regressions for both treatment and control groups to recover OLS
estimates of
0
ˆ

and
1
ˆ

.
23
Estimation of the unobserved counterfactual coefficient vector
*

is
dependent on the choice of a weighting matrix. There has been a line of research focused on the
appropriate value for Ω to define the counterfactual, including Oaxaca (1973) (
OX
Ω
=1 for treated
group price structure;
OX
Ω
=0 for control group price structure), Reimers (1983) (
(0.5)
RIΩ
)
and Cotton (1988) (
CsIΩ
, where s captures relative group size). Oaxaca and Ransom (1994)
23
The coefficient estimates for each group are an intermediate step and are provided in Table C.3 in the online appendix for reference.

23
argue that each of those weights are arbitrarily chosen and theoretically derive a weighting matrix
given as:
' 1 '00
( ) ( )
OR 
Ω X X X X
(9)
where
X
is a matrix of observations for a pooled sample (i.e. both treated and control observations)
and
0
X
is the observation matrix for the control group. The cross-product matrices are used as
weights for the coefficient vectors estimated separately for each group as a better representation
of the counterfactual then simply choosing an arbitrary weight. The weighting matrix in (9)
interprets the regression estimate from a pooled model over both groups as the counterfactual price
structure that would exist in the absence of treatment.
For this empirical work, it is difficult to determine whether the treated or control properties are
more representative of the housing market in absence of treatment, eliminating Oaxaca (1973) as
a potential weighting strategy. The unbalanced nature of the data set eliminates Reimer (1983) as
a feasible option. Cotton (1988) is intuitively appealing but as Oaxaca and Ransom (1994) note, it
is an arbitrary choice. Therefore, I utilize Oaxaca and Ransom’s (1994) theoretically derived
weighting matrix for my preferred model specification (but also test the other weights for
robustness). The reference coefficients for (8) can then be expressed as follows:
* 1 0
ˆˆ
()
OR OR
I
  
  ΩΩ
(10)
where (9) defines
OR
Ω
, and
1
ˆ

and
0
ˆ

are the estimates for the treatment and control groups,
respectively. With the counterfactual now defined, the estimation proceeds to identify the two parts
of the last line of equation (7) – the explained component of the decomposition

24
 
*| 1 | 0
o
i i i i
E Price Dune E Price Dune
   
  
   
and the net effect of dune construction on
housing prices
 
1*
|1
i i i
E Price Price Dune



.
5. Estimation of the Policy Effect
The data are a pooled cross-section of housing sales on LBI observed over time.
24
All estimation
strategies used in this work utilize a combination of neighborhood, spatial proximity, and flood
zone fixed effects to control for time-invariant unobservables while year and quarter fixed effects
control for time-varying unobservables and adjustments in the housing market. My preferred
model uses the OB estimator with the Oaxaca and Ransom (1994) weighting matrix and
bootstrapped standard errors clustered by neighborhood and year and month sold, and a semi-log
function form for the hedonic price function. Table 4 presents the results from this estimation in
Panel A, showing the estimated differential in housing prices between the two groups and the
decomposition. The estimated net effect of dune construction is a 3.6 percent increase in housing
prices.
25
The estimate is highly significant and precisely estimated with a narrow 95% confidence
interval (CI). The effect translates to an average capitalization in the range of $27,222 - $33,511,
or an annualized benefit of $2,623 - $3,229 per home.
26
Panel B displays the net effect of treatment
under alternative weighting schemes for comparison. Both Oaxaca using the control group price
structure as the counterfactual (i.e.
OX
Ω
=0) and Cotton (
CsIΩ
) produce statistically significant
point estimates within the CI of my preferred estimate albeit with slightly less precision (i.e. wider
CIs).
24
The lack of a substantial number of repeat sales in this data precludes the use of property-level fixed effects.
25
This empirical result compares favorably to the magnitude of the increase in high-value coastal properties in a numerical model of subsidized
beach nourishment found by McNamara et al. (2015).
26
Results based on median ($765,492) and mean ($942,344) sales prices in the sample.

25
With my primary result of a 3.6 percent increase in housing prices attributable to dune
construction, it is important to return to the question of interpretation raised in section 2. At the
very minimum, this research design generated estimates of the capitalization of the dune policy
into the local housing market, reflecting a measure of the economic value of adaptation policy. In
order to compare my results to estimates where the sufficient conditions of Kuminoff and Pope
(2014) for interpreting capitalization as MWTP are met, I estimate six single-year hedonic price
functions for 2007 – 2012 (after 1st dune was constructed). I utilize the exact model specification
as the full sample for each year. Results are displayed in Panel C of Table 4 and show a range of
significant policy effects from 2.9 percent to 5.7 percent. These results suggest the wedge between
capitalization and MWTP from this policy intervention is likely to be small, allowing cautious
interpretation of capitalization effect as an ex- post MWTP. At a minimum, this effect has a
welfare interpretation as a lower bound of Hicksian equivalent surplus (Banzhaf 2015).
The preferred model specification is subject to assumptions on treatment assignment, functional
form, and the specific spatial-temporal landscape for the housing market. To further demonstrate
the robustness of the main result, the remainder of this section describes alternative specifications
of the preferred model.
Arguments were presented previously on the plausibility of assumptions of both treatment
exogeneity and selection on observables. My first robustness check is to use the alternative (and
more conventional) estimators implied by those assumptions. Panel A of Table 5 displays these
results. A DID model estimated with robust standard errors clustered by neighborhood, year sold
and month sold shows a significant impact of 5.2 percent on housing prices from the dunes. The
second alternative uses a bias-corrected nearest neighbor matching estimator (Abadie and Imbens
2002). This model relies on the assumption that treatment is random conditional on covariates and

26
model restrictions reduce the potential confounds of unobservables. Operationally, exact matches
are required for year sold to control for time-varying unobservables. Additional matching variables
are chosen to correspond to characteristics that are plausibly driving selection into treatment as
seen in Table 3 (i.e. distance to public access and lot size) along with age, neighborhood, and
spatial proximity band. The model is specified to require four matches per observation with bias-
adjusted robust standard errors. Results from this estimator yield a marginally significant 6.3
percent treatment effect.
The three estimators provide a range of 3.6 – 6.3 percent for the average treatment effect with
the preferred OB estimate providing a lower bound. It is interesting to note that OB estimate
results in tighter CIs than the more traditional DID and matching estimator approaches. This
supports the finding of Sloczynski’s (2015) empirical Monte Carlo simulations that suggest the
OB estimator performs very well compared to other potential treatment effects estimation
strategies. Yet, results with Oaxaca’s (1973) original weighting scheme or the sample size
weights of Cotton (1998) are closer in magnitude with similarly wide CIs to results of the other
estimators. This demonstrates that the choice of the weighting matrix for the counterfactual
coefficients may have a measurable effect on the magnitude and precision of the estimate for the
outcome of interest.
The preferred model specification utilizes a semi-log form for the hedonic price function. Panel
B of Table 5 provides results from using a more flexible Box-Cox form. Maximum likelihood
estimation rejects the linear, multiplicative inverse, and log specifications of the model and yields
a positive and significant transformation parameter on sales price of approximately 0.09. This
transformed dependent variable is then used in the Oaxaca-Blinder estimation and yields a very
similar percentage effect of treatment (3.3 percent) as the model using the semi-log price function.

27
Next, I check for any spatial boundary constructs that may influence the main result. Since the
dune construction is discontinuous during the time frame of analysis, I systematically test the
impact of proximity to the boundaries of the dune. Five models are presented, each with a different
subset of the original data and at distances that highlight where the localized amenity value
deviates from the mean policy effect. The OB estimator is run with data that includes only
transactions within 1 mile, ½ mile, and ¼ mile of the dune boundaries and data that excludes all
transactions within ¼ mile and 500 foot boundary of the dune edge. The range of effects found in
the inclusion models is small (3 – 3.5 percent) and similar to the full sample estimate of 3.6 percent.
The exclusion models have a range of 3 – 4.3 percent, with the large impact occurring when homes
very close (less than 500’) to the dune edge are excluded. All the inclusion and exclusion models
tested lead to a range of estimates well within the 95% CI of the full sample specification. There
is suggestive evidence that the benefits may be less on the dune boundaries due to the
discontinuities but the magnitude of the change is relatively small.
Lastly, two models are run with data restricted to observations around the construction of the
first dune intervention in Surf City, from 2004-2009 and from 2005-2008, respectively.
27
The
results imply that the dune effect increases marginally compared to the preferred estimate and the
effect become largest when the time frame is narrowed to three years (4.6 percent). To summarize,
Figure 5 displays my preferred estimate of the policy effect with 95% CI on the far left and the
point estimates and CIs for all robustness checks performed. The light grey shaded area represents
the range of the confidence interval for my preferred estimate. Note that this band includes nearly
27
The focus here is on the Surf City dune due to data limitations for performing the same checks around the Harvey Cedars and Brant Beach
dunes. The number of observations for the treated group is much smaller for these interventions due to the temporal constraints of the market
definition.

28
all of the point estimates of the robustness checks, suggesting my preferred specification is
estimating a relatively precise impact of dune construction on housing values in LBI.
5.1 Back-of-the-Envelope Policy Implications
To place the above results into perspective for policy, I calculate a simple back-of the envelope
benefit costs analysis. The following calculations compare capitalized benefits of the dunes to
property owners and construction costs only. Applying the results of the analysis to the three
communities receiving the dunes, the capitalized benefits are approximately $170 million.
However, engineering costs for those three dune systems to date exceed $261 million. Although
my back-of-the-envelope calculation above abstracts from certain benefits (e.g. wider beaches
increase recreation opportunities and value) and costs (e.g. replenishment alters nearshore
dynamics and may eliminate surfing opportunities and destroy benthic habitat) associated with this
policy intervention, the dunes do not appear to be providing enough benefits to justify the large
costs of implementation. That said, it is important to note that the policy effect measured here is
likely a lower bound on these values as they are measured using revealed preferences before Sandy
and LBI had not experienced a major storm event prior to Sandy since the Ash Wednesday storm
of 1962.
Despite this result, the information provided by Superstorm Sandy highlighted the effectiveness
of the dunes ex-post to minimize damage and the risks associated with being located in an
unprotected town. Intuitively, post-Sandy individuals are likely to value the benefits of storm
protection more so then their pre-Sandy counterparts. Furthermore, the realization of storm
protection benefits post-Sandy appear large. Adding dunes and wider beaches limited the liabilities
of the federal government (through NFIP and FEMA post-disaster aid) as shown through post-
Sandy expenditures. For example, compare Surf City (treated) and neighboring Ship Bottom

29
(untreated). The towns share a border and are very similar in terms of population, demographics,
and housing stock but the federal government had $46 million less in post-Sandy liabilities in Surf
City as compared to Ship Bottom.
28
This outcome suggests a strong incentive to undertake these
projects from a public finance perspective given the pre-existing subsidy and disaster aid
programs. It also raises questions regarding the government’s role in adapting to climate change
and insuring against fat-tail events. These questions remain important areas for future research.
That said, the apparent disconnect between the net cost of the policy and the ex-post realization of
large protection benefits during Sandy provide further motivation for decomposing the dune effect
into amenity flows.
6. Spatial Considerations and Decomposition of the Treatment Effect
The previous section presented results from models identifying the average policy effect of a
federal shoreline stabilization intervention. Here, I first explore the spatial extent of the
capitalization with both Oaxaca-Blinder and DID models. Second, I attempt a decomposition of
the average policy effect into three amenity streams (storm protection, ocean viewshed, and visitor
traffic/privacy concerns) using a DID model with high-resolution spatial data and a GIS viewshed
tool to quantify the amenities.
6.1 Spatial Extent of the Capitalization
It is naïve to assume that the impact of dune construction and beach renourishment is homogenous
for all homes on the island. For example, the existing literature shows the value of beach width
capitalization varies with proximity to the beach (e.g. Landry and Hindsley 2011; Gopalakrishnan
et al. 2011; Landry and Allen 2016). Therefore I run additional Oaxaca-Blinder and DID models
28
A breakdown of these federal expenditures is provided in Table C.1 in the online appendix.

30
with the same specifications as the previous models to determine the extent of the policy effect. I
use the spatial proximity bands defined in Section 3 to account for distance to the policy
intervention. Specifically, I run five separate OB models defining the two groups in each model as
treated and control transactions for each spatial proximity band (i.e. oceanfront, ocean block, 2nd
block, 3rd, block, and bayfront). Results are presented in Table 6. The point estimates indicate a
spatial limit to the effects and non-monotonic pattern to the capitalization of benefits from the
dune. Oceanfront homes experience a positive impact higher than the average policy effect of
approximately 4.4 percent. Importantly, the homes in the ocean block experienced higher gains of
approximately 6.6 percent. The positive and significant impact extends to homes in the second
block (3.1 percent) and turns negative and insignificant after that point.
29
The non-monotonicity
of this result is surprising at first glance but rather intuitive given the potential for multiple amenity
flows being impacted by the dunes. That is, ocean block homes still receive protection benefits but
may not experience a reduction in ocean views of the same magnitude as similar oceanfront homes.
I also estimate a single DID model with the policy effect interacted with each spatial proximity
band. The results are shown in the second column of Table 6. As with the overall policy effect, the
DID estimates are larger than the OB model, but the non-monotonic pattern and the attenuation of
the effect after the second block are both clear and significant in this model as well. Ocean block
homes have the largest significant capitalization (17.7 percent), followed by oceanfront homes
(13.5 percent), and second block home (4.0 percent). As with the OB models, the policy effect
attenuates after the second block and becomes insignificant. Both modeling strategies point to
homes in the first two bands capturing a large percentage of the benefits from the constructed
dunes.
29
Note that oceanfront, ocean block, and second block homes are on average, 16 feet, 244 feet, and 881 feet from the dunes respectively.

31
6.2 Decomposition of Benefits and Ancillary Costs
My original motivation for this decomposition relates to my interpretation of USACE dune
projects from Section 2 as an ex-ante public adaptation to climate change with potential to have
ancillary benefits or costs. Two empirical findings here, the simple benefit-cost analysis showing
the policy as a net cost and the non-monotonicity of the policy effect across space, further
encourage this effort to assess the determinants driving those results. As noted earlier, this
decomposition focuses on storm protection, ocean views, and visitor traffic/privacy concerns.
Since the strength of the Oaxaca-Blinder estimator is identifying average differences between two
groups, decomposition of the impact of the dune into heterogeneous amenity effects requires a
more flexible modeling strategy. Therefore, I estimate the amenity decomposition with a DID
framework by adding interaction terms with the treatment variable for each affected amenity as
follows:
   
   
1 2 1 2
1 2 1 2
lnPrice P_ _
ijt it it it it it
it it it it it it t j i i ijt
Dune Area Dune P Lateral Dune
View Dune BW Dune
     
          
     
          XL
(11)
This model takes a semi-log form with the natural log of the Price of house i in municipality j at
time t as the dependent variable. Duneit is the policy variable of interest equal to one if the
transaction is assigned treatment status while
_it
P Area
,
_it
P Lateral
it
View
and
it
BW
are variables
quantifying protection area, lateral protection, first-floor ocean views, and beach width,
respectively. The vectors X and L represent housing and location characteristics of each transacted
home. The model also contains fixed effects for time (𝜏𝑡), neighborhood (𝜂𝑡), flood zone (
i

), and
spatial proximity band (
i

).
Estimation of (11) requires high-resolution spatial data to quantify the amenities impacted by
the dune. The level of storm protection is directly related to the size and positioning of the dune.

32
This barrier potentially provides both frontal area and lateral protection. In terms of frontal
protection, FEMA classifies a dune as an effective barrier to the wave action associated with a
100-year storm event if the cross-sectional area of the frontal dune is greater than 540 square feet
(FEMA-540 rule). This frontal dune area is triangular with the base being determined by the 100
year flood elevation and the height defined with a vertical line from the peak of the dune (see
Figure 2). Barone et al. (2009) measured this area at 250’ intervals on LBI in 2005 prior to any
USACE replenishment project. I utilize this data set to proxy for protection level in the control
communities assigning the nearest interval measurement to each transaction. For the control group
transactions, the mean (max) value for the cross-sectional area of the frontal dune is 41.2 (159.9),
or about 7 (30) percent of the level required for adequate protection. Ideally, this measure would
vary temporally in the control communities, but I assume a constant dune area over time due to
data limitations (i.e., data were only captured in 2005). For treated parcels, a value of 540 is
assigned as the dunes were constructed by the USACE to satisfy the FEMA-540 rule and likely do
not vary spatially. The second aspect of storm protection is lateral protection. The discontinuous
nature of the dunes opens up the potential that treated homes on or near the boundaries may be
more susceptible to storm surge and are thus less protected than homes located at the interior of
the dune. This observation, motivated by Smith et al. (2014), is incorporated into the analysis as a
measure of distance from each parcel to the nearest lateral edges of the three dune systems on the
island.
Ocean view is defined in terms of the degrees of Atlantic Ocean visible from an observer on
the first floor of each home, with a marginal change as a result of the policy expressed as a one
degree decrease. Quantifying this presents a challenge since I cannot directly observe these
views for each property in the sample. However, availability of rich spatial data for the study

33
area combined with geo-processing techniques allow for the estimation of an approximate
viewshed for each observation both before and after treatment. A viewshed tool was developed
following a similar methodology outlined in both Bin et al. (2008) and Crawford et al. (2014).
30
Details on the development of the viewshed tool are provided in the online appendix.
The USACE project also results in the addition of sand to increase the size of the beach berm
seaward of the dune in the treated communities. A wider beach has the potential to increase
storm protection, recreation opportunities and tourist visitation making beach width a common
proxy variable in the coastal hedonic literature (e.g. Gopalakrishnan et al. 2011; Landry and
Hindsley 2011; Landry and Allen 2016). There are some unique features of this empirical setting
that should be noted here to understand what beach width may be a proxy for in equation (11).
Not all oceanfront homes have direct beach access, with some having private community dune
walkovers and others using the nearest public walkover. Dune construction did not alter access
as homes with direct access maintained that right. There also was no increase in public access,
parking, public restrooms, or any other service for visitors to the beach. The local concern is
therefore that increases in beach width associated with the policy may increase visitation, reduce
privacy, and create traffic and congestion (due to limited public facilities and parking) on the
island. Therefore, I interpret the increase in beach width as a result of the USACE policy as a
proxy for visitor access/privacy concerns for LBI homeowners. The variable for beach width
corresponding to each parcel is a measure, in feet, of the distance from the nearest public access
point to the Atlantic Ocean shoreline, including both beach berm and dunes. Due to the dynamic
nature of erosion, GIS shoreline features at different points in time during the study time frame
30
Crawford et al. (2014) find that using yearly viewshed measures in a coastal housing market did not produce significantly different effects on
sales prices compared to a single viewshed for multiple years. Considering this result and the near build-out density of development on LBI, this
analysis focuses on producing two viewshed measures (pre-dune and post-dune) to capture the policy impact on ocean views.

34
(2002, 2007, and 2012) are used to provide some variation in the beach width measure.
Transactions are grouped into three periods, 2000-2004, 2005-2009, and 2010-2012 and assigned
a width value based on the 2002, 2007, and 2012 shoreline, respectively. A marginal change is
defined as a one-foot increase in width.
With the four variables representing the amenity flows quantified, estimation of (11) can
proceed. Relating this back to equation (3), the value of adaptation is linked to the storm
protection results while the ancillary impacts of the policy are connected to the viewshed and
beach width variables. Table 7 displays the results of three model specification of equation (11).
The first two models vary the specification for the protection variable while the third model
restricts the model to the spatial proximity bands with significant policy effects. Model (1)
simply uses an indicator variable to control for protection while also decomposing the viewshed
and beach width changes associated with the policy. Results suggest the dune itself, when
decomposed from ancillary impacts, could raise home values 20.8 percent. The model finds
ocean views and beach width to be positive and statistically significant amenities when
considered in levels but when interacted with the dune policy, the impacts turn negative and
significant, offsetting some of the gains in protection services. The model finds a 0.42 percent
decline in property value per degree of lost ocean view and a 0.07 percent decline per foot of
increased beach width. Given the average change in these variables associated with the policy,
those marginal changes imply a 5.5 percent decline in value from 13 degrees of ocean view lost
and a 8.3 percent decline from the 118 foot gain in beach width. In levels, the coefficient
estimates on ocean view (0.0029) and beach width (0.0003) are comparable to other findings in
the literature. Two previous studies measuring ocean views that find a 0.3 and 0.34 percent
increase for additional degree of view, respectively (Bin et al. 2008; Crawford et al. 2014). For

35
beach width, my estimate implies a capitalized value of $283 per foot of beach width, well
within the range of values commonly found (e.g. Landry and Allen 2016).
Model (2) replaces the indicator variable for the dune with the cross-sectional area of the frontal
dune to control for protection. The results are quite similar as compared to Model (1) but with
slightly different interpretation of the protection results. In this case, results suggest a 0.05 percent
increase in property value per square foot increase in the area of the frontal dune. Given the large
average change between treated and control communities as a result of dune construction, this
implies a 26 percent increase in home value attributable to the average change in dune area.
Protection offered by lateral distance from the dune discontinuity does not appear to be
economically or statistically significant in this model or the other spatial specification. Similar to
Model (1), the impact of ocean views and beach width related to the policy capitalize as ancillary
costs that partially offset the protection gains.
31
The spatial restriction model (3) limits the analysis to transactions in the oceanfront, ocean block,
and 2nd block proximity bands to match the extent of the capitalization of the policy effect. Results
for the policy interactions with beach width and ocean view are nearly identical and the value of
protection increases slightly (0.07 percent per unit increase in dune area). Beach width and ocean
views as independent variables increase to 3.9 percent per degree of view and $377 per foot of
beach width, again consistent with previous literature. Model (3) results suggest that the
capitalization of protection is proximity dependent and the models here are moving in the same
manner as the models for the average policy effect.
31
To address the potential for amenity-specific spillover effects in control communities, I conduct a falsification test estimating equation (10) with
all treated communities dropped and treatment falsely assigned to the neighboring control communities (i.e., Ship Bottom, Loveladies, and
Brighton). The coefficients on the policy indicator interacted with each amenity show no statistically significant effects. Results are presented in
Table C.2 in the online appendix.

36
The results provide suggestive evidence supporting the theory that large public adaptation efforts
may have ancillary impacts that may enhance or detract from the policy goal. In this case, lost
ocean views and visitor traffic/privacy impacts partially offset the policy’s goal of providing storm
protection services. The decomposition also provides some insight on the apparent disconnect
between the findings that the policy is a net cost and the ex-post realization of cost savings resulting
from protection services. In other words, the results indicate that if the dunes only provided storm
protection services, the policy may generate substantial net benefits. It also provides empirical
evidence supporting the non-monotonicity of the capitalization since the ancillary costs likely
impact oceanfront homeowners more so than their ocean block counterparts.
7. Conclusion
The federal policy response to vulnerable coastal communities has recently shifted to include
the construction of natural infrastructure (i.e. dunes) for hazard protection. This research provides
a quasi-experimental evaluation of this federal shoreline stabilization policy and contributes to
growing literatures on valuing coastal beaches and dunes (e.g. Landry and Allen 2016; Qui and
Gopalakrishnan 2016) and the broader implications of climate change and coastal management
(e.g. Gopalakrishnan et al. 2016a, Gopalakrishnan et al. 2016b). The values for dune construction
found in this work can be interpreted as values of an ex-ante public adaptation to climate change
and results suggest the potential for ancillary costs from such investments. This latter finding
points to the need for more empirical work related to coastline stabilization and public policies
aimed at providing climate adaptation services.
This paper evaluates this federal policy using residential housing data from a barrier island in
New Jersey. Using the logic of the Oaxaca-Blinder estimator, I find the value of the dune policy

37
to be strictly positive (3.6 percent). Despite the apparent benefits, the policy appears to be a net
cost to society. To investigate the ancillary impacts of public adaptation and the determinants
leading to the net cost policy implication, I decompose the policy into impacts associated with
storm protection, ocean views, and visitor traffic/privacy concerns. These results indicate that the
storm protection services of the policy are partial offset by ancillary costs related to the other
amenities impacted by dune construction.
Practically speaking, the dune policy valuation estimates from this research are likely to transfer
to similar coastal areas on the Eastern seaboard. However, the decomposition revealed the ancillary
impacts in NJ were negative, reducing the effectiveness of the intervention. These ancillary effects
are likely to be location specific, as demonstrated by the negative coefficient on the interaction of
the policy and beach width reflecting landowner concerns about traffic and privacy from more
visitors. Fortunately, the estimation method and data generation techniques used in this work are
straightforward to replicate in other coastal areas where dunes are planned.
Lastly, it is also important to note a distributional concern related to the USACE policy
intervention. According to the Project Cooperation Agreement signed in 2005 between NJDEP
and USACE, dune construction and beach replenishment geared toward storm protection has the
following cost share: 65% federal and 35 % non-federal. 75% of non-federal expenditures are
covered by NJDEP, leaving approximately 8.75 % of total costs the responsibility of the
communities receiving the dune and beach replenishment (USACE 2005). This cost-sharing
arrangement implies an annualized household cost of the dune of approximately $168 per year in
Surf City. The remaining cost of $1748 per home is spread among federal and state taxpayers. In
essence, federal and state taxpayers are subsidizing the protection of assets for a very small number
of individuals who have chosen to reside in a high-cost, high-risk location. My results here point

38
to the need for re-evaluation of how these projects are funded. The annualized capitalization per
home ($3,229) could form the basis for a property tax surcharge after dune construction, forcing
property owners to internalize the risks associated with their housing and location decisions.
REFERENCES
Abadie, A., G.W. Imbens. 2002. Simple and Bias-Corrected Matching Estimators for Average
Treatment Effects. NBER Working Paper T0283.
Abbott, J.K., H.A. Klaiber. 2011. An Embarrassment of Riches: Confronting Omitted Variable
Bias and Multi-Scale Capitalization in Hedonic Price Models. Rev. Econ. Stat. 93(4), 1331-42.
Banzhaf, H.S. 2015. Panel Data Hedonics; Rosen’s First Stage and Difference-Differences as
“Sufficient Statistics.” NBER Working Paper 21485.
Banzhaf, H.S., R.P. Walsh. 2013. Segregation and Teibout Sorting: The Link Between Place-based
Investments and Neighborhood Tipping. J. Urban Econ. 74(1), 83-98.
Barone, D.A., K.K. McKenna, and S.C. Farrell. 2014. Hurricane Sandy: Beach-dune Performance
at New Jersey Beach Profile Network Sites. Shore and Beach, 82(4): 13-23.
Barone, D.A., S.C. Farrell, R.V. Koch, B.S Howard, M.H. Gruver, M.J. Mihalasky. 2009. “Beach-
Dune System Susceptibility Assessment: Long Beach Island, New Jersey,” Dataset.
Bento, A., M., C. Lang. 2015. Who Benefits from Environmental Regulation? Evidence from the
Clean Air Act Amendments. Rev. Econ. Stat. 97, 610-22
Bin, O., T.W. Crawford, J.B. Kruse, C.E. Landry. 2008. Viewscapes and Flood Hazard: Coastal
Housing Market Response to Amenities and Risk. Land Econ. 84(3), 434-48.
Bin, O., J.B. Kruse. 2006. Real Estate Market Response to Coastal Flood Hazards. Nat. Hazards
Rev. 7(4), 137–44.
Bin, O., S. Polasky. 2004. Flood Hazards on Property Values: Evidence before and after Hurricane
Floyd. Land Econ. 80(4), 490-500.
Blinder, A.S. 1973. Wage Discrimination: Reduced Form and Structural Estimates. J. Hum.
Resour. 8(4), 436-55.
Boslett, A., T. Guilfoos, C. Lang. 2016. Valuation of Expectation: A Hedonic Study of Shale Gas
Development and New York’s Moratorium. J. Environ. Econ. Manage. 77, 14-30.

39
Currie, J., L. Davis, M. Greenstone, R. Walker. 2015. Environmental Health Risks and Housing
Values: Evidence from 1,600 Toxic Plant Openings and Closings. Am. Econ. Rev. 105, 678-709.
Cotton, J. 1988. On the Decomposition of Wage Differentials. Rev. Econ. Stat. 70(2), 236-43.
Crawford, T.W., O. Bin, J.B. Kruse, C.E. Landry. 2014. On the Importance of Time for GIS View
Measures and Their Use in Hedonic Property Models: Does Being Temporally Explicit Matter?
Trans. GIS 18(2), 234-52.
Davis, L.W. 2011. The Effect of Power Plants on Local Housing Values and Rents. Rev. Econ.
Stat. 93, 1391-1402.
Davlasheridze, M., K. Fisher-Vanden, and H.A. Klaiber. 2017. The Effects of Adaptation
Measures on Hurricane Induced Property Losses: Which FEMA investments have the highest
returns? J. Environ. Econ. Manage. 81, 93-114.
Gopalakrishnan, S., D. McNamara, M.D. Smith, and A.B. Murray. 2016a. Decentralized
Management Hinders Coastal Climate Adaptation: The Spatial-dynamics of Beach
Nourishment. Environ. Resour. Econ. Forthcoming.
Gopalakrishnan, S., C.E. Landry, M.D. Smith, and J.C. Whitehead. 2016b. Economics of Coastal
Erosion and Adaptation to Sea Level Rise. Annu. Rev. Resour. Econ. 8, 119-139.
Gopalakrishnan, S., M.D. Smith, J.M. Slott, A.B. Murray. 2011. The Value of Disappearing
Beaches: A Hedonic Pricing Model with Endogenous Beach Width. J. Environ. Econ. Manage.
61(3), 297-310.
Guo, C., C. Costello. 2013. The Value of Adaption: Climate Change and Timberland Management.
J. Environ. Econ. Manage. 65(3), 452-68.
Hall, T.M., A.H. Sobel. 2013. On the Impact Angle of Hurricane Sandy’s New Jersey Landfall.
Geophys. Res. Lett. 40, 2312-15.
Hallstrom, D.G., V.K. Smith. 2005. Market Responses to Hurricanes. J. Environ. Econ. Manage.
50(3), 541-61.
Halvorsen, R., and Raymond Palmquist. 1980. The Interpretation of Dummy Variables in
Semilogarithmic Equations. Am. Econ. Rev. 70(3), 474-75.
Landry, C.E. and T. Allen. 2016. Hedonic Property Prices and Coastal Beach Width. Working
Paper: SSRN: http://ssrn.com/abstract=2474276.
Landry, C.E., P. Hindsley. 2011. Valuing Beach Quality with Hedonic Property Models. Land
Econ. 87(1), 92-108.

40
Kahn, M.E. and R.P. Walsh. 2014. Cities and the Environment. NBER Working Paper 20503.
Kelly, D.L., C.D. Kolstad, G. Mitchell. 2005. Adjustment Costs from Environmental Change. J.
Environ. Econ. Manage. 50(3), 468-95.
Kline, P.M. 2011. Oaxaca-Blinder as a Reweighting Estimator, Am. Econ. Rev. 101(3), 532-37.
Kline P.M., E. Moretti. 2014. Local Economic Development, Agglomeration Economics and the
Big Push: 100 Years of Evidence from the Tennessee Valley Authority. Q. J. Econ. 129, 275-331.
Kousky, C., H. Kunreuther. 2013. Addressing Affordability in the National Flood Insurance
Program. Resources for the Future and Wharton Risk Management and Decision Processes
Center Issue Brief 13-02.
Kuminoff, N.V., J.C. Pope. 2014. Do Capitalization Effects for Public Goods Reveal the Public’s
Willingness to Pay? Int. Econ. Rev. 55(4), 1227-50.
Mansur, E.T., R. Mendelsohn, W. Morrison. 2008. Climate Change Adaptation: A Study of Fuel
Choice and Consumption in the US Energy Sector. J. Environ. Econ. Manage. 55(2), 175-93.
McMillian, D.P. 2008. Changes in the distribution of house prices over time: Structural
characteristics, neighborhoods, or coefficients? J. Urban Econ., 64(3): 573-589.
McNamara, D.E., S. Gopalakrishnan, M.D. Smith, A.B Murray. 2015. Climate Adaptation and
Policy-induced Inflation of Coastal Property Value. PLoS ONE 10(3), e0121278.
Mendelsohn, R. 2000. Efficient Adaptation to Climate Change. Clim. Chang., 45(3-4), 583-600.
Michaels, R.G., V.K. Smith. 1990. Market-Segmentation and Valuing Amenities with Hedonic
Models: The Case of Hazardous Waste Sites. J. Urban Econ. 28(2), 223–42.
Muehlenbachs, L., E. Spiller, C. Timmins. 2015. The Housing Market Impacts of Shale Gas
Development. Am. Econ. Rev. 105(12), 3633-59.
Oaxaca, R.L. 1973. Male-Female Wage Differentials in Urban Labor Markets. Int. Econ. Rev.
14(3), 693-709.
Oaxaca, R.L., M.R. Ransom. 1994. On Discrimination and the Decomposition of Wage
Differentials. J. Econom. 61(1), 5-21.
Palmquist, R.B. 1992. Valuing Localized Externalities. J. Urban Econ. 31, 59–68.
Qin, Y. H. Zhu, and R. Zhu. 2016. Changes in the distribution of land prices in urban China during
2007-2012. Reg. Sci. Urban Econ., 57: 77-90
Qiu, Y. and S. Gopalakrishnan. 2016. Shoreline Defense against Climate Change and the
Capitalized Impact of Beach Nourishment, Working Paper.

41
Rather, J. 2001. Corps Drops Sand-Replenishment Plan for Fire Island. NY Times, April 15, 2001.
Reimers, C.W. 1983. Labor Market Discrimination against Hispanic and Black Men. Rev. Econ.
Stat., 65(4), 570-79.
Rosenbaum, P.R., D.B. Rubin. 1983. The Central Role of the Propensity Score in Observational
Studies for Causal Effects. Biometrika 70(1), 41-55.
Smith, M.D., A.B. Murray, S. Gopalakrishnan, A.G. Keeler, C.E. Landry, D. McNamara, L.J.
Moore. 2014. Geoengineering Coastlines? From Accidental to Intentional. Duke Environmental
and Energy Economics Working Paper EE 14-02.
Sloczynski, T. 2015. The Oaxaca-Blinder Unexplained Component as a Treatment Effects
Estimator. Oxford B. Econ. Stat., 77(4), 588-604.
Smothers, R. 2006. Wary of Bureaucrats, Beachfront Property Owners Oppose Efforts to Stop
Erosion. NY Times, April 15, 2006.
Trenberth, K.E., J.T. Fasullo, T.G. Shepherd. 2015. Attribution of climate extreme events. Nature
Clim. Chang. 5, 725-30.
Urgo, J.L. 2006. A Standoff over Sand: A State Beach Project Requires More Public Access, but
… It's Meeting Some Resistance. The Philadelphia Inquirer, June 11, 2006.
USACE. 1999. Final Feasibility Report and Integrated Environmental Impact Statement.
(http://www.nap.usace.army.mil/Portals/39/docs/Civil/LBI/LBI_FeasRpt_Complete.pdf)
_______. 2005. Project Cooperation Agreement between The Department of the Army and The
State of New Jersey Department of Environmental Protection: Barnegat Inlet to Little Egg Inlet,
Long Beach Island, NJ Shore Protection Project. (http://www.nj.gov/dep/ec/lbi_pca.pdf)

42
Figure 1. Study Area and Location of USACE Constructed Dunes
Notes: Black dots indicate housing transactions in the analysis. The spatial location and extent of the three USACE dunes in Surf City, Harvey
Cedars, and Brant Beach are displayed as grey bands in front of the communities.

43
Figure 2. Schematic of a Dune System
Notes: Figure adapted from Barone et al. (2009).
100 Year Stillwater Flood Elevation
Sea Level
Beach
Primary Frontal
Dune
Vertical Line from Peak of Frontal Dune
Cross-Sectional Area of
Frontal Dune
(Subject to FEMA-540 Rule)
Backshore

44
Figure 3. Simple Difference-in-Differences Estimation
Notes: Simple statistical exercise where the treated communities are defined as Surf City and Harvey Cedars and control communities
are all others (except Brant Beach, which is removed from this simple analysis because dune construction was completed only 4
months prior to Sandy). The baseline period is 2003 – 2005, before any dune construction and the post-dune period (2010-2012) is
after construction of the dunes. The difference-in-difference estimate is represented by δ.

45
Figure 4. Price Trends in Treated and Control Communities
Notes: Black dots indicate individual housing transactions by month. Linear price trends for the two primary treated communities compared to all control communities both before and after
dune construction activities. Year dates on the horizontal axis indicate the beginning of each year (i.e. 2004 = January 2004; 2011 = January 2011, etc.). Shaded areas represent dune
construction activities and transactions during this time are removed to avoid any impacts this activity may have on home prices. Dotted lines are added to the shaded area to demonstrate the
trend from pre-dune to post-dune periods in each group. A similar figure is presented in the appendix (Figure C.3) for the third treated community of Brant Beach with limited transactions
due to proximity of dune completion and Sandy.

46
Figure 5. Robustness Checks on Average Treatment Effects
Notes: The estimate on the far left is my preferred estimate and the grey box corresponds to the 95% CI of that estimate. The ten other point estimates and 95% CIs from the various robustness checks
described in the text are also shown. Nearly all point estimates for the robustness checks fall within the range of my preferred estimate’s 95% CI (approximately 1-6 %).
-0.02
-0.01
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
0.11
0.12
0.13
0.14
Coefficient Estimates

47
Table 1: Summary Statistics for Residential Home Sales on Long Beach Island from 2000-2012
Source: Author calculations from housing deed records, tax assessor data and GIS processes. * Ocean views are only present in 7 percent of the sample and this mean is for all homes that have
some ocean view.
Full Sample (N=4,827)
Treated (N=357)
Control (N=4,470)
Variable
Mean
Std. Dev.
Mean
Std. Dev.
Mean
Std. Dev.
Sales Price (2012 $)
$942,344
$698,399
$887,886
$540,672
$946,693
$709,354
Bedrooms
3.79
1.18
3.95
1.08
3.78
1.19
Bathrooms
2.71
1.09
2.79
1.00
2.70
1.10
Square Footage
1746
796
1814
718
1740
802
Lot Size (ft2)
5700
4915
5237
2267
5736
5065
Age of Home
35.7
23
36.2
23.1
35.6
23.1
Distance to Ocean (ft)
1145
740
1297
868
1133
728n
Distance to Bay (ft)
808
590
897
673
801
582
Distance to Public Access (ft)
935
704
1038
826
926
693
Distance to Comm. Property (ft)
507
515
546
528
504
513
Beach/Dune Width (ft)
285
69
394
59
276
62
Protection (Dune Area in ft2)
78.1
134
540
0
41.2
29.6
Ocean View (°) *
59.3
41.9
47.4
38.4
60.3
42.1
Oceanfront
0.06
0.23
0.09
0.28
0.06
0.23
Oceanfront Block
0.17
0.37
0.20
0.40
0.16
0.37
Second Block
0.26
0.44
0.14
0.35
0.27
0.45
Third Block
0.26
0.44
0.36
0.48
0.25
0.43
Fourth Block
0.13
0.34
0.14
0.35
0.13
0.34
Bayfront
0.12
0.33
0.07
0.26
0.13
0.33
Condo
0.06
0.24
0.01
0.09
0.06
0.24
Garage
0.42
0.49
0.52
0.50
0.41
0.49
Hot Tub
0.09
0.29
0.06
0.23
0.09
0.29

48
Table 2: Results of Simple Statistical Difference-in-Differences Estimation
Notes: Simple statistical exercise demonstrating the research design for this analysis. Baseline period is 2003 – 2005, before any dune
construction and the post-dune period (2010-2012) is after construction of the dunes in the treated communities of Surf City and
Harvey Cedars. Brant Beach is removed from this simple analysis because dune construction was completed only 4 months prior to
Sandy. ** Significant at the 5 percent level. * Significant at the 10 percent level.
Outcome Variable:
Log Sales Price (2012$)
Coefficient
Standard Error
t-stat
Baseline (2003 – 2005)
Control
13.772
Treated
13.769
Difference
-0.003
0.025
-0.11
Post-Dune (2010 – 2012)
Control
13.662
Treated
13.729
Difference
0.067**
0.033
2.02
Difference-in-Differences
0.069*
0.041
1.68

49
Table 3: Summary Statistics for Oceanfront Homes by Treatment Status
Notes: Treatment status is assigned to all transactions behind a dune after completion construction AND transactions between the date
that knowledge of the project was publicly known and the end of construction are removed to reduce potential confounds associated
with uncertainty of the timing of dune construction.
Treated (N=31)
Control (N= 250)
Sales Price (2012 dollars)
$2,042,123
$2,152,182
Bedrooms
4.32
4.23
Bathrooms
3.68
3.5
Square Footage
2,523
2,124
Lot Size (feet2)
7,798
10,911
Age
27.3
27.5
Distance to Ocean
155
126
Distance to Bay
1,522
1,364
Distance to Public Access
77.7
184

50
Table 4: Oaxaca-Blinder Estimation of Impacts of Dune Construction on Housing Prices
N=4,827
Coefficient
Bootstrap
Std. Error
95 % C.I.
Panel A. Estimation with Oaxaca-Ransom Weights Ω =
' 1 '00
( ) ( )

XX X X
Differential
Treated
13.558***
0.026
13.504
13.612
Control
13.574***
0.018
13.539
13.609
Difference
-0.016
0.032
-0.078
0.047
Decomposition
Explained Variation
-0.051*
0.030
-0.111
0.008
Net Effect of Treatment
0.036***
0.011
0.014
0.057
Panel B. Net Effect of Treatment under Alternative Weights
Oaxaca (1973) Ω = 0
0.051**
0.020
0.012
0.090
Oaxaca (1973) Ω = 1
0.092
0.065
-0.035
0.220
Reimers (1983) Ω = (0.5)𝐼
0.072
0.046
-0.019
0.162
Cotton (1988) Ω = 𝑠𝐼
0.054***
0.019
0.017
0.092
Panel C. Net Effect of Treatment by Year using Preferred Specification
2007 (N=392)
0.044**
0.018
0.008
0.080
2008 (N=284)
0.038**
0.017
0.005
0.072
2009 (N=279)
0.013
0.016
-0.019
0.045
2010 (N=328)
0.055***
0.019
0.018
0.092
2011 (N=321)
0.029**
0.013
0.004
0.054
2012 (N=354)
0.055***
0.012
0.031
0.079
Notes: Point estimates for the treatment effect of the dune obtained from Oaxaca-Blinder regressions of log housing prices on a variety
of housing characteristics, distances to important features, spatial amenities, and fixed effects. Models use bootstrapped standard errors
with 50 replications clustered by neighborhood, year sold, and month sold. Coefficients for each group and the counterfactual for the
preferred specification are given in Table C.3 in the online appendix. *** Significant at the 1 percent level. ** Significant at the 5
percent level. * Significant at the 10 percent level.

51
Table 5: Robustness Checks on Dune Effects
Notes: Oaxaca-Blinder regressions are run with bootstrapped standard errors. Difference-in-differences regressions are run with robust standard errors
clustered by neighborhood and year and month. Nearest neighbor matching estimator uses four matches per observation with exact matches required
for year sold with robust standard errors. Percentage effects for the dune dummy are calculated following Halvorsen & Palmquist (1980). Temporal
restrictions are centered on the dune intervention in Surf City. *** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant
at the 10 percent level.
Coefficient
Standard Error
95% C.I.
Percentage Effects
Panel A. Alternative Estimators
Oaxaca-Blinder
0.036***
0.011
0.014
0.057
3.62
DID
0.051***
0.018
0.016
0.085
5.20
NNM
0.062*
0.036
-0.008
0.132
6.34
Panel B. Alternative Functional Form (Linear Box-Cox)
Oaxaca-Blinder
0.033***
0.002
0.030
0.037
3.36
DID
0.053***
0.003
0.046
0.059
5.44
NNM
0.015
0.020
-0.024
0.054
1.51
Panel C. Oaxaca-Blinder Spatial Restriction Models
Within 1 Mile
(N=2,481)
0.034***
0.009
0.017
0.051
3.46
Within ½ Mile
(N=1,560)
0.034***
0.010
0.015
0.054
3.45
Within ¼ Mile
(N=778)
0.030*
0.018
-0.005
0.064
3.04
Exclude ¼ Mile
(N=4,151)
0.030***
0.011
0.007
0.052
3.04
Exclude 500 feet
(N= 4,635)
0.042***
0.011
0.021
0.064
4.28
Panel D. Oaxaca-Blinder Temporal Restriction Models
2004 – 2009
(N=2,136)
0.040***
0.014
0.013
0.067
4.09
2005 – 2008
(N=1,450)
0.045***
0.017
0.011
0.078
4.55

52
Table 6: Spatially-Delineated Effect of Dune Construction
Notes: Oaxaca-Blinder models are estimated separately for each spatial proximity band. The DID model is estimated as a single model with spatial proximity band interacted with the policy
indicator. *** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level
Oaxaca-Blinder
(Separate Model for Each Spatial Band)
Difference-in-Differences
(Single Model; N= 4,827)
Observations
Coefficient
Std. Error
Coefficient
Std. Error
Dune*Oceanfront
281
0.043***
0.015
0.127**
0.061
Dune*Oceanfront Block
804
0.066***
0.014
0.163***
0.022
Dune*Second Block
2,517
0.030**
0.013
0.039*
0.041
Dune*Third Block
632
-0.033
0.024
0.022
0.033
Dune*Bayfront
593
-0.036
0.040
-0.064
0.080
Aggregate Policy Effect
(from previous models
for comparison purposes)
4,827
0.036***
0.012
0.051***
0.018

53
Table 7: Decomposition of the Average Treatment Effect of Dune Construction
*** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level
Model (1)
Dune Indicator
Model (2)
Dune Area
Model (3)
Spatially-Restricted Dune Area
Coefficient
Std. Error
Coefficient
Std. Error
Coefficient
Std. Error
Treated*Dune Indicator
0.1893*
0.1088
-
-
-
-
Treated*Dune Area (ft2)
-
-
0.0005*
0.0003
0.0007**
0.0003
Treated*Lateral Distance (ft)
-
-
0.00002
0.00001
0.00001
0.00001
Treated*Ocean View (°)
-0.0042***
0.0011
-0.0044***
0.0012
-0.0042***
0.0011
Treated*Beach Width (ft)
-0.0007**
0.0003
-0.0007**
0.0003
-0.0007**
0.0003
Dune Area
-
-
-0.0001
0.0002
-0.0003
0.0002
Lateral Distance
-
-
-6.66e-06***
1.56e-06
-4.21e-06**
1.87e-06
Ocean View
0.0029***
0.0006
0.0029***
0.0006
0.0039***
0.0007
Beach Width
0.0003***
0.0001
0.0003***
0.0001
0.0004***
0.0001
Spatial Proxmity FE
X
X
X
Neighborhood FE
X
X
X
Year and Month FE
X
X
X
Observations
4,827
4,827
3,602
R-Squared
0.806
0.806
0.771

A54
Online Appendix
Appendix A. Climate Adaptation Example
I present a modification to the stylized model of adaptation valuation in Guo and
Costello (2013) to demonstrate a key difference related to public adaptation.
Assume a social planner seeks to maximize the net benefits of a function of housing
values H in response to a policy change:
**
( ) max ( ,g, ) ( ( ), [ ( )], )
d
H f z f z g z
    

(A.1)
where

is an exogenous environmental parameter (e.g. erosion), z is a discrete
choice policy variable with two outcomes,
01
{ , }z z z
, g is a private good (e.g.
ocean views) impacted by the change in z, and
*()z

represents the optimal policy
given

. I follow Guo and Costello (2013) by assuming the function f is strictly
concave in

and g, differentiable in

and g and the envelope function H is
continuous and differentiable. A small change in

impacts the value of adaptation
directly through the optimal choice of z and indirectly through the impacts on g
resulting from that optimal choice.
*
() '[ ( ( )]
dH f dz g z dz f
f g z
d z d z d

    
   
    
   
(A.2)
If some threshold level of erosion (

) is not reached, the policy is not
implemented and housing values are only affected by the change in erosion (
f



).

A55
If the change in erosion crosses a threshold and results in constructed dunes, the
value associated with the policy is now impacted by the value of adaptation (
f dz
zd



) in the form of storm protection benefits and the ancillary benefits or costs
(
*
'[ ( ( )] g z dz
f g z zd




). This simple extension highlights the potential for bias
resulting from interpreting the value of adaptation policy as the value of adaptation
if the ancillary impacts are unaccounted for. The empirical analysis in this paper
demonstrates that both of the post-policy impacts are economically and statistically
significant and provides a clear connection to this simple analytical model.
Appendix B. Development of Viewshed Algorithm
High-resolution (1’ pixels) color orthophotos are utilized to construct building
footprints and centroids for all transacted homes in the sample. Coastal LiDAR
elevation data from 2005 and 2010 were then obtained from NOAA’s Digital Coast
Center. LiDAR is a remote sensing technology that can measure different types of
elevation by beaming lasers from low-flying aircraft and analyzing the reflected
light. The data provide both baseline and first-returns digital elevation models
(DEM) for Long Beach Island smoothed with an inverse distance weighting
algorithm at 6’ spatial resolution. The baseline DEM depicts the elevation of the
barrier island and the first-returns DEM represents the top of all buildings and
vegetation on the island. The 2005 first-returns LiDAR provides a pre-dune

A56
baseline for the analysis since construction on the first dune did not begin until
September 2006. The most recent LiDAR data available (2010) contains both the
Surf City and the Harvey Cedars dune. The Brant Beach dune, built in 2012, was
incorporated into the 2010 first-returns DEM raster in ArcGIS using the concept
burn streams into DEM to complete the post-dune DEM for analysis. This process
is designed to add decrements in DEMs for streams and other water features. I
simply altered the decay coefficient algorithm to instead add increments of
elevation to the shoreline in Brant Beach where the dune was eventually
constructed. The process can be characterized by the following equation:
DE = E + (G / (G + D))k × H (B.1)
where DE is the newly calculated elevation representing the dune, E is the old
elevation from the DEM, G is the grid resolution, D is the distance from the dune
peak, k is the decay coefficient, and H is the elevation increment. An appropriate
facsimile of the Brant Beach dune was generated with k = 2 for the decay coefficient
and H = 10 for the average increase in the shoreline height from the construction of
the dune.
An iterative geo-processing algorithm utilizes the data described above to capture
the degree of ocean view from each home. For each parcel in each time period, the
building footprint is zeroed out to the baseline elevation. This step ensures that the
observer “sees” past the confines of the house. Then, an observation point is defined
for both the first (10’) and second floor (20’) of the house. The tool then determines

A57
the amount of the Atlantic Ocean that can be seen from each observation point
across the first-returns DEM, with a maximum view of 180°. This view is then
calculated with the following formula:
View° = ArcLength/π * 180° (B.2)
where ArcLength is the value returned by the tool capturing the arc of a circle in
the ocean that is visible from each observation point. Once both views are recorded,
the algorithm then replaces the building footprint in the first-returns DEM and
moves on to the next parcel.
Figure B.1 provides an example of the algorithm results for first floor view from
an oceanfront parcel in Surf City. The results indicate a nearly 62° reduction in this
view as a result of the dune construction in 2006.

A58
Figure B.1. First Floor Viewshed from a Surf City Oceanfront Parcel
Notes: The red dot is the centroid of the housing footprint for the oceanfront parcel. The half circle
outlined in black represents extent of possible ocean view from the first floor of that specific parcel.
Yellow areas represent the area of the ocean visible to an observer on the first floor of the home.
The degree measures of ocean view shown above were calculated as follows: View° = Arc Length/π
* 180.
2005: Before the Dune
2010: After the Dune
47.3°
109.1°

A59
Appendix C. Additional Figures and Tables
Table C.1: Damage and Federal Aid for Sandy Relief on Long Beach Island
Notes: Surf City, Harvey Cedars received dunes and beach replenishment prior to Sandy. Barnegat Light has a substantial natural dune system and data is provided for
comparison purposes Damages are considered minor under $8,000. Damages are considered major between $8,000 and $28,800. Damages are considered severe above
$28,800. NFIP does not break down payments by storm, but discussion with local officials confirms a majority of these payments were related to Sandy.
Source: Data in Panels A and B from the NJ Department of Community Affairs. Panel C data from NFIP: http://bsa.nfipstat.fema.gov/reports/1040.htm#34
Long Beach
Township
Harvey
Cedars
Surf City
Ship
Bottom
Beach
Haven
Barnegat
Light
Panel A. Housing Damages related to Sandy
Housing Units
8418
1218
2583
2031
2549
1252
Homes Damaged
810
64
608
829
260
17
Rentals Damaged
198
12
55
110
109
2
Percent Impacted
12.0
6.2
25.7
46.2
14.5
1.5
Major Impact (%)
8.5
5.0
2.0
4.0
10.2
1.2
Severe Impact (%)
2.0
0.7
0.6
2.5
2.2
0.2
Panel B. Non-NFIP Federal Funds Distribution for Sandy Relief
Total Federal
Funds
$11,088,034
$1,269,886
$2,092,094
$7,364,868
$12,972,642
$549,457
per capita
$3,618.81
$2,475.41
$1,689.90
$6,707.53
$10,892.23
$988.23
Panel C. National Flood Insurance Data (1/1/78 to 5/31/14)
Total Loss
Claims
7,519
857
1,381
1,616
2,609
238
Total Payments
$214,831,368
$10,188,404
$20,877,434
$61,387,954
$83,761,839
$2,589,196

A60
Table C.2: Results of Falsification Tests
Note: Models are specified in the exact same manner as the preferred models. All treated observations are removed from the
analysis. Treatment is falsely assigned to three communities that share a border with the treated communities (Ship Bottom,
Loveladies, and Brighton). Both the falsification test with the full OB and DID models and the amenity decomposition DID
model do not indicate any statistically significant spillover effects into control communities.
Falsification Test
Full Model
Falsification Test
DID decomposition model
Coefficient
Std. Error
Coefficient
Std. Error
Dune Treatment (OB)
-0.013
0.010
Dune Treatment (DID)
-0.025
0.020
Treated*Dune Area
-
-
-0.0002
0.0003
Treated*Lateral Distance (ft)
-
-
-6.08e-06
4.02e-06
Treated*Ocean View (°)
-
-
0.0002
0.0015
Treated*Beach Width (ft)
-
-
0.0003
0.0002

A61
Table C.3: Reference Coefficients for Oaxaca-Blinder Model
*** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level
Treated Group (
1
ˆ

)
N= 357
Control Group (
0
ˆ

)
N=4,470
Reference (
*

)
Coefficient
Std.
Error
Coefficient
Std.
Error
Coefficient
Std.
Error
# of Bedrooms
0.048
0.047
0.160***
0.018
0.152***
0.018
Rooms2
0.004
0.005
-0.018***
0.002
-0.017***
0.002
# of Bathrooms
0.077
0.051
0.103***
0.014
0.103***
0.014
Baths2
-0.021***
0.008
-0.007***
0.002
-0.007***
0.002
Square Footage
0.00005***
0.036
0.0002***
0.000
0.0002***
0.000
SqFeet2
-1.0e-07***
2.76e-08
-7.6e-09***
8.3e-10
-7.1e-09***
9.3e-10
Lot Size (sq. ft)
0.033***
0.002
8.7e-06***
1.7e-06
8.7e-06***
1.7e-06
LotSize2
-1.3e-09
1.3e-09
3.7e-10***
9.1e-11
3.5e-10***
9.1e-11
Age of home
-0.009***
0.002
-0.007***
0.001
-0.007***
0.001
Age2
5.9e-05**
2.5e-05
0.000***
5.9e-06
0.000***
5.8e-06
Oceanfront
0.420**
0.203
0.300***
0.047
0.309***
0.047
Oceanfront Block
0.424***
0.105
0.076**
0.030
0.088***
0.030
Second Block
0.068
0.083
0.068***
0.023
0.061***
0.023
Third Block
0.090*
0.048
0.045***
0.016
0.045***
0.016
Bayfront
0.332***
0.073
0.329***
0.019
0.329***
0.019
Condo
-0.304*
0.166
-0.440***
0.024
-0.449***
0.023
Garage
0.042*
0.024
0.046***
0.008
0.048***
0.008
Hot Tub
0.113**
0.045
0.029**
0.012
0.030**
0.012
Fireplace
0.050*
0.027
0.030***
0.007
0.035***
0.007
Dist. to Ocean (ft)
0.0003
0.0004
-0.0000***
0.000
-0.0000***
0.000
Dist. To Bay
-0.0003***
0.0001
-0.0001***
0.000
-0.0001***
0.000
Dist. to Public Access
-0.0002
0.0003
-0.0000
0.000
-0.0000
0.000
Dist. to Comm. Distr.
0.0001**
0.0000
0.0002***
0.000
0.0001***
0.000
Quarter & Year FE
Y
Y
Y
Neighborhood FE
Y
Y
Y
Flood Zone FE
Y
Y
Y
Constant
13.712***
0.554
12.414***
0.085
12.468***
0.084

A62
Figure C.1. Transactions Prices in Surf City and Ship Bottom: PCA Signing
Notes: Black line is Surf City and grey line is Ship Bottom. The trend lines are estimated separately for the time periods
before and after the signing of the Project Cooperation Agreement between USACE and the NJDEP using nonparametric
regressions with a tri-cube weighting function and a bandwidth of 0.5. The vertical line represents the date the Project
Cooperation Agreement was signed (August 2005).

A63
Figure C.2: Timeline of Important Events Related to Dune Construction on LBI
Sept. 1999
Oct. 2006
April 2010
October 29, 2012
Feb. 2007
Mar. 2012
June 2012
USACE Feasibility
Report Issued
Surf City
Dune Construction
Post-Tropical Cyclone
SANDY
Harvey Cedars
Dune Construction
Brant Beach
Dune Construction
Aug. 17, 2005
Project Cooperation
Agreement Signed by
USACE & NJDEP
Sept. 30, 2011
Dredging Contract
signed for Brant
Beach Project
July 15, 2008
Harvey Cedars
Approves use of Eminent
Domain on Easement Holdouts
June 16, 2006
NJDEP sues holdouts in
Surf City, paving way for
project to begin
Sept. 2009

A64
Figure C.3. Price Trends in Brant Beach v. Control Communities
Notes: Black dots indicate individual housing transactions by month. Linear price trends for the two primary treated communities compared to all control communities
both before and after dune construction activities. Year dates on the horizontal axis indicate the beginning of each year (i.e. 2011 = January 2011, etc.). Shaded areas
represent dune construction activities and transactions during this time are removed to avoid any impacts this activity may have on home prices. Dotted lines are added
to the shaded area to demonstrate the trend from pre-dune to post-dune periods in each group.