Weathering the storm: measuring household willingness-to-pay for risk-reduction in post-Katrina New Orleans.
Landry, Craig E. ; Hindsley, Paul ; Bin, Okmyung 等
1. Introduction
Hurricane Katrina made landfall on the Louisiana-Mississippi border
of the Gulf Coast on August 29, 2005, leaving behind widespread
devastation on the Alabama, Mississippi, and Louisiana coasts. Although
the eyewall of Katrina did not pass directly over New Orleans, wind
driven waves and storm surge breached several points in the levee
system, demonstrating that the city was ill equipped for a storm of
Katrina's magnitude. Insufficient artificial and natural storm
protection, in conjunction with New Orleans's highly vulnerable
physical and human geography, contributed to devastation throughout the
city.
The Louisiana Coastal Protection and Restoration Plan (LACPR 2009)
and Mississippi Coastal Improvements Program (MsCIP 2009) were created
in response to a U.S. Congressional directive to develop plans for
hurricane risk-reduction and coastal restoration in both Louisiana and
Mississippi. The LACPR Plan Formulation Atlas considered measures that
could be combined to form an exhaustive 200 million alternatives. The
final technical report presents four to six alternatives for each of
five planning units. The plans consider "Category 5"
hurricanes and storm surge resistant levees, the mitigative role of
coastal landscapes, livable communities, cultural resources, and risk.
Our study investigates a general and limited set of options.
We examine individuals' willingness-to-pay (WTP) to reduce
flood risk in New Orleans through application of a stated preference
choice experiment. In so doing we offer a different perspective to LACPR
in a fairly simple framework. We give a measure of the public will (both
national and local) to protect human and physical capital in this
vulnerable location. The choice experiment focuses on hypothetical
projects that propose funding lines of defense in the form of coastal
restoration and Category 5 levees, as well as modernizing existing
transportation networks in New Orleans. Through the application of a
stratified sampling procedure, we investigate rebuilding preferences for
individuals in the New Orleans metropolitan area and U.S. tax-payers in
general.
Our results indicate that levee flood protection designed to
withstand a Category 5 storm is the most highly valued rebuilding
feature. New Orleans metropolitan area residents are willing to pay a
substantial amount, while the average U.S. household is willing to pay
more. Surprisingly, WTP for coastal restoration was not statistically
significant for the New Orleans or U.S. samples but was significant for
a model that combines the two samples. A latent class model reveals that
households who view coastal restoration as an important part of
rebuilding New Orleans and have higher income are willing to pay for
coastal restoration, while those that do not see coastal restoration as
important and have lower income are not willing to pay. New Orleans
metropolitan area residents are willing to pay for modernized
transportation in the New Orleans metropolitan area, while the average
U.S. household is not. Again, the latent class model reveals some
differences in economic value across groups, with higher income U.S.
households who view coastal restoration as important harboring a
negative WTP for improvements in transportation.
2. Background
In the aftermath of Hurricane Katrina, the public has been forced
to make difficult decisions concerning how to rebuild. The geographic
and social vulnerabilities of New Orleans contribute to the complexity
of determining how government will allocate public funds for rebuilding.
There was an estimated $10 billion in damage to roads, bridges, and the
utility system in New Orleans alone. In Orleans Parish, 134,344 housing
units (71% of the housing stock) were damaged, making rebuilding no
small feat. New Orleans borders water on three sides, making protection
a significant task. Also, much of New Orleans lies below sea level,
essentially making the city a bowl between Lake Pontchartrain and the
Mississippi River. When the levees fail, as they did after Katrina, this
bowl can fill up, leaving much of the city underwater. New Orleans
relies heavily on a system of levees and pumps that hold back Lake
Pontchartrain and the Mississippi River and remove water when it enters
the bowl. (1) Clear evidence of this vulnerability is the 27 major
flooding disasters that have occurred in New Orleans over its roughly
300-year history (Kates et. al 2006).
During Katrina, over 80% of New Orleans was flooded, largely as a
result of failed levees. A preliminary analysis by the University of
California at Berkeley and the American Society of Civil Engineers
determined that these levees failed before they were overtopped,
indicating design failure (Seed et al. 2006). The potential damage from
a major hurricane had received considerable attention from the media and
academics prior to Katrina. (2) Unfortunately, there was insufficient
political will to heed these warnings and protect the city in time. The
existing system did not perform up to its projected Category 3
storm-protection standard.
There are a number of reasons why federal, state, and local
governments failed to adequately fund levees and other flood protection
measures. The U.S. Army Corps of Engineers faced cost increases and
design changes stemming from technical issues that limited their ability
to fund new construction projects. A Corps fact sheet from May 2005
stated that the appropriated funds for fiscal year 2005 were
insufficient to cover new construction projects, including levee
enlargement to enhance protection in the New Orleans metropolitan area.
In addition, socio-political issues, including environmental concerns,
legal challenges, and local opposition to some aspects of the flood
management plan, complicated initiation and completion of some projects
(U.S. GAO 2005). The contentious environment surrounding levee
maintenance and augmentation combined with the high price tag limited
initiative to address flood hazard in New Orleans, not only for
President Bush but also previous administrations. Kunreuther and Pauly
(2006) refer to this phenomenon as the not in my term of office
syndrome.
In addition to man-made structures, natural coastal features such
as wetlands and barrier islands provide additional storm protection for
coastal regions. Previous estimates from Hurricane Andrew suggest that a
kilometer of coastal marsh can reduce storm surge by roughly 7.9 cm
(Lovelace 1994). Louisiana has experienced significant losses of coastal
wetlands. Hurricanes Katrina and Rita destroyed 217 square miles of
coastal wetlands in a single season. The destruction of coastal wetland
in the New Orleans area due to the single event of Katrina would
normally be expected to take a span of 50 years (LACPR 2009). While
flooding from Katrina was largely the result of failed levees, degraded
coastal wetlands played a significant role in the disaster.
The degradation of Louisiana's coastal environment stems from
individual and government action at various levels within the
Mississippi River basin. Kousky and Zeckhauser (2006) term the
associated losses in ecosystem services as JARing actions (where JAR
stands for jeopardized assets that are remote). The construction of
levees, jetties, and canals in the Mississippi River basin significantly
changed sediment transport in the system. Alterations in sediment
transport have starved wetlands (Turner 1997). In addition, land
subsidence, either naturally or due to hydrocarbon extraction, and
rising sea levels threaten low lying coastal areas (Morton et al. 2002).
Decreased sediment flow and resource extraction have imposed external
costs on New Orleans and other Gulf Coast cities in the form of a
degraded natural environment and reduced storm protection.
In addition to natural and man-made flood protection, transit and
highway infrastructure play a key role in evaluating the vulnerability
of coastal populations. The capacity and resilience of transit and
highway infrastructure affect how successfully transit can be used in
emergency evacuation and disaster response. In a special report, the
Transportation Research Board (2008) recommended that "Federal
funding should be provided for the development of regional evacuation
plans that include transit and other public transportation
providers." Further, public transit fills a unique role in
providing a mode of evacuation for populations that are
transit-dependent and may require special assistance.
3. Preferences for Rebuilding New Orleans
The main purpose of this article is to evaluate individual
preferences for the reconstruction of New Orleans. The rebuilding plans
constitute a series of local public goods; we estimate individual WTP
for these public goods. Since many decisions have yet to be made on
restoring New Orleans, we employ hypothetical choice experiments (CEs)
to assess preferences for rebuilding. CEs are a stated preference method
that can be used to value the characteristics of rebuilding projects. In
a CE, subjects are asked to express a preference over several
alternatives. Each alternative is characterized by an array of
attributes that describe the alternative. The levels of each attribute,
for example, the number of acres of wetlands restored under a particular
rebuilding plan, can vary across alternatives, and each choice can
include a status quo or "no choice" option. The attributes
that describe each alternative and the levels that each attribute can
take are chosen by the researcher to address the valuation question at
hand. By observing respondents" choices over a number of choice
sets, we can learn about the tradeoffs individuals are willing to make
in terms of a rebuilding plan for New Orleans.
Our principal sample is composed of New Orleans metropolitan area
households the primary beneficiaries of rebuilding efforts. We employ a
random digit dialing survey that uses paired comparisons--status quo
rebuilding plan versus an alternative that can exhibit improvements in
flood control, coastal restoration, and/or transportation
infrastructure. The paired comparison approach was deemed necessary
because visual aids were difficult to employ with a telephone survey. By
focusing on status quo versus an alternative in each choice set, we
minimize the amount of information that respondents must process, since
the status quo was constant across all choice sets. We use an
experimental design that allows us to maximize statistical performance
while maintaining task simplicity. In addition to the New Orleans
subjects, we also gathered choice data from a sample of U.S. households.
Experimental Design
Our choice experiment investigates rebuilding options using four
primary attributes: (i) levee augmentation, (ii) coastal restoration,
(iii) transportation system improvements, and (iv) a funding mechanism
in the form of a one-time increase in federal income tax payments. As
indicated in Table 1, each program attribute has two levels, while the
tax attribute has four levels. The initial level of each program
attribute is described as the status quo level in order to facilitate
the pairwise choice design. Similar to previous work in the
environmental literature (Adamowicz, Louviere, and Williams 1994;
Adamowicz et al. 1998; Layton and Brown 2000; McGonagle and Swallow
2005: Ladenburg and Olsen 2008), the choice experiment focuses on
preferences for public goods--in our case, this is rebuilding or
improving public works rather than preferences for private goods, such
as funds for rebuilding private property (which would primarily benefit
individual households and businesses). We focus on public projects that
decrease existing vulnerabilities (levee augmentation and coastal
restoration) or enhance evacuation possibilities (improvements in
transportation infrastructure). Examples of conjoint choice sets can be
found in Appendix A.
Respondents were given a choice between two levels of flood
protection. The status quo option was to ensure that all levees were
capable of withstanding the wind, waves, and storm surge that would
accompany a Saffir-Simpson Category 3 storm. The alternative option
would fortify all levees to be capable of withstanding the wind, wave
action, and storm surge consistent with a Saffir-Simpson Category 5
hurricane. (3) By congressional mandate, the LACPR offers multiple
planning options capable of providing this level of protection. As such,
we chose to focus on this level of storm protection, which will provide
a sense of the magnitude of the maximum benefits that storm protection
could provide. This estimate would be an upper bound on other levels of
storm protection, all else being equal.
The choice sets include an option for restoration of
Louisiana's coastal wetlands. The status quo option is no coastal
restoration, and the alternative is to invest in restoring coastal
wetlands. Improvements in coastal wetlands would provide additional
protection against hurricane force winds and storm surge. In addition,
restoring coastal wetlands would provide for additional environmental
benefits, such as fisheries habitat and other ecosystem services. These
additional benefits were not noted in the survey, but we suspect that
many coastal residents are aware of these additional benefits.
The survey also asked respondents to consider improvements in New
Orleans's transportation infrastructure. The status quo option
entails limited bus service, street cars, and conventional roads. The
alternative is modernized transportation infrastructure that includes
expanded bus and light rail service and improved road networks. The
modernized transportation system would provide for improved transit
through the city on a day-to-day basis and would enhance the ability of
citizens to evacuate in the event of a hurricane.
The payment vehicle was a compulsory, one-time increase in federal
income tax payments for all U.S. households. The status quo was provided
at zero additional cost, while the tax payment associated with the
alternative varied at $50, $150, $300, or $450 per household. The survey
explicitly states that all money raised by this one-time tax would go
directly to rebuilding projects in New Orleans and restoration projects
in coastal Louisiana.
Hypothetical bias is a potential limitation of our CE research
method. This bias can arise within a stated preference framework due to
the hypothetical nature of the exercise; lacking real incentives for
choice, subjects may not be sufficiently motivated to expend cognitive
effort to search their preferences. Evidence of hypothetical bias in CEs
is mixed (Carlsson and Martinsson 2001: Lusk and Schroeder 2004;
Johansson-Stenman and Svedsater 2008). Lusk and Schroeder (2004) find
suggestive evidence that CEs are capable of producing unbiased estimates
of marginal willingness-to-pay (MWTP), while there may be bias in
estimation of total WTP. There is evidence that hypothetical bias can be
attenuated through application of a "cheap talk" script, which
focuses respondent attention on the phenomenon of bias and encourages
them to respond as if the exercise were real (Carlsson, Frykblom, and
Lagerkvist 2005; List, Sinha, and Taylor 2006). We, thus, employ a
variant of cheap talk that is similar to the original language in
Cummings and Taylor (1999) but shortened to fit within the context of a
telephone survey and changed to reflect differences in the nature of the
good being valued. The cheap talk script is included in Appendix B.
In applying a CE, the researcher designs the profiles of
alternatives that are shown to subjects (i.e., deciding which levels of
attributes are to be combined in a single alternative). These profiles
and how they are combined define the choice sets that individuals will
consider when participating in the experiment, and they determine the
matrix of independent variables that are used in analysis of the CE data
(described below). As such, the design of profiles influences the
efficiency of parameter estimates. With our proposed attributes and
levels, a full factorial design has 32 alternative profiles ([2.sup.3] x
4 = 32). The full factorial design, however, includes options that would
be dominated by the status quo (e.g., status quo conditions at zero vs.
positive price). As such we chose only a fraction of the full array of
possible profiles, restricting the dominated options from consideration;
for our problem, fully efficient designs (i.e., those that minimize
variance of parameter estimates) for linear models can be constructed
with 8 or 16 alternative profiles. We chose 16 profiles, which
represents a fractional factorial design from which main effects can be
estimated. We follow Huber and Zwerina (1996) in constructing a linear
experimental design that is orthogonal (levels of each attribute vary
independently of one another so that attribute levels are not
correlated) and balanced (levels of each attribute appear with equal
frequency). We employ SAS macros %MktEx and %ChoiceEff to design an
efficient fractional factorial design of 16 pairwise choice sets
(Kuhfeld 2010). In all choice sets, the status quo at zero additional
tax is offered against an alternative plan that has at least one
improvement in program attributes and a higher tax.
Since our econometric model, however, is non-linear we cannot claim
that our design is in fact fully efficient (which would require advance
knowledge of unknown parameters). Huber and Zwerina (1996) claim that
using linear designs for choice experiments is a reasonable approach in
situations for which no prior knowledge of parameter estimates is
available. In order to lessen the burden on subjects, we use a blocked
design of the 16 choice sets, employing only four choice sets per
respondent. The %MktBlock SAS macro was used to efficiently partition
our 16 choice sets into four blocks of four choice sets (status quo vs.
alternative). An example of one of the blocks is included in Appendix A.
The sequencing of the choice sets within each block was alternated
across respondents in order to control for order effects, producing a
total of 16 choice sets--four blocks of four choice sets, each with four
sequences.
Survey Questionnaire and Administration
Our survey targeted two populations, residents of the New Orleans
metropolitan statistical area (MSA) and U.S. residents not in the New
Orleans MSA. Each survey had three primary sections, and we estimated it
would take between 10 and 15 minutes to administer. We conducted a
series of focus groups to pretest the survey instrument. The focus
groups were composed of subjects from various racial, ethnic, and
socio-economic backgrounds, and individual responses to survey questions
were noted and explored in an effort to learn how subjects may interpret
questions. The first section of the New Orleans survey elicits
information concerning the respondent's family attachment to New
Orleans, whether the respondent experienced Hurricane Katrina, and
whether this event and the aftermath would influence their decision to
stay in the area. The first section includes a series of Likert-scale
questions designed to assess the subjects' perceptions of various
attributes of the rebuilding plan, including the importance of crime
control, housing availability, job creation, flood protection, coastal
wetland restoration, improved transportation, and cultural preservation.
For the U.S. survey, the first section gauges individuals'
familiarity and experience with New Orleans, in addition to the
assessment of perceptions of the importance of rebuilding factors. The
second section of the survey administers the choice experiment. Our
blocked experimental design offered four choices to each respondent,
with subjects choosing either the status quo at $0 additional federal
taxes per household or an alternative scenario that offers improvements
in the rebuilding plan in exchange for a one-time payment of additional
federal taxes for each U.S. household. Subjects were instructed to treat
each choice set as if it were an independent referendum that should be
considered in isolation from the other choices. In each survey, we
precede the four hypothetical choices with a cheap talk script (see
Appendix B). The third part of the survey elicits information on
socio-demographic factors, including sex, ethnicity, whether the
respondent considers her/himself Latino or Cajun, level of education,
employment status, age, income, and household size.
4. Data
Our sample was collected via a stratified random digit dial (RDD)
of telephone numbers in the New Orleans MSA and other U.S. households.
The survey was administered between May 2007 and June 2008 by
individuals in East Carolina University's Community Research Lab.
Postcards were sent to mailing addresses associated with the phone
numbers, and those returned as undeliverable were eliminated from the
sample. Calls were placed, and non-working numbers and ineligible
numbers (businesses, fax numbers, etc.) were also eliminated. After this
process, there were roughly 500 eligible phone numbers located in the
New Orleans MSA. An equal number of eligible phone numbers were located
in the rest of the United States.
Successful contact rates were low for the New Orleans MSA; this
likely reflects displaced households. Contact was established with 298
households in the New Orleans MSA compared with 355 in the rest of the
United States. The final dataset includes information from 128
households in the New Orleans MSA and 220 U.S. households not in the New
Orleans vicinity. The corresponding response rates are 25.6% for the New
Orleans MSA and 44% for the U.S. sample. Once contact was established
with the household, the cooperation rates were 43% and 62%,
respectively. Because of incomplete information, only 120 households in
the New Orleans MSA and 217 U.S. households not in the New Orleans
vicinity are used in the choice models.
The potential biases in all telephone surveys are magnified in the
wake of a disaster like Katrina (Galea et al. 2008: Kessler et al.
2008). Neighborhoods housing the poorest, least educated residents
usually suffer the most damage and take the longest to recover essential
services like telephones. Relocation within the city creates additional
challenges. RDD samples help address these issues, but potential bias
remains. In an effort to address potential response bias, we develop a
weighting scheme to adjust data to match characteristics from the 2006
American Community Survey (U.S. Census Bureau). Our inverse probability
weights are based on observable demographic factors--sex, race, Latino
status, education level, marital status, and income. Table 2 depicts the
weighted and unweighted descriptive statistics for the New Orleans and
U.S. strata. We estimate choice models for both strata and combine the
strata in order to estimate a single model, applying weights so that the
results reflect population proportions.
The average New Orleans respondent had been living in the
metropolitan area 41 years, and 76% of households contacted have at
least one set of parents from the New Orleans area. Eighty-one percent
were in New Orleans just before Hurricane Katrina struck. Thirty-two
percent of households have considered leaving New Orleans in the wake of
the disaster, with 22% indicating they are very likely or somewhat
likely to leave. Turning to the U.S. respondents, about 7% indicated
that they have visited New Orleans, and 15% responded that they either
visit on a regular basis or plan to visit in the future. Eleven percent
of U.S. respondents indicated that they have friends or family in the
New Orleans area.
Tables 3-5 report results on individual perceptions of the
importance of various factors in the rebuilding plan for the New Orleans
and U.S. samples in the form of a weighted frequency table. Our results
indicate that individuals in both samples believe that flood protection
is very important, but a higher proportion of individuals in the New
Orleans sample feel that both coastal wetland restoration and improved
transportation are very important.
5. Methods
We use the random utility model (RUM) as a theoretical basis for
our choice experiment. We assume that individuals choose rebuilding
projects for New Orleans that yield the highest level of utility.
Individual n's utility associated with a choice i in choice set t,
denoted [U.sub.nit], is a function of project characteristics,
[x.sub.nit], and associated cost, [c.sub.nit]. Utility can be decomposed
into an observable portion, [V.sub.nit]([x.sub.nit],[c.sub.nit],
[??][??]), and an unobservable portion known only by the subject,
[[epsilon.sub.nit]:
[U.sub.nit] = [V.sub.nit] (c.sub.nit], [x.sub.nit]; [??][??]) +
[[epsilon].sub.nit], (1)
where [??] and [??] are unknown parameter vectors to be estimated.
The probability of individual n choosing a project i over other choice j
in set t is, thus,
[P.sub.nit] = Pr[[V.sub.nit]([c.sub.nit], [x.sub.nit]; [??], [??])
+ [[epsilon].sub.nit] [greater than or equal to] [V.sub.njt]
([c.sub.njt,] [x.sub.njt]; [??], [??]) + [[epsilon].sub.njt]]. (2)
We assume the observable portion of utility is additive:
[V.sub.nit]([c.sub.nit], [X.sub.nit]; [??], [??]) = [??]cnit, +
[??][x.sub.nit]). (4) Under the assumption that the error terms in
Equation 2, [[epsilon].sub.nit] are independent and identically
distributed (i.i.d.) extreme value variates for all n, i, and t, the
choice probabilities take the closed-form expression [P.sub.nit] = exp
([??][c.sub.nit] + [??][x.sub.nit]) / [summation over (j)]exp
([??][c.sub.njt] + [??] [x.sub.njt]). (3)
Under this pooled logit formulation, the multinomial logit (MNL)
model can be used to estimate the normalized unknown parameters, [alpha]
= [??]/[sigma] and [beta] = [??]/[sigma], where [sigma] is the scale
parameter of the extreme value distribution.
It is widely recognized, however, that MNL incorporates taste
variation in a potentially restrictive manner, limits substitution
patterns, and does not allow for correlation across repeated individual
choices. Thus, in our application of RUM, we employ the repeated mixed
logit (RXL) model (Train 1998; Herriges and Phaneuf 2002) and the latent
class (LC) or finite mixture model (Train 1998; Boxall and Adamowicz
2002), each of which incorporates unobserved individual heterogeneity by
allowing the [alpha] and/or [beta] parameters to vary within the sample.
The variability of utility parameters incorporates taste heterogeneity,
provides for more complex substitution patterns, and allows correlation
across individual choices.
For the RXL model, the [epsilon]nit are i.i.d, extreme value
variates for all n, i, and t, and the choice probabilities for any
period t are conditional on an individual-specific vector
[[beta].sub.n]. Including an alternative specific constant for the
status quo alternative, the conditional choice probabilities are given
by
[P.sub.nit]([psi],[alpha],[beta]) = exp([psi][d.sub.nit] +
[alpha][c.sub.nit] + [beta]'n[x.sub.nit]) / [summation over
(j)]exp([psi][d.sub.njt] + [alpha][c.sub.njt] +
[beta]'n[x.sub.njt]) (4)
where [d.sub.njt] = 1 for status quo, zero otherwise. We assume
[[beta].sub.n] ~ [phi]([beta]|[mu], [OMEGA]), where [phi] is a
multivariate normal probability density with mean [mu] and covariance
matrix [OMEGA]. (5) Since our experiment is designed to estimate main
effects, we restrict [OMEGA] to be diagonal; covariance parameters would
only be identified based on assumed functional form. Since
[[epsilon].sub.nit] are i.i.d, for all t, the conditional probabilities
for a series of choices i = {[i.sub.1,] ... [i.sub.T]} is given by the
product of Equation 4 across choice occasions:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (5)
Under the formulation of RXL, the unconditional choice
probabilities are
[P.sub.ni] = [integral]
[P.sub.ni]([psi],[alpha],[beta])[phi]([beta]|[mu],[OMEGA])dB. (6)
The likelihood function is the product of Equation 6 over all
individuals in the sample. The means of the [psi] and [alpha]
parameters, as well as the means and variance terms for [beta], are
recovered from simulated maximum likelihood estimates.
The LC model differs from the RXL in that it incorporates
unobserved individual heterogeneity through the use of discrete rather
than continuous mixing distributions. In this model, it is hypothesized
that individual-specific characteristics ([s.sub.n]) sort individuals
into K groups. Each group potentially has different preferences over
project choices, so that the probability of Equation 2 conditional on
membership in group k is
[P.sup.k.sub.nit] = exp([[psi].sub.k][d.sub.nit] +
[[alpha].sub.k][c.sub.nit] + [[beta]'.sub.k][x.sub.nit]) /
[summation over (j)] exp([[psi].sub.k][d.sub.njt] +
[[alpha].sub.k][c.sub.njt] + [[beta]'.sub.k][x.sub.njt]) [for
all]k. (7)
Since the unobserved errors are i.i.d, extreme values across t, the
conditional probabilities for a series of choices i = {[i.sub.1,]....
[i.sub.T]} by type k is given by the product of Equation 7 across choice
occasions
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (8)
Group membership is unknown to the researcher. The conditional
choice probabilities in Equation 8 are weighted by logit probabilities
for class membership, which take the form
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (9)
where the vector [s.sub.n], contains demographic variables that
influence class membership according to unknown parameters [delta]k.
Identification requires that parameters for one k [member of] K are
normalized to zero. The unconditional probability for a series of
choices by individual n is obtained by a weighted sum of Equation 8 over
the k groups, where the weights are given by Equation 9:
[P.sub.ni] = [summation over k[member of]K]
[[pi].sub.nk]([[delta].sub.k]) x [P.sup.k.sub.ni]
([psi],[alpha],[beta]). (10)
The parameters of the model in Equation 10 are estimated by maximum
likelihood.
We use compensating variation (CV) to measure the incremental
welfare change, also known as marginal willingness-to-pay (MWTP),
associated with program attributes for rebuilding New Orleans.
Conditional on [[beta].sub.nj], CV for a rebuilding program attribute j
is defined as
[CV.sub.nj] = [DELTA]xj([[beta].sub.nj]), (11)
for each j element of the vector x. Given the discrete nature of
program attributes, [DELTA][x.sub.j] = 1. For the RXL model, Equation 11
is simulated for all n respondents by taking R draws from the posterior
distribution of [3, calculating CV, and averaging across the R
calculations. For the LC model, Equation 11 is calculated for each of
the k segments (replacing [[beta].sub.nj] with [[beta].sub.kj]). Mean CV
can be calculated as the weighted average across segments, where the
weights are given by Equation 9. The Krinsky-Robb procedure (1986) is
used to produce standard errors of CV. Krinsky-Robb is a parametric
bootstrap method that takes random draws from the multivariate normal
distribution of parameters using information from the vector of
estimated parameters and the variance-covariance matrix. In our
application we take 10,000 random draws in order to develop both 90% and
95% confidence intervals of MWTP.
6. Results
The RUMs are estimated using Matlab and NLOGIT (Greene 2007). (6)
We estimate three models using the RXL estimator, corresponding with New
Orleans, United States, and combined datasets. Each model includes dummy
variables for projects with Category 5 levees, coastal restoration, and
modernized transportation system. For the U.S. and combined models, all
of these parameters are assumed to be drawn from a normal distribution
with diagonal covariance matrix. For the New Orleans sample, the
coefficient for the Category 5 levee and modernized transportation are
assumed fixed: estimated standard deviations for these parameters under
the assumption of normality were not statistically significant. (7) The
coefficients for the alternative specific constant representing the
status quo option and the tax variable are assumed fixed. Models were
estimated using maximum simulated likelihood based on 1000 Halton draws.
(8) Table 6 presents the parameter estimates for RXL choice models.
In each of the three models, the constant representing the status
quo is not statistically significant. As anticipated, the coefficient on
the one-time tax increase is negative and statistically significant at
0.1% chance of a type I error in each model. For each model, the
coefficient representing Category 5 levees is positive, implying that
individuals prefer projects that implement the maximum level of storm
protection. Each coefficient representing Category 5 levee protection is
statistically significant at the 1% level. Among project attributes,
Category 5 levee protection has the largest coefficient, indicating that
the average individual believes this project attribute is important
relative to other program attributes. Under the assumption of normality,
the standard deviation for this coefficient suggests that most
individuals exhibit positive preferences for this attribute, but
significant preference heterogeneity does exist for U.S. and combined
models.
In allowing for a random parameter for coastal restoration, the
standard deviation was found to be statistically insignificant for the
New Orleans model. Employing a fixed coefficient, the mean utility
effect for coastal restoration in the New Orleans models is
statistically significant (at the 10% level), and we estimate a positive
parameter. Results for U.S. and combined models suggest that utility
values for coastal restoration encompass both negative and positive
values. The mean coefficient for coastal restoration is not
statistically significant in these models, but the standard deviations
are statistically significant at the 1% level. We interpret these
results as indicating that some individuals in the broader population
value coastal restoration while others perceive it as something that
should not be funded through general taxation.
The coefficient for modern transportation is positive in each model
but statistically significant only for the New Orleans (5% level) and
combined (10% level) estimates. Since variability in the random
parameter was not statistically significant, the New Orleans model is
estimated with a fixed parameter. The estimated mean effects for New
Orleans and combined model are positive, as expected. In the combined
and U.S. samples the standard deviations for the distribution of
coefficients for modern transportation are statistically significant at
the 10% and 5% level, respectively. Much like coastal restoration,
results from the combined and U.S. samples indicate that some
individuals favor rebuilding projects with modernized transportation
while others favor projects without it.
In an effort to investigate determinants of preference
heterogeneity within our samples, we also estimated LC models for both
the New Orleans and U.S. samples. While these efforts were inconclusive
for the New Orleans sample, the approach did reveal potential sources of
variation in preferences among U.S. residents. We focus on a similar
specification for the LC model, with a status quo alternative specific
constant, a project tax variable, and indicator variables for Category 5
levees, coastal restoration, and modernized transportation systems.
Sociodemographic variables defining the finite mixture probabilities are
comprised of household income and the Likert-scale response indicating
the importance of coastal wetland restoration. Table 7 presents the
results of the latent class model for the U.S. sample. (9)
For the U.S. LC model, respondents are endogenously divided into K
= 2 groups, with posterior probabilities suggesting that roughly 35% of
the sample falls into the first group and the remaining 65% in the
second group. The class membership probability parameters indicate that
the first group views coastal wetland restoration as less important than
the second group. The negative sign on the income variable indicates
that the first group is represented by lower income households.
The status quo variable is positive and statistically significant
for the first group but insignificant for the second group. For each
group, the coefficient for Category 5 levees is positive and
statistically significant at the 1% level, implying that individuals in
both groups prefer projects that employ the maximum level of storm
protection. Individuals in the second group respond positively to
projects that include coastal restoration, while choices in the first
group were not affected by coastal restoration. The coefficient for
modern transportation was positive for the first group, but negative for
the second group (in both cases statistically significant)! Lastly, the
coefficient on tax is negative and statistically significant at the 1%
level for each group, as expected. The negative impact of cost, however,
is four times larger for those in the first group. This pattern of
results suggest consistency in the data and internal validity of the LC
model, since individuals with less concern over coastal restoration and
lower income are more likely to vote against improvements in the
rebuilding plan for New Orleans, less likely to support coastal
restoration initiatives, and more sensitive to the magnitude of the tax
increase.
Table 8 presents MWTP estimates for rebuilding attributes that
mitigate future risks to New Orleans and its citizens. Figures 1 3
depict the confidence intervals of MWTP for rebuilding attributes in the
different samples. Our estimates indicate that the average individual in
the New Orleans sample is willing to pay $301 for Category 5 levee
protection versus $509 for the average individual in the U.S. sample.
The average individual in the combined sample is willing to pay $449 for
Category 5 levees. The confidence intervals, estimated with the
Krinsky-Robb procedure, indicate that all welfare estimates for Category
5 levee protection are statistically significant at the 1% level. The
latent class model allows us to examine welfare estimates for discrete
groups of U.S. residents. The first group, identified as likely to
include lower income individuals who view coastal restoration as less
important in the rebuilding plan, is associated with a WTP of $433 for
Category 5 levees. An average individual from the second group
(counterpart to the first group) is willing to pay $514 for Category 5
levees. The difference between these two welfare estimates for the LC
model is not statistically significant. The LC MWTP measure aggregated
across the two groups is $485. As indicated in Figure 1, all estimates
(except New Orleans) exhibit significant overlap and similar central
tendencies.
[FIGURE 1 OMITTED]
Turning to coastal restoration values, we do not obtain
statistically significant measures of MWTP for the New Orleans and U.S.
samples for the RXL model. In the former case, this result likely
reflects the low level of significance for the fixed coastal restoration
parameter, while in the latter it reflects wide variability in this
random parameter. The average individual in the combined model is
willing to pay $103 for coastal restoration, and this estimate is
significant at the 5% level. Estimates from the LC model indicate an
average individual from the second group in the U.S. sample is willing
to pay $214 for coastal restoration. Figure 2 indicates that only the
estimates associated with the combined RXL model and group two for the
LC model have distributions with sufficient mass above zero.
Lastly, we find that the average individual ill the New Orleans
sample is willing to pay $137 for modernized transportation (significant
at the 10% level), while MWTP for the U.S. sample is not statistically
significant in the RXL results. Households in the combined sample are
willing to pay $103 for modernized transportation (significant at the
10% level). With the LC model, MWTP for modernized transportation is
positive but insignificant for group one, but negative and statistically
significant for group two! The average person in group two--more likely
to include higher income households and individuals that view coastal
restoration as important has a negative MWTP of -$93.45 (significant at
the 5% level), implying that they view modernized transportation as an
economic "bad." The point estimate for aggregate MWTP for the
LC model is negative but not significantly different from zero. These
distributions of MWTP are depicted in Figure 3.
[FIGURE 2 OMITTED]
7. Discussion and Conclusions
Employing choice experiments via a random digit dialing telephone
survey, we produce estimates of economic value for public projects that
reduce risk from severe storms. Our experiment offers improvements in
levee flood protection, coastal restoration, and improvements in
transportation infrastructure. Each alternative improvement scenario is
associated with higher one-time payment of federal taxes for all U.S.
households. These improvements are valued in pairwise comparisons with
status quo conditions, and thus our estimates represent MWTP for
risk-reducing projects. Each subject evaluates four pairwise choice sets
of the total 16 choice sets, which were designed using efficient
algorithms for linear models. The choice experiment was implemented as a
referendum with majority rules provision, and subjects were instructed
to treat each choice as independent of other choices.
[FIGURE 3 OMITTED]
In general, respondents find traditional engineered flood
protection structures to be the most valued line of defense. The local
and national sentiment indicates that bolstering levees to withstand a
Category 5 storm represents a valuable public investment. One
explanation for the high valuation of Category 5 levee protection is
that it may be viewed as certain protection, since 5 is the highest
rating on the Saffir-Simpson scale. Experimental evidence from Wakker,
Thaler, and Tversky (1997) demonstrates that people require a
disproportionally high discount in order to accept probabilistic
insurance (insurance that does not pay with 100% certainty). This is
seen as an effect of decision weighting in prospect theory. Coastal
restoration garners some support but not to the degree that engineered
flood protection systems received. Lastly, improved transportation
systems are supported but not as strongly as levee improvement and
coastal restoration.
Results of the repeated mixed logit model indicate that households
in the New Orleans metropolitan area are willing to pay $301 per
household for Category 5 levee protection and $137 per household to
modernize the New Orleans metropolitan transportation system. In
addition to households' values for Category 5 levee protection,
which primarily reflects a form of hazard mitigation, benefits from
modernized transportation also represent an improvement to quality of
life via better day-to-day transportation options. Estimates of value
for coastal restoration for New Orleans residents are not statistically
significant. Aggregating over all New Orleans tax-paying households,
estimated economic value for Category 5 flood protection is
approximately $118 million (95% confidence interval: $54-181 million).
(10) The aggregate economic value of modernized transportation
infrastructure for tax-paying New Orleans households is $54 million (90%
confidence interval: $7-100 million).
We also present results for a sample of U.S. households that were
offered the opportunity to vote in the same choice experiment.
Surprisingly, U.S. residents are willing to pay $509 per household for
Category 5 levees in New Orleans. This mean estimated economic value
exceeds New Orleans residents' mean MWTP by 69%. Comparing opinions
on flood protection, 84% of U.S. respondents feel it is "very
important" to protect New Orleans from floods, compared to 98% of
New Orleans residents. Thus, this economic value could indicate a true
preference for flood protection in this vulnerable and culturally
distinct location. The difference could reflect a higher income for the
U.S. population relative to New Orleans residents, assuming flood
protection is a normal good.
Accounting for preference heterogeneity via the repeated mixed
logit model, we do not find a statistically significant economic value
for U.S. households that can be attributed to coastal restoration in
South Louisiana. Further investigation, however, using the latent class
model allows us to endogenously divide the U.S. sample into two distinct
groups based on observable factors. The first group is more likely to
include lower income households that do not view coastal wetland
restoration as important, while the second group is characterized by
those with higher incomes and who place greater importance on coastal
restoration. WTP for coastal restoration for the first group is not
significantly different from zero, but the average individual in the
second group is willing to pay $214 for coastal restoration. Members of
the first group may be less familiar with coastal wetlands, in general,
and unaware of the storm protection provided by coastal marshes.
Fifty-two percent of U.S. respondents consider coastal restoration as
"very important," considerably less than the 86% of New
Orleans residents that express this view. Using posterior probabilities,
we estimate that the average likelihood of individuals in our sample
belonging to the first group is around 35%.
Lastly, with the repeated mixed logit model, U.S. respondents'
WTP for improvements in transportation infrastructure is not
statistically significant; again, the LC model reveals different
results. While we did not find a significant result for the first group,
parameters for the second U.S. group exhibited a negative and
statistically significant WTP for modernized transportation. This result
may indicate that these types of individuals disapprove of development
in high risk areas and do not want to create an incentive for expanded
redevelopment in the form of modernized transportation. Public services,
such as utilities and public transportation, act as de facto land use
policy since they provide access to more locations. This, in effect, can
create incentives for development because a larger proportion of the
population can access more remote areas. Without modern transportation,
people may be dissuaded from developing in remote or high risk
locations.
Combining the two samples and reweighting for representation at the
national level and to correct for response bias based on observable
factors, we produce tentative estimates of economic value for
risk-reduction in New Orleans. Under the assumption that this sample is
a reasonable approximation of national preferences, the average U.S.
household is willing to pay $449 for upgrading New Orleans's levee
system to withstand a Category 5 storm, and the average WTP for coastal
restoration is $103 per household. The average U.S. household is WTP
$103 for modernized transportation in New Orleans. Aggregating over all
U.S. taxpaying households, economic values for rebuilding New Orleans
are approximately $50 billion (95% confidence interval: $29-71 billion)
for Category 5 flood protection in New Orleans, $12 billion (95%
confidence interval: $2-21 billion) for coastal restoration, and $12
billion (90% confidence interval: $0.3 21 billion) for modernized
transportation. (11)
Although a comprehensive cost benefit analysis is beyond the scope
of this study, these estimates could provide valuable information for
policymakers as they analyze risk-reducing projects for post-Katrina New
Orleans and southern Louisiana. While there are no definitive or
inclusive estimates of costs to rebuild New Orleans, Congressional
reports suggest that the total cost of various rebuilding projects could
be close to $200 billion. (12) Such high cost estimates raise important
questions as to whether rebuilding New Orleans makes economic sense. To
date, the federal government has provided billions of dollars in
assistance to the Gulf Coast to repair and rebuild damaged public
infrastructure. An article in the Washington Post reports that the cost
of rebuilding New Orleans's levees has been about $10 billion,
although the cost may increase to fully protect the entire region
("Levee Repair Costs Triple" March 31, 2006).
The cost to protect and restore coastal wetland in Louisiana has
been estimated at $14 billion over a 30-year period (National Research
Council of the National Academies 2006). (13) While details remain to be
settled, our benefit estimates suggest that at least selective
rebuilding on the basic infrastructure could be justified from an
economic efficiency perspective. Hopefully, this study will stimulate
future research on the costs and benefits of rebuilding New Orleans as
more carefully constructed estimates become available.
Appendix A: Choice Experiment
Remember, the current plan is (i) limited bus service, street cars,
and conventional roads; (ii) no restoration of coastal wetlands: (iii)
repair the levees to withstand a Category 3 hurricane: and (iv) no
additional taxes.
1. Transportation and the levees would be the same as the current
plan. This alternative plan proposes to restore coastal wetlands. This
plan would cost each U. S. household an extra $300. Would you vote for
the current plan or this new plan?
2. The levees and the coastal wetlands would be the same as the
current plan. but the new plan would include improvements in the
transportation system. This plan would cost each U.S. household an extra
$450. Would you vote for the current plan or this new plan?
3. The transportation and the coastal wetland restoration would be
the same as the current plan, but the new plan would include
improvements in the levees to protect the city against a Category, 5
hurricane. This plan would cost each U.S. household an extra $50. Would
you vote for the current plan or this new plan?
4. In this plan, the transportation system would be improved, the
coastal wetlands restored, and the levees improved to protect the city
against a Category 5 hurricane. This plan would cost each U.S. household
an extra $150. Would you vote for the current plan or this new plan?
Example of Conjoint Choice Set Alternatives for Block 1.
Set Transportation Coastal Restoration Levee Tax
1 Conventional Yes Category 3 $300
2 Modern No Category 3 $450
3 Conventional No Category 5 $50
4 Modern Yes Category 5 $150
Appendix B: Cheap Talk Script
We would now like to ask you about four rebuilding plans for New
Orleans. The plans differ in the types of improvements that are made to
the city and the cost to tax-payers. Suppose that each of the plans are
put up for a vote, you may vote for or against each plan or choose not
to vote--majority rules.
Before we get to the vote, consider the following information. In a
recent study, groups of people participated in a vote just like the one
you are about to participate in. The improvements and costs of the plan
for these groups were not real, just as they will not be real for you.
No one had to pay money if the vote passed, and most voted .for the
plan.
Other groups of similar people participated in the same vote, but
payment was real and everyone really did have to pay the cost if the
vote passed. In these groups most voted against the plan. We call this
difference between the way people say they would vote and the way they
really vote "bias."
Sometimes when we hear about a vote that involves doing something
that is basically good--helping people in need, improving air and water
quality, or anything else--our reaction in a hypothetical situation is
to think: sure, 1 would do this. I really would vote to spend the money.
But when the vote is real, and we would actually have to spend our
money if it passes, we think a different way. We still would like to see
good things happen, but when we are faced with having to spend money, we
think about our options: if I spend money on this, that's money I
don't have to spend on other things. We vote in a way that takes
into account the limited amount of money we have.
I would like for you to think about your votes just like you would
think about a real vote, where if enough people vote for the plan,
you'd really have to pay and so would everyone else. Please keep
this in mind as you answer the four voting questions.
For the purpose of these questions, the current rebuilding plan for
New Orleans will be
--Limited bus service (routes, transfer points, and hours of
service buses), limited use of street cars, and conventional road
network
--No restoration program for Louisiana's coastal wetlands
--Repair the levee system to withstand a Category 3 hurricane
--You pay $0 in additional tax money for one year
We will now give you the opportunity to vote on four separate plans
for rebuilding. Each of the four plans differs in the type of
improvements that are made and the associated costs. Money to fund the
plan would come from a one-time tax on all U.S. households. The tax
amount differs due to the nature of the rebuilding plan and because we
are uncertain about what the actual costs would be. Assume that all
money raised would go directly to rebuilding New Orleans.
Please consider each plan separately in relation to the current
plan and indicate whether or not you would vote for this plan if the
vote were real.
References
Adamowicz, Wiktor, Jordan Louviere, and Michael Williams. 1994.
Combining revealed and stated preference measures for valuing
environmental amenities. Journal of Environmental Economies and
Management 26(3):271-92.
Adamowicz, Wiktor, Peter Boxall, Michael Williams, and Jordan
Louviere. 1998. Stated preference approaches for measuring passive use
values: Choice experiments and contingent valuation. American Journal of
Agricultural Economies 80(1):64-75.
Boxall, Peter. and Wiktor Adamowicz. 2002. Understanding
heterogeneous preferences in random utility models: A latent class
approach. Environmental and Resource Economies 23:421-16.
Carlsson, Fredrik, and Peter Martinsson. 2001. Do hypothetical and
actual marginal willingness to pay differ in choice experiments? Journal
of Environmental Economies and Management 41:179-92.
Carlsson, Fredrik, Peter Frykblom, and Carl-Johan Lagerkvist. 2005.
Using cheap talk as a test of validity in choice experiments. Economics
Letters 89(2):147-52.
Cummings, Ronald G., and Laura O. Taylor. 1999. Unbiased value
estimates for environmental goods: Cheap talk design for the contingent
valuation method. American Economic Review 89:649-65.
Fischetti, Mark. 2001. Drowning New Orleans. Scientific American
(October). Available at
http://www.scientificamerican.com/article.cfm?id-drowning-new-orleans.
Galea, Sandro, Andrea Maxwell, and Fran Norris. 2008. Sampling and
design challenges in studying the mental health consequences of
disasters. International Journal for Methods in Psychiatric Research
17(S2):S21-8.
Greene, William H. 2007. NLOGIT 4.0 reference guide. Plainview, NY:
Econometric Software.
Herriges, Joseph A., and Daniel J. Phaneuf. 2002, Inducing patterns
of correlation and substitution in repeated logit models of recreation
demand. American Journal of Agricultural Economies 84(4):1076-90.
Huber, Joel, and Klaus Zwerina. 1996. The importance of utility
balance in choice designs. Journal of Marketing Research 33:307-17.
Johansson-Stenman, Olof, and Henrik Svedsater. 2008. Measuring
hypothetical bias in choice experiments: The importance of cognitive
consistency. The B.E. Journal of Economic Analysis and Policy 8, Article
41.
Kates, Robert, Craig Colten, Shirley Laska, and Stephen Leatherman.
2006. Reconstruction of New Orleans after Hurricane Katrina: A research
perspective. Proceedings of the National Academy of Sciences of the
United States of America 103(40): 14653-60.
Kessler, Robert C., Terence M. Keane, Robert J. Ursano, Ali Mokdad,
and Alan M. Zaslavsky. 2008. Sample and design considerations in
post-disaster mental health needs assessment tracking surveys.
International Journal of Methods Psychiatric Research 17(S2):S6-20.
Kousky, Carolyn, and Richard J. Zeckhauser. 2006. JARing actions
that fuel floods. In On risk and disaster: Lessons from Hurricane
Katrina, edited by Ronald J. Daniels, Donald F. Kettl, and Howard
Kunreuther. Philadelphia: University of Pennsylvania Press.
Krinsky, Itzhak, and A. Leslie Robb. 1986. On approximating the
statistical properties of elasticities. The Review of Economies and
Statistics 86:715-19.
Kuhfeld, Warren. 2010. "Marketing research methods in SAS:
Experimental design, choice, conjoint, and graphical techniques."
SAS 9.1 version. Accessed 18 January 2011. Available
http://support.sas.com/techsup/technote/mr20l0.pdf.
Kunreuther, Howard, and Mark Pauly. 2006. Rules rather than
discretion: Lessons from Hurricane Katrina. Journal of Risk and
Uncertainty 33:101-16.
Ladenburg, Jacob, and Soren B. Olsen. 2008. Gender-specific
starting point bias in choice experiments: Evidence from an empirical
study. Journal of Environmental Economics and Management 56(3):275-85.
Layton, David F., and Gordon Brown. 2000. Heterogeneous preferences
regarding global climate change. The Review of Economics and Statistics
82(4):616-24.
List, John A., Paramita Sinha, and Michael H. Taylor. 2006. Using
choice experiments to value non-market goods and services. Advances in
Economic Analysis and Policy 6(2):1-37.
Louisiana Coastal Protection and Restoration (LACPR). 2009.
"Draft Final Technical Report, March 2009." U.S. Army Corp of
Engineers, New Orleans District. Mississippi Valley Division. Accessed
18 January 2011. Available
http://lacpr.usace.army.mil/default.aspx?p=LACPR_Draft_Technical_Report.
Lovelace, John K. 1994. Storm-tide elevations produced by Hurricane
Andrew along the Louisiana coast, August 25-27, 1992. Open File Report
94-371. Baton Rouge, LA: U.S. Geological Survey.
Lusk, Jayson, and Ted Schroeder. 2004. Are choice experiments
incentive compatible? A test with quality differentiated beef-steaks.
American Journal of Agricultural Economics 85:840-56.
McGonagle, Michael P., and Stephen K. Swallow. 2005. Open space and
public access: A contingent choice application to coastal preservation.
Land Economics 81(4):477-95.
Mississippi Coastal Improvements Program (MsCIP). 2009. "Draft
technical report February 2009." U.S. Army Corp of Engineers,
Mobile District. Accessed 18 January 2011. Available
http://lacpr.usace.army.mil/Report/DTR/Draft%20LACPR%20Technical%20Report% 20Mar%2009.pdf.
Morton, Robert A., Noreen A. Buster, and M. Dennis Krohn. 2002.
Subsurface controls on historical subsidence rates and associated
wetland loss in Southcentral Louisiana. Transactions Gulf Coast
Association of Geological Societies 52:767-78.
National Research Council of the National Academies. 2006. Drawing
Louisiana's new map: Addressing land loss in coastal Louisiana.
Committee on the Restoration and Protection of Coastal Louisiana, Ocean
Studies Board, Division on Earth and Life Studies. Washington, DC: The
National Academies Press.
Seed, Robert B., Robert G. Bea, and Remon I. Abdelmalak, et al.
(2006). "Investigation of the Performance of the New Orleans Flood
Protection Systems in Hurricane Katrina on August 29, 2005."
University of California, Berkeley 1-2. Accessed 18 January 2011.
Available http://www.ce.berkeley.edu/~new_orleans/.
Train, K. 1998. Recreation demand models with taste differences
over people. Land Economics 74:230-39.
Train, Kenneth. 2003. Discrete choice methods" with
simulation. New York: Cambridge University Press.
Transportation Research Board. 2008. "The role of transit in
emergency evacuation." National Academy of Sciences Transportation
Research Board Special Report 294, ISBN 978-0-309-11333-5. Accessed 18
January 2011. Available
http://onlinepubs.trb.org/Onlinepubs/sr/sr294.pdf.
Turner, R. Eugene. 1997. Wetland loss in the northern Gulf of
Mexico: Multiple working hypotheses. Estuaries 20(1):1-13.
U.S. Government Accountability Office (U.S. GAO). 2005.
"History of the Lake Pontchartrain and vicinity hurricane
protection project." GAO-06-244T. Accessed 18 January 2011.
Available http://www.gao.gov/new.items/d06244t.pdf
U.S. GAO. 2007. "Preliminary Information on Gulf Coast
Rebuilding." GAO-07-809R. Accessed 18 January 2011. Available
http://www.gao.gov/new.items/d07809r.pdf.
Wakker, Peter P., Richard Thaler, and Amos Tversky. 1997.
Probabilistic insurance. Journal of Risk and Uncertainty 15:7-28.
(1) The state's levee system was founded in the Louisiana
constitution, which created local levee and drainage districts to build
and maintain levees. Since Katrina, class action suits have been brought
against the Orleans Levee District, the Lake Borgne Basin Levee
district, the East Jefferson Levee District, and their respective boards
of commissioners, as well as the U.S. Army Corps of Engineers.
(2) Between June 23 and 27, 2005 the New Orleans Times-Picayune ran
a series entitled "Washing Away" that was critical of federal,
state, and local government flood risk management in south Louisiana.
The vulnerability of New Orleans was also mentioned in the U.S.
Commission for Ocean Policy, as well as in a Scientific American piece
titled "Drowning New Orleans" (Fischeni 2001).
(3) An anonymous reviewer points out that the Saffir-Simpson scale
does not directly take storm surge into account, and thus our
description of stoma protection may be somewhat ambiguous in this
regard.
(4) As noted by Train (2003, p. 41), "Under fairly general
conditions, any function can be approximated arbitrarily closely by one
that is linear in parameters. The assumption is therefore fairly
benign."
(5) Other distributional assumptions are possible. For example, the
parameters can be sign-restricted by assuming that they follow a
log-normal distribution. Since we are attempting to learn about the
preferences of individuals, we choose not to impose signs on the
parameters and thus employ a normal distribution.
(6) The mixed logit was estimated using code written by H. Allen
Klaiber for the "'Micro-Econometrics In and Out of Markets: A
Second Training Workshop on Micro-Econometrics in Environmental
Economics." This workshop was developed and funded by the Center
for Environmental and Resource Economic Policy (CEnREP) at North
Carolina State University and the U.S. Environmental Protection Agency.
(7) While the standard likelihood ratio test is biased toward
accepting the null hypothesis of zero standard deviation of
coefficients, the results are suggestive, and given the complexity of
the model and the small sample size for NOLA we find it useful to
restrict the model.
(8) See Train (2003) for a discussion of using Halton sequences to
draw from densities in mixed logit models.
(9) Results for the combined model are very similar.
(10) According to the 2005 2007 American Community Survey 3-Year
Estimates, there are 392,659 households in the New Orleans MSA.
(11) According to the 2005 2007 American Community Survey 3-Year
Estimates, there are 111,609,629 households in the United States.
(12) The Congressional Budget Office estimated the value of capital
stock destroyed by Hurricanes Katrina and Rita in the range of $70-130
billion, and the State of Louisiana estimated that the cost to the state
alone could reach $200 billion (U.S. GAO 2007).
(13) The extent of the damage caused by Hurricanes Katrina and Rita
was not fully determined at the time of this report. The report also
provides that the use-value of wetlands estimated by the State of
Louisiana is in excess of $37 billion by 2050.
Craig E. Landry, * Paul Hindsley, ([dagger]) Okmyung Bin,
([dagger]) ([dagger]) Jamie B. Kruse, [section] John C. Whitehead,
[parallel] and Ken Wilson (#)
* Department of Economics and Center for Natural Hazards Research,
East Carolina University, Greenville, NC 27858, USA; E-mail
[email protected]: corresponding author.
([dagger]) Department of Environmental Studies, Eckerd College, St.
Petersburg, EL 33711, USA: E-mail hindslpr@ eckerd.edu.
([dagger]) ([dagger]) Department of Economics, East Carolina
University, Greenville, NC 27858, USA; E-mail
[email protected].
([section]) Department of Economics and Center for Natural Hazards
Research, East Carolina University, Greenville, NC 27858, USA; E-mail
[email protected].
([parallel]) Department of Economics, Appalachian State University,
Boone, NC 28608, USA; E-mail whiteheadjc@ appstate.edu.
(#) Department of Sociology, East Carolina University, Greenville,
NC 27858, USA; E-mail
[email protected].
Thanks are given to participants of Katrina Research Symposium
(organized by Tulane University), seminar participants at Mississippi
State University, Department of Agricultural Economics, and two
anonymous reviewers. This research was supported by the National Science
Foundation--Small Grants for Exploratory Rcscarch (award SES0554987).
Received August 2009: accepted May 2010.
Table 1. Choice Experiment Design
Attribute Levels
Levee protection Category 3 storm (status quo)
Category 5 storm
Coastal restoration No (status quo)
Yes
Transportation system Conventional (status quo)
Modernized
One-time tax payment for all U.S. $0 (status quo)
households $50
$150
$300
$450
Each choice set was a pairwise comparison. with the status quo at zero
additional tax offered against an alternative with at least one
improvement and a higher tax.
Table 2. Descriptive Statistics by Strata
NOLA NOLA
(unweighted) (weighted)
Mean (standard Mean (standard
Obs deviation) deviation)
Female 119 0.7395 0.5788
(0.4408) (0.4958)
White 116 0.7155 0.6154
(0.4531) (0.4886)
African American 116 0.2500 0.3725
(0.4349) (0.4856)
No high school 117 0.0684 0.1433
(0.2535) (0.3519)
College degree 117 0.3419 0.0983
(0.4764) (0.2989)
Married 119 0.5630 0.4933
(0.4981) (0.5021)
Income (<15K) 116 0.2500 0.2111
(0.4349) (0.4099)
Income (15-30K) 116 0.3621 0.1878
(0.4827) (0.3922)
Income (>100K) 116 0.0776 0.1534
(0.2687) (0.3619)
United States United States
(unweighted) (weighted)
Mean (standard Mean (standard
Obs deviation) deviation)
Female 215 0.5860 0.4954
(0.4937) (0.5011)
White 214 0.7617 0.7657
(0.4271) (0.4246)
African American 214 0.1822 0.1518
(0.3869) (0.3597)
No high school 216 0.0509 0.0920
(0.2204) (0.2896)
College degree 216 0.3426 0.2721
(0.4757) (0.4461)
Married 212 0.5236 0.5472
(0.5006) (0.4989)
Income (<15K) 122 0.2295 0.1996
(0.4223) (0.4014)
Income (15-30K) 122 0.3443 0.2250
(0.4771) (0.4193)
Income (>100K) 122 0.0246 0.0793
(0.1555) (0.2714)
This table represents the weighted and unweighted descriptive
statistics for the NOLA and U.S. strata.
Table 3. Weighted Frequencies for Importance of Flood Protection
U.S. Sample NOLA Sample
Importance of Flood Protection Frequency % Frequency %
Not important 7.088 2.64 0 0
Somewhat important 35.250 13.12 2.044 1.55
Very important 224.599 83.62 128.582 97.69
No response 1.646 0.61 1 0.76
This table reports weighted frequencies to correct for non-response.
Table 4. Weighted Frequencies for Importance of Coastal Wetland
Restoration
U.S. Sample NOLA Sample
Importance of Coastal
Wetland Restoration Frequency % Frequency %
Not important 34.508 12.85 5.367 4.08
Somewhat important 94.325 35.12 2.130 9.22
Very important 139.750 52.03 113.129 85.95
No response 0 0 1 0.76
This table reports weighted frequencies to correct for non-response.
Table 5. Weighted Frequencies for Importance of Improved
Transportation
U.S. Sample NOLA Sample
Importance of Improved
Transportation Frequency % Frequency %
Not important 10.863 4.04 4.768 3.62
Somewhat important 109.772 40.87 36.044 27.38
Very important 145.814 54.29 89.814 68.23
No response 2.133 0.79 1 0.76
This table reports weighted frequencies to correct for non-response.
Table 6. Repeated Mixed Logit Models
New Orleans United States
Status quo 0.5324 0.3663
(0.3989) (0.4026)
Category 5 1.3801 *** 3.4436 ***
(0.2957) (0.6981)
Category 5 standard deviation -- 3.3284 ***
-- (0.7477)
Coastal restoration (CR) 0.5177 * 0.5845
(0.3088) (0.3682)
CR standard deviation 1.6057 *** 2.4609 ***
(0.5096) (0.6770)
Modern transportation (MT) 0.6295 ** 0.5507
(0.2766) (0.3420)
MT standard deviation -- 1.5446 ***
-- (0.6650)
Tax -0.0046 *** -0.0068 ***
(0.0007) (0.0014)
Individuals 120 217
Observations 480 868
Null In likelihood -497.355 -765.094
In likelihood -345.7521 -465.5882
Halton draws 1000 1000
Combined
Status quo 0.7076
(0.4028)
Category 5 2.9463 ***
(0.5836)
Category 5 standard deviation 2.0667 ***
(0.6571)
Coastal restoration (CR) 0.6755
(0.4562)
CR standard deviation 2.6503 ***
(0.7274)
Modern transportation (MT) 0.6778 *
(0.3495)
MT standard deviation 1.3564 *
(0.7155)
Tax -0.0066 ***
(0.0014)
Individuals 336
Observations 1347
Null In likelihood -1775.89
In likelihood -1523.98
Halton draws 1000
Standard errors are in parentheses. Mixing distribution assume
normality.
*** Statistical significance for I% chance of type I error.
** Statistical significance at 5%.
* Statistical significance at 10%.
Table 7. Latent Class Model: U.S. Sample
Group 1 Group 2
Status quo 1.441 ** -0.0736
(0.623) (0.1541)
Category 5 3.799 *** 1.014 ***
(1.176) (0.111)
Coastal restoration -0.220 0.421 ***
(0.873) (0.138)
Modern transportation 0.990 ** -0.184 **
(0.461) (0.075)
Tax -0.0088 *** -0.002 ***
(0.0022) (0.0004)
Class probability parameters
Constant 4.881 *** 0
(1.176)
Coastal wetland -32.558 *** 0
Importance (3.266)
Income -0.057 *** 0
(0.0166)
Individuals 217
Observations 868
Null In likelihood -598.8808
In likelihood -479.6496
Rho-square 0.199
Iterations 78
Standard errors are in parentheses.
*** Statistical significance for 1% chance of type I error.
** Statistical significance at 5%.
* Statistical significance at 10%.
Table 8. Welfare Measures (MWTP)
Repeated Mixed Logit Models
Combined U. S. NOLA
Sample ($) Sample ($) Sample ($)
Category 5 448.75 509.16 300.87
95% confidence (263.16, (329.53, (138.54,
interval 634.34) 688.79) 463.20)
90% confidence (292.98, (358.40, (164.63,
interval 604.52) 659.92) 437.11)
Coastal 102.88 86.43 112.86
restoration
95% confidence (18.69, (-30.31, (-28.26,
interval 187.16) 203.17) 253.98)
90% confidence (32.15, (-11.55, (-5.58,
interval 173.62) 184.41) 231.30)
Modernized 103.24 81.42 137.23
transportation
95% confidence (-16.36, (-26.83, (-4.22,
interval 222.84) 189.67) 278.68)
90% confidence (2.86, (-9.43, (18.51,
interval 203.62) 172.27) 255.95)
Latent Class Models
U.S. U.S. U.S.
Sample ($) Sample ($) Sample ($)
Group 1 Group 2 Aggregate
Category 5 432.56 514.39 485.75
95% confidence (269.12, (291.69, (330.43,
interval 596.00) 737.09) 641.07)
90% confidence (295.38, (327.49, (355.40,
interval 596.74) 701.29) 616.11)
Coastal -25.05 213.59 130.06
restoration
95% confidence (-266.99, (0.78, (-32.20,
interval 216.89) 426.47) 292.32)
90% confidence (-228.11, (34.93, (-6.13,
interval 178.01) 392.25) 266.25)
Modernized 112.77 -93.45 -21.27
transportation
95% confidence (-52.91, (-181.57, (-103.29,
interval 278.45) -5.33) 60.75)
90% confidence (-26.28, (-167.41, (-90.11,
interval 251.82) -19.49) 47.57)
Statistically significant MWTP estimates in bold.
