Abstract
Despite mounting threats from rising sea levels, adaptation to sea level rise (SLR) is often challenged by limited funding and understanding of residents’ preferences. Using an online choice experiment, we investigate residents’ preference for three SLR adaptation strategies: building seawalls, replenishing the beach, and installing stormwater pumps in Miami-Dade County. We control the preference, scale, and alternative heterogeneity using generalized multinomial logit models with error components. Results show that residents prefer additional adaptation strategies to the status quo, and valuations of adaptation attributes are correlated with residents’ sociodemographics. Accounting for alternative heterogeneity also significantly improves model performance.
1. Introduction
The mean global sea level is estimated to rise by 65±12 cm (26±5 in.) above the 2005 mean sea level by 2100 (Nerem et al. 2018). This process is caused by the expansion of seawater from global warming and water mass input from land ice melt as well as land water reservoirs (Nicholls and Cazenave 2010). The consequence of sea level rise (SLR) includes inundation and flooding, loss of coastal land, erosion of beaches, saltwater intrusion to surface water, and loss of coastal wetlands and saltmarshes, which will inflict considerable economic loss and ecological damage (Caldwell and Segall 2007; Nicholls and Cazenave 2010; Nicholls et al. 2011; Reidmiller et al. 2018). The projected economic losses from SLR could be massive without mitigation or adaptation efforts. Using Florida as an example, inaction to climate change could result in a total loss of $92 billion by 2050 (in 2006 U.S. dollars) and $345 billion by 2100 (5% of the state’s GDP) in tourism revenue, commercial and residential property values, and cost of living (Stanton and Ackerman 2007). Adaptation strategies to reduce the vulnerability of economic, social, and biological systems may be cost-effective in the long run to prevent more significant damage from SLR.
Adaptation strategies to SLR can be categorized into protection, accommodation, retreat, and avoidance (FDEP 2018). Protection involves providing hard artificial (e.g., seawalls and bulkheads) or natural barriers (e.g., sand dunes and mangrove wetlands) between water and land. Accommodations allow economic activities to continue with the rising sea by elevating roads and infrastructures or improving planning and warning systems. Retreat includes relocation to upper land while abandoning or demolishing damaged properties (Fankhauser 1995; Hallegatte 2009). Avoidance involves barring development in hazard zones. The costs of adaptations vary widely, with some being quite expensive. For example, the U.S. Geological Survey estimates that protecting the vulnerable 1,710 miles along the 4,000 miles of Florida’s shoreline with seawalls would cost $9 billion (Lu, Peng, and Du 2012). Without beach replenishment, the rising sea level in south Florida could accelerate the rate of beach erosion as high as 30%–50% (Finkl 1996). The costs of elevating roads and structures and replenishing beaches could be as high as $21.6 billion along the Gulf Coast and $7.8 billion along the Atlantic Coast, without considering the rising cost of sand to replenish beaches (Titus et al. 2009). The feasibility of various adaptations to SLR is site-specific because coastal flooding hazards are influenced by location-specific factors such as local elevation, change of elevation, regional variations in the rate of SLR, and exposure to storm surges induced by local extreme weather events (Wdowinski et al. 2016). Because there is little literature focusing on the empirical estimate of the economic benefits of local SLR adaptation and the perceived importance of different attributes of the adaptation strategies demanded by local residents, this study will provide essential information for adaptation planning and cost-benefit analysis.
Previous studies have used preference elicitation methods such as discrete choice experiments to examine the economic benefits and welfare changes for nonmarket goods such as climate change adaptation strategies (Shoyama, Managi, and Yamagata 2013; Andreopoulos et al. 2015; Remoundou et al. 2015; Ščasný et al. 2017). However, few have examined the public preferences for trade-offs of adaptation strategies and the extent to which specific attributes of adaptations influence preferences. Obtaining such information is essential for developing effective SLR adaptation strategies because adaptation strategies differ significantly in costs, effectiveness, and expected planning horizons. For instance, beach nourishment needs to be replenished periodically as they erode over time, and the nourishment costs are expected to increase due to erosion and a limited supply of natural sand. Armoring shorelines with hard structures, such as seawalls, helps protect beaches along the artificial barrier but increases beach erosion in nearby areas and endangers wildlife habitats. Furthermore, it reduces beach access and induces further economic development in at-risk areas, which may exacerbate future damages. Installing stormwater pumps to mitigate flooding damages has failed to prevent flooding in low-lying, populous coastal areas such as Miami (FDEP 2018).
This article contributes to the literature on the valuation of SLR adaptations by using the adaptation strategies adopted in Miami-Dade County, Florida, as an example. Specifically, we focus on three protective adaptation strategies to elicit homeowners’ preferences for these adaptations via a choice experiment in an online survey of representative homeowners in the county. The three strategies are building seawalls, replenishing the beach, and installing stormwater pumps.
We focus on Miami-Dade County because this area is highly economically vulnerable to SLR (Melillo, Richmond, and Yohe 2014). The rate of SLR has accelerated in southeast Florida, from 3 ± 2mm (0.12 ± 0.08 in.) per year pre-2006 to 9 ± 4mm (0.35 ± 0.16 in.) post-2006 (Wdowinski et al. 2016). After 2006, the frequency of rain-induced and tide-induced flooding has grown by 33% and 400%, respectively (Wdowinski et al. 2016). Miami-Dade County is also the seventh most populous county in the United States (U.S. Census Bureau 2019). Projections show that the number of people affected by a 1.8 m SLR in Broward and Miami-Dade Counties in Florida accounts for more than a quarter of the affected population by SLR in the United States (Hauer, Evans, and Mishra 2016). Furthermore, Florida’s economy is heavily reliant on tourism and real estate, especially in southeast Florida. In 2018, Florida beaches had about 810 million beach day visits (Florida Trend 2019). The loss of property values from SLR inundation between 2005 and 2016 is estimated to be $465 million (McAlpine and Porter 2018). SLR is also expected to reduce the values of the properties prone to flooding even further.
In response to the mounting risks from SLR, Miami voters supported the $400 million Miami Forever general obligation bond in November 2017 (MiamiForever.org). The bond will fund public projects to reduce flooding risks, improve stormwater infrastructure, increase affordable housing, and upgrade cultural services to the most vulnerable populations. In 2018, Miami Beach, one of the cities in Miami-Dade County, promoted a $500 million plan to elevate roads and install pumps to adapt to SLR (Harris 2018). This is noteworthy, as Florida lacks state-coordinated adaptation planning despite the scientific agreement on the projections of accelerated SLR (Ariza et al. 2014). Understanding residents’ perceived benefits of SLR adaptations can provide essential insights for the local government to identify key attributes of adaptation strategies valued by local residents and communicate the trade-offs of different SLR adaptations.
Our study differs from existing work in the following ways. First, we focus on specific attributes that can describe the benefits of adaptation strategies and are easier to use to communicate with residents. The attributes include the expected duration of protection and the expected change in flood insurance premiums. Estimates on residents’ valuation of these attributes can inform future reform on incentive programs to encourage and support adaptation strategies. Second, whereas previous studies focus only on preference heterogeneity, we address preference heterogeneity, scale heterogeneity, and alternative heterogeneity by using a generalized multinomial logit model (G-MNL) with an error component (EC) that allows the alternatives to have specific individual or group effects for different adaptation strategies (Scarpa, Ferrini, and Willis 2005; Caputo, Scarpa, and Nayga 2017). This model helps decompose the heterogeneity that may come from preference, scale, and alternatives. As a result, it provides better estimates of the preferences for SLR adaptation strategies. Third, we estimate the marginal willingness to pay (MWTP) for each attribute with respect to each respondent’s sociodemographics and home property characteristics to identify factors affecting the perceived benefits from adaptation strategies. Last, we estimate the total MWTP that homeowners are willing to pay by increasing the stormwater utility fee, providing a reference point for the local government to design appropriate programs for SLR adaptation using public funds.
2. SLR Adaptation Strategies
Rising sea levels accelerate beach erosion. Beach, as a natural resource, is a valuable asset providing both use and nonuse economic values to residents. Numerous studies have documented that residents are willing to support the replenishment of beaches (beach nourishment) to obtain greater recreational benefits, support wildlife, or improve property values (see review in Landry 2011; Penn et al. 2016; Pascoe 2019). For example, the MWTP for beach nourishment is estimated to be between $100 and $8,000 per square foot in Georgia, North Carolina, and South Carolina (Pompe 1999; Landry, Keeler, and Kriesel 2003; Gopalakrishnan et al. 2011). Several studies have examined the recreational benefits and nonuse economic values of beaches in Florida using contingent valuation and travel cost methods. The study by Bell and Leeworthy (1990) estimates that the value of a beach day in Florida is $34 (1984 U.S. dollars) for those households traveling great distances. Shivlani, Letson, and Theis (2003) characterize three south Florida beaches and elicit visitors’ MWTP for beach nourishment using the contingent valuation method (CVM). They find that the MWTP for using beach nourishment to maintain turtle nesting habitat is $2.12 per visit, and the MWTP for beach recreational activities alone is $1.69 per visit.
Previous studies have examined residents’ preference for adaptation strategies to combat the adverse effects of climate change, such as SLR and coastal flooding (Holladay, Kunreuther, and Stahl 2016; Johnston and Abdulrahman 2017; Johnston et al. 2018; Landry, Shonkwiler, and Whitehead 2018; Makriyannis, Johnston, and Whelchel 2018; Hindsley and Yoskowitz 2020). Using CVM, Holladay, Kunreuther, and Stahl (2016) find that respondents in New York with homes damaged by Hurricane Sandy are willing to pay $7 per month more to fund a comprehensive flood control system than those with undamaged homes. Residents in New Orleans prefer hard physical (artificial) structures to coastal wetland restoration (Landry et al. 2011), and residents in North Carolina strongly oppose shoreline armoring and are willing to support beach nourishment and retreat (Landry, Shonkwiler, and Whitehead 2018). Earlier studies have found that respondents’ income, residence, and familiarity with beaches and the presence of sand dunes are positively correlated with respondents’ MWTP for beach erosion control in the northeastern region of the United States (Lindsay et al. 1992). Other studies have used the discrete choice experiment (DCE) to assess the value of climate change risk reduction methods and adaptation strategies (Johnston and Abdulrahman 2017; Johnston, Makriyannis, and Whelchel 2018; Makriyannis, Johnston, and Whelchel 2018; Hindsley and Yoskowitz 2020). Using the DCE method, those studies evaluated the trade-offs of different attributes related to coastal hazard adaptations and provided implications for future studies on climate change.
None of the studies have focused on one of the most populous, economically prospective, yet vulnerable counties, such as Miami-Dade, and investigated the residents’ perceived values for SLR adaptations in Florida. In contrast to the existing studies, this article examines the perceived benefits of adaptation strategies concerning two key attributes: increasing the duration of protection and offsetting growth in flood insurance premiums by reducing flooding risks. We explore Florida residents’ MWTP for these attributes and their distribution based on sociodemographics. The information on these two attributes can be used to provide valuations on new adaptation strategies yet to be implemented to reduce flooding risks.
3. Choice Experiment Design and Survey
We designed a choice experiment with three adaptation strategy options (beach nourishment, seawalls, and stormwater pumps) and one status quo option. In each choice set, respondents were asked to select their most preferred one among the alternatives. If the respondents were unsure which adaptation strategy to choose or simply disliked all the alternatives, they could choose the status quo option, which is the default. The adaptation strategies vary in three attributes: (1) additional years of protection, (2) the percentage change of flood insurance premium, and (3) additional monthly stormwater utility fees. We selected the attribute levels of the status quo, beach nourishment, seawalls, and stormwater pumps based on the literature and discussions with the experts of SLR adaptation and local stakeholders in Miami-Dade County.
All the attributes and corresponding levels of the three adaptation strategies are summarized in Appendix Table A1. The status quo is set to have an additional eight years of protection, with an average of 15% increase in flood insurance premium and zero additional monthly stormwater utility fees. The status quo and its attribute levels are estimated based on the adaptation strategies currently being implemented in Miami-Dade County. For the three adaptation strategies (beach nourishment, seawalls, and stormwater pumps), the attribute of additional years of protection includes duration levels of 16, 24, 32, and 38 years, which are 8, 16, 24, and 30 years longer, respectively, than that of the status quo (8 years). The levels are selected based on the information that the life span of seawalls, stormwater pumps, and beach nourishment ranges from 6 to 50 years (Frenning 2001; Phillips and Jones 2006; Caracas et al. 2014; McInerney et al. 2017). That is, the longer the years of protection, the longer the lifespan of the project. The percent change of flood insurance premium associated with the adaptations includes an increase of 2%, 5%, and 8%, which are estimated based on the reported average increase rate from the increased flooding risks associated with SLR (Hurtibise 2016). It is expected that flood insurance premiums for residential and commercial properties will increase in the future, affecting homeowners who have flood insurance and those who do not. If additional adaptations are implemented, the increase in flood insurance premiums will be slower than the increase under the status quo (15%). Additional monthly stormwater utility fees to fund the adaptation strategies include $4, $8, $12, and $16 per month compared with the status quo ($0). Established in 1991, stormwater utility fees are the primary funding collected by the local government to provide adequate flood protection (Regulatory Economic Resources 2019). These fees are mandated for both developed residential and nonresidential properties and are calculated as a function of equivalent residential units (ERUs). As of 2019, each property is charged $4 per month for 1 ERU (set at 1,548 square feet). We choose to use utility fees as the payment vehicle because Florida has no state income tax, and the Department of Water and Sewer in Miami-Dade County serves approximately 85% of the local population.
Because each type of adaptation has unique characteristics, we use a labeled choice experiment design to help respondents better understand the attribute information (Louviere, Hensher, and Swait 2000; Lusk and Schroeder 2004; de Bekker-Grob et al. 2010). The design also allows us to estimate consumer preference for the adaptation strategies by estimating the alternative specific constants. The D-efficient design is used to generate the choice experiment by minimizing the D-error of the determinant of the asymptotic variance-covariance matrix of the parameters from a conditional logit model using Ngene 1.1.2 (ChoiceMetrics 2014). We generate the choice experiment assuming there is no prior information (using zero as the prior parameter values). This is because little information is known about preferences for SLR adaptation strategies (Huber and Zwerina 1996; Kessels, Goos, and Vandebroek 2006). Designing the choice experiment without prior information is an appropriate choice because using low-quality prior information might result in an even less optimal design than the one without any prior information (Ferrini and Scarpa 2017). Ngene generates 20 choice questions, and the design ends with an MNL D-error of 0.3145.1 To reduce the cognitive burden on respondents, we randomize the 20 choice questions into two blocks, with each having 10 choice tasks (Carson et al. 1994; Swait and Adamowicz 2001). Therefore, respondents are randomly assigned to one block and answer the ten choice questions in that block.
Appendix Table A2 illustrates an example of a choice question presented to the respondents: the premium of flood insurance is likely to increase by 15% in eight years if no additional adaptation actions are taken (status quo in row 1, column 5). In this case, without more adaptations, the increase in monthly stormwater utility fees is zero, and the years of protection offered by the current level of adaption are eight years. Beach nourishment, seawalls, and pumps have different levels for the attributes correspondingly.
An online survey with the choice experiment was administered by a market research firm (Qualtrics) to their panel members with residency in Miami-Dade County in May 2017, following a one-week pretest in April and before the highly publicized “Miami-Forever” ballot in November 2017. Conducting a pretest during the survey development period using focus groups was recommended in the study by Johnston et al. (2017). We developed and tested the survey design with extension specialists and researchers working on climate resilience and SLR issues in the state, graduate students, and faculty members working in the field to ensure respondents’ understanding and comprehension of the survey questions. After the survey instruments were finalized, we soft-launched the study by collecting 10% of the sample in the online panel. The soft launch ensured that the online survey flow was executed correctly and there were no major problems with our survey instruments. The soft launch results did not indicate any major issues with our survey.
The survey was designed in both English and Spanish because around 64% of Miami residents speak Spanish, and 28% of them speak English as of 2010.2 Survey invitations were sent to panel members until the quota was completed, and qualified responses were recorded. Participants of the survey were required to be (1) year-round residents of Miami-Dade County, (2) at least 18 years old, and (3) homeowners. The sample was stratified to ensure that the survey participants represent the adult population in Miami-Dade County in terms of gender and racial composition.
After screening, participants answered questions on their perception and knowledge about SLR and their experience with storm-induced and sunny-day flooding. Next, they were presented with information about the benefits and limitations of the three adaptation strategies (beach nourishment, seawalls, and stormwater pumps). To ensure respondents’ comprehension of the benefits and limitations, we asked them to answer a set of true-or-false questions after information about the adaptation strategies was presented. We used the cheap talk script to reduce potential hypothetical bias while introducing the adaptation strategies before conducting the choice experiment (Champ, Moore, and Bishop 2009; Penn and Hu 2019; Wuepper, Clemm, and Wree 2019). The cheap talk script is presented in Appendix Figure A1, and an example of the choice experiment question is presented in Appendix Figure A2. At the end of the survey, participants provided information about their sociodemographics, such as household income and home property values, and whether they had purchased flood insurance in the past 12 months.
4. Methodology
Econometric Models
Following the random utility theory (McFadden 2001), we assume that individuals will choose an alternative with the attribute bundle that maximizes their utility. The individual n’s utility of choosing the alternative i from a choice set of J alternatives in a situation t can be specified as
1
where Xnit is the attribute variables in the study. β is a vector of unknown preference coefficients that weight the exogenous attributes. εnit is the stochastic component of utility, which captures the unobserved factors that affect the utility. A variety of models can be used based on the assumption of the distribution of the preference coefficients and error terms (Bazzani et al. 2017). Equation [1] can be estimated using a conditional logit (CL) model when εnit is independent and identically distributed (i.i.d.) with Gumbel distribution (Meas et al. 2014). However, CL assumes homogeneous preference and independence of irrelevant alternatives (IIA). Previous research studying consumer preference has concluded that heterogeneity needs to be considered from both the methodological and empirical standpoints (Lusk, Roosen, and Fox 2003; Greene, Hensher, and Rose 2006; Ortega et al. 2011; Greene and Hensher 2013; Wongprawmas and Canavari 2017). When heterogeneity exists, the CL model could bring biased results, and it needs a more flexible model. Revelt and Train (1998) propose the mixed logit model (MIXL) that relaxes the IIA assumption and captures the heterogeneity in preference by allowing preference coefficients to vary across individuals. In MIXL, coefficient vector β is assumed to follow a density function f(β|θ) where θ is the parameter of the distribution. Typically, β can be specified as 
, where 
is the population mean and ui is the individual-specific heterogeneity, with a mean of 0 and standard deviation of 1 (Greene 2012). Under the assumption of intra-respondent homogeneity of the MIXL model, the choice probability can be specified as a weighted average of a product of logit formulas evaluated at a different level of β, given the weight by the density function f(β|θ).
In most applications, the coefficient vector is assumed to follow a multivariate normal distribution (Liu et al. 2019). However, research has also found that the multivariate normal distribution may not correctly reflect the real choice behavior (Louviere and Meyer 2008; Louviere et al. 2008). In addition to preference heterogeneity, a scale effect can contribute to the heterogeneity in attribute coefficients. For some respondents, the scale of the error term for some respondents is more substantial than for other respondents. This is also referred to as scale heterogeneity, which may occur when choice behavior has greater dispersion for some respondents. A scaled multinomial logit model (S-MNL) can be implemented to capture scale heterogeneity (Fiebig et al. 2010). Fiebig et al. (2010) further developed a G-MNL that nests MIXL and S-MNL. The models accounting for both scale and preference heterogeneity (namely, G-MNL) perform better than MIXL in all 10 data sets examined in Fiebig et al. (2010). Greene and Hensher (2010) also conclude the improvement of using the generalized mixed logit model (same as G-MNL) over the standard mixed logit model. Following Fiebig et al. (2010), Greene and Hensher (2010), and Greene (2012), we apply the G-MNL model to take into account the preference heterogeneity and scale heterogeneity. The utility function in G-MNL is specified as
2
where εnit is i.i.d. with the Gumbel distribution, and βi is specified as
3
and ui follows a certain distribution, 0 ≤ γ ≤ 1:
4
The G-MNL model is nested with all the major models used for the choice experiment data. To be specific, when τ = 0 and L ≠ 0, the G-MNL results in a MIXL model; when γ = 0 and L = 0, the G-MNL results in S-MNL. The model can be estimated using a simulated maximum likelihood method (Ouma, Abdulai, and Drucker 2007).
To further improve the robustness of the model and its accuracy in reflecting respondents’ choice behavior in this specific study, we take additional steps to refine the model. In this study, respondents are asked to make a decision (choose) among three adaptation strategies along with the status quo option. Although the adaptation strategies vary by attribute levels, the status quo option remains the same in all 10 choice sets. Research has concluded that a higher utility variance exists for the designed alternatives than the utility of the status quo (Scarpa, Ferrini, and Willis 2005; Caputo and Nayga 2017; Caputo et al. 2018). This implies that the unobserved utility of the adaptation strategies might have a more considerable discrepancy than the status quo (Caputo, Nayga, and Scarpa 2013; Bazzani et al. 2017). Therefore, the adaptation strategies may share similar alternative specific random effects that are unobservable, which indicates a higher correlation in the error terms of the adaptation strategies than the status quo (Caputo et al. 2018). In our study, we further improve the G-MNL model by adding the alternative specific ECs with zero mean to capture the potential correlated error terms across different adaptation strategies (Scarpa, Ferrini, and Willis 2005; Gao et al. 2019). The utility function for the four alternatives in this study is specified as
5
6
7
8
where ECn,i (i = 1, 2, 3, 4) is the EC associated with individual n and alternative i. εn,i,t is assumed to be i.i.d. with Gumbel distribution (i = seawall, beach, pump, status quo). αn,i is the individual n’s alternative specific constant of choosing alternative i, and βn,i is the preference coefficient weighting the attributes in the choice experiment. Un,status_quo,t does not include a constant because the status quo is used as the base category to avoid multicollinearity. The definition of the variables in the utility functions is provided in Appendix Table A1 as well.
Different EC structures can be specified based on theoretical justification and empirical estimation. We hypothesize three EC structures given the study context:
Two independent ECs (ECn,123, ECn,4), where ECn,1 = ECn,2 = ECn,3 = ECn,123 ≠ ECn,4.
Three independent ECs (ECn,13, ECn,2, ECn,4), where ECn,1 = ECn,3 = ECn,13 ≠ ECn,2 ≠ ECn,4.
One independent EC for status quo and three random alternative specific constants for three adaptation strategies, where ECn,1 = ECn,2 = ECn,3 = 0 and ECn,4 ≠ 0.
The ECs capture the “alternative specific random individual effects that account for choice situation invariant variation” (Greene 2012, N-587). Therefore, the three cases imply different relationships between the adaptation strategy-specific random effects. Case 1 assumes that the adaptation strategies share the alternative specific random effects. Therefore, the utilities of the strategies are correlated but independent from that of the status quo. Case 2 assumes that the hard structure adaptation strategies (seawalls and stormwater pumps) share the same alternative specific random effect. In contrast, the soft structure adaptation strategy (beach nourishment) and status quo have unique alternative specific random effects. Case 3 assumes each alternative and status quo has independent alternative specific random effects. Thus the utilities of all three adaptation strategies and the status quo are independent. Since prior information on the structure of ECs is unknown, model performance measures such as the log-likelihood, Akaike information criterion (AIC), and pseudo R-squared are compared with select the model with the most appropriate structure.
Estimation of MWTP
We estimate a G-MNL with an EC in the MWTP space. In the preference space, MWTP can be obtained by taking the negative ratio of the coefficients of the nonprice attribute and the cost attribute (−βn.k / βn, fee). MWTP can be estimated directly in the MWTP space model (Louviere, Hensher, and Swait 2000; Train and Weeks 2005; Hynes, Hanley, and Scarpa 2008; Shi, Xie, and Gao 2018). MWTP space estimation may reduce the goodness of fit of the models more than preference space estimation in some cases. This inferiority is minimized significantly when scale and preference heterogeneity are identified using the G-MNL model (Hensher and Greene 2011). In addition, models in MWTP space result in a more behaviorally plausible MWTP range and a better statistical fit (Scarpa, Ferrini, and Willis 2005; Scarpa, Thiene, and Train 2008; Balcombe, Fraser, and Chalak 2009; Shi, Xie, and Gao 2018).
When estimating the models, the coefficient of stormwater utility fees (βn, fee) are fixed as 1 and Feen,i,t is specified as the negative value of the additional monthly stormwater utility fees of the choice experiment. As a result, the coefficients (βn, year and βn, insurance) in the utility function can be directly interpreted as the MWTPs for corresponding attributes (year and insurance). For comparison, the mixed logit models with or without ECs in preference space are presented along with the G-MNL models with or without ECs in MWTP space.
5. Results
Summary of Survey Response
A total of 267 residents completed the survey.3 Table 1 presents the sociodemographic profiles of the sample in the current study and compares it with the 2011–2015 American Community Survey (ACS) 5-Year Estimates and ACS 2016 1-Year in Miami-Dade County. The sample in this study represents the local population in Miami-Dade County compared with the U.S. census data in terms of gender (52.4% vs. 51.5% females). Because the respondents are homeowners, the sample over-represents the middle-aged and elder groups (35–74 years old) compared with the U.S. Census data. In general, respondents have higher educational attainment than the average Miami-Dade County resident (33.3% and 35.6% of the respondents have a bachelor’s degree or graduate degree compared with 18% and 10%, respectively). Using $50,000 as a cutoff value, 21.4% of the respondents have an annual household income below $49,999, which is lower than the average of households in Miami-Dade County (55.8%). The middle-income households ($50,000 to $99,999) account for 35.5% of the sample compared with 25.9% of the households in Miami-Dade County. The high-income range ($100,000 and above) represents 55.8% of the sample, compared with 23% of the households in Miami-Dade County. The divergence in income between the sample and the general population in Miami-Dade county could come from the fact that we required the participants of the survey to be homeowners.
Comparison of Sample and Population’s Sociodemographics
Perception and Preference
We first measured respondents’ awareness, perception, and attitude toward SLR using a five-point Likert scale based on their level of agreement with a set of statements, as presented in Figure 1. It shows that around 76.5% (37.5% + 39.0%) of the respondents in the sample strongly agree or agree that the sea level is rising, and 67.8% (31.1% + 36.7%) agree that the rate of SLR is increasing. Only 25.9% (9.4% + 16.5%) of the respondents never thought about SLR before taking the survey. About two-thirds of the respondents (63.7%) agree or strongly agree that more frequent flooding is due to SLR, and most (71.2%) agree that Florida should minimize future development on the oceanfront. More than one third (43.8%) of the respondents agree that the intensity of hurricanes has increased due to SLR. These results indicate that homeowners in Miami-Dade County have a relatively strong awareness of SLR and its potential consequences.
Respondents’ Perceptions about Sea Level Rise
Using pictures as illustrations, we presented respondents with the major functionality, implementation methods, advantages, and limitations of each adaptation strategy. Following this presentation, respondents were asked about their familiarity with each adaptation strategy. Figure 2 shows that many respondents (55.8% and 45.3%, respectively) have seen seawalls and beach nourishment prior to the survey. Respondents are relatively less familiar with stormwater pumps, with most (44.2%) hearing about them and only 27.0% having seen them before the survey. About 8.6%, 4.5%, and 6.4% of the respondents agree that stormwater pumps, seawalls, and beach nourishment projects, respectively, are implemented in their neighborhoods. At the end of this section of the survey, participants were asked to rank the strategies based on their preference, irrespective of the costs. Beach nourishment is chosen as the most preferred strategy, followed by seawalls and stormwater pumps (Figure 3).
Familiarity of the Three Adaptation Strategies
Ranking Preference of the Three Adaptation Strategies
Regression Results
The final choice experiment data include a total of 2,670 observations based on the 267 respondents answering 10 choice questions. In total, eight models or four sets of mixed logit models in the preference space and generalized multinomial logit models in the MWTP space, respectively, are estimated using Nlogit 6.0 (Greene 2012). The first set of models includes standard models (MIXL and G-MNL), the second set includes models with two ECs (MIXL+2ECs and G-MNL+2ECs, case 1), the third set includes models with three ECs (MIXL+3ECs and G-MNL+3ECs, case 2), and the fourth set includes models with one EC and three random alternative specific constants for three adaptation strategies (MIXL+1EC and G-MNL+1EC, case 3).
All the models are estimated using Halton draws with 500 simulations in Nlogit 6.0 (Revelt and Train 1998; Meas et al. 2014). In the MIXL, all the nonprice attributes are assumed to follow normal distributions, and the coefficient of Fees is specified as a fixed coefficient so that the MWTP for the nonprice attributes is normally distributed as well (Hensher and Greene 2003; Gao and Schroeder 2009). Assuming fixed price parameters are a strong behavioral assumption; the results of these models in this study are only used as the benchmark for comparison with the models in MWTP spaces. As discussed in previous sections, the price coefficient is fixed at one in the G-MNL model in the MWTP space.
Table 2 reports the log-likelihood, AIC, and pseudo R-squared of various models. We find that G-MNL models perform better than MIXL models when they have the same error structure (e.g., G-MNL vs. MIXL, G-MNL+2ECs vs. MIXL+2ECs). G-MNL models have larger log-likelihood and pseudo R-squared and smaller AIC than the corresponding MIXL models. This is consistent with previous studies that showed the G-MNL model in the MWTP space results in a better model fit than the MIXL model in the preference space (Scarpa, Ferrini, and Willis 2005; Scarpa, Thiene, and Train 2008; Balcombe, Fraser, and Chalak 2009). In addition, G-MNL models with ECs perform better than the corresponding MIXL and G-MNL models without ECs (e.g., G-MNL+2ECs/3ECs/1EC vs. G-MNL). These results show that including the ECs improves the model fit and demonstrate the importance of accounting for the alternative specific random individual effects of different options in choice experiments.
Model Fitness Comparisons
All the G-MNL models with ECs are very similar regarding model performance. The pseudo R-squared of G-MNL+2ECs/3ECs/1EC models are 0.310, 0.311, and 0.310, respectively. The AICs of these models are 5,136.2, 5,133.9, and 5,132.0, respectively. After considering the model performance measures and the significance of the ECs of the three models (G-MNL+2ECs/3ECs/1EC), we decided to use the results of the G-MNL+2ECs model. This is because the ECn,2 associated with the soft structure (beach nourishment) in the G-MNL+3ECs model is not statistically significant (Appendix Table A3, row 9, column (9)), indicating a model with two ECs is more appropriate. We select G-MNL+2ECs with a significant EC shared by the three SLR adaptation strategies (Table 3) because G-MNL+1EC (Appendix Table A4) assumes that the utilities of the adaptation strategies are independent. The significant EC shared by the adaptation strategies in G-MNL+2ECs implies that the utilities or preferences of the strategies are correlated. Our results confirm that the SLR adaptation strategies share the same unobservable EC that is different from the status quo. The much larger standard deviation of the EC of the status quo indicates that respondents have a more heterogeneous preference for the status quo than for the adaptation strategies in the current study (Table 3). The more divided opinions on the status quo imply that some residents may consider the current adaptation strategies good enough while others think quite differently.
Regression Results with Two Error Components
Because the G-MNL+2ECs model performs equivalently to the other two G-MNL models with ECs but gives more behaviorally reasonable results, the discussions that follow focus on the G-MNL+2ECs model (Table 3, columns (8) and (9)). Table 3 also includes the MIXL, MIXL+2ECs, and G-MNL models for comparison. Results in Table 3 show that the coefficients of all the attributes of the SLR adaptation strategies are statistically significant. The coefficients of Fees are significantly negative in both MIXL and MIXL+2ECs, implying a negative marginal utility of cost. Seawall, Beach, and Pump are all positive, indicating that respondents in Miami-Dade County are willing to pay more for additional SLR adaptations than maintaining the status quo. The coefficients of Protection (Years) are also positive, which means that respondents are willing to pay more for additional Years of protection associated with the adaptation strategies. The coefficients of insurance are all negative, implying that respondents are willing to pay more to reduce flood insurance premiums brought by additional SLR adaptations. The significance of the standard deviation of the random parameters in all four models in Table 3 reveals the existence of preference heterogeneity regarding the SLR adaptations and their different attributes (Hensher, Rose, and Greene 2005; Meas et al. 2014).
MWTP for SLR Adaptation Strategies
Since the G-MNL+2ECs model is estimated in the MWTP space, the mean estimates of SLR adaption attributes in Table 3 can be directly interpreted as MWTP. Table 3 illustrates that respondents are willing to pay the most for additional seawalls ($21.32/month), followed by beach nourishment ($18.99/month) and stormwater pumps ($16.09/month). Respondents are also willing to pay $0.51/month for each extra year of protection bought by these adaptation strategies. The WTP for an additional 1% decrease in the flood insurance premium is about $1.34/month.
The relative dispersion in MWTP can be measured by the coefficient of variation (CV), the ratio of the standard deviation to the mean.4 The CV can be interpreted as the measurement of preference heterogeneity. A higher CV indicates a more heterogeneous preference for certain adaptation strategies or attributes, and a lower CV indicates a more homogeneous preference. Results in Table 3 illustrate that the MWTP for seawalls is more heterogeneous (CV = 0.54) than those for beach nourishment (CV = 0.46) and stormwater pumps (CV = 0.42). Moreover, years of protection has a higher CV (1.00) than percent change of flood insurance premium (0.88), indicating that the preference for the years of protection provided by SLR adaptation strategies is more heterogeneous than the preference for the changes in flood insurance premium associated with the adaptation strategies.
In Miami-Dade County, 858,289 households are homeowners, at a homeownership rate of 52.2% (U.S. Census Bureau 2018). The total MWTP for seawalls would roughly be $18.3 million per month and $219.6 million per year.5 Total MWTP for beach nourishment and stormwater pumps would be approximately $195.6 million and $165.7 million per year, respectively. Stormwater utility fees are currently collected and used to support the planning, operation, and maintenance of the existing stormwater management systems in Miami-Dade County (Regulatory Economic Resources 2019). In the fiscal year 2018, a $33 million utility fee was collected in Miami-Dade County to support the maintenance of the existing systems (Miami-Dade County Bond Administration 2018). In the same year, the county promoted a $500 million plan for SLR adaptation (Harris 2018). The $500 million budget is slightly smaller but close to the total annual MWTP ($219.6M + $195.6M + $165.7M = $581M) for all three adaptation strategies examined here. The results indicate that public support for SLR adaptation is strong, and there is a potential to generate continuous funding support through stormwater utility fees for additional SLR adaptations.
Factors Influencing Individual MWTPs
To further explore the potential distribution effect of implementing adaptations, we examine the extent to which individual-level MWTPs are influenced by demographic variables, property characteristics, and familiarity with the adaptation strategies. The individual-level MWTPs for each attribute are calculated based on the individual-specific posterior distribution derived from the sequence of observed choices in the experiments (Train 2009). Following Lindsey et al. (1992), we focus on factors related to respondents and properties. Seemingly unrelated regressions are used where the dependent variables are the individual-level MWTP for building a seawall, installing pumps, and replenishing beaches, the additional years of protection for the adaptation strategies, and additional decreases in the flood insurance premium. The use of seemingly unrelated regressions is a convenient approach for comparing the means of MWTPs among different types of respondents while accounting for multiple correlated hypotheses since the individual-level MWTPs are generated from the same model estimates. The independent variables of the model are respondents’ ethnicity, sociodemographics (age, gender, income), home property characteristics (location and value), and familiarity with the SLR adaptation strategies.6 The results are presented in Table 4.
Seemingly Unrelated Regressions on Individual-Level Marginal Willingness to Pay
For the MWTP for the three SLR adaptation strategies (columns (1)–(3)), we find that ethnicity does not significantly affect MWTP for the three specific SLR adaptation strategies, except that Hispanic respondents prefer the pumping system more than other ethnic groups (column (3), third row). Older respondents have a significantly higher MWTP for an additional decrease in flood insurance premium (column (6), fourth row). Male respondents have a significantly lower MWTP for additional years of protection than do females (column (5), fifth row). Household income significantly positively correlated with respondents’ MWTP for seawall and additional decreases in flood insurance premiums (columns (1) and (6), sixth row). Our results indicate that lower-income and young households are less willing to pay extra fees to control the increase in their flood insurance premiums compared with higher-income and older households. The coefficients statistically significant at the 10% significance level are age for additional years of protection (+), male for beach nourishment (+) and additional decrease in flood insurance (−), home values for seawalls (−), flood insurance holder for seawalls (−), and pumping (+). These results indicate that the preferences for different SLR adaptation strategies and the attribute of adaptation strategies depend largely on sociodemographics, such as ethnicity, age, gender, and household income.
The degree of residents’ familiarity with adaptation strategies also influences their MWTP for the strategies differently. For instance, whether heard or seen, the three types of adaptation strategies studied here do not affect residents’ MWTP for all the strategies. Respondents who experienced seawalls being implemented in the neighborhood strongly dislike seawalls. Those who have beach nourishment implemented in the neighborhood strongly prefer seawalls. Respondents whose neighborhoods have pumping implemented strongly dislike beach nourishment. In summary, the distributional effects of SLR adaptations will potentially depend on the extent to which they will affect flood insurance premiums, the neighborhoods with greater concentrations of older and female populations, and higher-income households.
6. Discussion and Conclusion
Discussion
Miami-Dade County is actively developing plans for SLR adaptation. In 2018, Miami Beach, one of the cities in Miami-Dade County, promoted a $500 million plan to elevate roads and install stormwater pumps to adapt to SLR (Harris 2018). Understanding residents’ preference for SLR adaptation is vital to sustaining these projects because the local governments primarily undertake the projects. Local governments engaging in strategies to protect the at-risk areas need to consider residents’ preferences of different SLR adaptations to identify effective outreach and communication efforts and gather public funding support.
Using survey data from 267 homeowners in Miami-Dade County, this study finds that residents are generally willing to support additional SLR adaptation strategies rather than maintaining the status quo. Specifically, our results show that residents prefer seawalls over beach nourishment, and both are preferred over reactionary stormwater pumping systems. Stormwater pumping, as a protective measure, is described as reactionary because it only plays a significant role during flooding and severe SLR (Crider et al. 2014; London 2017). We estimated that the MWTP for seawalls, beach nourishment, and stormwater pumps are $21.32/month, $18.99/month, and $16.09/month, respectively. We found that respondents value the longevity of the adaptation strategies by offering $0.51/month for each extra year of protection. Meanwhile, homeowners care about their flood insurance premium and are willing to pay $1.34/month to avoid a 1% increase in the insurance premium. The results are consistent with previous findings that people generally prefer seawalls to beach nourishment (Fankhauser 1995; Betzold and Mohamed 2017; Rulleau and Rey-Valette 2017). However, it is hard to directly compare the MWTP in our studies with what has been found previously because most of these studies have been conducted in very different settings, using different attributes, across different countries, and for different types of participants. Previous studies in New Zealand, Japan, and Germany have found the MWTP for seawalls ranging from $34.51 (for 0.3 km) to $101.05 (for additional 5 m higher seawalls) annually, and beach nourishment ranging from $11.26 to $48 (Phillips and Council 2010; Matthews, Scarpa, and Marsh 2017a; Meyerhoff, Rehdanz, and Wunsch 2021; Omori 2021). Our study is the few conducted in the United States, and the MWTP estimates are relatively higher than the previous results. This is because we focus on the homeowners and conduct the survey in Miami-Dade County, a place prone to flooding and with a higher level of awareness of SLR due to local campaigns to adapt to SLR. In contrast, most previous studies focus on tourists or residents (Imamura et al. 2016; Matthews, Scarpa, and Marsh 2017b; De Salvo et al. 2018; Omori 2021).
For preference among different strategies, we find that homeowners prefer hard barrier structures (seawalls), especially among higher-income communities, even though the survey provided a description of the benefits and limitations of seawalls. Previous research shows that safety is an essential priority in considering an adaptation strategy (de Bruin et al. 2009). It is likely that homeowners still perceive seawalls as providing the best protection against flooding for their properties. This highlights the need for better communication on the trade-offs of adaptation strategies. For example, hard structure protection may not be the optimal solution from environmental, social, and engineering aspects. Beach access and habitat provide vital market and nonmarket values for the local economy and support residents’ livelihoods in Miami-Dade County (Shivlani, Letson, and Theis 2003; Klein, Osleeb, and Viola 2004). If not well planned, building seawalls may potentially bring increased erosion, visual blight, and loss of public beaches, negatively affecting ecosystems and more impoverished communities in the long run (Caldwell and Segall 2007). Seawalls on some sections of the shoreline may disrupt the natural landscape and cause the degradation of natural shorelines and nearby wetlands (Gittman et al. 2016). In addition, given the porous limestone bedrock under Miami-Dade County, even with seawalls, seawater can still seep through the soil, bypassing the seawalls and disrupting their effectiveness (Goodell 2013; Crider et al. 2014). Conversely, soft shoreline management approaches, such as beach replenishment and restoration of mangroves, offer ecological and recreational benefits (Spurgeon 1999; Vo et al. 2012). However, the availability of sand for replenishment within the range of the nourishment site could be a problem, and it could become costly for large projects (Dobkowski 1998; Parsons and Powell 2001). Hard barrier structures and soft shoreline management approaches have advantages and limitations. The trade-off of different types of adaptation strategies should be examined with consideration of the local residents’ preferences when planning for SLR adaptation. The distributional effects from adaptation should be considered in the planning process. Moreover, a more holistic evaluation of the adaptation strategies is needed when there is a mismatch between public choice and optimal strategies that are more scientifically sound.
We found heterogeneous preferences among different groups of residents. Hispanic residents generally prefer the stormwater pumping system, and other groups of communities seem to be indifferent to the adaptation strategies in the study. Unlike those respondents with higher household incomes who strongly prefer seawalls to other strategies, respondents who have more expensive home properties and have flood insurance tend not to like seawalls. It is possible that homeowners perceive seawalls as reducing the aesthetic value of highly valued properties and being a substitute for flood insurance. Meanwhile, we found that flood insurance holders prefer stormwater pumping, indicating a complementary effect between flood insurance and stormwater pumping rather than the substitutional effect between flood insurance and seawalls. The reasons may be that in Miami-Dade County, stormwater pumps were used as a reactionary strategy. That is, these pumps have limited ability to prevent flooding. However, they can help businesses and communities return to normal faster after hazards in a large area by pumping away large volumes of water during and after heavy rain or flooding (Crider et al. 2014; London 2017). As such, stormwater pumps are more likely to benefit residents who are in the flood zones. Flood insurance holders who are more likely to be in the flood zones are willing to pay more for stormwater pumps as an adaptation strategy against SLR.
The familiarity with SLR adaptation strategies generally has little effect on their preference for those strategies. These results indicate that residents’ familiarity and experience with a particular adaptation strategy do not necessarily lead to its support. Residents may have been making trade-offs based on the actual and perceived effectiveness of the strategy. Thus, their perception and experience need to be examined for adaptation planning. Regarding the preference of adaptation attributes, older and female respondents prefer to have long years of protection and more reduction in flood insurance premiums (in %). Households with higher incomes are the most sensitive to changes in flood insurance premiums (in %). Potential explanations are that older and female respondents are more risk-averse and price-sensitive than younger and male respondents (Kousky 2011).
Contribution to the Literature
A comparison with previous studies demonstrates the significance and contribution of this study, mainly from three different aspects. First, a handful of previous research has put a substantial emphasis on the comparison of residents’ preference and MWTP for hard structure (seawall) versus soft infrastructure (beach nourishment) or other alternatives such as managed retreat (Rulleau and Rey-Valette 2017; De Salvo et al. 2018; Johnston, Makriyannis, and Whelchel 2018; Makriyannis, Johnston, and Whelchel 2018; Oliveira and Pinto 2021; Omori 2021). Little research investigates preferences and MWTP for stormwater pumping and compares three adaptation strategies in one study. Some previous studies only focus on the adaptation strategies without including the attributes of these strategies (Rulleau and Rey-Valette 2017; De Salvo et al. 2018; Makriyannis, Johnston, and Whelchel 2018). For those studies that differentiate adaptation strategies by attribute, the most common attributes are construction characteristics such as height and length of seawalls and area of beach nourishment (Imamura et al. 2016; Johnston, Makriyannis, and Whelchel 2018; Meyerhoff, Rehdanz, and Wunsch 2021; Omori 2021), or attributes related to the common good such as the environment (Makriyannis, Johnston, and Whelchel 2018; Oliveira and Pinto 2021). Given that we have three adaptation strategies with various physical characteristics (for example, it is difficult to compare a one-foot increase in seawall height with a one-foot increase in beach length), we focus on easy-to-communicate attributes related to residents’ livelihood, such as years of protection, change in insurance premium, and stormwater utility fees. Therefore, we contribute to the literature by including both hard and soft strategies and stormwater pumping, which has been rarely studied in previous work.
The second contribution that distinguishes our study from previous research is that we generated insightful policy implications for the local government. Previous studies usually estimate the individual-level MWTP, whereas our research calculates the total MWTP to provide more information for the local government’s decision making. In particular, by recruiting homeowners in Miami-Dade County as participants, we show that residents are willing to support SLR adaptations because the stormwater utility fee they pay is one of the primary sources of funds for the cost of SLR adaptations. In the monetary aspect, our results estimate an annual fund of $581 million for the three adaptation strategies, which is substantial compared with the $500 million budget for SLR adaptation in Miami-Dade County (Harris 2018). The total MWTP estimates demonstrate a feasible continuous funding resource for SLR adaptation in the area.
Last, in addition to the empirical results and policy implications, the significance of the EC models offers insights into an issue that many studies have overlooked when valuing heterogeneous preferences for SLR adaptations. Previous work has shown the importance of scale and preference heterogeneity in choice experiment settings through various experiments (Fiebig et al. 2010; Greene and Hensher 2010; Hensher and Greene 2011). SLR adaptation strategies may have unique features and characteristics that might not be fully captured by elicited attributes in the choice experiment. We use a labeled design of different strategies as alternatives to address this problem, considering that adaptation strategy alternatives might not have identical error terms across different options. The results in the current study demonstrate the importance of incorporating alterative-specific random ECs in modeling people’s choices of different options to adapt to SLR or climate change. First, consistent with previous studies, our results show that consumers’ preference for the SLR adaptation strategies is more homogeneous than for the status quo (Scarpa, Ferrini, and Willis 2005; Caputo, Scarpa, and Nayga 2017). We also find a strong correlation in respondents’ preference for the three SLR adaptation strategies, and the preference is significantly different from that for the status quo. Second, we find respondents have more divided opinions about the status quo than for the adaptation strategies. Although it is safe to include only two alternative specific ECs (one for the status quo and the other for all other options) in our study, future work could explore more complex structures and use information-based criteria to select appropriate structures of the ECs.
Conclusion and Future Research
We demonstrate that homeowners in Miami-Dade County are willing to pay the highest stormwater utility fees to support additional protection provided by seawalls, followed by beach nourishment and stormwater pumps. The total MWTP is slightly higher than the $500 million budget for SLR adaptation in Miami-Dade County (Harris 2018), indicating strong financial support by the residents for SLR adaptations in the area. There are heterogeneous preferences for adaptation strategies primarily influenced by respondents’ sociodemographics and their exposure to potential flooding hazards. This result implies that adopting any single or combination SLR adaptation strategies would result in heterogeneous welfare change for different groups of residents. A comprehensive evaluation of the cost and benefit of different SLR adaptation strategies should be conducted before their implementation. Notwithstanding the results of this study, there are implications for future work. First, the proximity to coastlines might play a critical role in people’s choices of strategies to adapt to SLR (Schmidt et al. 2013). Our random sample is not geographically distributed, limiting this study to explore the effects of proximity to coastline and storm surge areas. Future studies may examine these geographical factors and explore how spatial heterogeneity affects the preference for adaptation strategies. Second, our study focuses on homeowners in Miami-Dade County. Future work may extend the scope to include other coastal communities for comparison with this article. Third, we use stormwater utility fees as a financing mechanism for coastal communities because the damages from floods and storm surges related to SLR are more severe in coastal areas than in inland areas. However, damages from SLR and climate change would affect the inland areas regardless of interconnected labor, assets, and financial markets (Sukop et al. 2018). There is a need to identify effective communication and financing strategies to encourage adaptation to SLR and climate change across heterogeneous social groups and regions, including inland and coastal areas (United Nations Development Programme 2007).
Acknowledgments
The findings and conclusions in this article are those of the authors and do not represent the views of the U.S. Government Accountability Office. We thank the editor and anonymous reviewers for their comments. This study was supported by the Florida Agricultural Experiment Station and by the Climate Change and Florida’s Agricultural, Natural Resources and Human Systems Seed Fund from the UF/IFAS Senior Vice President for Agriculture and Natural Resources and Dean for Research and USDA National Institute of Food and Agriculture, Multistate Hatch project (CRIS:005439).
Footnotes
Appendix materials are freely available at http://le.uwpress.org and via the links in the electronic version of this article.
↵1 After collecting about 10% of the samples from the soft launch of the survey, we estimated the conditional logit model and used the parameter estimate as the prior to generate a new choice experiment. However, the new choice experiment was very close to our original choice experiment, with similar D-errors. Therefore, we decided to use the original choice experiment.
↵2 See https://apps.mla.org/map_data_results&state_id=12&place_id=45000&cty_id=.
↵3 The sample size is large enough for efficiently estimating the population preference parameters, with a 95% confidence interval and a 0.05 margin of error (Dillman, Smyth, and Christian 2014), assuming respondent’s choice of an option is an 80/20 split. The 80/20 split assumption is reasonable considering that each alternative in the choice experiment has a 25% (20/80) chance of being selected. Besides, the sample size is much larger than the minimum sample size requirement if we focus on generating statistically significant parameter estimates for the choice experiments. Based on Rose and Bliemer (2013) and de Bekker-Grob et al. (2015), we only need 100 respondents to generate statistically significant parameter estimates for our choice experiment.
↵4 The standard deviation can demonstrate the deviation and dispersion in MWTP estimates. However, since the scale of the MWTP can be quite different, a relative dispersion needs to be calculated to better compare the dispersion across different scales of MWTP. Therefore, relative dispersion in MWTP is important.
↵5 The sum of $18.3 million is obtained by multiplying the total number of households (858,289) by the average MWTP for seawalls ($21.32).
↵6 Based on the familiarity and experiences with the adaptation strategies, two dummy variables were created. They equal 1 if a respondent has heard or seen a particular strategy; or the strategy is implemented in their neighborhood.
References
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.