Grantee Research Project Results

Final Report: A Consistent Framework for Valuation of Wetland Ecosystem Services Using Discrete Choice Methods

EPA Grant Number: R831598
Title: A Consistent Framework for Valuation of Wetland Ecosystem Services Using Discrete Choice Methods
Investigators: Milon, J. Walter , Weishampel, John F. , Scrogin, David
Institution: University of Central Florida
EPA Project Officer: Hahn, Intaek
Project Period: April 1, 2004 through March 31, 2006 (Extended to December 31, 2007)
Project Amount: $313,797
RFA: Valuation for Environmental Policy (2003) RFA Text |  Recipients Lists
Research Category: Environmental Justice

Objective:

The overall goal is to develop and test a consistent framework to estimate wetland services values. The diverse nature of wetland services negates complete valuation through a single method or data source. Our approach uses a dual modeling strategy to utilize revealed preferences (RP) from housing market models and stated preferences (SP) from a choice model for wetland ecosystem services inCentral Florida. There are four interrelated objectives to implement this strategy: 1) To estimate the revealed preference demand for proximity to wetlands and other water resources using hedonic pricing and discrete choice models of residential property values; 2) To estimate the demand for ecosystem services from different types of wetlands using a stated choice survey; 3) To develop a general valuation function for wetland ecosystem services, and 4) To estimate the implicit value of ecosystem services in wetland mitigation banking markets.

Summary/Accomplishments (Outputs/Outcomes):

Revealed Preferences for Wetlands

Housing Market Data and GIS Analysis

The property sales data used in the hedonic price and discrete housing choice analyses covered the period January 2000 through May 2005. The data were obtained from property tax appraisers in three Central Florida counties: Orange (Orlando), Polk (Lakeland), and Volusia (Daytona Beach). In addition to identifying the sales prices of single-family residential properties, dates of sale, and geographic locations, the data contained a variety of physical attributes commonly included in property value analyses. These included the number of bedrooms and bathrooms, the square footage of the structure under central air/heating, the square footage of the land (or parcel), the age of the home, and the presence of a pool. The sample of property sales and individuals used in the analysis was homeowners who received a ‘homestead exemption’ for property taxes. In Florida the exemption is assigned to a property if it is owned by a Florida resident who uses the property as their primary residence. With this sample of homeowners, we exclude nonresidents and individuals, partnerships, and corporations who purchase properties solely for investment purposes. This definition of the sample is also consistent with that employed in the stated choice wetland valuation survey.

To generate the spatial/environmental variables included in the hedonic and discrete choice models, property sales data were overlaid with GIS land use maps and maps identifying a variety of natural and human-made spatial attributes. Digital land cover/use maps were acquired from the two regional water management districts with jurisdiction in the respective counties: St. John’s River Water Management District (SJRWMD) for Orange and Volusia counties and Southwest Florida Water Management District (SFWMD) for Polk county. These maps include data based on medium and low altitude flight imagery collected in 2000 and 2003 at scales of 1:24,000 and 1:6,000 resulting in an image resolution of 1 meter. The data were analyzed and interpreted into cover and land use types by specialists within the water management districts based on the Florida Land Use and Cover Classification System (FLUCCS).

Using mapping tools in ArcInfo 9.0, single family residential property sales were identified within the landscape and geo-coded. The Euclidean distance between the centroid of each parcel and the edge of the nearest patch of each of the four wetland types and to the nearest lake, patch of upland forest, and patch of agricultural or rangeland. The wetland types are: (1) Hardwood Forests (loblolly bay, tupelo, and bottomland hardwoods); (2) Coniferous Forests (cypress, pond pine, and cabbage palm); (3) Freshwater Marshes (sawgrass, cattail, and other aquatic vegetation); and (4) Wet Prairies (emergent and sparse vegetation).

In addition to classifying wetlands according to the vegetation type, the wetlands were also classified by their potential to be transformed into another land use. Prior research suggests that the permanency of protection for environmental amenities and open space may influence the value of surrounding properties. Protected wetlands appear in two forms compared to Unprotected wetlands.  First, wetlands contained in the Florida Natural Areas Inventory (FNAI) are designated as being regionally significant and are not subject to development in the future because they are owned by a state or local government entity (FNAI Protected).  In addition, other protected wetland areas were identified using property appraisers’ land use codes that designated land set aside by developers and owned by neighborhood homeowner associations; these include conservation easements, undeveloped lands, and open spaces not available for future development (Other Protected).

The three counties compare in a number of ways. Orange is the smallest and most population-dense, Volusia is the most commercial-dense, and Polk is the largest and most agricultural-dense. In all three counties, about one-sixth of the landscape is comprised of wetlands; upland forest and water comprise the smallest portion of the total landscape coverage. In all counties, the average distance to the nearest wetland patch is more than 1,000 meters. While some properties are located adjacent to each of the wetland types, the nearest wetlands are miles away from some properties, most notably in Polk county. Across counties the acreage of the nearest patches of Hardwood Forest and Coniferous Forest wetlands associated with the average property sold are considerably larger on average than the nearest patches of Freshwater Marsh and Wet Prairie wetlands. In all cases, the smallest nearest patches are designated as Unprotected.

Comparing the property parcels, the average price of a property sold in Orange county (expressed in 2000 dollars) was about $45,000 greater than that of Volusia county and about $55,000 greater than Polk county. In terms of physical characteristics, the heated area of the mean property sold was comparable across the counties; however, the area of the land was notably smaller in population-dense Orange county relative to Volusia and Polk counties. The mean number of bedrooms and bathrooms was similar across counties, as was the proportion of properties with pools.

Hedonic Property Models and Implicit Prices for Wetlands

Hedonic property price models (linear and double log) were estimated using the property sales and GIS data for the three counties. In general, the hedonic property models performed well with high explanatory power. Results for the structural and location attributes of the properties indicated a large number of statistically significant relationships that were consistent with the general literature on hedonic property models. To test for differences in the implicit prices of wetland proximity between the three protection status categories, the distance variables were interacted with dummy variables that identify whether the respective wetland was publicly protected from future development (FNAI Protected) or informally protected through an easement or collective ownership by a neighborhood organization (Other Protected).

For the wetland relationships, results indicated that there were considerable differences in the property price effects of wetland types and protection status across the three counties. One wetland type may be either an amenity or disamenity depending on protection status and/or county.  The extent of surrounding residential and commercial development and whether the properties were located in a flood zone had little effect on these relationships. Due to the number of empirical specifications, it is difficult to summarize the individual coefficient results across all wetland types, protection status, and county. Complete details are reported in Section 2, Table 2 – 6, and Appendix 2 – C of the full report.

The most direct interpretation of the relationship between wetland types and property prices is provided by the ‘implicit prices’ of wetland proximity derived from the estimated coefficients in the hedonic property model. These implicit prices account for the multiple interactions within the model and measure the mean incremental ‘willingness to pay’ (positive or negative adjustment to the sales price) needed to acquire a marginal (one unit) change in an attribute. Table ES – 1 reports some of the mean implicit prices and 99% confidence intervals for a one-meter reduction in distance to each of the four wetland types, by protection status, across the three counties. Positive (negative) willingness to pay coupled with confidence intervals that contain solely positive (negative) values indicate amenity (disamenity) effects from proximity to the respective wetland type.

The results in Table ES – 1 indicate that wetland proximity tends to have relatively small overall effects on property prices regardless of the type of wetland. In some cases, however, wetland proximity had relatively larger effects that were directly related to the category of protection status. For example, larger amenity effects were found for proximity to Hardwood Forest ($3.41 per meter) and Coniferous Forest ($6.14 per meter) wetlands designated as FNAI Protected in Orange county. Alternatively, Hardwood Forest wetlands with FNAI protection had larger disamenity effects (-$5.26 per meter) in Polk county. For Freshwater Marsh wetlands, larger disamenity affects were found for proximity to Unprotected patches in Orange county (-$4.97 per meter), FNAI Protected patches in Volusia county (-$6.80 per meter), and Other Protected patches in Polk county (-$3.58).  Lastly, the most consistent–and perhaps most striking–findings reported in Table ES – 1 were associated with proximity to Wet Prairie wetlands. In all cases, proximity to FNAI Protected patches led to larger disamenity effects on property prices, most notably in Polk county (-$18.40 per meter). These implicit prices indicate that the effect of wetlands on property prices was very location specific and depended on the wetland type and protection status. It was not possible to conclude that wetlands, in general, were either amenities or disamenities in the Central Florida region.

Table ES – 1. Mean Implicit Price of 1 Meter Reduction in Property Distance to Nearest Wetland Type by Protection Status

Discrete Choice Property Models

Discrete choice analysis was also used with the housing data described earlier as an alternative method to estimate demand and implicit prices for wetlands near residential property. The approach was to model the probability of housing choice as a function of the same structural and spatial attributes in the hedonic property price models. To generate choice sets for each property sale, a home buyer’s choice set was defined as the chosen alternative and 249 randomly drawn properties purchased/sold within a three-month window around the purchase/sales date of the respective property within the same county. For example, if an individual purchased a property in Orange county in April, then the choice set contains 250 properties, each of which sold in either March, April, or May within the same year.

Similar to the hedonic price analysis, conditional logit models were estimated using both linear and log specifications. Due to the variety of empirical specifications, it is also difficult to summarize the discrete choice model coefficient results across all wetland types, protection status, and county. Complete details are reported in Section 2, Table 2 – 7, and Appendix 2 – D of the full report. Overall, the bulk of the wetland and protection status coefficients were not statistically significant, indicating that the probability of choosing a property was not affected by the proximity of the property to wetlands nor of the protection status of the nearest wetlands. In fact, none of the coefficients were significant for the models estimated with the Volusia county data. However, two notable effects were found in Orange and Polk counties where the nearest Wet Prairie wetlands were perceived as a disamenity in the respective housing markets. This result was comparable to the disamenity effects for Wet Prairies in the hedonic price analysis.

Similar to estimation of the implicit prices from the hedonic models, the estimated discrete choice models may be used to estimate the compensating surplus (willingness to pay) associated with incremental changes in distance to wetlands using estimates of the marginal utilities of distance and income. However, as the wetland distance coefficients were insignificant in most cases and the price coefficient was insignificant in all cases (see Table 2-8 and Appendix 2-D), surplus estimation from the discrete choice models was not pursued. Also, one of the project’s original objectives to combine property-based discrete choice models with stated choice models for wetland valuation was not pursued.

Stated Preferences for Wetland Ecosystem Services

Choice Survey Design

A stated preference choice survey was developed to estimate residential property owner preferences and willingness to pay for wetland ecosystem services through a public land acquisition program across Orange, Polk and Volusia counties. The approach followed random utility theory and employed a conjoint-based (pairwise choice) survey to elicit preferences for public acquisition of specific land parcels. Each parcel was comprised, in part, of the same four wetland types used in the property valuation models (Hardwood Forest, Coniferous Forest, Freshwater Marsh and Wet Prairie). Each parcel also provides unique ecosystem services, such as serving as habitats for wildlife and recharging groundwater recharge supplies. To mitigate potential hypothetical biases that are a common criticism of stated preferences experiments, the survey design presented choice scenarios in the context of a land acquisition program representative of established programs in central Florida. In addition, the pairwise choice sets contained actual parcels of undeveloped private lands sampled from the landscape rather than hypothetical parcels constructed through factorial survey design methods commonly employed in stated choice surveys.

The sample included 600 homeowners recruited from the three counties using a random sample from the same property appraiser information described previously for the property valuation models. Subjects completed the survey at various university facilities equipped with desktop computers and internet connections to access a server that presented the survey and recorded responses.

The survey was comprised of three sections. First, the introductory section greeted participants with a short voice-scripted and visual description of the state and local land conservation initiatives and solicited opinions on state programs and spending. The section concluded with visual and verbal descriptions of the four wetland types and presented an example of one of the pairwise choice sets to instruct subjects about the response format. The second section contained a menu of six pairwise choices of parcel sites selected from a total of sixty-eight sites identified across the three counties. Pairwise combinations included within home county sites, home county and other county sites, and only other county sites to test whether the location (jurisdiction) of the sites influenced the site choice. For a given pair of sites, participants were presented with information sequentially through maps and aerial photos about: i) the county in which the sites were located; ii) the total site acreage and the number and percent of acres in each of the eight land categories; iii) the number and percent of acres in each of the four wetland categories; and iv) the ecosystem rating (or ratings) for each site. The maps and aerials photos were over-written with multi-media ‘flash’ technology to highlight and distinguish between the site attributes and to identify their location within the surrounding landscape and the respective counties. A final choice screen summarized the site attributes and cost per household/acquisition prices (randomly assigned from $10 to $125 in $5 increments) and allowed participants to review the maps and aerial photographs. After reviewing the information, subjects were asked to choose their most preferred of the two sites or to choose ‘neither’ site. Finally, the third section solicited socio-economic information from participants.

A key element of the survey design was the ratings for ecosystem services at each of the 34 parcel sites included as part of the pairwise choice information. Ratings for the thirty-four sites were developed for three services: Wildlife Habitat Quality, Groundwater Recharge Potential, and Surfacewater Connectivity. The rating protocol was based on Uniform Mitigation Assessment Method (UMAM) developed by the Florida Department of Environmental Protection for wetland mitigation assessment and utilized a variety of GIS data, assessment information from environmental agencies within Florida, and personal visits by the research team to each site. Site ratings were assigned on a 1 (lowest) to 10 (highest) scale for each of the three ecosystem services. Complete details on the rating protocol are provided in Section 3.2.2 of the full report. To assess the significance of these ecosystem service ratings in site choices, a ‘short’ version of the survey included only the Wildlife Habitat service and a ‘long’ version included all three ecosystem service ratings as part of the descriptive information for each parcel site.

Stated Choice Survey Results and Implicit Prices for Wetland Ecosystem Services

Response data were used to estimate econometric models for the short and long versions of the survey using a nested logit choice specification to account for the neither option in the site choices. Also, three different specifications of the land composition variables were used along with the socioeconomic characteristics of each respondent in the nested logit models.

As expected, the ‘price’ individuals would pay to acquire a site–cost per household–had a negative and significant effect on the probability of site choice in all cases. Similarly, the ecosystem service rating coefficients for Wildlife Habitat and Groundwater Recharge differed significantly from zero in all cases indicating that these ecosystem services were perceived as amenities by respondents. In contrast, the Surfacewater Connectivity coefficients were not significant. Results from the long version also indicate that Wildlife Habitat was perceived as a relatively more important ecosystem service than Groundwater Recharge.

Additional results for both the short and long version of the survey indicated that respondents preferred larger wetland parcel sites over smaller sites. However, this effect diminished in significance and magnitude in the short version. This suggests that respondents gave relatively more weight to a full set of ecosystem service ratings versus a single rating as a determinant of the site choice. Also, results indicated that only two types of land composition, Wetland Hardwood Forest and Wetland Prairie acreage, had significant and positive effects on the probability of site choice with both versions of the survey. Of the remaining six land types identified in the site choice information, only Upland Forest acreage had a significant effect in most of the estimated models.

Another interesting result was that with the short version of the survey, the probability of site choice diminishes significantly in all cases as the proximity of a site to the respondent’s residence increases. In contrast, no significant proximity effects are found with the long version. One possible explanation is that Wildlife Habitat and diversity were viewed as a local public good so proximity matters in the absence of any additional information about the services offered by the site. But, to the extent that Groundwater Recharge services were perceived as providing benefits that were independent of site location (at least regionally), the amenity effects of proximity may diminish when subjects were informed about the presence and quality of these additional services.

Other location issues that were tested included whether the level of development surrounding a parcel site (and visible to respondents in the aerial photos used in the survey) influenced site choice. Results indicated no significant difference in preferences for sites that were in lightly developed or rural areas. However, some models indicated that the choice probability was significantly less for urban sites with adjacent development relative to rural sites.

Finally, county-specific effects of parcel site location were examined with the hypothesis that respondents would be more likely to choose a site located within their county of residence versus a site located in either of the other two counties. Results did not support this hypothesis. Complete details on the estimated models are report

Journal Articles:

No journal articles submitted with this report: View all 5 publications for this project

Supplemental Keywords:

watersheds, groundwater, ecosystem indicators, cost benefit,, RFA, Scientific Discipline, Economic, Social, & Behavioral Science Research Program, Economics & Decision Making, Economics, Urban and Regional Planning, decision-making, Social Science, Ecology and Ecosystems, public values, deliberative policy, revealed preference, policy analysis, economic tradeoffs, environmental values, public policy, surveys, decision analysis, wetlands preservation, discrete choice, aquatic ecosystems, web-based methods, stated preference, econometric analysis, environmental policy

Progress and Final Reports:

Original Abstract

2004 Progress Report

2005

2006 Progress Report