2006 Progress Report: A Consistent Framework for Valuation of Wetland Ecosystem Services Using Discrete Choice MethodsEPA Grant Number: R831598
Title: A Consistent Framework for Valuation of Wetland Ecosystem Services Using Discrete Choice Methods
Investigators: Milon, J. Walter , Scrogin, David , Weishampel, John F.
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 Period Covered by this Report: April 1, 2006 through March 31, 2007
Project Amount: $313,797
RFA: Valuation for Environmental Policy (2003) RFA Text | Recipients Lists
Research Category: Economics and Decision Sciences
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 joint modeling strategy to integrate revealed preferences (RP) from a discrete choice model of a housing market and stated preferences (SP) from a choice model for wetland ecosystem services. There are four interrelated objectives to implement this strategy: (1) to estimate the demand for proximity to wetlands and other water resources using discrete choice and hedonic pricing models of residential property values; (2) to estimate the demand for ecosystem services from different types of wetlands that are not in proximity to residential property using a stated choice survey; (3) to develop and test a combined discrete choice model from the RP and SP data to produce a general valuation function for wetland ecosystem services; and (4) to estimate the implicit prices of wetland services in wetland mitigation banking markets.
Work during year two of the project focused on completing the integration of the GIS land-use maps with the three-county property sales datasets for the hedonic property price and discrete housing choice analyses and the development of the browser-based SP land acquisition survey.The data reveal sizable differences in the composition of the landscape within a given county and across the three counties. However, in each county about 15 percent of the landscape is comprised of wetlands (17.9% in Orange County, 16.7% in Volusia County, 13.9% in Polk County). Considering the wetland categories, Orange and Volusia Counties have comparable percentages of wetlands coverage in all cases, though Volusia County has about 30,000 more wetland acres than Orange County. For these counties, wetland hardwood forests and wetland coniferous forests are the dominant wetland types, comprising more than 70% of the total acreage defined as wetlands. Similar to Orange and Volusia Counties, more than 70% of the total wetland acreage is attributed to wetland hardwood forests and coniferous forests, with about 50% of the total acreage defined as wetland coniferous forest. The discussion below considers the RP and SP portions of the project, in turn.
Hedonic Property Price and Discrete Housing Choice Analyses
The property sales data used in the hedonic price and discrete housing choice analyses span the three county study area (Orange County, Volusia County, and Polk County) over the period January 2000–May 2005. All unqualified sales and other sales that did not appear to be arm’s-length transactions were discarded in constructing the datasets for estimation. A set of property attributes that were common and directly comparable between the three counties was included in the analyses. These include the number of bedrooms and bathrooms, the square footage of the physical structure under central air/heating, the square footage of the land (or parcel), the age of the home, and the existence of a pool.
For generation of the spatial/environmental variables to include in the hedonic and discrete choice models, the GIS land-use maps were overlaid with the tax appraiser property sales data and the digitally coded GIS maps identifying a variety of natural and human-made spatial attributes. Using mapping tools in ArcInfo 9.0, single family residential property sales were geo-located. The Euclidean distances from the centroid of each parcel to the edge of the nearest of the four types of wetlands, and to other natural amenities (e.g., lakes and upland forests), were measured. In addition, the sizes of these nearest natural amenities were also measured.
The econometric analysis focused initially upon the hedonic property price model. Preliminary estimates of alternative specifications of the hedonic models revealed some counterintuitive results, which led to concern about possible mismeasurement of the distance and size (area) variables. To investigate, we selected a sample of properties to test for consistency in the distance calculations and identification of the correct land use. It was discovered that several factors could be attributed to potential mismeasurement of the variables. These included the temporal lags between the land-use maps and the date property sales, the measurement of the centroids of the parcels, and inconsistencies between the GIS land-use maps obtained from the water management districts and the county property appraiser tax rolls. The lags between the creation of the land-use maps by the state water management districts relative to the continuous urban and rural developments in the three counties were believed, in some cases, to result in the identification of upland forests and wetlands within the landscape that no longer existed due to residential and commercial development.
In addition to the land-use distance and area calculations, the composition of the land surrounding the residential parcels was measured. This process employed Computer-Aided Drafting and Design (CADD) software to partition the counties into grids spanning 1 square mile. ArcGIS was then used to aggregate the landscape into the four original land uses: residential lands, commercial/industrial lands, agricultural lands, and undeveloped lands and water. The percentage of each land type contained in the square-mile grids was then calculated. The residential parcel map was then overlaid with the land-use grid so as to link the land uses within the grids to the individual parcels. With this step, the final residential property datasets were complete, and estimation of the hedonic price and discrete housing choice models could proceed.
Estimation results from selected specifications of the hedonic property price and discrete housing choice (conditional logit) models indicate that the sign and the size of the effect of proximity to a given type of wetland differs between counties, and within a given county, the sign and the size of the effect of proximity differs between the types of wetlands. (The choice sets used for estimation of the discrete choice models were defined in terms of the sales dates of the properties, similar to Smith and Banzaf , “Meta analysis in model implementation: choice sets and the valuation of air quality improvements,” forthcoming in Journal of Applied Economics.) For example, in Orange and Volusia Counties, the hedonic estimation results indicate that, all else constant, the sales price of the average property is negatively related to the distance to the nearest wetland hardwood forest. However, the average sales price in Polk County increases as the distance to the nearest wetland hardwood forest increases. Alternatively, in Orange and Polk Counties, the average sales price is positively related to the distance to the nearest wetland marsh, whereas in Volusia County, the relation is negative. The hedonic results also indicate that the sign and the size of the effect of the total area of the nearest wetland differ across counties for a given category of wetland and within counties across the categories of wetland. As with the distance coefficients, the average property price is positively related to the size of the nearest wetland in some cases (e.g., all wetland categories in Orange County) and negatively related in other cases (e.g., wetland hardwood and coniferous forests and wetland prairies in Volusia County). Consistent with the hedonic price analysis, the conditional logit estimation results indicate that in some cases, wetlands may be considered an amenity, while in other cases they appear to be a disamenity. Further, for a given wetland type, the effects differ between counties in many cases, and for a given county, the effects differ between wetland categories.
SP Land Acquisition Analysis
For development of the stated choice survey, the first task was to select a representative sample of wetlands from the Central Florida landscape that could be used to estimate willingness to pay to acquire wetlands with different types and levels of ecosystem services. After selecting the sites and identifying their composition (i.e., measuring the acreage in each of the land-use categories), the sites were visited to validate the information attained from the aerial photographs and GIS maps and to acquire additional information about their characteristics (e.g., the ability to sustain wildlife habitats). Given the available data, the development of the survey instrument was undertaken in cooperation with the University of Central Florida’s (UCF’s) Institute for Simulation and Training (IST), and draft versions of the instrument were discussed in focus groups before being submitted to the University’s Institutional Review Board (IRB).
The Sampling Strategy and Site Selection. ArcGIS software was used to generate wetland density maps for the three counties. These maps were constructed from the county land-use maps that exclude areas comprised of dense residential lands, commercial and industrial lands, and large lakes. The sites defined for the pairwise choice survey were selected by generating 60 random points within the study area of each county. One kilometer square buffer areas were then generated for each random point. The buffers were then joined with the parcel ownership data from the county property appraiser tax rolls described in the previous section to identify the ownership of the parcels and the land composition within each buffer. Parcel ownership was then viewed individually within each buffer to exclude lands that were already developed or which were located within well-known, publicly owned tracts (e.g., state parks and preserves).
Site Visitations and the Wetland Assessment Methodology. The selected sites were visited and visual inspections were conducted in order to verify the current status of the lands (i.e., that they were not under development) and to assess the ecological and wetland conditions at sites. The site classification methodology followed the Uniform Mitigation Assessment Method (UMAM). The approach was originally developed by the Florida Department of Environmental Protection for mitigation purposes; however, the approach differs from the mitigation assessment methodologies that have traditionally been employed in wetland assessments. The quantitative ecological assessment of the sites consisted of three scored categories. These include wildlife habitat, groundwater recharge, and surface water connectivity.
The Browser-Based Pairwise Choice Survey. The repeated pairwise choice survey was developed in cooperation with the UCF’s IST. The survey instrument appears in a voice-scripted, interactive, Internet-based format, allowing all responses to be logged directly to an Internet server managed by IST. The survey is comprised of three parts that include:
- An introduction section that contains: (i) a set of “warm-up” questions soliciting selected information about participants (e.g., their county of residence and the number of years they have lived in Florida) and their opinions on selected public programs; and (ii) a background information section describing the various types of wetlands contained in Central Florida and the format of the pairwise choice experiment comprising part 2, described below.
- A section containing the repeated pairwise choice instrument that presents a total of six pairs of sites, one pair at a time, the attributes of the sites (e.g., the land composition and ratings of the sites’ wildlife habitat, groundwater recharge, and surface water connections) described visually and numerically, and the choice “screen” for each pair.
- A conclusion section that solicits selected socio-economic information from participants, including age, gender, educational attainment, membership in and contributions to selected environmental groups, and household income.
Upon the completion of the programming of the draft version of the survey instrument by the UCF IST during summer 2006, focus groups were conducted. Edits were then made to the survey format and the script. As required, the final version of the survey instrument was submitted to the UCF IRB for approval prior to being administered in the field. The survey was approved by the IRB in December 2006. One of the two long versions of the survey to be administered in Orange County may be accessed from the Internet for review by performing the following steps:
- Set the screen resolution to 800x600, and turn on the computer speakers in order to hear the voiced script.
- Launch the Internet browser, and type the following URL into the address box: http://xlandsurvey.crata.ucf.edu. Press return to access the survey site.
- In the “User name” box that appears, type “landsurvey” (exclude quotes), and in the “Password” box, type “view!Survey”. Press enter.
- In the next dialogue box that appears, type “OR5000” in the Survey Code box, and type a four-digit number in the personal identification number (PIN) box. Press enter. The survey will then begin.
Note: A given four-digit PIN may be only used once, so if the PIN that you enter is rejected (i.e., you get an error message that says “Your PIN must be distinct”), then it means that the PIN you have entered has already been used. You will therefore need to enter another four-digit number.
Current work is focusing upon completing the RP hedonic property price and discrete housing choice analyses and the interview surveys with residential homeowners using the SP survey. Both hedonic and discrete choice models are being investigated in order to evaluate the relative differences in the implicit values for different types of wetlands that result from the alternative modeling approaches. After the completion of the stated choice surveys, a joint RP and SP model will be developed and estimated after combining the RP and SP datasets. The model will provide a general valuation function for wetland ecosystem services as described in Objective (3) (see page 1).