Jump to main content.

Improving Air Quality Benefit Estimates from Hedonic Methods

Quick Links

Research Funding

Metadata


Objectives:
The objective of the research proposed herein is to critically examine the relative importance of data aggregation, attribute tradeoffs, and variation caused by space and time within a hedonic benefit study using a single, pooled cross-section, time-series data set. The primary focus will be on the hedonic price of air quality. The analysis will be conducted in the South Coast Air Basin, which consists of the California counties Los Angeles, Orange, Riverside, and San Bernadino for the period 1980-1996. These counties contain over one hundred cities, which will generate sufficient spatial variation to test the relative importance of community characteristics on hedonic price estimation. The extensive time series nature of the data will allow the required temporal variation.

Approach:
We will use a hierarchical linear model (sometimes called a mixed model or a multilevel model) for our analysis of the relationship between air quality and housing prices. Our empirical analysis will employ the two techniques that are generally favored in the theoretical literature, full maximum likelihood and restricted maximum likelihood.

Expected Results:
The important outputs of our research will be: (1) an extensive pooled cross-section, time-series data set that will include approximately 50,000 observations per year over the study period, 1980-96; (2) the application of a new procedure to the task of estimating hedonic price functions; and (3) estimated marginal willingness to pay values that will be more suitable for benefits transfer. A set of case studies will become the benchmark for future benefits transfer work.

Metadata

EPA/NSF ID:
R825826
Principal Investigators:
Thayer, Mark
Murdoch, James C.
Beron, Kurt
Technical Liaison:
Research Organization:
San Diego State University
Texas-Dallas, University of
Funding Agency/Program:
EPA/ORD/Valuation
Grant Year:
1997
Project Period:
October 1, 1997 to September 30, 1998
Cost to Funding Agency:
$124,931
Project Status Reports:
      There were five major accomplishments over the reporting period.

      First, we conducted an extensive literature review of approximately sixty journal articles on the hedonic price method. For each article we provided a detailed review, a discussion of the article's relevance, and information regarding data used and conclusions drawn concerning air pollution.

      Second, we created a unique cross-section, time series data set consisting of approximately 1.6 million observations over the period 1980-95. An observation relates to a specific sale of an owner occupied single family home in our study area. The dependent variable in the empirical analysis is the home sale price of these dwellings. The independent data set includes variables that correspond to four types of attributes: house quantity and quality, neighborhood, community, and environment. House size or quantity is described through such variables as square footage of living space, number of bathrooms and bedrooms, and lot size or land area. House quality is depicted by variables such as the presence of pool, number of stories, roof type, number of fireplaces, etc. Neighborhood quality is based primarily upon neighborhood characteristics contained in the data tapes for both the 1980 and 1990 census. Community variables such as school quality and the crime rate are measured at the city level. Air pollution is measured by both pollutant concentration readings taken at monitoring stations and visibility readings from local airports. The pollution data were obtained from two sources: the South Coast Air Quality Management District (SCAQMD) and the National Climatic Data Center (NCDC). Variables that depict neighborhood and community influences are matched to the housing data using common location indicators. For example, each subset of the data set is coded with GIS coordinates allowing accurate matching of attributes at the various levels of aggregation. However, the air pollution data require the following multi-step procedure in order to assign a specific single family home the appropriate pollution measures: (1) the air pollution data, obtained from monitoring station or airport readings is aggregated into a summary statistic (e.g., annual average, median, etc.); (2) these summary data are entered into the Surfer computer program to generate isopleth contours; (3) the isopleths are utilized to create pollution levels at grid points that cover the entire study area; (4) each census tract is assigned the pollution level of the grid point that is closest to its centroid. Each single family home in a specific census tract is assigned the same pollution value.

      Third, we have estimated cross-sectional benchmark hedonic equations for each year in our sample. The results indicate that air pollution, as measured by ozone, total suspended particulates, and visibility, is a significant determinant of home sale price. We then examined the sensitivity of the benchmark equations by utilizing alternative pollution measures, using more detailed neighborhood variables, and estimating other functional forms. We have also employed a hierarchical linear model (sometimes called a mixed model or a multilevel model) to study the relationship between air quality and housing prices. These tests indicate that air pollution has a robust impact on home sale prices.

      Fourth, we estimated inverse demand curves for visibility for the entire sample period using two different approaches: (1) ordinary least squares; and (2) two stage least squares. The dependent variable in the demand estimation is the individual marginal willingness to pay for a change in visibility, determined as follows. The hedonic equation is differentiated with respect to visibility using the characteristics corresponding to each individual data point. The hedonic prices are converted to constant 1995 dollars using the consumer price index and aggregated to the census tract level since data on individual homeowner attributes (e.g., income, education, and ethnicity) are not available. Thus, the dependent variable represents the average hedonic price or marginal willingness to pay in the census tract. This procedure produces approximately 2000 data points per year. Given these implicit prices, the inverse demand curve is estimated by regressing price against quantity (visibility) and other household shift variables, such as income and education. The independent variable set performs as expected and the estimated demand curves are generally robust to sensitivity analysis.

      Fifth, we integrate the estimated demand curves to produce an estimate of the benefits that accrue to households given a change in visibility. Our preliminary analysis indicates that previous studies, based both on the hedonic price method and the contingent valuation method, have seriously underestimated the economic value of visibility improvements.

      In conclusion, the research conducted to date contains several innovations relative to the existing literature. First, this report contains the most comprehensive review of the previous work attempting to determine benefits from housing data. Second, the data set assembled for this project is the most comprehensive ever assembled. The number of observations exceeds that used in any previous hedonic price study. In addition, the quality of the data and the application of Surfer to provide detailed assignment of pollution to individual homes are unprecedented. The time-series element of the data, which allows the identification of time-dependent market segments, has also not been used before in the study of urban air pollution. Third, our estimation procedures represent the latest innovations in econometrics. In the estimation of the hedonic price function we use both traditional (OLS, fixed effects) and more innovative (random effects, hierarchical linear model) procedures. Likewise, the demand curve estimation uses both OLS and instrumental variables. The result is a benefit assessment study that is state of the art.

      Publications/Presentations:

      We have presented our preliminary results regarding visibility to the South Coast Air Quality Management District. We are currently preparing a journal article on the value of visibility.

      Future Activities:

      Our research has identified three specific areas that require more detailed investigation. First, we will continue to explore the use of the hierarchical linear model to determine the relative importance of aggregation levels. Second, our hedonic price function results indicate that neighborhood effects, measured by such variables as distance to beach, school quality, etc., have a significant impact on accurate determination of the individual effect of the pollution variables. These variables, in contrast to the house specific variables, are generally measured with error, which may interfere with the researcher's ability to ascertain the independent influence of the air pollutants. The solution to this problem may lie in the use of a spatial autoregressive framework. Third, our two stage least squares demand curve estimates are based on the use somewhat uncertain instrumental variables. We intend to conduct a thorough search of potential instrumental variables.

Project Reports:
Final

Objective: The objective of the research was to critically examine the relative importance of data aggregation, attribute tradeoffs, and variation caused by space and time within an air quality hedonic benefit study using a single, pooled cross-section, time-series dataset. The analysis was conducted in the South Coast Air Basin, which consists of the California Counties of Los Angeles, Orange, Riverside, and San Bernardino, for the period 1980?1995. These counties contain over 100 cities; this is sufficient spatial variation to test the relative importance of community characteristics on hedonic price estimation. The extensive time series nature of the data provides the required temporal variation.

Three specific estimation approaches were used. We used traditional methods to estimate benchmark hedonic models. These results then were compared to results derived from both hierarchical linear and spatial econometric models.

Summary/Accomplishments: Four specific research tasks were completed. In Task 1, we conducted a comprehensive literature review concerning the hedonic method, data sources, and the magnitudes of estimates from the hedonic method. Approximately 60 journal articles were reviewed. For each article, we provided a detailed review, a discussion of the article's relevance, and information regarding data used and conclusions drawn concerning air pollution.

In Task 2, we assembled the multilevel data necessary for the estimation of the hedonic models. The data were assembled at the site, neighborhood, school district, and environmental levels. All data were geo-referenced and maintained in a geographic information system (GIS). The dataset consisted of approximately 1.6 million observations over the period 1980?1995. An observation relates to a specific sale of an owner-occupied single family home in our study area. The dependent variable in the empirical analysis is the home sale price of these dwellings. The independent dataset includes variables that correspond to four types of attributes: house quantity and quality, neighborhood, community, and environment. House size or quantity is described through such variables as square footage of living space, number of bathrooms and bedrooms, and lot size or land area. House quality is depicted by variables such as the presence of a pool, number of stories, roof type, number of fireplaces, and so on. Neighborhood quality is based primarily upon neighborhood characteristics contained in the data tapes for both the 1980 and 1990 census. Community variables such as school quality and the crime rate are measured at the city level. Air pollution is measured by both pollutant concentration readings taken at monitoring stations and visibility readings from local airports. The pollution data were obtained from two sources: the South Coast Air Quality Management District (SCAQMD) and the National Climatic Data Center (NCDC). Variables that depict neighborhood and community influences are matched to the housing data using common location indicators. For example, each subset of the dataset is coded with GIS coordinates, allowing accurate matching of attributes at the various levels of aggregation. However, the air pollution data require the following multi-step procedure to assign a specific single family home the appropriate pollution measures: (1) the air pollution data, obtained from monitoring station or airport readings, are aggregated into a summary statistic (e.g., annual average, median, and so on); (2) these summary data are entered into the Surfer computer program to generate isopleth contours; (3) the isopleths are utilized to create pollution levels at grid points that cover the entire study area; and (4) each census tract is assigned the pollution level of the grid point that is closest to its centroid. Each single family home in a specific census tract is assigned the same pollution value.

In Task 3, we used traditional methods to estimate benchmark hedonic models. Results indicate that air pollution, as measured by ozone, total suspended particulates, and visibility, is a significant determinant of home sale price. We also examined the sensitivity of the benchmark equations by utilizing alternative pollution measures, using more detailed neighborhood variables, and estimating other functional forms. The benchmark results are compared to results derived from a hierarchical linear hedonic model (HLM) that specifically recognizes the multilevel structure of the data. The HLM results indicate that air pollution has a robust impact on home sale prices. Task 3 also was extended to explicitly investigate spatial econometric models. There were two reasons for this extension to the research plan. First, the spatial econometric model provides an alternative methodology for analyzing neighborhood effects. Thus, we believed that our HLM analysis would benefit from a comparison with results from an explicitly spatial model. Second, generalized method of moments (GMM) estimators for spatial models were developed during the research period that enabled us to estimate spatial models with large numbers of observations. The spatial econometric model results also indicate that air pollution has a robust impact on home sale prices.

In Task 4, we provided numerical estimates of the monetary benefits of changes in air pollution in the South Coast Air Basin. These estimates are presented throughout our full final report and are summarized in Chapter VI for total suspended particulates (TSP).

Conclusions: The research conducted in this project contains several innovations relative to the existing literature. First, this report contains a comprehensive review of the previous work attempting to determine benefits from housing data. Second, the dataset assembled for this project is the most comprehensive ever assembled. The number of observations exceeds that used in any previous hedonic price study. In addition, the quality of the data and the application of Surfer to provide detailed assignment of pollution to individual homes are unprecedented. The time-series element of the data, which allows the identification of time-dependent market segments, also has not been used before in the study of urban air pollution. Third, our estimation procedures represent the latest innovations in econometrics. In the estimation of the hedonic price function, we use both traditional (OLS, fixed effects) and more innovative (random effects, hierarchical linear model, spatial econometric) procedures. Likewise, the demand curve estimation uses innovative econometric methods. The result is a benefit assessment study that is state of the art.

Book Chapter:

Beron K, Hanson Y, Murdoch J, Thayer M. Hedonic price functions and spatial dependence: implications for the demand for urban air quality. In: Florax R, Anselin L, eds. New Advances in Spatial Econometrics. Springer-Verlag, 1999. (An analysis of the impact of spatial dependence on estimates for the demand for air quality.)

Dissertation/Thesis:

Hanson Y. Proper estimation of hedonic models. Ph.D. Dissertation, University of Texas at Dallas (expected 1999). (A systematic treatment of the effects of different modeling strategies for the air pollution, spatial weights matrix, and parametric specification of the hedonic model.)

Journal Articles:

Beron K, Murdoch J, Thayer M. Hierarchical linear models with application to air pollution in the South Coast Air Basin. American Journal of Agricultural Economics. A specific treatment of the HLM model in estimating the marginal willingness to pay for air quality.

Beron K, Murdoch J, Thayer M. The benefits of visibility improvement: new evidence from the Los Angeles metropolitan area. Journal of Real Estate Finance and Economics. (An examination of the impact of a specific aspect of air quality-visibility, or the ability to clearly see distant objects-on housing values. Our analysis is based on data for the period 1980 through 1995 with visibility, other air pollution data, and other characteristics.)

Presentations:

Murdoch J. Hedonic price function and spatial dependence: implications for the demand for urban air quality. Presented to the International Regional Science Association, Santa Fe, NM, November 1998.

Murdoch J. Hierarchical linear models with application to air pollution in the South Coast Air Basin. Presented to the American Agricultural Economics Association, Nashville, TN, August 1999.

Thayer M, Murdoch JC, Beron K. Improving air quality benefit estimates from hedonic models. Presented to the South Coast Air Quality Management District, 1998.

Thayer MA. The benefits of visibility improvements. Presented to the South Coast Air Quality Management District, Diamond Bar, CA, June 1998.

Supplemental Keywords: property values, environmental damage, stigma.


Local Navigation


Jump to main content.