Grantee Research Project Results

Final Report: The Role of Locational Equilibria and Collective Behavior in Measuringthe Benefits of Air Pollution Policies

EPA Grant Number: R828103 aka R826609
Title: The Role of Locational Equilibria and Collective Behavior in Measuringthe Benefits of Air Pollution Policies
Investigators: Smith, V. Kerry , Sieg, Holger
Institution: North Carolina State University
EPA Project Officer: Chung, Serena
Project Period: February 21, 2001 through February 20, 2004
Project Amount: $199,948
RFA: Decision-Making and Valuation for Environmental Policy (2001) RFA Text |  Recipients Lists
Research Category: Environmental Justice

Objective:

The overall objectives of this research project were to: (1) extend a new framework (developed by Epple and Sieg [1999]) for estimating consumer preferences for spatially delineated public goods so that the approach could be evaluated for use with environmental resources; (2) develop the data and implement a large-scale application of the framework for estimating consumer preferences for air quality in southern California; (3) evaluate the feasibility of computing the general equilibrium responses of housing prices (because of household adjustment) to exogenous changes in air quality conditions; (4) compare general and partial equilibrium welfare measures for these exogenous changes in local public goods; and (5) compare the results from the locational equilibrium framework with results from hedonic and random utility models previously used in policy analysis.

The Epple-Sieg approach describes household behavior within a framework that has a long history in economics. It suggests that when public goods are spatially distinct, it should be possible to observe households moving to different locations seeking public goods that more closely align with their preferences. If analysts are prepared to specify how public goods such as air quality contribute to well being, the Epple-Sieg analysis suggests that if there are measures of the spatial amenities and disamenities along with housing costs, it should be possible to estimate consumer preferences for these amenities.

Summary/Accomplishments (Outputs/Outcomes):

This research developed four extensions to the Epple-Sieg approach: (1) a methodology for measuring price indices that adjusts for heterogeneous housing; (2) a framework that includes a range of restrictions implied by locational adjustment that are used with aggregate data on the types of households located in specific communities; (3) a method for using the preference estimates to compute the new prices that would result from an exogenous change in amenities reflecting the adjustments households would make in response to that change; and (4) an analysis of the issues that arise in measuring partial and general equilibrium welfare measures (willingness to pay [WTP]) for large-scale policy changes.

To our knowledge, this research is the first effort to provide a practical resolution for measuring changes in welfare when there are general equilibrium implications of large changes in spatially differentiated public goods.

Methods and Data

The empirical implementation of the model focuses on the Los Angeles, CA, metropolitan area, which consists of the area west of the San Gabriel Mountains and includes parts of five counties: Los Angeles, Orange, Riverside, San Bernardino, and Ventura. We assume that the school district corresponds to the community selected by a household when making its locational decisions.

A comprehensive database on housing markets in the Los Angeles metropolitan area was assembled based on housing transactions collected by a firm, Transamerica Intellitech, that sells access to these data to realtors and financial institutions. These data contain housing characteristics and transaction prices for virtually all housing transactions in southern California between 1989 and 1991. Table 1 reports means of the main variables in the housing sample by counties for these years.

 

Table 1. Descriptive Statistics of the Housing Sample

Variable	Los Angeles	Orange	Riverside	San Bernardino	Ventura
Number of Observations	59,466	21,560	19,644	16,304	7,851
Market Value of House	245,325	262,142	143,450	151,845	253,684
Number of Bathrooms	1.89	2.14	2.06	2.09	2.22
Number of Bedrooms	3.01	3.31	3.25	3.26	3.46
Lot Size (square feet)	7,758	6,761	10,118	8,866	9,364
Building Size (square feet)	1,537	1,718	1,614	1,595	1,800
Swimming Pool (proportion)	0.16	0.16	0.12	0.13	0.14
Fireplaces (proportion)	0.53	0.26	0.93	0.79	0.79
Age	37.6	23.7	9.4	16.2	17.3
1 Mile of Coast (proportion)	0.025	0.054	0.0	0.0	0.026
3 Miles of Coast (proportion)	0.073	0.153	0.0	0.0	0.146
Commute Time (minutes)	26.7	26.2	32.8	29.8	26.3

The data used to derive the housing price indices are based on housing sales and do not necessarily match the housing stock in the area. According to the 1990 U.S. Census (the year closest to our sample), 70 to 80 percent of all households change houses within 10 years. In our sample, 15 percent of the houses were built within 1 year, whereas in the 1990 Census for the area, only 3 percent were built in 1989-1990.

One approach to address differences between our transaction sample and the U.S. Census is to construct weights for the transaction sample. The basic idea is to assign large weights for the observations in the transaction sample that appear to be undersampled based on estimates from the 1990 U.S. Census and small weights to observations that are relatively oversampled. We construct these weights such that the transaction sample matches three attributes of the stock of all owner-occupied housing in the U.S. Census: location, the number of bedrooms, and the age of the house. Hence, the observations on the housing sales are weighted by the relative weight of their school district. Conditional on school districts, they are weighted by the number of bedrooms (in 5 categories) and the age of the house (8 categories), assuming age and bedrooms are independent (the correlation between bedrooms and age was not available). We estimate the price indices and the quartiles for housing using both weighted and unweighted samples (as discussed below, air pollution readings are assigned to each house based on its location in relation to air pollution monitors. The weights applied for housing expenditures also were applied in estimating the average pollution readings in each school district). We estimate the community-specific price indices using ordinary least squares using fixed effects for each community. Another aspect of our research investigated the effects of different approaches for measuring the price indices. This analysis supported the use of the fixed effects and is summarized in Sieg, et al. (2002).

Two spatially delineated public goods, air quality and education, are considered in our model. Data for observed concentrations of ozone and particulate matter less than 10 microns in diameter (PM10) were obtained from the California Air Resources Board monitoring records. Southern California has one of the most extensive air quality monitoring systems in the world. In the five counties of interest, no fewer than 45 monitors were measuring ozone each year from 1987 to 1992 (after eliminating monitors active on less than 50 days). Beginning in 1987, less than 19 were measuring PM10.

There is not a one-to-one correspondence between air quality monitors and school districts. However, the large number of monitors enabled us to interpolate air quality measures for each school district. In Riverside County, for example, one-half of all houses are within 4.5 miles of at least one monitor, and 90 percent are within 9.9 miles. Air pollution is measured using a centered 3-year average (about the sales year) of pollution readings for the nearest monitor to each house in each year of the average. Our temporal window is 1989-1991. We computed the community estimate by using the Census owner weighted and unweighted averages for the houses sold in each community.

Three measures of air pollution, ozone concentration, ozone exceedences, and particulate matter (PM10), are considered for evaluating the effects of air quality. Ozone is measured in parts per million (ppm) as the average of the top 30 1-hour daily maximum readings at a given monitor during a year. The average of the 30 highest ozone readings was the most consistently important air pollution measurement in our estimates and the primary focus of the analysis.

To measure the quality of education, we constructed measures of performance from standardized test scores for each school district using the 1992-1993 California Learning Assessment System Grade Level Performance Assessment test. Each student taking this exam is assessed at one of six performance levels (with six being the highest level).

Table 2 summarizes a few key features of the data, price index estimates, ozone readings, as well as our estimates for the price and income elasticity demand for housing using the two different community definitions. Our estimation strategy assumes that housing can be “unbundled” into a set of structural characteristics of a house (including the lot size) and the site-specific public goods. Hedonic models construct a price index to account for the first influence and measure the residual due to community effects as the fixed effect. This process allows us to interpret our estimates for the housing demand parameters as relating to standardized units of housing. In the case of the 93 school districts version of the model, we reported both Census weighted and unweighted results for comparison. Asymptotic standard errors for the estimates are given in parentheses below the estimated elasticities.

 

Table 2. Summary of Key Features of Sieg, Smith, Banzhaf, and Walsh Model Results

	Counties		Communities
Area	Los Angeles (59,466 mi2)a Orange (21,560) Riverside (19,644) San Bernadino (16,304) Ventura (7,851)		93 to 103 school districts
Prices	Transactions Price 1989-1991		143,450-262,142b
Price Index	Fixed Effects (unbundle community from structure)		1-6.672 (weighted) 1-7.010 (unweighted)
Ozone	Average of 30 highest hourly readings		0.150 0.091-0.232
Elasticities	93c		103c
	weighted	unweighted	weighted
Price	-0.39 (0.015)	-0.116 (0.043)	-0.037 (0.017)
Income	0.734 (0.011)	0.762 (0.009)	0.729 (0.014)
aThe numbers in parentheses are the number of observations for housing sales from each county used in our analysis.
bThis range is the average sales price range across counties in nominal dollars.
cThese numbers correspond to the number of communities used in the model. The values in parentheses below the estimated elasticities are the estimated asymptotic standard errors.

Evaluating Changes in Ozone

Two different policy analyses were conducted with the model. The first, reported along with the details of developing the estimates in Sieg, et al. (2003), considers the observed change in ozone conditions from the time period used to estimate the model 1990 to 1995. The second considers the U.S. Environmental Protection Agency's (EPA) policy evaluation of the impact of the Clean Air Act Amendments for Los Angeles as part of the First Prospective Analysis.

Table 3 summarizes a few of the findings from our analysis of the Hicksian WTP for the change in ozone from 1990 to 1995. Two features of this comparison are important. First, the price flexibility of income () varies with household and is larger than the income elasticities for housing reported in Table 2. Second, there are marked differences in what we have defined as partial and general equilibrium measures for the improvement in ozone. These estimates are reported as positive values. The minus sign in the table legend implies we are reporting the absolute magnitude of the proportionate reduction in ozone experienced by households in this area from 1990 to 1995.

 

Table 3. Some Illustrative Results for 1990 to 1995 Change in Ozone^a

County/Area	1990-95 -Δozone	WTPPE	WTPGE	S
Study Area	0.193	1,210	1,371	0.021	13171
Los Angeles	0.208	1,472	1,556	0.023	1.184
Orange	0.180	901	1,391	0.022	1.187
Riverside	0.207	834	372	0.012	1.099
San Bernadino	0.163	738	367	0.012	1.100
Ventura	0.062	164	725	0.018	1.150
a s = WTPGE/m  = across school districts 1.052 to 1.441

Results

As the results in Table 3 suggest, there can be large differences in the partial equilibrium (PE) and general equilibrium (GE) measures of WTP. Moreover, the differences are not always in one direction. Households originally living in Ventura County experienced a gain measured with WTPGE for the smallest improvement in ozone conditions (about 6 percent reduction in the concentrations), but the overall adjustment also allowed them to realize a 1.2 percent reduction in their housing prices. Taken together, these changes imply a $561 difference per household between the WTPGE and WTPPE in Ventura County.

The reverse situation characterizes households in Riverside County. Although they had one of the largest improvements in ozone conditions (a 20.7 percent reduction), their WTPGE is less than the WTPPE measure. This outcome arises because their housing costs rose by nearly 6 percent.

We conducted two types of comparisons to benchmark our results with others to attempt to address these concerns. Our estimates of annual WTPGE for improved air quality (reduced ozone) fall between 1 and about 2.3 percent of annual income. These estimates would be low in comparison to the health-based assessments of the changes in all air pollutants in the Los Angeles area. However, this assessment is largely judgmental; we did not conduct a formal assessment. The EPA Retrospective and Prospective assessments offer benefit estimates for changes in these same pollutants based largely on their health impacts. We used them as a basis to form our evaluation.

Our interpretation of these results implies that a prudent assessment of our results would treat the effects attributed to the ozone measure as reflecting air pollutants that tend to be jointly present. For the case of our study area, this would likely include particulate matter. The second more specific comparison is discussed in the next section and relates to a comparison of the model's assessment of the benefits measured for the scenarios considered in the EPA's Prospective scenarios relative to the average benefit measures they report based on health effects.

One of the objectives of this research was to assess whether the locational equilibrium framework could be used to provide a localized assessment of large-scale policy effects. Such efforts would parallel the detailed regional and national ambient air quality assessments the EPA undertakes as part of its regulatory assessments. EPA's regional scale methodology relied on the Urban Airshed Model (UAM) and the variable grid-UAM (UAM-V) for ozone concentrations, which also were the primary focus of our policy analyses. UAM with higher resolution was applied to the Los Angeles area. EPA's Los Angeles modeling closely matched the area of our economic model. It is a 65 by 40 array of 5-kilometer resolution grid cells. The domain includes the South Coast Air Basin from Los Angeles to beyond Riverside.

EPA's analysis relates to the ambient concentrations of ozone that result from emissions in future years (2000 and 2010). EPA developed both a base year level of emissions and projected growth of polluting activities for 2000 and 2010 in the absence of Clean Air Act Amendments (CAAA) requirements. These projections were modified to reflect control assumptions in each of these two "future years." For volatile organic compounds (VOCs) and nitrogen oxide (NOx), both contributors to ambient ozone, the national assessment estimates a 27 percent reduction because of CAAA's effects on VOCs in 2000 and a 35 percent reduction in 2010. For NOx, the reduction is comparable for 2000 (i.e., 26 percent) and a little larger in 2010 (39 percent).

The EPA projections, with and without CAAA, considered the conditions for two separate 3-day periods (June 23-25, 1987, and August 26-28, 1987), with baseline conditions augmented by the emission profiles associated with each scenario.

Table 4 summarizes the VOC and NOx emission totals for each scenario in Los Angeles. With the assistance of the EPA staff and its contractors, the three analyses undertaken for Los Angeles were interpolated to a consistent 5-kilometer by 5-kilometer grid cell pattern for the study area included in our analysis.

Our analysis had access to the results of three sets of simulations from the air diffusion models: (1) baseline runs for ambient ozone concentrations in 1990; (2) ambient concentrations in 2000 and 2010, when air pollution regulations are "frozen" at federal, state, and local controls corresponding to their 1990 levels of stringency and effectiveness; and (3) concentrations in 2000 and 2010, when federal, state, and local rules promulgated under the 1990 CAAA are implemented (the ozone concentrations estimated to correspond to the reduced emission levels presented in Table 4).

Table 5 reports for each scenario the averages for the area as a whole (in the first row under each scenario), each county, and then for two of the school districts we profiled in describing the projected changes in ozone concentrations. Our policy comparison solves the locational equilibrium model under the 2000 and 2010 business-as-usual ozone levels (ba00 and ba10). Then, for comparison, the 2000 and 2010 equilibria are computed starting from the ba cases, under the assumption that the 1990 CAAA regulations are implemented in the years 2000 and 2010 (ct00 and ct10).

 

Table 4. EPA Projected VOC and NOx Emissions for Los Angeles (Tons Per Day)

Souce/Pollutant	Base 1990	Without CAAA		With CAAA
		2000	2010	2000	2010
VOC
Area	758	770	817	607	7000
On-Road Mobile	1,179	999	1,168	410	213
Point
Low Level	197	196	196	196	158
Elevated	1	3	2	3	2
Total	2,135	1,968	2,237	1,216	1,073
NOx
Area	450	467	529	453	463
On-Road Mobile	993	1,280	1,573	879	626
Point
Low Level	216	186	186	139	139
Elevated	19	19	12	18	8
Total	1,678	1,953	2,300	1,489	1,236

 

The first column reports the average proportionate reduction in ozone implied by replacing the baseline 2000 (and 2010) "without CAAA" with the projected "with CAAA controls" for the initially assigned school district (Δozone). Δp in column 2 (labeled "actual") corresponds to the proportionate changes in prices experienced by households whose initial assignment was the identified county or school district. Thus, they reflect the result of any movement between school districts. ΔP̃ in the next column corresponds to a measure for what happened to prices in the original location, also reported as a proportionate change. The remaining statistics correspond to the average value for general equilibrium WTP, the WTPGE relative to income, and the WTPPE.

Averaged over all school districts, the discrepancy between the WTPGE and WTPPE measures is small, under $100 per household on average. This result is not surprising because the analysis holds the number of households constant. They resort among a finite set of alternatives under the changed conditions. Improvements in the amenity conditions at one location lead to movements to that area (adjustment is assumed costless), price adjustment results, and the balancing effect of the market reduces the gains available.
 

 

Table 5. Alternative Benefit Estimates With (ct) and Without (ba) CAAA Regulations for 2000 and 2010

	ozone	Actual Δ	Initial ΔP̃	WTPGE	WTPGE/m	WTPPE
2000
Area-wide	-0.124	0.001	0.002	798	0.012	708
County
Los Angeles	-0.114	-0.005	-0.004	933	0.013	819
Orange	-0.134	-0.007	-0.011	869	0.013	534
Riverside	-0.154	0.042	0.046	179	0.006	517
San Bernadino	-0.179	0.045	0.045	238	0.007	613
Ventura	-0.097	-0.009	-0.008	497	0.011	344
Selected School Districts
Long Beach (89)	-0.046	-0.022	-0.026	499	0.011	49
Claremont Unified (75)	-0.019	-0.011	0.028	607	0.010	1151
2010
Area-wide	-0.105	0.000	0.000	789	0.012	723
County
Los Angeles	-0.094	-0.006	-0.002	941	0.012	940
Orange	-0.106	0.010	-0.016	821	0.012	413
Riverside	-0.183	0.063	0.059	186	0.006	547
San Bernadino	-0.143	0.028	0.029	270	0.008	487
Ventura	-0.094	-0.012	-0.026	539	0.011	21
Selected School Districts
Long Beach (89)	0.061	-0.031	-0.041	488	0.011	-236
Claremont Unified (75)	-0.142	-0.031	0.005	481	0.011	524

All of these estimates were developed using the parameter estimates for the 103 school district models (with Census weights for the hedonic estimates of community specific prices). On average, they are comparable to the estimates derived in EPA's national assessment on a per household basis. Table 6 compares the per household health-related benefits for all criteria pollutants (in 1990 dollars) for 2000 and 2010, with the per household estimates from the Sieg, et al., model for Los Angeles using only the ozone improvements. This type of comparison is not intended to validate the locational model. Rather, it provides an order of magnitude comparison. Despite the parsimony in the parameterization of the Los Angeles analysis, it does appear to gauge the improvements in air quality conditions to be about the same level of importance as the EPA assessment.

 

 

Table 6. Comparing the Average Household Gains Estimated in U.S. EPA's Analysis with the Los Angeles Model

Source	Year
	2000	2010
1. U.S. EPA Prospective Analysis Health Benefits
Aggregate (billion)	68	108
Per household (dollars)	648	1,028
2. GE Benefits for Los Angeles Area
Area Wide Average per household	798	789
Range	-47 to 6,136	-118 to 7,347

The last row in Table 6 also highlights a point illustrated at the school district level in Table 7. It reports the range of estimates for the WTPGE across households in these school districts. Households in the lowest income districts do experience losses, despite the fact that the air quality in these communities does improve. The losses stem from increases in their housing costs.

There is an important caveat to this conclusion. The locational equilibrium model requires that we define a finite and fixed set of alternatives. The lowest income households have "nowhere to go" in the model. In a real world setting, there may be areas outside the domain defined by the market area that would provide locations they could "adjust to." Of course, this outcome is not guaranteed. The model assumes costless adjustment, and the low-income households may well be most sensitive to this assumption. Thus, it is not certain that adjustment to areas outside the assumed "market" would mitigate or eliminate these estimated losses.

Overall, these analyses demonstrate that it is feasible to develop models of household preferences for housing and site-specific amenities at a level of spatial resolution that is consistent with the EPA's air quality assessments for policy. Thus, it would be possible in principle to develop micro-consistent models of household behavior at the urban scale level that would be consistent with the more detailed urban scale assessments undertaken for the analysis of air quality.

This research demonstrated that it is possible to estimate household preferences for site-specific air quality conditions that are based on the aggregate predictions implied by a locational equilibrium model. Moreover, it is illustrated how to address questions about the implications of the equilibrium effects on benefit measures derived from property values for changes in spatially delineated public goods. To illustrate how the framework could be used in a specific policy, estimates were developed for the average household's annual WTPGE for the reductions in ozone concentrations in Los Angeles between 1990 and 1995. The measure considered both the ozone improvement and the changes in prices and other exogenous public goods households would make through relocation. They were estimated at about 2 percent of a household's annual income.

It is important to note that the models developed are not full-scale GE models. They do not consider adjustment in factor markets along with commodity markets as is done in the national computable GE models. Nonetheless, they do offer the ability to link economic adjustment to spatial features of economic activities and, as such, to consider one aspect of household adjustment to spatially differentiated amenities. The Los Angeles analysis is the first large-scale application, and considerable additional research would be needed before economic urban scale models, consistent with policy needs, would be available. Specific attention would need to be given to the effects of employment opportunities in locational choice and to evaluate the importance of transactions costs for general equilibrium adjustments. Although these are important issues, the information necessary to address these questions can be assembled. Moreover, some independent research by Bayer, et al. (2003), has suggested related modeling structures that may help to resolve some of the methodological issues associated with implementing the models.

Journal Articles on this Report : 7 Displayed | Download in RIS Format

Publications Views

Other project views:	All 31 publications	10 publications in selected types	All 9 journal articles

Publications

Type	Citation	Project	Document Sources
Journal Article	Banzhaf HS, Smith VK. Meta analysis in model implementation: choice sets and the valuation of air quality improvements. Journal of Applied Econometrics 2007;22(6):1013-1031.	R828103 aka R826609 (Final) R829508 (2004)	Abstract: Wiley Online-Abstract Exit Other: Ideas-Abstract Exit
Journal Article	Sieg H, Smith VK, Banzhaf HS, Walsh R. Interjurisdictional housing prices in locational equilibrium. Journal of Urban Economics 2002;52(1):131-153.	R828103 aka R826609 (Final)	Abstract: Science Direct - Abstract Exit
Journal Article	Sieg H, Smith VK, Banzhaf HS, Walsh R. Estimating the general equilibrium benefits of large changes in spatially delineated public goods. International Economic Review 2004;45(4):1047-1078.	R828103 aka R826609 (2001) R828103 aka R826609 (Final)	Abstract: Wiley - Abstract Exit Other: Ideas - Abstract Exit
Journal Article	Smith VK, Pattanayak S. Is meta-analysis a Noah's ark for non-market valuation? Environmental and Resource Economics 2002;22(1-2):271-296.	R828103 aka R826609 (2001) R828103 aka R826609 (Final)	Full-text: London's Global University - pdf Exit Abstract: RTI International-Abstract Exit
Journal Article	Smith VK, Banzhaf HS. A diagrammatic exposition of weak complementarity and the Willig condition. American Journal of Agricultural Economics 2004;86(2):455-466.	R828103 aka R826609 (2001) R828103 aka R826609 (Final)	Abstract: American Journal of Agricultural Economics-Abstract Exit Other: Ideas-Abstract Exit
Journal Article	Smith VK, Sieg H, Banzhaf HS, Walsh R. General equilibrium benefits for environmental improvements: projected ozone reductions under EPA's Prospective Analysis for the Los Angeles air basin. Journal of Environmental Economics and Management 2004;47(4):559-584.	R828103 aka R826609 (Final)	Full-text: ScienceDirect-Full Text HTML Exit Other: ScienceDirect-PDF Exit
Journal Article	Smith VK, Banzhaf HS. Quality adjusted price indexes and the Willig condition. Economics Letters 2007;94(1):43-48.	R828103 aka R826609 (Final) R829508 (2003) R829508 (2004)	Abstract: Science Direct-Abstract Exit

Supplemental Keywords:

air pollution benefits, environmental equity, benefit transfer practices, locational equilibrium, ozone improvement, household adjustment, general equilibrium benefit measures, nonmarket valuation of public goods, spatially delineated public goods, optimizing behavior., RFA, Scientific Discipline, Economic, Social, & Behavioral Science Research Program, Economics, decision-making, Social Science, Economics & Decision Making, air pollution policy, ecosystem valuation, locational equilibria, collective choice process, models

Progress and Final Reports:

Original Abstract

2001 Progress Report

2002