Chapter 3. Revealed preference methods
Revealed preference (RP) methods refer to a range of valuation techniques which all make use of the fact that many (non-market) environmental goods and services are implicitly traded in markets, which allows then for RP methods to uncover these values in a variety of ways, depending on the good in question and the market in which it is implicitly traded. For example, demand for nature recreation can be estimated by looking at the travel costs associated with this activity, with recent developments linking this to geographical information systems to improve accuracy in mapping natural attributes at recreational sites or distances to those sites. Another prominent application are hedonic price techniques which value environmental goods and services as attributes or characteristics of related purchases, notably of residential property, or are used to evaluate the relationship between wages and the occupational risk of death and injury. Finally, averting behaviour and defensive expenditures approaches occur when individuals take costly actions to avoid exposure to a non-market bad. An important development in RP is the ever-growing sophistication of econometric methods brought to bear, reflecting a broader interest in much of applied economics on crucial matters such as causal inference.
3.1. Introduction
An established theme in the appraisal of public policies is the desirability of quantifying in monetary terms the intangible impacts of these proposals (where relevant and practicable) on the well-being of the public. For example, within the domains of environmental or health policy, it is increasingly recognised that these intangible impacts are likely to comprise a meaningful component of the total benefits of policy interventions. However, many of these impacts are non-market goods (or bads). This means that the value that the public places on these impacts cannot simply be observed with reference to market information, such as price and consumption levels. This has given rise to the proliferation of methods that have sought to uncover, in a variety of ways, the value of non-market goods. Some of the more prominent of these methods have been around for many years. Yet, their increasing use – most notably in environmental policy – has provided an additional impetus both in respect of, on the one hand, ever greater sophistication in application and, on the other hand, scrutiny regarding validity and reliability of these methods. This chapter provides an overview of one of the most popular approaches to value non-market goods: revealed preference methods.
The unifying characteristic of revealed preference methods is the valuation of non-market impacts by observing actual behaviour and, in particular, purchases made in actual markets. The focus is solely on use values. To use the terminology of Russell (2001) these methods seek to quantify the market “footprint” of non-market goods (or bads). There are a number of different approaches that have been proposed to fulfil this objective. Boyle (2003) provides a review of the main three methods, summarised in Table 3.1: i) hedonic pricing; ii) travel cost; and iii) averting or defensive behaviour.
Table 3.1 (column 2) outlines the specific aspects of revealed economic behaviour that each method has sought to examine. This might entail the observation of purchases of durable goods such as property in the case of hedonic pricing, or double-glazed windows in the case of defensive expenditures. In most cases, individual or household behaviour is the main focus. Behaviour in each of these markets is thought to reveal something about the implicit price of a related non-market good (or bad). However, the conceptual framework underpinning each approach is different (Table 1: column 3). For example, the purchase of a property can be conceived of as buying a differentiated good whose price depends on a number of characteristics of the property, including the prevalence and quality of environmental amenities its vicinity. In the case of defensive expenditures, this could entail the purchase of a substitute market good such as double-glazed windows in order to compensate for the existence of a non-market bad such as road traffic noise.
RPMs have been applied in a variety of contexts (Table 3.1, column 4). The strength of these approaches is that they are based on actual decisions made by individuals or households. This is in contrast to stated preference methods (Chapters 4and 5) which ask people how they would hypothetically value changes in the provision of non-market goods. For some commentators this, in principle, makes the findings of market based studies the more reliable indicator of peoples’ preferences. This is because they provide actual data on how much people are willing to pay to secure more of a non-market good or to defend themselves against the harm caused by a non-market bad. Of course, the reality is somewhat more complicated. For example, it is not necessarily straightforward to uncover these values in practice. Superiority – relative to alternative valuation methods – in practice might better be considered on a case-by-case basis.
This chapter provides an overview of the conceptual bases of a range of revealed preference approaches to the valuation of non-market economic impacts. The most important issues underpinning the theory of each approach, and implications for their practical application, are highlighted. In each case, give one or more case study examples are given, to illustrate the way the approach has been used, and how some of the theoretical or empirical issues were tackled. The chapter also reviews the latest methodological developments for each of the techniques. The objective is that these discussions will serve to suggest how applicable each of the approaches might be to the valuation of non-market economic impacts in areas other than those in which they have already been used.
3.2. Hedonic price method
The Hedonic Price Method (HPM) (Rosen, 1974) estimates the value of a non-market good by observing behaviour in the market for a related good. Specifically, the HPM uses a market good via which the non-market good is implicitly traded. The starting point for the HPM is the observation that the price of a large number of market goods is a function of a bundle of characteristics. For instance, the price of a car is likely to reflect its fuel efficiency, safety and reliability; the price of a washing machine might depend on its energy efficiency, reliability and variety of washing programmes. The HPM uses statistical techniques to isolate the implicit “price” of each of these characteristics. There are many possible applications of the HPM (e.g. the market for wine, Gustafson et al., 2011), but two types of markets are of particular interest in non-market valuation: a) property markets; and, b) labour markets.
In terms of the housing market, the HPM uses housing market transactions to infer the implicit value of the house’s underlying characteristics. We can describe any particular house by its structural characteristics (e.g. the number and size of rooms, the presence and size of a garden), its location/accessibility (e.g. proximity to schools, shops, roads), neighbourhood characteristics (e.g. crime rate) and the local environment and nearby amenities (e.g. air quality, proximity to green spaces). The price of a house is determined by the particular combination of characteristics it displays, so that properties possessing more and better desirable characteristics command higher prices and those with larger quantities of bad qualities command lower prices, everything else being constant. The HPM is concerned with unbundling the contributions of each significant determinant of house prices in order to identify marginal willingness to pay for each housing characteristic. The method has been used extensively in real estate research (Herath and Maier, 2010).
Rosen (1974) presents the theoretical rationale for this analysis, showing that the utility benefit of marginal changes in one component of the bundle of attributes in a composite good like housing can be monetised by measuring the additional expenditure incurred in equilibrium. For example, we might assume that, in general, people would prefer a quiet residential environment to a noisy one, but since no market exists for the amenity “peace and quiet”, we have no direct market evidence on how much this amenity is valued where people live. However, peace and quiet can be traded implicitly in the property market. Individuals can express their preference for a quiet environment by purchasing a house in a quiet area. A measure of the value of peace and quiet is then the premium that is paid for a quieter house compared with a noisier but otherwise identical one. These firm foundations in economic theory and observable market behaviour, rather than on stated preference surveys, make the method desirable from an environmental policy perspective.
The HPM involves collecting large amounts of data on prices and characteristics of properties in an area, and applying statistical techniques to estimate a “hedonic price function”. This function is a locus of equilibrium prices for the sample of houses. These prices result from the interaction of buyers and sellers in the property market in question. If the array of housing characteristics in the market is approximately continuous, then we can say that buyers will choose levels of each characteristic so that its implicit price is just equal to buyers’ marginal valuation of the characteristic. Then, the slope of the hedonic price function with respect to each characteristic is equal to the implicit price. The appropriate functional form for this regression specification is arguable, but many empirical studies have estimated semi-logarithmic regression models of the form:
[3.1]
where the dependent variable (
) is the natural logarithm of the sale price for each property transaction i in labour market φ in period t. The independent variables might include structural housing characteristics sit, neighbourhood characteristics nit, environmental characteristics xit, unobserved labour market characteristics fj, and other unobserved components eit. In recent years, the use of geographical information systems (GIS) and the availability of GIS data on environmental features and neighbourhood characteristics have increased the detail, flexibility and accuracy with which these attributes can be linked to house locations (Kong et al., 2007; Noor et al., 2015).
There is a long tradition of studies using the HPM to estimate the effect of environmental amenities and disamenities on property prices, with the first environmental study, an application to air pollution, dating back to 1967 (Ridker and Henning, 1967). Since then, a very large number of studies have analysed the price impacts of a wide range of environmental amenities such as water quality (Walsh et al., 2011; Leggett and Bockstael 2000; Boyle et al., 1999), air quality (Smith and Huang 1995; Bayer et al., 2009) preserved natural areas (Correll et al., 1978; Lee and Linneman 1998), wetlands (Doss and Taff 1996; Mahan et al., 2000), forests (Garrod and Willis 1992; Tyrvainen and Miettinen 2000; Thorsnes 2002), beaches (Landry and Hindsley 2011), agricultural activities (Le Goffe 2000), nature views (Benson et al., 1998; Paterson and Boyle 2002; Luttik 2000; Morancho 2003), urban trees (Anderson and Cordell 1985; Morales 1980; Morales et al., 1983) and open spaces (Cheshire and Sheppard 1995, 1998; Bolitzer and Netusil 2000; Netusil 2005; McConnell and Walls 2005). These environmental hedonic studies typically focus on a single or a very limited number of environmental attributes, thereby possibly failing to account for the interplay between multiple environmental amenities and housing preferences. A recent study by Gibbons, Mourato and Resende (2014) breaks the mould by simultaneously considering a large number of natural amenities (Box 3.1).
Gibbons, Mourato and Resende (2014) use the HPM to estimate the amenity value associated with proximity to several habitats, designated areas, domestic gardens and other natural amenities in England. Unlike previous studies, that mostly tended to focus on a single environmental good, this analysis measured the value associated with a large number of natural amenities in England, on a national scale. It is important to know if the link usually found between environmental characteristics and house prices remains discernible when conducting the analysis over a much wider geographical area with a greater environmental diversity. Moreover, an analysis at a wider geographical scale potentially permits the investigation of the value of larger scale environmental variables, such as different habitats or ecosystems and different types of protected areas.
The authors analyse a sample of 1 million housing transactions from 1996 to 2008, with information on location at full postcode level. The data set includes sales prices and several internal and local characteristics of the houses. Internal housing characteristics are property type, floor area, floor area-squared, central heating type (none or full, part, by type of fuel), garage (space, single, double, none), tenure, new build, age, age-squared, number of bathrooms (dummies), number of bedrooms (dummies), as well as year and month dummies. The authors use Travel to Work Area (TTWA) fixed effects to control for unobserved labour market variables (such as wages and unemployment rates) and other geographical factors. Including the TTWA dummies in the hedonic function regression, means the model utilises only the variation in environmental amenities and housing prices occurring within each TTWA (i.e. within each labour market) and so takes account of more general differences between TTWAs in their labour and housing market characteristics. Other control variables included: distances to various types of transport infrastructure, distance to schools, distance to the centre of the local labour market (TTWA), land area of the ward, population density and local school quality.
In terms of local environmental characteristics, Gibbons et al. (2014) use nine broad habitat categories, describing the physical land cover in terms of the share of the 1km x 1km square in which the property is located: (1) Marine and coastal margins; (2) Freshwater, wetlands and flood plains; (3) Mountains, moors and heathland; (4) Semi-natural grasslands; (5) Enclosed farmland; (6) Coniferous woodland; (7) Broad-leaved/mixed woodland; (8) Urban; and (9) Inland Bare Ground. An additional six land use share variables are also used, depicting the land use share, in the Census ward in which a house is located, of the following land types: (1) Domestic gardens; (2) Green space; (3) Water; (4) Domestic buildings; (5) Non-domestic buildings and (6) “Other” (incorporating transport infrastructure, paths and other land uses). Finally, five “distance to” variables describing distance to various natural and environmental amenities (in 100s of kilometres), were also included: (1) distance to coastline, (2) distance to rivers, (3) distance to National Parks (England and Wales), (4) distance to National Nature Reserves (England and Scotland), and (5) distance to land owned by the National Trust (the UK’s leading independent conservation organisation managing large areas of British countryside, coasts and properties). Additionally, the authors used two variables depicting designation status: the proportion of Green Belt land and of National Park land in the Census ward in which a house is located. The idea is to see whether knowledge that certain habitats are protected from development has a value to homebuyers.
Gibbons and colleagues use a semi-log hedonic price function specification and the estimates are fairly insensitive to changes in specification and sample. This provides some reassurance that the hedonic price results provide a useful representation of the values attached to proximity to environmental amenities in England.
A summary of key findings for England is presented in Table 3.2. Results reveal that the effects of many environmental characteristics on house prices are highly statistically significant, and are quite large in economic magnitude. Gardens, green space and areas of water within census wards all attract a considerable positive price premium. There is also a strong positive effect from freshwater locations, broadleaved woodland, coniferous woodland and enclosed farmland (with urban land cover as a base). Increasing distance to natural amenities such as rivers, National Parks and National Trust sites is unambiguously associated with a fall in house prices. Each 1km increase in distance to the nearest National Park lowers prices by 0.24% or GBP 465. This implies that being inside a National Park (i.e. at zero distance from it), combined with 100% of the ward as a National Park, implies a huge GBP 33 686 premium relative to the average house in England (which is 46.7 km from a National Park). In turn, Green Belt designation becomes important when looking at major metropolitan areas. The results indicate a WTP of around GBP 7 000 for houses in Green Belt locations, which offer access to cities, coupled with tight restrictions on housing supply.
Overall, the authors conclude that the house market in England reveals substantial amenity value attached to a number of habitats, designations, private gardens and local environmental amenities.
The effect of disamenities and “bads” has also been investigated via the HPM including road, railway and airport noise (Andersson et al., 2009; Day et al., 2006; Wilhelmsson 2000; Pope 2008a), wind turbines (Gibbons 2015; Hoen et al., 2011), electric power plants (Davis 2011), shale gas exploration (Muehlenbachs et al., 2015; Gibbons et al., 2016) and floods (Beltrán-Hernández, 2016). Finally, the method has also been used to evaluate the effects of environmental policies such as the Clean Air Act (Chay and Greenstone, 2005) and the Superfund programme for the clean-up of hazardous waste sites (McCluskey and Rausser, 2003).
The most common methodological approach in these studies has been to include distance from the property to the environmental amenity or disamenity as an explanatory variable in the model. More recently the use of GIS has improved the ability of hedonic regressions to explain variation in house prices by considering not just proximity but also amount and topography of the environmental amenities, for example by using as an explanatory variable the proportion of an amenity existing within a certain radius of a house.
For the most part, this large body of literature has consistently shown an observable effect of environmental factors on property prices, supporting the assumption that that the choice of a house reflects an implicit choice over the nearby environmental amenities so that the value of marginal changes in proximity to these amenities is reflected in house prices.
The HPM has also been used to estimate the value of avoiding risk of death or injury. It has done this by looking for price differentials between wages in jobs with different exposures to physical risk. That is, different occupations involve different risks (in that, for example, being a firefighter entails, on average, very much higher risks of injury or worse than does a desk-bound occupation). Employers must therefore pay a premium to induce workers to undertake jobs entailing higher risk. This premium provides an estimate of the market value of small changes in injury or mortality risks (Kolstad, 2010). Hedonic methods have thus been applied to labour markets in order to disentangle such risk premia from other determinants of wages (e.g. education etc.). An example of this approach is shown in Box 3.2.
Marin and Psacharopoulos (1982) undertook one of the first studies of the relationship between wages and occupational risk in the UK. The motivations for the work were twofold. First, the study aimed to test the theory that earnings should be higher in higher risk jobs, taking account of non-competitive factors such as unionisation. Second, the objective was to provide an estimate of the value of changes in mortality risk for use in project and policy appraisal.
The authors used data from the General Household Survey and data on occupational risk to estimate an earnings function – a type of hedonic price function – which included variables such as numbers of years of schooling and work experience (“human capital” variables), occupational risk, the extent of unionisation, and a ranking of occupational desirability. Two risk index series were constructed. The first considered the overall relative risk of dying in each occupational group. This measure would by implication include those risks for which compensation might not be required because they are willingly borne (e.g. publicans), the effect of risks borne in other occupations (e.g. higher mortality rates for above-ground mine workers, reflecting health problems contracted in previous (below-ground) employment), and chronic (e.g. cancer) risks which employees are likely to have difficulty in assessing. The measurement problems resulting from this first index led the authors to prefer a more specific risk variable, relating to the risk of death through an accident at work, as a more immediate, less “desirable”, and more easily perceived risk measure. Thus, this measure was seen as a more labour market-specific risk compared with the first measure which referred to deaths in general.
Marin and Psacharopoulos also included in their earnings function a variable to consider any interaction effect between occupational risk and the extent of unionisation, with no prior expectation of whether the estimated coefficient should be positive or negative. A positive effect could result if unions had better knowledge of risks than individual workers, and used this in collective wage bargaining. A negative effect could occur for a number of reasons. Collective bargaining could take place at a broader level than the occupations considered by the authors, reducing the sensitivity of that bargaining to measures of occupational risk. Second, unions might bargain directly for the implementation of measures to improve on-the-job safety, making risk less important as a bargaining tool in earnings determination.
The results of the analysis confirmed that higher risks of death were associated with higher earnings in the UK. The relationship with workplace accident mortality was stronger than that with overall occupational mortality, as expected. The union-risk interaction term was found to be negative, suggesting (although not strongly) that unionisation tended to weaken the compensating differential between more and less risky jobs.
The implicit value of mortality risk can be translated into a population-level measure, commonly called the “value of statistical life”, by calculating the differential of the earnings function with respect to risk. For all workers in the sample (n = 5 509), the value of statistical life computed to GBP 603 000 or GBP 681 000 in 1975 prices (GBP 3.14 m or GBP 3.54 m in 2001 prices), depending on whether union-risk interaction was included. Due to the nature of the hedonic price function, as a locus of equilibrium prices resulting from the interaction of buyers and sellers, these figures also provide estimates of the cost to firms of reducing workplace risk.
Marin and Psacharopoulos also performed estimations on subsamples of the total sample of workers, to consider whether compensating differentials varied across professional, non-manual and manual workers. Estimations for professionals were less successful, given the very low level of risk (and hence casualties) in the associated occupational groups. The authors suggested that this reflects high values of safety on the part of these workers, and low costs to firms of reducing risks for sedentary workers. Estimations for non-manual and manual workers were more satisfactory, and resulted in values of statistical life of around GBP 2.25m for the former group (reflecting non-manual workers higher average income and higher estimated risk coefficient), and figures for the latter group very close to those estimated for the whole sample (GBP 619 000-GBP 686 000). The non-manual value of statistical life translates into a figure of around GBP 11.7 m in 2001 prices.
3.2.1. Limitations
Not surprisingly there are a number of issues surrounding the practical application of the HPM. Firstly, HPM only measures use values, as reflected in property prices. And it is based on a number of assumptions, namely of property markets that are competitive and in equilibrium, requiring that individuals optimise their house choices based on the prices in various locations. It also assumes free mobility, i.e. that individuals are able to adjust the different levels of each characteristic of interest by moving property, with no transaction costs.
Moreover, the HPM assumes perfect information. In reality, individuals might not have perfect information. In the case of wage-risk premia, this means that workers may not be fully aware of the accident risks they face in the workplace, so that their wage-risk choices do not accurately reflect their true valuation of risk. Estimates of the value of risk obtained from observing these choices will then be biased. In the case of environmental variables, house buyers might not be aware of issues such as land contamination or probability of flooding in which case such elements will not be accurately reflected in house prices. Pope (2008b) investigates information asymmetries about flood risk, where sellers are typically better informed than buyers, making it attractive for sellers to wait for an uninformed buyer to make a bid on the house. After the introduction of a seller disclosure law in North Carolina, requiring sellers to disclose flood risks so that buyers are fully informed, the author estimates a 4% decline in housing prices in flood zones. Notably, before the disclosure law came into force, there appeared to be no impact of flood plain designation on housing prices. Pope’s results suggest that asymmetric information between buyers and sellers caused an underestimation of the estimated marginal values for flood plains prior to the disclosure law.
The HPM estimation procedure also faces some well-known econometric problems such as the arbitrary choice of a functional form for the hedonic price function, multicollinearity, heteroscedasticity, defining the spatial and temporal extent of property markets, and omitted variable bias. Moreover, the standard approach in the past literature has involved using cross-sectional data, which poses numerous identification problems, having to rely on controls for the large number of factors that affect house prices, many of which are unobservable.
In terms of multicollinearity, non-market characteristics tend to move in tandem: e.g. properties near to roads have greater noise pollution and higher concentrations of air pollutants. This means that it is frequently difficult to “tease out” the independent effect of these two forms of pollution on the price of the property. In many cases, researchers have tended to neglect the issue, omitting a potentially important characteristic from the analysis, and producing biased estimates as a result. See Box 3.3 for an example.
Leggett and Bockstael (2000) address the issue of multicollinearity directly in their study of the impact of varying water quality on the value of waterside residential property in Chesapeake Bay, USA. Water pollution in Chesapeake Bay can be produced by sewage treatment works and other installations which could also have a negative impact on visual amenity. The potential for bias thereby stems from the fact that properties closest to these installations could suffer both worse water quality and worse visual amenity, making it difficult to determine the price effect of each.
However, to overcome this potential problem, the authors were able to take advantage of a natural feature of Chesapeake Bay. The Bay has a varied coastline, with many localised inlets and a diverse pollution-flushing regime. As a result, it was possible to find a property located on an inlet which suffered from poor water quality but with no direct line of sight to the associated pollution source, and hence no visual disamenity. Similarly, a property located close to a sewage treatment works would not necessarily suffer from poor water quality if the flushing regime in that particular inlet was benign. The natural features of Chesapeake Bay thereby broke the potentially collinear relationship between visual amenity and water quality, allowing both characteristics to be included in the estimation equation without causing statistical problems.
In hedonic property studies, as with most studies of the value of environmental resources, some consideration needs to be given of the appropriate way to measure the environmental variable of interest. For instance, laypeople often respond most readily to the visual appearance of water, tending to attach higher values to water of greater clarity. However, biological water quality – which reflects the ecological potential of a water body – is not necessarily related to water clarity. Further, chemical water quality is more important for determining whether a water body is suitable for swimming or other sports where contact with the water is a possibility. Chemical water quality might not be well understood by members of the public, however.
Leggett and Bockstael used reported faecal coliform levels as their measure of water quality. This indicates that it was in general the recreational value of being located close to Chesapeake Bay which was being estimated in their study. These data were advertised in local newspapers and at information points, and the limit at which beaches would be closed for public health reasons was also clearly stated. The authors also obtained good evidence for believing that existing and prospective Chesapeake residents took an active interest in local water quality, providing further support for the possibility of a positive relationship between property values and water quality.
The authors found that standard locational variables had the expected signs in their estimated hedonic price equation. Increased acreage, reduced commuting distance, and proximity to water all had positive impacts on property prices, compared with the average estimated USD 350 000 per one acre plot. The closer a property was to a pollution source, the lower the price would tend to be. Local faecal coliform levels were also negatively related to property prices. For every unit increase in median annual concentration reported at the nearest measuring station, property value was observed to fall by USD 5 000 (average concentration in the sample was one count per ml, with a range of 0.4-23/ml). This could be used as an estimate of the marginal value of small changes in water quality in the Chesapeake area, and elsewhere.
Leggett and Bockstael emphasise that their results cannot be used to estimate the value of significant changes in water quality (as might occur through the introduction of new environmental standards, for instance). This is because a significant change would constitute a shift in the supply of environmental quality to the Chesapeake Bay housing market, and hence would induce a shift in the hedonic price function, as buyers and sellers renegotiated to obtain new optimal house purchase outcomes. This is an important qualification to the policy use of non-market value estimates obtained via the HPM.
The potential for omitted variables in hedonic modelling has long concerned researchers (Kuminoff et al., 2010). Omitted variable bias occurs as there may be unobservable housing characteristics that matter to households that are correlated with the environmental amenity of interest. This potential misspecification of the hedonic price function could result in biased value estimates. In an influential study, Cropper et al. (1988) showed that, in the presence of omitted variables, simpler functional forms such as linear, log-linear, log-log perform best. As a result, most studies published since then have used these simpler models in order to minimise the potential for omitted variable bias.
Care also needs to be taken to specify the extent of the property market accurately. The extent of the market is defined for any one individual house buyer by that individual’s search. If properties are included in the analysis which are outside of the individual’s market, hedonic price estimates will be biased. If properties are excluded which are in the market, the resulting estimates will be unbiased but inefficient. Unfortunately, with many different individuals searching for property in a given locality, the resulting house purchase data are likely to be drawn from a large number of overlapping markets. In this case, it has been argued that it is probably better to underestimate the extent of the market under study, rather than overestimate it (Palmquist, 1992).
Finally, the overwhelming majority of the HPM literature estimate only the marginal implicit prices associated with the characteristics of interest, as the typical policy question of interest is whether the current stock on a local non-market good is capitalised in the property market. But this is only the first stage of the HPM. Most studies do not go on to estimate demand functions for the characteristics of interest, i.e. the second stage of the HPM, which would allow the estimation of the value of non-marginal and non-localised changes. This is because estimating demand relationships from hedonic price data is theoretically and analytically challenging and requires extensive information. Day et al. (2006) provide a rare example.
3.2.2. Recent developments
In recent years, research into omitted variable bias and resulting endogeneity problems in hedonic price models has led to many econometric developments. In a review of the effects of omitted variable bias in HPM studies, Kuminoff et al. (2010) find that studies using large cross-sectional data sets have started to include spatial fixed effects in the hedonic price function (e.g. fixed effects for travel to work areas such as in Gibbons et al., 2014, or for school districts) in order to control for spatially clustered omitted variables. And as panel data sets and repeated cross-section data sets became increasingly available, researchers have been able to adopt quasi-experimental methods such as fixed effects, first differences and difference-in-differences to address the problem of omitted variables and accurately identify non-market values (e.g. Horsch and Lewis, 2009; Gibbons, 2015; Gibbons et al., 2016). Some authors have also used repeat sales data to address the issue (e.g. Beltrán-Hernández, 2016). Kuminoff et al. (2010) argue that, when spatial fixed effects are used to control for omitted variables, the seminal result by Cropper et al. (1988) regarding the superiority of simpler hedonic price functional forms no longer holds. Instead, they show that there are large gains in estimation accuracy by moving to more flexible specifications of the hedonic price function (such as the quadratic Box-Cox model) when using quasi-experimental identification, spatial fixed effects, and/or temporal controls for housing market adjustments.
Horsch and Lewis (2009) propose a quasi-random experiment to identify the effects of milfoil, an invasive aquatic species, on property values. Milfoil is spread by the movement of boats and since boaters are more likely to visit nice lakes with desirable (and often unobservable) amenities, the likelihood of a lake being invaded by milfoil is correlated with the error term in an hedonic price function (endogeneity). As a result, standard ordinary least squares (OLS) estimation of cross-sectional hedonic price data is likely to produce positively biased coefficient estimates on variables related to the presence of milfoil. Using time series data, that include data on lakes before and after milfoil invasions, the authors propose a difference-in-differences (DiD) analysis, with lake fixed effects. This estimation strategy allows the identification of the effect of milfoil on property values because the fixed effects control for all observable and unobservable lake amenities that affect property values, while the DiD specification exploits the natural experiment features of the dataset, that contains before-and-after data on milfoil invasions. Results indicate that milfoil invasions reduce average property values by around 8%.
In another recent example, Beltrán-Hernández (2016) uses the difference-in-differences approach to measure the ex-post economic benefits of structural flood defences in England, constructed between 1995 and 2004. The study is based on a large panel data set, with over 12 million property transactions, including sale prices of houses that have been sold multiple times. These data were then merged with GIS data containing the spatial location and main characteristics of a total of 1,666 flood defences constructed in England during the period of analysis. The author uses a repeat-sales model to look at the capitalisation of the flood defence infrastructure between two sales of the same property. The repeat-sales model is akin to a first-differences specification of the DiD model. This specification permits the evaluation of the price effect of flood defence construction, which is not uniform across properties, while controlling for time-invariant characteristics. The results suggest that flood defences result in increases in property prices ranging between 1% and 13%, depending on the level of risk and on the type of property (i.e. GBP 2 000 to GBP 30 000, for a median-priced house in 2014). However, in the case of flats (not affected by floods) and rural properties (where flood defences may result in loss of amenity value), the construction of flood protection infrastructure results in significant negative impacts that range from a price discount of -1% to -9% (-GBP 3 000 to -GBP 10 000).
In order to deal with endogeneity, recent studies have also used an instrumental variable approach. An example is Bayer et al.’s (2009) hedonic price study of air quality. Because air pollution is likely to be correlated with unobserved local characteristics, such as economic activity, that also affect property prices, standard estimates of willingness to pay are likely to be biased downwards. To tackle the issue the authors instrument for local air pollution, using the contribution of distant sources to local air pollution as an instrument. This strategy works because many air pollutants come from distant sources and those distant sources are unlikely to be correlated with local economic activity. Instrumenting for air pollution greatly increases the magnitude of the coefficient on air pollution (in this case particulate matter PM10) concentration in the hedonic price regression.
3.3. Travel cost method
The travel cost method (TCM) is a technique that has been developed to estimate recreational use values of non-market goods, typically outdoor natural areas but applicable to any location used for recreational purposes (Clawson and Knetsch 1969; Bockstael and McConnell 2007; Parsons 2017). For example, natural areas are frequently the focus of recreational trips (e.g. parks, woodland, beaches, rivers, lakes etc.). Such natural areas, for a number of reasons, typically do not command a price in the market and so we need to find an alternative means of appraising their value.
The basis of the TCM is the recognition that individuals produce recreational experiences through the input of a number of factors. Amongst these factors are the recreational area itself, travel to and from the recreational area and, in some cases, staying overnight at a location and so on. Typically, while the recreational area itself is an unpriced good, many of the other factors employed in the generation of the recreational experience do command prices in markets, such as travel costs. Travel costs could therefore be used as a proxy for the value of accessing the site.
Most of the early research using the TCM approach was indeed motivated by estimating the value of visits to recreational sites. In time, the method was adapted to be able to also value quality changes. Indeed, the last 50 years have witnessed a considerable evolution of travel cost method techniques, from simple aggregate demand models to very sophisticated analysis of individual level choices.
Parsons (2003) usefully differentiate between travel cost models that estimate demand for a single recreational site and models that estimate demand for multiple sites. We will now consider these two categories of models in turn.
3.3.1. Single site models
The single site TCM derives from the observation that travel and the recreational area are (weak) complements such that the value of the recreational area can be measured with reference to values expressed in the market for trips to the recreational area. To estimate the TCM, therefore, we need two pieces of information: a) the number of trips that an individual or household takes to a particular recreational area over a period of time (e.g. a year); and b) how much it costs that individual or household to travel to the recreational area, which acts as a proxy for the price of visiting the site.
The costs of travelling to a recreational area, in turn, include two elements: i) the monetary costs in return fares or petrol expenses, wear and tear and depreciation of the vehicle and so on; and ii) the cost of time spent travelling. Time is a scarce resource to the household. Time spent travelling could be spent in some other activity (e.g. working) that could confer well-being. In other words, the individual or household incurs an opportunity cost in allocating time to travel. Put more simply, demand for trips will be greater if it takes less time to travel to the recreational area, independent of the monetary cost of travel.
Of course to implement this procedure we require a value for the (shadow) price of time. One possible value for the price of time to an individual is their wage rate (Cesario, 1976). If individuals can choose the number of hours they spend working then they will choose to work up to the point at which an extra hour spent at work is worth the same to them as an hour spent at leisure. At the margin, therefore, leisure time will be valued at the wage rate. In the real world, individuals can only imperfectly choose the number of hours they work and the equality between the value of time in leisure and the wage rate is unlikely to hold. Empirical work has been undertaken that has revealed that time spent travelling is valued at somewhere between a third and a half of the wage rate and travel cost researchers frequently use one or other of these values as an estimate of the price of time (Czajkowski et al., 2015).
The information used in the TCM is usually collected through surveys carried out at the recreational site. With these data, a demand curve for access to the recreational site can be estimated, which explains the number of visits (i.e. the quantity) as a function of travel costs (i.e. the price) and other relevant explanatory variables. This demand curve is typically downward sloping as the number of trips normally declines the higher the costs of the trip. Higher costs are normally associated with people living further away from the site. The points along the demand curve indicate consumer willingness to pay to visit the site. The non-market value associated with the recreation benefits at the site is estimated as the consumer surplus, i.e. the area under the demand curve between an individual’s WTP and their travel cost expenditure.
Initial applications of the TCM used what is known as the zonal TCM (Parsons, 2003). Zonal TCM calculated aggregate visit rates (i.e. number of visits from an area divided by the population of that area) and average cost trips from different pre-defined geographical ̳zones surrounding the recreational site of interest. This permitted the estimation of number of visits per capita for each of the zones considered. The approach therefore looked at the average behavior of groups of visitors rather than at individual choices. Because of its lack of consistency with economic theory the use of the zonal model has declined over time.
Today, the most commonly applied variant of the single-site TCM is the individual TCM. This approach makes use of individual-level rather than aggregate data, namely, the number of individual visits to a recreational site over a period of time (e.g. a year) and their respective costs. The method has been applied to value a wide range of outdoor recreation pursuits such as forest recreation (Christie et al., 2006), lake visits (Corrigan et al., 2007), recreational fishing (Shrestha et al., 2002), ski centre visits (Steriani and Soutsas, 2005), mountain biking (Chakraborty and Keith, 2000), National Parks (Heberling and Templeton, 2009), deer hunting (Creel and Loomis, 1990) and many more.
In early individual TCM models the number of visits was treated as a continuous variable and OLS regression methods were typically used, leading to biased estimates. In the late 80’s researchers started to use instead more appropriate count data models such as Poisson and negative binomial regression models which take into account the nature of the visitation data: i.e. visits are non-negative integers; data is often truncated at zero due to on-site sampling meaning that respondents will have at least one visit; and the visit distribution tends to be typically skewed towards small numbers of trips (Parsons, 2017).
A limitation of the individual TCM is that it does not easily accommodate the presence of substitute recreational sites. In many real-world situations individuals are faced with a wide range of substitute recreational sites: e.g. choice of which beach to go to, which river to go fishing in, which ski resort to visit, or even, choices between different types of sites, say whether to go to a woodland or a national park. In such cases, we require an approach capable of adequately modelling the discrete choice that consumers make between sites rather than an approach that focus on the “continuous” choice of how many trips to make to single site. The next section presents the model typically used in such cases, the Random Utility Model.
3.3.2. Multiple sites models
The standard method applied in the case of multiple sites is the Random Utility Model (RUM) (Bockstael et al., 1987). The RUM is a discrete choice modelling technique where, in the presence of multiple recreational sites, individuals are assumed to choose which site to visit based on the site characteristics as well as the costs of travelling to the different substitute sites. Although the RUM is often described as an extension of the TCM it is in fact more akin to a theory of choice rather than a valuation technique and can be applied in any situation in which households’ make discrete choices that involve combinations of market goods and environmental goods and services (Maddison and Day, 2015).
In recent years, the popularity of random utility modelling for recreational choice has boomed, in parallel with a decrease in application of more traditional travel cost models. It is now the dominant revealed preference method for recreation demand estimation (Phaneuf and Smith, 2005) and has been applied to a very extensive range of recreational experiences including fishing, swimming, climbing, boating/cannoing/kayaking, hunting, hiking, skiing, and park/forest/river visits, amongst others. For policy and management purposes, the RUM approach is very useful as it allows the estimation of the value of changes in site quality as well as site closures, in multiple sites. Phaneuf and Smith (2005) and Parsons (2017) offer detailed overviews of the evolution of RUM and its applications to recreational demand. Box 3.4 contains an application to the choice of game parks in South Africa (Day, 2002).
Day (2002) provides a relatively sophisticated application of the multiple site travel cost method to four of South Africa’s game parks. These internationally renowned games reserves – Hluhluwe, Umfolozi, Mkuzi and Itala – each cover vast land areas of roughly several hundred square kilometres and are managed by the KwaZulu-Natal Parks Board (KNPB).
The premise for Day’s approach is that a visit to any one of these game reserves reflects a choice between four key cost determinants: i) the economic cost of travel to the site; ii) the cost of time while travelling; iii) the cost of accommodation at the site; and, iv) the cost of time whilst on-site. Most travel cost approaches have focused only on costs i) and ii). For many recreational sites this is sufficient. However, Day argues that overnight trips are an important feature of visits to the reserves that he examines in this study. In order to take account of this trip characteristic, Day extends the RUM framework often used in recreational contexts to predict that an individual will choose to make a given visit to a particular site rather its alternatives because the chosen site provides that individual with the most utility (or well-being) from the options available. Such a model is thus ideally suited to explaining a visitor’s decision with reference to the qualities of alternative sites (e.g. number and variety of fauna and flora) as well as the different costs of travelling to these sites. Day further extends this framework in order to take account of visitor choice of accommodation and length of stay at the site.
The data used in this study are based upon a (random) sample of 1 000 visitors to the four different reserves. For each of these visitors, this included information on, for example, length of stay, size of party and how much. In total, that the visit cost each household. It is worth noting that this study did not need to use on-site surveys say of visitor total travel costs and demographic/ socioeconomic characteristics. For example, with respect to physical distance travelled, this was calculated by the author with reference to data on visitor addresses combined with a Geographical Information Systems (GIS) model in order to calculate the distance that each visitor travelled “door-to-door”. Only visitors living in South Africa were sampled to minimise the problem of multipurpose trips.
An interesting feature of Day’s study is the determination of the money value to be assigned to an hour spent travelling relative to an hour spent on-site. Day demonstrates quite reasonably that an hour spent travelling is likely to be valued less highly than an hour spent on-site at the reserve. Furthermore, he argues that the former is likely to be valued more than time in general because there could be a significant disutility associated with time spent travelling. In other words, people enjoy time travelling a lot less than most other uses of time and so this activity has a high opportunity cost. By contrast, the latter is likely to be valued less than time in general because there could be a significant utility associated with time spent on-site. In terms of proportions of the wage rate, Day concludes that his analysis justifies valuing travel time at 150% of the household wage rate while on-site time is valued at 34% of the wage rate. Whereas the latter seems consistent with previous findings in the literature (see discussion above) the former is somewhat higher than conventionally assumed by travel cost practitioners.
Day uses assembled data on cost, trip duration and accommodation decision variables as well as other trip characteristics as inputs to a sophisticated statistical analysis of the determinants of the choice to take a given trip to a particular reserve (using a nested logit model). Ultimately, the findings of this detailed analysis can be used to derive policy relevant information on the benefits provided by the reserves. For example, Day calculates the amount of money that would have to be given to affected households in South Africa following the (hypothetical) closure of one of the reserves in order to fully compensate them for the loss of this recreational amenity. Since only South African visitors were considered in the analysis, the estimates, these estimates do not include the welfare costs that would be associated with the loss of visitors from abroad. A summary of these findings is presented in Table 3.3.
Why are these data important? Day argues that one response to this question is that the KNPB is finding itself under increasing pressure to justify the substantial public funding that it receives. Demonstrating the monetary value of the recreational benefits provided by the KNPB might be one crucial way in which this body can make its case for public funds. Thus the values in Table 3.3 (column 2) can be thought of as the per trip benefits attributable to the current management regime at each reserve. Alternatively, this is the (yearly) per trip loss of welfare or well-being. In money terms, that occurs if the reserve were to be closed “tomorrow”.
Column 3 in Table 3.3 illustrates the total annual welfare losses for each reserve: i.e. the per trip value multiplied by the number of trips which would no longer be taken over a year if the reserve was closed. In effect, this column provides policy-makers with one basis for assessing the dollar magnitude of the (non-market) recreational benefits generated by public expenditure on each reserve. Finally, it is interesting to note that the final row in Table 3.3 indicates that if both Hluhluwe and Umfolozi (i.e. the most highly valued) reserves were to close then the combined welfare loss is greater (than the sum of individual values in column 3, rows 2 and 3). The intuitive explanation for this is that these two parks are in close proximity to each other. Removing one or other would mean that many households would most likely just switch their visits to the remaining reserve. However, if both of these sites were to be no longer available for visits then the loss for households would be disproportionately greater reflecting the absence of remaining substitutes.
Application of the RUM is data intensive and requires data on individuals’ choice of site, place of residence, socio-economic and demographic characteristics, frequency of visits to the site of interest and other similar sites, as well as trip cost information. These data can be collected from either an on-site or off-site survey. Data are also required on the characteristics of the different recreation sites under consideration, and their quality. These can either be collected from objective datasets (e.g. water quality measurements) or be based on subjective perceptions of quality by visitors.
The RUM models the probability of visiting a particular site as a function of the characteristics of the sites in the choice set of possible sites to visit. The estimated model controls for visitors’ socio-economic characteristics, travel costs and travel time, and site quality characteristics to estimate the benefit derived from a recreation visit. The value of a change in environmental quality is then estimated by relating the estimated model coefficient for environmental quality to the costs of a visit, as inferred from the travel costs.
3.3.3. Limitations
The TCM has narrow applicability to the estimation of recreational use values and requires the availability of large data sets on recreational activities, including extensive GIS analysis of travel cost data and site characteristics (for RUM studies).
Some of the limitations associated with the single-site TCM model, such as the lack of consideration for substitute sites can be resolved by the use of the RUM variant. But one issue remaining is the problem of multiple purpose trips (Parsons, 2017). Many recreational trips are undertaken for more than one purpose. For example, standard travel cost methods cannot easily be applied to trips undertaken by international tourists since such tourists will usually visit more than one destination. One solution to this problem has been to ask visitors (as part of the on-site survey) to estimate the proportion of the enjoyment they derived from their entire trip that they would assign to visiting the specific recreational area of interest. Total travel costs for the entire trip are multiplied by this amount and this can be used as the basis for assessing travel costs at the recreational site.
Other challenges include the valuation of travel time as often results are very sensitive to the assumptions made. As noted above, TCM researchers need to make assumptions about how visitors would have used their time in other welfare raising activities, if they had not been travelling for recreation. Such assumptions are mostly ad-hoc, typically based on using a fraction of the wage rate, and difficult to validate in empirical studies. Critics of the wage-based value of time approach also note that it makes little sense for people without wages (e.g. students or homemakers), to assume that their marginal utility of time is zero (Czajkowski et al., 2015).
3.3.4. Recent developments
As with the hedonic price method discussed above, many of the innovations in recreational demand modelling have been in the econometric analysis methods used. Discrete choice models in particular have witnessed a literal revolution in recent years, with ever increasing sophistication in estimation. Examples of such developments include new approaches to deal with the incorporation of unobserved heterogeneity (e.g. via mixed logit models and latent class models), instrumental variables, models for handling on-site sampling, and dealing with corner solutions (Phaneuf and Smith, 2005; Parsons, 2017). Moreover, recreation models could also benefit from modern quasi-random experimental designs when evaluating changes in policies and management practices in recreational sites (Phaneuf and Smith, 2017).
In parallel, there has been a move to integrate TCM models with stated preference data (Adamowicz et al., 1994; Englin and Cameron, 1996; Whitehead et al., 2000; Landry and Liu, 2011). The advantage of the combined approach is the ability to measure changes in the quality of recreation sites that have not yet happened. Most of the efforts have concentrated on the single site TCM model, using it in combination with contingent valuation or contingent behavior questions. For example, Corrigan et al. (2007) combined an individual TCM with a contingent behaviour question to estimate the value of improved water quality at Clear Lake in Iowa (USA). In addition to reporting how many visits they took in the past year, households surveyed were also asked how many trips they would have taken if water quality at the lake been improved, as per a contingent scenario described in the survey. The inclusion of a contingent behaviour question allowed the estimation of willingness to pay values for improvements in water quality at the lake. The average value of water quality improvements at Clear Lake was estimated to be around $140 per household per year for a small improvement and $350 per household per year for a large improvement. Analysis of the combined dataset typically involves stacking data from the two different sources and estimating a single model using the two types of observations.
Finally, the treatment of the opportunity cost of travel time in TCM models has become an area of active research in an attempt to overcome the limitations posed by the commonly used wage-based value of time assumptions (Czajkowski et al., 2015). Several authors have used stated preference methods to elicit stated values of time (Álvarez-Farizo, Hanley, and Barberán, 2001; Ovaskainen, Neuvonen, and Pouta, 2012; Czajkowski et al., 2015). Álvarez-Farizo et al. (2001) found a significant variation in leisure time values. Other authors have focused on revealed valuations of travel time. For example, Fezzi, Bateman, and Ferrini (2014) used a natural experiment to identify the value of time, where individuals had a choice of travelling via a toll road, which is faster, or not paying a toll and taking more time to reach the recreation site. Finally, Larson and Lew (2014) proposed a system of joint labour-recreation equations to capture the fact that the demand for time depends on whether individuals can freely substitute recreation for work or whether they have instead fixed work hours.
3.4. Averting behaviour and defensive expenditures method
Methods based on averting behaviour take as their main premise the notion that individuals and households can insulate themselves from a non-market bad by selecting more costly types of behavior (Dickie, 2017). These behaviours might be more costly in terms of the time requirements they imply, or of the restrictions they impose on what the individual would otherwise wish to do. Alternatively, individuals might be able to avoid exposure to non-market bads via the purchase of a market good. These financial outlays are known as defensive expenditures. The value of each of these purchases represents an implicit price for the non-market good or bad in question.
There are numerous instances which provide an illustration of these methods to value non-market goods and bads. Garrod and Willis (1999) offer the example of households installing double-glazed windows to decrease exposure to road traffic noise. Essentially, double-glazing is a market good which, in this example, acts as a substitute for a non-market good (peace and quiet, in the sense of the absence of road traffic noise). If noise levels decrease for other reasons – perhaps as a result of a local authority’s implementation of traffic calming measures – then households will spend less on these defensive outlays. Changes in expenditures on this substitute good provide a good measure of households’ valuations of traffic calming policies that decrease noise pollution (a bad) and, correspondingly, increase the supply of peace of quiet (a good). Many other examples exist as reviewed in Dickie (2017), the majority of which are applications to the valuation of reduced mortality and morbidity. Provins (2011) reviews recent empirical applications of the defensive expenditures method to value health impacts from water services, particularly focusing on drinking water quality. Box 3.5 summarises a well-known application to bicycle helmets and children safety (Jenkins, Owens and Wiggins, 2001).
There is growing policy interest in actions which reduce health risks to children and addressing how these benefits should be handled within a cost-benefit framework (see Chapter 15). A study by Jenkins, Owens and Wiggins (2001) provides a simple but interesting example of the application of a revealed preference approach – specifically defensive expenditure – to this question. The authors argue that there is no reason why it should simply be assumed that the value of reductions in, say, children’s mortality risks can be approximated with reference to values derived in the context of mortality risks faced by adults. On the one hand, children have (on average) greater life years remaining than the typical adult in such studies. Furthermore, it is plausible that society places a premium on the safety of children, especially very young children. On the other hand, children are not currently economically productive nor will they be in the near future. In other words, Jenkins and colleagues argue that there are a number of reasons to believe that the value that society would choose to assign to a given mortality risk faced by a child will diverge (in possibly offsetting ways) from how an adult would value their own risk of death.
The case examined by Jenkins, Owens and Wiggins (2001) is the purchase of safety products that target children. Specifically, the authors look at the US market for bicycle helmets which significantly reduce the wearer’s risk of death as a result of head injury. This product, they argue, has a number of desirable properties for indicating the implicit price of a person’s safety. For example, the good provides a benefit to the wearer only (unlike other defensive purchases such as smoke alarms which protect all those living within a home). This is useful if what are wanted are values of reducing individual (as opposed to household) risks. In addition, the authors claim that bicycle helmets do not generate diverse joint products to the same extent as other defensive goods (such as air conditioning or double-glazed windows). This is not to say that complications do not exist. For example, a bicycle helmet not only protects its wearer from risk of fatal head injuries: clearly, it reduces the risk of non-fatal injury as well.
The basis of this study’s use of the cost of bicycle helmets as a proxy for the value of fatal injury risk reduction is the assumption that a consumer purchases a helmet when his or her value for reducing risk is greater than the (net) cost of the product. Of course, in the case of the purchase of a child’s helmet it is typically the parent that is the buyer and hence the decision-maker. In other words, Jenkins and colleagues are concerned with evaluating the revealed preferences of parents for their children’s safety. To restate the logic of this approach in this context: a parent purchases a helmet when he or she perceives that the value of reducing risks to his or her child is greater than the (net) cost of the product. The authors use this insight as the basis for estimating the (implied) value of a statistical life for the typical helmet wearing bicycling child. This is defined as the (annualised) cost of the helmet divided by the change in the probability of death due to the purchase of the helmet.
Jenkins et al.’s study estimated that the value of a statistical life for US children aged 5 to 9 years old was roughly USD 2.9 million in 1997. The calculation underpinning this finding is the following. Firstly, it is reckoned that the annualised cost of a helmet is about USD 6.50. Secondly, the authors calculate that wearing helmets when cycling most (but not all) of the time amongst this age group results (nationally) in about 32 fewer deaths. Given that the 5 to 9 year old bicycle riding population was about 14.3 million in 1997, this gives an annual fatal risk reduction of 0.0000024: i.e. 32/14.3 million (or 1 in 446 875). The value of a statistical life for the typical 5 to 9 year old child of bicycle helmet purchasing parents is calculated as USD 6.50/0.0000024 or USD 2.9 million. Note that this assumes that the only benefit that the good provides is the reduction of fatal head injuries. In practice, wearing a helmet will reduce non-fatal head injury risks as well. The authors deal with this issue by arbitrarily assuming that the desire for reducing a child’s risk of a fatal injury accounts for one half of the decision to purchase a helmet: i.e. multiply USD 2.9 million by 0.5 to obtain a more conservative figure for the value of statistical life of USD 1.5 million (GBP 933 000 in 2001 prices).
While the broad logic of this approach is sound, it assumes that helmet-buying parents are extremely well informed about the risks to children of cycling, and are motivated to purchase market goods which provide only apparently minor reductions in risk. It might be argued that, in reality, parents see an annual USD 6.50 as a small price to pay for any reduction in risk for their children. On the other hand, GBP 933 000 is not a high value of statistical life in comparison with estimates obtained via other market and non-market methods. This might then serve as a reminder that, through their purchases, parents are revealing a value of a statistical life (of their child) of at least this amount, when in practice, their maximum valuation might be a lot higher than this. In other words, the defensive expenditure approach reveals lower-bound estimates of the value of the non-market good in question.
Examples of defensive expenditure focus on the purchase of market goods which act as a substitute for a non-market good. However, individuals might change their behaviour in costly but perhaps less obvious ways in order to avoid an adverse impact on their well-being. Freeman et al. (2014) use the example of an individual who spends additional time indoors to avoid exposure to outdoor air pollution. In this case, the allocation of time to avoiding a non-market bad (i.e. the risk of adverse health impacts like asthma attacks, or coughing and sneezing episodes) is typically not observable and the substitute item is itself a non-market good (i.e. time that could have been used more productively). Nevertheless, the avoidance costs of spending time indoors could be evaluated by asking people directly about their time-use. Moreover, time use has a market analogue in the form of wages that would be paid to an individual if the time spent indoors could otherwise be spent working (see discussion of value of time in the travel cost method section above). Box 3.6 presents an example of the application of the method in the face of averting behaviour to reduce the health risks associated with air pollution (Bresnahan, Dickie and Gerking, 1997).
Bresnahan, Dickie and Gerking (1997) examine behaviour and changes in health risks. Specifically, these health risks arise from exposure to concentrations of ground-level ozone with sunlight (ground-level ozone in cities can arise from a combination of certain pollutants, emitted as a result of energy generation and use of motor vehicles). Acute health impairment particularly in response to peak concentrations of ozone has been documented in a number of epidemiological and medical studies. Moreover, the authors note that spending less time outdoors on bad air quality days – e.g. days when ozone concentrations exceed recommended standards – can effectively decrease exposure to pollution for certain at-risk groups. Their study seeks to evaluate the extent of actual defensive expenditure and averting behaviour amongst members of these groups living in the Los Angeles area.
Data were drawn from repeated survey responses of a sample of (non-smoking) Los Angeles residents living in areas with relatively high concentrations of local air pollutants. In addition, the sample contained a high proportion of individuals with compromised respiratory functions. Respondents were asked a range of questions about, for example, their health status, purchase of durable goods that might mitigate indoor exposure to ground-level ozone, their outdoor behaviour in general and on bad air quality days in particular.
The findings of the Bresnahan, Dickie and Gerking (1997) study were that two-thirds of their sample reported changing their behaviour in some meaningful way on days when air quality was poor. For example, 40% of respondents claimed either to re-arrange leisure activities or stay indoors during such days, and 20% of respondents increased their use of home air conditioning units. Furthermore, those respondents who experienced (acute) air pollution-related symptoms tended to spend less time outside on bad air quality days. Finally, The authors found tentative evidence that averting behaviour increases with medical costs that would otherwise be incurred if a respondent became ill.
In summary, bad air quality days appeared in this study to lead to significant changes in behavior (although these findings do not capture permanent decisions to take recreation indoors regardless of air quality on particular days). It is reasonable to speculate that these behavioural changes impose non-trivial economic costs on respondents. For example, these burdens might take the form of the purchase and running of air conditioning with an air purifying unit or the inconvenience imposed by spending time indoors. However, Bresnahan and colleagues do not attempt to put a monetary value on these actions. As Dickie and Gerking (2002) point out, this would not necessarily be a straightforward exercise. For example, and as we have already argued, time spent indoors avoiding exposure to air pollution is not necessarily time wasted. In other words, there is no simple way of valuing a person’s time when time which an individual would have spent enjoying outdoor leisure activities is substituted for time spent enjoying indoor leisure activities.
In terms of policy, the fact that individuals can take significant action to minimise their exposure to environmental risks and/or incur in defensive expenditures will have an impact on the measurement accuracy of the physical effect of the environmental risks. Accounting for these behavioural responses is therefore essential to accurately measure the effect of changes in environmental risks and provide information on the economic benefits of pollution control. Ignoring these actions can lead to severely biased estimates, namely an underestimation of the physical damage that would result from increases in environmental risk factors (Neidell, 2009; Dickie, 2017).
3.4.1. Limitations
A number of complications arise in the practical application of averting behaviour and defensive expenditure approaches to valuing non-market goods. Dickie (2017) argues that the challenges confronting the method are responsible for its more limited impact on practical policy analysis when compared with that of other revealed preference methods discussed in this chapter.
Four problems in particular, are worth noting here. First, defensive expenditures typically represent a partial or lower bound estimate of the value of the impact of the non-market bad on well-being. For example, in the double-glazing case, greater indoor tranquillity may be achieved, but gardens will still be exposed to road traffic noise at the same levels, so double-glazing will not help homeowners to avoid the costs of road traffic noise completely. Moreover, households’ willingness-to-pay for tranquillity might exceed what they paid for double glazing.
Second, many averting behaviours or defensive expenditures create joint products. For instance, time spent indoors avoiding air pollution is not otherwise wasted. This time can also be put to other productive uses that have value, such as undertaking household chores, indoor leisure activities or working from home. The double-glazing case also creates joint products – e.g. energy conservation. It is the net cost of the expenditure or change in behaviour – that is, the cost after taking account of the value of alternative uses of time, for instance, or energy savings – which is the correct measure of the value of the associated reduction in the non-market bad. However, distinguishing the determinant of behaviour that is of interest, and the costs of the various components, might not be an easy matter in practice.
Third, it is not easy to assign a monetary value to behavioural changes associated with defensive actions. Dickie (2017) cites the example of keeping a child indoors to avoid exposure to outdoor air pollution. The monetary cost of keeping a child indoors rather than outdoors is not easily estimated, particularly as the wage-based value of time approach would not be applicable for children.
Finally, it can be difficult to causally identify the effects of the disamenity and of averting behaviour on the outcome of interest, in the presence of unobserved factors related to both the behaviour and the outcome (Dickie, 2017). Consider again the case of health and the defensive behaviour where children are kept indoors to avoid exposure to outdoor air pollution. Some children will have poorer health and be more susceptible to air pollution, and might therefore be kept indoors more often. When there are unobserved effects (i.e. an omitted variable), such as children’s heterogeneous natural resistance to illness, that are related to both the health outcome and the averting behaviour, then the impact of pollution on health and the impact of averting behaviour on health are both badly measured due to endogeneity. The problem of causal identification of the effect of the disamenity, and of the behaviour on the outcome of interest, is one of the key challenges of the averting behaviour approach.
3.4.2. Recent developments
As was the case with other revealed preference methods discussed in this chapter, there have been some significant econometric developments of relevance for the averting behaviour approach. In particular, recent years have witnessed improvements in causal identification strategies. Together with the increasing availability of detailed and comprehensive data sets, namely on health and pollution (e.g. Deschênes and Greenstone, 2011; Ziving et al., 2011), these developments are expected to lead to more accurate estimations and consequently, growing influence of averting behaviour methods in applied policy analysis.
There are many examples of econometric approaches used to tackle the identification challenge. For example, Neidell (2009) investigates the behavioural response to the provision of information about asthma risks associated with exposure to ozone in Southern California. To identify the effect of information provided via smog alerts, a regression discontinuity design is employed, by exploiting the deterministic selection rule used for issuing the alerts. The author finds that smog alerts significantly reduce daily attendance at two large outdoor facilities: the Los Angeles Zoo and the Griffith Park Observatory. Then, using daily time-series regression models, that include time and area fixed effects, Neidell examines the impact of ozone on asthma hospitalisations. He finds that the estimates of the effect of ozone on hospital admissions of children and the elderly, accounting for the information effects, are significantly larger than estimates where such effects are not considered (by about 160 per cent for children and 40 per cent for the elderly). The author concludes that failure to account for the substantial actions that individuals may take to reduce their exposure to air pollution in the face of information, such as decreasing the amount of time spent outside, will lead to biased estimates of air pollution damages.
Deschênes and Greenstone (2011) estimate the impact of climate change on mortality, and the impact of defensive expenditures against the impacts of climate change. Energy consumption (via air conditioning, used to protect against high temperatures) is utilised as the measure of self-protection. The analysis benefits from a large comprehensive dataset on mortality, energy consumption, weather, and climate change predictions for the whole continental United States. The identification strategy to deal with possible omitted variable bias relies on random yearly local variation in temperature, and the statistical models used include county and state-by-year fixed effects to adjust for differences in unobserved health across the country, due to sorting. The authors find a statistically significant relationship between mortality and daily temperatures, with extremely cold and hot days being associated with elevated mortality rates. But the effect is smaller than what would be predicted based on previous heat waves. Deschênes and Greenstone also find substantial heterogeneity in the behavioural responses to extreme temperatures across the country. They conclude that the weaker than expected mortality-temperature relationship is at least partially due to the self-protection provided by individuals’ avertive (cooling) behavior, as reflected in increased energy consumption.
Finally, a number of authors have started to combine averting expenditures/defensive behavior methods with stated preference methods (e.g. Rosado et al., 2006), and/or attitudinal/perception data from survey questions. As an example of the latter, Lanz (2015) investigates averting expenditures to deal with tap water hardness and aesthetic quality in terms of taste and odour. The averting expenditures include water softener devices, bottled water, water filter devices, or adding squash or cordial before drinking water. Via a survey, he finds that 39% of respondents report at least one such behaviour, with mean yearly expenditure around GBP 92 (substantial vis-à-vis a yearly average household bill of GBP 186 for water services). Lanz argues that it is the perceived (rather than actual) failure to reach the desired water quality that will determine these averting expenditures: failure to control for perceptions may therefore generate biased estimates. To fix this, he includes information on both objective and perceived water quality (elicited through the survey). Unobserved factors might affect both the averting behaviour and the water quality perception, leading to biased identification of marginal WTP. To control for this possible endogeneity, Lanz models the relationship between objective and subjective water quality in a first stage regression, and then includes instrumented subjective quality as part of the valuation function. Results confirm that perceived water quality is endogenous, and the associated marginal WTP estimates are biased downwards; instrumenting perceived quality with objective quality yields marginal WTP estimates that are approximately two times higher for water hardness and three times higher for aesthetic quality.
3.5. Conclusions
Economists have developed a range of approaches to estimate the economic value of non-market or intangible impacts. Those which we have considered in this chapter share the common feature of using market information and behaviour to infer the economic value of an associated non-market impact.
These approaches have different conceptual bases. Methods based on hedonic pricing utilise the fact that some market goods are in fact bundles of characteristics, some of which are intangible goods (or bads). By trading these market goods, consumers are thereby able to express their values for the intangible goods, and these values can be uncovered through the use of statistical techniques. This process can be hindered, however, by the fact that a market good can have several intangible characteristics, and that these can be collinear. It can also be difficult to measure the intangible characteristics in a meaningful way. Moreover, the potential for omitted variables and consequent misspecification of the hedonic price function is an on-going concern.
Travel cost and random utility methods utilise the fact that market and intangible goods can be complements, to the extent that purchase of market goods and services is required to access an intangible good. Specifically, people have to spend time and money travelling to recreational sites, and these costs reveal something of the value of the recreational experience to those people incurring them. The situation is complicated, however, by the presence of substitute sites, the fact that travel itself can have value, that some of the costs are themselves intangible (e.g. the opportunity costs of time), and that many trips are multipurpose.
Averting behaviour and defensive expenditure approaches are similar to the previous two, but differ to the extent that they refer to individual behaviour to avoid negative intangible impacts. Therefore, people might buy goods such as safety helmets to reduce accident risk, and double-glazing to reduce traffic noise, thereby revealing their valuation of these bads. However, again the situation is complicated by the fact that these market goods might have more benefits than simply that of reducing an intangible bad. Averting behaviour occurs when individuals take costly actions to avoid exposure to a non-market bad (which might, for instance, include additional travel costs to avoid a risky way of getting from A to B). Again, we need to take account of the fact that valuing these alternative actions might not be a straightforward task, for instance, if time which would have been spent doing one thing is instead used to do something else, not only avoiding exposure to the non-market impact in question, but also producing valuable economic outputs. Moreover, it is often difficult to causally identify the effects of the disamenity and of the averting behaviour on the outcome of interest.
This chapter has shown that revealed preference methods are widely used, in a range of environmental policy applications. Recent decades have witnessed substantial developments particularly in the sophistication of the econometric methods used to elicit causal relationships, in the detail, accuracy and comprehensive nature of the available data sets, and in methods being used in combination. Overall, we find that methods where the value of the environmental good or service is inferred through observations from real world market purchases have the potential to play a central role in policy analysis.
References
Adamowicz, W.L., J. Louviere and M. Williams (1994), “Combining Revealed and Stated Preference Methods for Valuing Environmental Amenities”, Journal of Environmental Economics and Management, Vol. 26, pp. 271-292, https://doi.org/10.1006/jeem.1994.1017.
Álvarez-Farizo, B., N. Hanley and R. Barberán (2001), “The Value of Leisure Time: A Contingent Rating Approach”, Journal of Environmental Planning & Management, Vol. 44/5, pp. 681-699, https://doi.org/10.1080/09640560120079975.
Andersson, H., L. Jonsson and M. Ogren (2009), “Property Prices and Exposure to Multiple Noise Sources: Hedonic Regression with Road and Railway Noise”, Environmental and Resource Economics, Vol. 45/1, pp. 73-89, https://doi.org/10.1007/s10640-009-9306-4.
Anderson, L.M. and H.K. Cordell (1985), “Residential property values improve by landscaping with trees”, Southern Journal of Applied Forestry, Vol. 9, pp. 162-166.
Bayer, P., N. Keohane and C. Timmins (2009), “Migration and hedonic valuation: The case of air quality”, Journal of Environmental Economics & Management, Vol. 58, pp. 1-14, https://doi.org/10.1016/j.jeem.2008.08.004.
Beltrán-Hernández, A. (2016), “Essays on the economic valuation of flood risks”, PhD Thesis, University of Birmingham.
Benson, E.D. et al. (1998), “Pricing residential amenities: The value of a view”, Journal of Real Estate Economics Finance, Vol. 16, pp. 55-73, https://doi.org/10.1023/A:1007785315925.
Bockstael, N.E. and K. McConnell (2007), Environmental and Resource Valuation with Revealed Preferences, Springer, Netherlands, https://doi.org/10.1007/978-1-4020-5318-4.
Bockstael, N.E., W.M. Hanemann and C.L. Kling (1987), “Estimating the value of water quality improvements in a recreational demand framework”, Water Resources Research, Vol. 23, pp. 951-960.
Bolitzer, B. and N.R. Netusil (2000), “The impact of open space on property values in Portland, Oregon”, Journal of Environmental Management, Vol. 59/3, pp. 185-193, https://doi.org/10.1006/jema.2000.0351.
Boyle, K.J. (2003), “Introduction to revealed preference methods”, in Champ, P.A., K.J. Boyle and T.C. Brown (eds.) (2003), A Primer on Nonmarket Valuation, Kluwer, Dordrecht.
Boyle, K.J., P.J. Poor and L.O. Taylor (1999), “Estimating the demand for protecting freshwater lakes from eutrophication”, American Journal of Agricultural Economics, Vol. 81/5, pp. 1118-1122, https://doi.org/10.2307/1244094.
Bresnahan, B.W., M. Dickie and S. Gerking (1997), “Averting behaviour and urban air pollution”, Land Economics, Vol. 73/3, pp. 340-357.
Cesario, F.J. (1976), “Value of Time in Recreation Benefit Studies”, Land Economics, Vol. 52/1, pp. 32-41.
Chakraborty, K. and J.E. Keith (2000), “Estimating the Recreation Demand and Economic Value of Mountain Biking in Moab, Utah: An Application of Count Data Models”, Journal of Environmental Planning & Management, Vol. 43/4, pp. 461-469, https://doi.org/10.5367/000000006776387097.
Chay, K.Y. and M. Greenstone (2005), “Does air quality matter? Evidence from the housing market”, Journal of Political Economy, Vol. 113, pp. 376-424, https://doi.org/10.1086/427462.
Czajkowski, M. et al. (2015), “The Individual Travel Cost Method with Consumer-Specific Values of Travel Time Savings”, Faculty of Economic Sciences, University of Warsaw Working Papers 12/2015(160), www.wne.uw.edu.pl/files/6714/2651/1660/WNE_WP160.pdf.
Cheshire, P.C. and S. Sheppard (1995), “On the Price of Land and the Value of Amenities”, Economica, Vol. 62, pp. 247-267.
Cheshire, P.C. and S. Sheppard, (1998), “Estimating the demand for housing, land and neighbourhood characteristics”, Oxford Bulletin of Economics and Statistics, Vol. 60, pp. 357-382, https://doi.org/10.1111/1468-0084.00104.
Cheshire, P.C. and S. Sheppard (2002), “The welfare economics of land use planning”, Journal of Urban Economics, Vol. 52, pp. 242-269, https://doi.org/10.1016/S0094-1190(02)00003-7.
Christie M. et al. (2006), Valuing forest recreation activities, Report to the UK Forestry Commission. www.forestry.gov.uk/pdf/vfrfcfinalreportv5.pdf/$file/vfrfcfinalreportv5.pdf.
Clawson, M. and J.L. Knetsch (1969), Economics of Outdoor Recreation, John Hopkins University, Baltimore.
Correll, M.R., J.H. Lillydahl and L.D. Singell (1978), “The Effects of Greenbelts on Residential Property Values: Some Findings on the Political Economy of Open Space”, Land Economics, Vol. 54, pp. 207-17.
Corrigan, J.R., K.J. Egan, and J.A. Downing (2007), “Aesthetic Values of Lakes and Rivers”, in Likens, G.E. (ed.), Encyclopedia of Inland Waters, Elsevier, Amsterdam.
Creel, M. and J. B. Loomis (1990), “Theoretical and empirical advantages of truncated count data estimators for analysis of deer hunting in California”, American Journal of Agricultural Economics, Vol. 72, pp. 434-441.
Cropper, M.L., L.B. Deck and K.E. McConnell (1988), “On the choice of functional form for hedonic price functions”, Review of Economics & Statistics, Vol. 70, pp. 668-675.
Davis, L.W. (2011), “The effect of power plants on local housing values and rents”, Review of Economics and Statistics, Vol. 93/4, pp. 1391-1402, https://doi.org/10.1162/REST_a_00119.
Day, B. (2002), “Valuing visits to game parks in South Africa,” in Pearce, D.W., C. Pearce and P. Palmer (eds.) Valuing the Environment in Developing Countries: Case Studies, Edward Elgar, Cheltenham.
Day, B., I. Bateman and I. Lake (2006), “Estimating the demand for peace and quiet using property market data”, CSERGE Working Paper EDM 06-03, University of East Anglia, www.econstor.eu/bitstream/10419/80288/1/511176120.pdf.
Deschênes, O. and M. Greenstone (2011), “Climate Change, Mortality, and Adaptation: Evidence from Annual Fluctuations in Weather in the US”, American Economic Journal: Applied Economics, Vol. 3/4, pp. 152-185, https://doi.org/10.1257/app.3.4.152.
Dickie, M. (2017), “Averting Behavior Methods”, in Champ, P.A. K.J. Boyle and T.C. Brown (eds.) A Primer on Nonmarket Valuation, 2nd Edition, Kluwer, Dordrecht.
Dickie, M. and S. Gerking (2002), “Willingness to Pay for Reduced Morbidity”, Department of Economics Working Paper 02-07, University of Central Florida.
Doss, C.R. and S.J. Taff (1996), “The Influence of Wetland Type and Wetland Proximity on Residential Property Values”, Journal of Agricultural and Resource Economics, Vol. 21/1, pp. 120-29, https://core.ac.uk/download/pdf/7060414.pdf.
Earnhart, D. (2001), “Combining revealed and stated preference methods to value environmental amenities at residential locations”, Land Economics, Vol. 77/1, pp. 12-29.
Englin, J. and T. Cameron (1996), “Augmenting travel cost models with contingent behaviour data”, Environmental & Resource Economics, Vol. 7/2, pp. 133-147, https://doi.org/10.1007/BF00699288.
Fezzi, C., I. J. Bateman and S. Ferrini (2014), “Using revealed preferences to estimate the Value of Travel Time to recreation sites”, Journal of Environmental Economics and Management, Vol. 67/1, pp. 58-70, https://doi.org/10.1016/j.jeem.2013.10.003.
Freeman, A.M. III, J.A. Herriges and C.L. Kling (2014), The Measurement of Environmental and Resource Values, 3rd Edition, Resources for the Future, Washington, DC.
Garrod, G.D. and K.G. Willis (1992), “The Environmental Economic Impact of Woodland: A Two State Hedonic Price Model of the Amenity Value of Forestry in Britain”, Applied Economics, Vol. 24/7, pp. 715-28, https://doi.org/10.1007/BF00304970.
Garrod, G.D. and K.G. Willis (1999), Economic Valuation of the Environment, Edward Elgar Publishing Ltd., Cheltenham.
Geoghegan, J. (2002), “The value of open spaces in residential land use”, Land Use Policy, Vol. 19/1, pp. 91-98.
Gibbons, S. (2015), “Gone with the Wind: Valuing the visual impacts of wind turbines through house prices”, Journal of Environmental Economics and Management, Vol. 72, pp. 177-196, https://doi.org/10.1016/j.jeem.2015.04.006.
Gibbons, S. et al. (2016), “Fear of Fracking? The Impact of the Shale Gas Exploration on House Prices in Britain”, SERC Discussion Paper 207, London School of Economics & Political Science, London, www.spatialeconomics.ac.uk/textonly/SERC/publications/download/sercdp0207.pdf.
Gibbons, S., S. Mourato and G. Resende (2014), “The amenity value of English nature: A hedonic price approach”, Environmental and Resource Economics, Vol. 57, pp. 175-196, https://doi.org/10.1007/s10640-013-9664-9.
Gustafson, C.R., T.J. Lybbert and D.A. Sumner (2011), “Consumer Characteristics, Identification, and Hedonic Valuation of Wine Attributes: Exploiting Data from a Field Experiment”, Centre for Wine Economics, RMI-CWE Working Paper Number 1102, http://ageconsearch.umn.edu/record/162517/files/cwe1102.pdf.
Heberling, M. and J. Templeton (2009), “Estimating the economic value of national parks with count data models using on-site, secondary data: The case of the great sand dunes national park and preserve”, Environmental Management, Vol. 43/4, pp. 619-627, https://doi.org/10.1007/s00267-008-9149-8.
Herath, S. and G. Maier (2010), “The hedonic price method in real estate and housing market research. A review of the literature”, SRE – Discussion Papers, 2010/03, WU Vienna University of Economics and Business, Vienna, http://epub.wu.ac.at/588/1/sre-disc-2010_03.pdf.
Hoen, B. et al. (2011), “Wind energy facilities and residential properties: The effect of proximity and view on sales prices”, Journal of Real Estate Research, Vol. 33/3, pp. 279-316, http://pages.jh.edu/jrer/papers/pdf/past/vol33n03/01.279_316.pdf.
Horsch, E.J. and D.J. Lewis (2009), “The Effects of Aquatic Invasive Species on Property Values: Evidence from a Quasi-Experiment”, Land Economics, Vol. 85/3, pp. 391-409, https://doi.org/10.3368/le.85.3.391.
Jenkins, R.R., N. Owens and L.B. Wiggins (2001), “Valuing reduced risks to children: The case of bicycle safety helmets”, Contemporary Economic Policy, Vol. 19/4, pp. 397-408, https://doi.org/10.1093/cep/19.4.397.
Kolstad, C.D. (2010), Environmental Economics, 2nd Edition, Oxford University Press, Oxford.
Kong F., H. Yin and N. Nakagoshi (2007), “Using GIS and landscape metrics in the hedonic price modeling of the amenity value of urban green space: A case study in Jinan City, China”, Landscape & Urban Planning, Vol. 79, pp. 240-252, https://doi.org/10.1016/j.landurbplan.2006.02.013.
Kuminoff, N.V., C.F. Parmeter and J.C. Pope (2010), “Which hedonic models can we trust to recover the marginal willingness to pay for environmental amenities?”, Journal of Environmental Economics and Management, Vol. 60, pp. 145-160, https://doi.org/10.1016/j.jeem.2010.06.001.
Landry C. and P. Hindsley (2011), “Valuing beach quality with hedonic property models”, Land Economics, Vol. 87/1, pp. 92-108, https://doi.org/10.3368/le.87.1.92.
Landry, C.E. and H. Liu (2011), “Econometric Models for Joint Estimation of Revealed and Stated Preference Site-Frequency Recreation Demand Models”, in Whitehead, J., T. Haab and J.-C. Huang (eds.) (2011) Preference Data for Environmental Valuation: Combining Revealed and Stated Approaches, Routledge, London.
Lanz, B. (2015), “Avertive expenditures, endogenous quality perception, and the demand for public goods: An instrumental variable approach”, Research Paper 36, Graduate Institute Geneva, http://repec.graduateinstitute.ch/pdfs/ciesrp/CIES_RP_36.pdf.
Larson, D.M. and D.K. Lew (2014), “The opportunity cost of travel time as a noisy wage fraction”, American Journal of Agricultural Economics, Vol. 96/2, pp. 420-437, https://doi.org/10.1093/ajae/aat093.
Lee, C.M. and P. Linneman (1998), “Dynamics of the Greenbelt Amenity Effect on the land market: The case of Seoul’s greenbelt”, Real Estate Economics, Vol. 26/1, pp. 107-29.
Leggett, C.G. and N.E. Bockstael (2000), “Evidence of the effects of water quality on residential land prices”, Journal of Environmental Economics and Management, Vol. 39/2, pp. 121-44, https://doi.org/10.1006/jeem.1999.1096.
Le Goffe, P. (2000), “Hedonic pricing of agriculture and forestry externalities”, Environmental and Resource Economics, Vol. 15/4, pp. 397- 401, https://doi.org/10.1023/A:1008383920586.
Luttik, J. (2000), “The value of trees, water and open space as reflected by house prices in the Netherlands”, Landscape and Urban Planning, Vol. 48/3-4, pp. 161-167, https://doi.org/10.1016/S0169-2046(00)00039-6.
Neidell, M. (2009), “Information, avoidance behavior, and health: The effect of ozone on asthma hospitalizations”, Journal of Human Resources, Vol. 44/2, pp. 450-478, https://doi.org/10.3368/jhr.44.2.450.
Maddison, D. and B. Day (2015), Improving Cost-Benefit Analysis Guidance: A Report to the Natural Capital Committee, London, UK: Natural Capital Committee, www.gov.uk/government/uploads/system/uploads/attachment_data/file/517027/ncc-research-improving-cost-benefit-guidance-final-report.pdf.
Mahan, B.L., S. Polasky and R.M. Adams (2000), “Valuing urban wetlands: A property price approach”, Land Economics, Vol. 76, pp. 100-113.
Marin, A. and G. Psacharopoulos (1982), “The reward for risk in the labor market: Evidence from the United Kingdom and a reconciliation with other studies”, Journal of Political Economy, Vol. 90/4, pp. 827-853.
McCluskey, J.J. and G.C. Rausser (2003), “Stigmatized asset value: Is it temporary or long-term?”, Review of Economics & Statistics, Vol. 85, pp. 276-28, https://doi.org/10.1162/003465303765299800.
McConnell, V. and M. Walls (2005), The value of open space: Evidence from studies of nonmarket behavior, Resources for the Future, Washington, DC, www.rff.org/files/sharepoint/WorkImages/Download/RFF-REPORT-Open%20Spaces.pdf.
Morales, D.J. (1980), “The contribution of trees to residential property value”, Journal of Arboriculture, Vol. 7, pp. 109-12.
Morales, D.J., F.R. Micha and R.L. Weber (1983), “Two Methods of Valuating Trees on Residential Sites”, Journal of Arboriculture, Vol. 9, pp. 21-24.
Morancho, A.B. (2003), “A hedonic valuation of urban green areas”, Landscape and Urban Planning, Vol. 66/1, pp. 35-41, https://doi.org/10.1016/S0169-2046(03)00093-8.
Muehlenbachs, L., E. Spiller and C. Timmins (2015), “The Housing Market Impacts of Shale Gas Development”, American Economic Review, Vol. 105/12, pp. 3633-3659, https://doi.org/10.1257/aer.20140079.
Netusil, N.R. (2005), “The effect of environmental zoning and amenities on property values: Portland, Oregon”, Land Economics, Vol. 81/2, pp. 227-246, https://doi.org/10.3368/le.81.2.227.
Netusil, N.R., S. Chattopadhyay and K.F. Kovacs (2010), “Estimating the demand for tree canopy: A second-stage hedonic price analysis in Portland, Oregon”, Land Economics, Vol. 86/2, pp. 281-293, https://doi.org/10.3368/le.86.2.281.
Noor, N.M., M.Z. Asmawi and A. Abdullah (2015), “Sustainable Urban Regeneration: GIS and Hedonic Pricing Method in determining the value of green space in housing area”, Procedia – Social and Behavioral Sciences, Vol. 170, pp. 669-679, https://doi.org/10.1016/j.sbspro.2015.01.069.
Ovaskainen, V., M. Neuvonen and E. Pouta (2012), “Modelling recreation demand with respondent-reported driving cost and stated cost of travel time: A Finnish case”, Journal of Forest Economics, Vol. 18/4, pp. 303-317, https://doi.org/10.1016/j.jfe.2012.06.001.
Palmquist, R.B. (1992), “Valuing localized externalities”, Journal of Urban Economics, Vol. 31, pp. 59-68.
Parsons, G.R (2017), “The Travel Cost Model”, in Champ, P.A., K.J. Boyle and T.C. Brown (eds.) (2017) A Primer on Nonmarket Valuation, 2nd Edition, Kluwer, Dordrecht, www.springer.com/us/book/9781402014451.
Paterson, R.W. and K.J. Boyle (2002), “Out of sight, out of mind? Using GIS to incorporate visibility in hedonic property value model”, Land Economics, Vol. 78/3, pp. 417-425, https://doi.org/10.2307/3146899.
Pearson, L.J., C. Tisdell and A.T. Lisle (2002), “The impact of Noosa National Park on surrounding property values: An application of the hedonic price method”, Economic Analysis & Policy, Vol. 32/2, pp. 155-171, https://doi.org/10.1016/S0313-5926(02)50023-0.
Phaneuf, D.J. and V.K. Smith, (2005), “Chapter 15: Recreation demand models”, in: Mäler, K. and J. Vincent (eds.) (2005), Handbook of Environmental Economics, pp. 671-761, https://doi.org/10.1016/S1574-0099(05)02015-2.
Poor, P.J., K.L. Pessagnob and R.W. Paul (2007), “Exploring the hedonic value of ambient water quality: A local watershed-based study”, Ecological Economics, Vol. 60/4, pp. 797-807, https://doi.org/10.1016/j.ecolecon.2006.02.013.
Pope, J.C. (2008a), “Buyer information and the hedonic: The impact of a seller disclosure on the implicit price for airport noise”, Journal of Urban Economics, Vol. 63/2, pp. 498-516, https://doi.org/10.1016/j.jue.2007.03.003.
Pope, J.C. (2008b), “Do seller disclosures affect property values? Buyer information and the hedonic model”, Land Economics, Vol. 84/4, pp. 551-572, https://doi.org/10.3368/le.84.4.551.
Provins, A. (2011), The Use of Revealed Customer Behaviour in Future Price Limits, Final Report to Ofwat, Eftec and Cascade Consulting, www.ofwat.gov.uk/wp-content/uploads/2015/11/rpt_com_201105eftec_ casc_reveal.pdf.
Randall, A. (2014), “Weak sustainability, conservation and precaution”, in Atkinson, G. et al. (eds.) Handbook of Sustainable Development, Edward Elgar Publishing, Cheltenham.
Ridker, R.B. and J.A. Henning (1967), “The determinants of residential property values with special reference to air pollution”, Review of Economics and Statistics, Vol. 49/2, pp. 246-257.
Rosado, M. et al. (2006), “Combining averting behavior and contingent valuation data: An application to drinking water treatment in Brazil”, Environment & Development Economics, Vol. 11/6, pp. 729-746, https://doi.org/10.1017/S1355770X0600324X.
Rosen, S. (1974), “Hedonic prices and implicit markets: Product differentiation in pure competition”, Journal of Political Economy, Vol. 82/1, pp. 34-55.
Russell, C.S. (2001), Applying Economics to the Environment, Oxford University Press, Oxford.
Shrestha, R.K., A.F. Seidl and A.S. Moraes (2002), “Value of recreational fishing in the Brazilian Pantanal: A travel cost analysis using count data models”, Ecological Economics, Vol. 42/1, pp. 289-299, https://doi.org/10.1016/S0921-8009(02)00106-4.
Smith, V.K. and J. Huang (1995), “Can markets value air quality? A meta-analysis of hedonic property value models”, Journal of Political Economy 103, pp. 209-227, https://doi.org/10.1086/261981.
Steriani, M.K. and K.P. Soutsas (2005), “Recreation demand model construction through the use of regression analysis with optimal scaling”, New Medit, Vol. 4, pp. 25-30, http://newmedit.iamb.it/share/img_new_medit_articoli/116_25steriani.pdf.
Thorsnes, P. (2002), “The value of a suburban forest preserve: Estimates from sales of vacant residential building lots”, Land Economics, Vol. 78/3, pp. 426-441, https://doi.org/10.2307/3146900.
Tyrvainen, L. and A. Miettinen (2000), “Property prices and urban forest amenities”, Journal of Environmental Economics and Management, Vol. 39/2, pp. 205-223, https://doi.org/10.1006/jeem.1999.1097.
Walsh, P.J., J.W. Milon and D.O. Scrogin (2011), “The spatial extent of water quality benefits in urban housing markets”, Land Economics, Vol. 87/4, pp. 628-644, https://doi.org/10.3368/le.87.4.628.
Whitehead, J.C., T.C. Haab and J.-C. Huang (2000), “Measuring recreation benefits of quality improvements with revealed and stated behavior data”, Resource & Energy Economics, Vol. 22/4, pp. 339-354, https://doi.org/10.1016/S0928-7655(00)00023-3.
Wilhelmsson, M. (2000), “The impact of traffic noise on the values of single-family houses”, Journal of Environmental Planning and Management, Vol. 43, pp. 799-815, https://doi.org/10.1080/09640560020001692.
Yusuf, H.R. et al. (1996), “Leisure-time physical activity among older adults, United States, 1990”, Archives of Internal Medicine, Vol. 156, pp. 1321-1326, https://doi.org/10.1001/archinte.1996.00440110093012.
Zivin, J.G., M. Neidell and W. Schlenker (2011), “Water quality violations and avoidance behavior: Evidence from bottled water consumption”, American Economic Review, Vol. 101/3, pp. 448-453, https://doi.org/10.1257/aer.101.3.448.