Environmental and Resource Economics

, Volume 69, Issue 3, pp 449–466 | Cite as

Benefits Transfer: Current Practice and Prospects

  • V. Kerry SmithEmail author


This paper introduces a special issue devoted to the benefits transfer methods used as part of benefit costs analysis for policy analysis. Benefits transfer methods, as they are applied for environmental policy analyses, use economic concepts together with existing empirical estimates to predict the incremental benefits from a change in some feature of an environmental resource. After giving two examples of the decisions that analysts confront in performing these analyses, I discuss the interconnections between the papers in this issue and the research challenges that emerged from discussions of them.


Benefit cost analysis Regulatory impact analysis Benefit transfer 

1 Introduction

This special issue of Environmental and Resource Economics was designed to expand the conceptual and empirical methods used in benefits transfer. All the papers are about using transfers as part of the benefit cost analyses for evaluating environmental regulations. My overview has three objectives: (1) to provide context for how the papers were developed; (2) to describe some themes that connect the papers; and (3) to suggest a few next steps that emerged from the papers and discussion at the conference where these papers were presented.

About 2 years ago, economists within EPA both in the National Center for Environmental Economics and in the Office of Water invited me to participate, along with Elena Besedin, in planning a research program to improve benefits transfer methods for Regulatory Impact Analyses (RIA’s).1 Our goal was to identify research issues that, if addressed, might enhance the reliability of these transfers.

Five papers were commissioned. Each author (or team of authors) was given nearly a year to develop their responses to one or more substantive research questions. Then, we held a conference where the papers were presented. Each paper had two discussants. One was an analyst from the U.S. Environmental Protection Agency who had direct experience in using benefit transfers and understood the constraints such assessments face. The second came from the academic community.2 The discussion at the conference stimulated several new papers. Three papers were added to the five conference papers for this special issue. Two were prepared by economists at EPA. The third paper, by Nicolai Kuminoff, expands upon his commentary on Matthew Turner’s paper from the conference. His analysis provides complementary insights to those developed by Turner in that Kuminoff discusses the importance of the assumptions used in models describing market equilibria and how they condition Turner paper’s insights for benefits transfer.3

After a brief discussion of the objectives of each paper in this issue, I outline two examples of problems where benefits transfers might be undertaken. Hopefully, these stylized examples illustrate some of the details that can arise in adapting the evidence available to match policy needs. After that, I comment on the evolution of transfer methods. Finally, after describing some of the links between the papers I suggest some research questions that if addressed would help in enhancing benefits transfers.

2 The Papers

2.1 Composition of the Special Issue

After my introduction, there are eight papers in this special issue. The first of these by Newbold, Simpson, Massey, Heberling, Wheeler, Corona, and Hewitt provides the context for benefit analyses within the U.S. Environmental Protection Agency. The authors have all been active in developing benefit cost analyses for major rules. Their experience spans analyses of air and water related rules. After reviewing the idealized steps in a benefits transfer, they take us “inside” the real world of developing analyses with limited time and resources. Thus, this background sets the stage for the five papers commissioned for this effort.

Four of the commissioned papers can be seen as different responses to a generic question can an economic model of individual choice inform benefits transfer practices? Kling and Phaneuf evaluate two features of willingness to pay measures that have been used to consider their credibility in other contexts. They ask whether scope and adding up tests can be used for evaluating benefits transfers. Blow and Blundell consider another set of insights from consumer theory and ask whether the logic of nonparametric revealed preference methods, together with preference restrictions, can be used to develop welfare bounds that would be relevant for benefits transfers. Turner uses a simple model for two regions to illustrate how the adjustments of firms and households, in response to an exogenous change, condition what can be learned from partial equilibrium measures of the effects of environmental regulations. Finally, McConnell and Siikamaki describe how large changes in environmental quality should induce changes in ancillary expenditures on market goods or different patterns of time use, as households adjust to these changes. Their arguments suggest that these types of observable adjustments would provide an indirect basis for judging the plausibility of benefit measures. That is, if these effects can be confirmed by existing experience, then some types of indirect responses might serve as predictions for what should be expected with a large policy change.

Most benefits transfers are based on a partial equilibrium framework. When these methods are applied to a large change in some dimension of environmental quality there may be little context for comparison. Some analysts compare household level benefits to household income and ask rhetorical questions about whether the fractions are reasonable. Their analysis implies that we might expect other types of adjustments. So their paper considers what we can find. As I discuss below it documents the limitations in the spatial and temporal resolution we have for linking changes in environmental quality and people’s behaviors.

These papers stimulated two other contributions that were not originally commissioned. Kuminoff considers the assumptions that are important to Turner’s conclusions about the relevance of the locational adjustments underlying his equilibrium model. The second by Newbold, Walsh, Massey, and Hewitt proposes a formal way to “build in” an adding up condition into structural models for benefits transfer. Finally, the last paper by Boyle and Wooldridge was commissioned as part of the project and addresses benefits transfer as an econometric exercise. It asks what can be learned about how we do transfers and interpret the results if we recognize that benefits transfers are a form of econometric prediction.

2.2 What is Benefits Transfer?

Benefits transfer relies on economic concepts together with existing empirical estimates so that this evidence can be used to predict the incremental benefits from a change in some feature of an environmental resource. The Boyle-Wooldridge paper is the first to my knowledge to describe transfers as predictions. Practitioners have acknowledged that estimates derived from transfers are random variables. What has not been done is to describe the variance in the estimates by drawing the link to the variance of a prediction. Benefits transfer can also be interpreted as applied economic analyses that utilize estimates of economic parameters to evaluate a policy issue. While there have been efforts to treat the resulting measures as subject to error, there are many ways judgments enter the process. As a result, confidence intervals for the estimates are incomplete. It is difficult to characterize the judgments in formal terms. Repeated simulations for all the possibilities are feasible. These approaches often lead to such wide intervals that they are not informative. Analyses of the sensitivity of the results to changes in the framing of the question underlying a benefits transfer provide another approach. These analyses do not substitute for the empirical confidence intervals but instead may be more helpful in understanding the importance of how questions are framed than extensive simulations.

Dudley et al.(2017) have translated insights that have evolved in using benefit cost analysis for regulations into ten tips. Four of these tips are especially relevant to understanding benefits transfer: (a) the core problem; (b) the alternatives presented; (c)the baseline conditions describing the starting point for any regulatory action; and (d) the increments in the conditions associated with the regulation being evaluated. To make my discussion of how these tips arise in the decisions associated with developing a transfer, consider two simple examples.
Fig. 1

Example of the demand for trips to a recreation site for fishing with a price change used to describe the effects of a site closure

Suppose an analyst is faced with estimating the foregone benefits from an episode of contamination in a fishing stream that prevented the stream from being used for one season.4 If she had an estimate of the seasonal demand for fishing at this stream for the typical recreational user, then one way to assess the economic loss from the closure for a season would be to treat it as a price change. That is, the closure is assumed to be equivalent to a change from the price faced by an average user, labeled P\(_{0}\) in Fig. 1, to the choke price, labeled P\(_{\mathrm{c}}\). This change is the same as a closure because pricing access at the choke price implies the user would choose to not participate. Using the demand function displayed in the figure, the loss would be P\(_{0} \hbox {AP}_{\mathrm{c}}\).

In most cases analysts will not know the site-demand function that is associated with the contaminated stream. They might have an estimate of the demand for fishing at another stream. The use of this demand function for the stream of interest inevitably raises issues. Are the two locations comparable? Are the user groups comparable? The answers to these questions are some of the details I indicated as distinguishing different types of transfers. What can be done depends on what is known. The determinants of recreation trips could include the travel costs to the site (as the implicit price), the travel costs to substitute streams, the users’ income, as well as the user’s fishing experience and demographic characteristics. There are least two matches that can be considered—comparing the impacted stream versus the one where demand is known; and comparing the two user groups (including the potential users—reflecting the participation decisions). Information to address these questions is not routinely available. At best, the analyst might have an estimate of the number of lost trips to the impacted site.5

Figure 1 illustrates the situation where we assume the known demand function matches the site with the closure reasonably well. There remain other decisions that need to be made. For example, the implicit prices for using the site may be quite different. If the typical or average travel costs for the impacted site are higher the effects of the closure depends on the upper portion of the demand. For a linear function it is simply the size of the triangle describing the loss. It is also important to acknowledge that the figure assumes we can treat the average income and other possible determinants of known demand as the same as these measures for users of the impacted site. The effects of any differences will depend on the specification for the demand. If it is linear, as in the figure, then changes in these additive variables will lead to parallel shifts in the function in or out, depending on how each of the variables affects the demand.

In cases where the information may be more limited—either for the known site or the impacted site, the consumer surplus per trip for the recreation areas where we have the demand information (i.e. \(\hbox {P}_{0}\hbox {AP}_{\mathrm{c}}/\hbox {Q}_{0})\) is often estimated. Then, if there is only the one demand function known, this measure together with an estimate of the number of lost trips is used to approximate the aggregate loss in consumer surplus. When estimates of consumer surplus per trip can be constructed for a number of sites it is possible to consider how to use the set of such estimates.

Should the analyst select one or more of these estimates to use? Would it be desirable to consider the criteria for a “best” match between one of these sites and the impacted stream? Alternatively, since these criteria are not necessarily agreed upon, would it be better to average the estimates across the available studies? As we noted the selected summary measure is often the average consumer surplus per trip. However, we could also consider averaging the estimates for the absolute magnitude of the own price elasticity. With many studies and associated estimates, regression methods are another approach that can be used to summarize them. These meta regression models seem to have become the preferred approach when there is a sufficient body of research to provide a sample with sufficient variation in the potential determinants of the variable to be summarized. In these circumstances, the model predicts the benefit measure for the transfer.

My second example considers one of the most widely accepted strategies for benefits transfer. This framework links three elements: (a) a geographically delineated network of point sources for emissions of these pollutants and an air diffusion model that links them to the areas where affected people live; (b) a set of concentration response functions that describe how changes in ambient concentrations of each pollutant change the mortality risks for different subpopulations; and (c) an estimate of the unit values people would place on changes in mortality risk. This last component is usually identified as the source of the benefit transfer. The unit benefit measure in this case is an estimate of a marginal rate of substitution (MRS) between fatality risks and wealth. Conventional practice expresses it so that MRS is transformed to describe the tradeoff that would be made for a risk change sufficient to avoid one expected death. When expressed this way it is labeled the value of a statistical life (VSL). This is the unit value “transferred” in these applications. Estimates for these tradeoffs are drawn from labor market models of the compensating income differentials people receive to work in risky jobs as well as from contingent valuation surveys. I will argue the full logic defines a platform for benefits transfer. An especially important part of it can be found in how the framework describes the extent of the market for the policy. I use this label to refer to the set of people who would be willing to pay for the reduced risks associated with policies that limit emissions of the air pollutants associated with these risks.

The logic defines the extent of the market based on physical criteria arising from how reduced emissions are converted into risk reductions. It assumes a new air quality rule would reduce emissions from the different point sources. These reductions are converted into changes in ambient concentrations of each pollutant at each location where people live using the air diffusion models. These results are then used with the concentration response models to compute the expected number of fatalities avoided and the VSL is applied to estimate what the populations in each location experiencing the pollution reductions would pay for the risk reductions.

How would this definition be provided for my first example? If we had a measure of the lost trips to the contaminated site, this estimate has an implicit answer to the question of the extent of the market built in. In most cases we don’t have that measure. Even if we would be prepared to assume the trips would have been “the same as last year”, these data are not usually collected unless there are entry restrictions. We might begin with an estimate of the average trips a person takes for this type of recreation during a year. To develop a measure for the total loss we would then need to know how many people would be users. The product of these two estimates yields the total lost trips. Defining the people who would be among the users amounts to a specification of the extent of the market.

Both the river closure and the reduction in emissions to improve air quality identify another issue that is implicitly defined in the ways I described for using existing research. For the river closure the example reduced the closure to lost trips and for the air quality it was reductions in the expected number of fatalities due to air pollution. Each amounts to transforming a tradeoff measure with a specific type of normalization so it matches the measure of the policy’s effects. This transformation implies there are potentially other normalizations that could be used to define unit benefit measures. To explain what I mean consider another way of looking at the case of the stream. Instead of a closure, suppose one wanted the asset value of the stream due to the contamination permanently damaging the water quality into perpetuity. Assume further we had the demand function for recreational use that was relevant to the stream and this was the only basis for its economic value. The analysis might use the same logic—treating existence of the stream (or control of access to the stream thru property rights) as the equivalent of a price change. There are many possibilities and I don’t want to get into the “weeds” of the alternatives—suppose I picked the comparison of the existing price to the choke price with the associated consumer surplus of \(\hbox {P}_{0}\hbox {AP}_{\mathrm{c}}\). By selecting a time horizon and discount rate the present value could be estimated. In this example one might ask how can these asset values be transferred to another stream—is it reasonable to divide by miles of stream for $/mile? In general, I would argue the answer is “no”. But this practice of normalizing a set of existing measures of economic benefits derived for one purpose and using them along with different physical measures describing the resource involved, has been used in some benefits transfers.6 The process seems analogous to dividing consumer surplus by trips (as a measure of the quantity of use). However, it is not the same. While trips can be repackaged over a season, the analogy breaks down for miles of a stream. That is, treating streams as analogous to bags of coffee—where it might be reasonable to purchase coffee beans by the ounce—this does not apply to treating miles of stream as interchangeable.

Equally important, it overlooks the fact that the task of many studies of losses or gains from a policy narrow the focus for the analysis to specific “things” that are assumed to be directly affected by that policy. My example of contamination did not consider nonuse values for the stream. It did not consider how the stream might support other water bodies in the region. These “things” affect both what an individual would pay to protect the stream in perpetuity and the people we would include as part of the extent of the market.

2.3 Can the Economic Model of Choice Inform Benefits Transfer?

Six of the papers in this issue explore theoretical conditions that can guide the development of credible benefit transfer. These insights range describing the potential properties one should expect for the value estimates be used in benefits transfers to characterizing bounds for estimates developed using on benefit transfer methods. In addition Turner’s paper stimulated discussion of whether the equilibrium properties of urban spatial models provide information that could inform benefit transfers.

The first of the papers using theory to guide transfers, by Kling and Phaneuf, considers whether criteria used for evaluating contingent valuation results can be applied for judging practices used in benefit transfer. This literature has focused on the properties analysts should expect to observe in estimates of willingness to pay for changes in one or more nonmarket resources—the scope and adding up tests.

A scope test considers a property of the willingness to pay (WTP) function describing how WTP changes with the size of a proposed change in a nonmarket good. Economic theory suggest a simple condition—the willingness to pay for a larger increment in either the amount or the quality of an environmental service should not be less than the willingness to pay for a smaller increment in the same thing.7 Efforts to evaluate this condition within CV surveys are labeled the scope tests.

The adding up test is a bit more challenging to describe accurately in simple terms. It is often presented as the implication that the total value of a change is equal to the sum of the values for each of a set of separate changes in components that “add up” to the whole change. Stated in these terms it may seem to be common sense, but details are important. The adding up condition would require that an individual’s willingness to pay for “something”, say A+B, would equal the sum of what she was willing to pay for each of two parts of that “something”, say for A and then for B, provided the sum of amounts of those parts corresponds to the total (A+B). Since the formal definition of willingness to pay is about the properties of a function, they relate to estimates that are consistent with that function. Explaining these details to survey respondents has proved difficult and as a result, the test has not been useful in judging the credibility of CV surveys. Moreover, without defining the A and the B as parts of a single good or service it is possible to offer examples that would contradict this logic.8 Thus, Kling and Phaneuf do not recommend it for judging transfers.

Kling and Phaneuf suggest that Whitehead’s (2017) scope elasticity has the potential for application in some types of benefits transfers. They argue that it may be especially useful in the case of use values. In these situations it is possible to construct estimates for these elasticities from the results reported in valuation studies. Thus, they could be used to provide a plausibility gauge, as a range of values for the size of willingness to pay estimates from a transferred benefit strategy based on the range of elasticity measures from the literature for a comparable resource. This role is analogous to how a price elasticity provides insights about what to expect as plausible reductions in quantity demanded when a commodity’s price is increased.

The second Newbold et al. paper explains why Kling and Phaneuf were not optimistic about using the adding up test in benefits transfers. As I noted earlier, each part of the whole and the value for the sum of those parts must be consistently connected through a unifying economic model and in the framing of the estimation of each component value. Newbold et al. show us a way in which the property can be used in a variation on structural benefits transfer.9 If an analyst is prepared to impose a functional form consistent with the economic model of choice on statistical summaries of willingness to pay estimates, then adding-up conditions can be satisfied through these imposed structural assumptions.

Their suggestion treats a marginal willingness to pay function as the starting point for a modeling estimates of the total willingness to pay that would be used in a meta model for benefits transfer. It uses the relationship between the virtual price function for a quasi-fixed good and the indirect utility function to derive a consistent economic expression for the willingness to pay function.10 This approach can be important in practice because many environmental rules have a phased-in implementation process with incremental changes. As a result, the economic analysis of the benefits associated with these increments should measure successive changes consistently, as both the baseline of each new increase and the sizes of the successive increments are modified.

Of course, benefit transfers based on constant unit values automatically satisfy this consistency condition because they are not a function of either the baseline or the size of the change in the environmental resource being considered. Other approaches based on reduced form meta-analyses need not satisfy this condition. The properties of these models will depend on how the environmental quality measures are represented and the functional forms selected.

The papers by Blow and Blundell and by McConnell and Siikamaki build on the economic insights that are derived when environmental quality is introduced formally into the specification of an individual’s preference function and specific linkages are made between these services and marketed goods. They show how these connections offer another approach for establishing the credibility of benefit transfers. While these judgments will depend on the maintained assumptions, once we recognize that all consumption involves some of our time and often combinations of goods and services, it should not seem unusual to suggest that the use of environmental services would be the same. As a result, the next step in this logic is to propose we use the choices of related market goods to inform a benefits transfer.

Blow and Blundell demonstrate how this might work. Their focus has been on adapting the bounds for demand responses that they derived for price changes. By drawing an analogy to the case of quality changes for private goods, they demonstrate how the preference restrictions conventionally associated with revealed-preference approaches to nonmarket valuation can be used to create bounds for compensating-variation measures of value. Their framework recognizes that environmental quality is a quasi-fixed good that is “used with” other purchased commodities. In this case we can use the same revealed preference logic to estimate bounds for what the value of changes in environmental quality would have to be in order to rationalize the observed changes in expenditures on the related private goods. Their logic imposes a restriction on the form of the marginal utility for the private good linked to environmental quality. Responses to a change in environmental quality can be used to infer the implied role of quality in preferences, because these linkages are used to estimate the values that assure the observed behavior in adjusting market goods would be consistent with preferences.11 Blow and Blundell assume there is direct access to measures for changes in environmental services along with information on the “associated” changes in market goods and services.

In the case of quality changes for private goods the linkages are virtually automatic. The purchase of an improved private good delivers the improved quality. This is not necessarily the case for private adjustments in response to environmental quality changes. So the research program commissioned the Mc Connell Siikamaki paper to investigate “how close is close enough”. That is the changes in expenditures on private goods are linked to the environmental quality changes by maintained assumptions (on the part of the analyst). If we want to observe their effects we need to consider the spatial and temporal resolution of what we can observe to detect these adjustments.

McConnell and Siikamaki’s paper considers the relationships between observed changes in environmental quality and changes in corresponding market commodities, the second Blow and Blundell condition. Their analysis does not impose the non-parametric methods on what is observed but instead asks whether the resolution of existing data allows uncovering any linkages between changes in private goods with the changes in environmental quality. Their applications selected spatially differentiated changes in specific dimensions of environmental quality and then considered the available measures of how we expect people might respond to them. I use the term “might” because the resolution of both the quality changes and the household responses in private goods are not as closely linked as envisioned in the Blow and Blundell conceptual framework. While they find some limited evidence supporting the connections between large changes in environmental quality and an adjustment of private goods, the prospects for using existing data sources seem limited. Readers should view the McConnell and Siikamaki paper as doing two things. First, it describes how the Blow and Blundell framework might be implemented by investigators with access to environmental and market data and are able to demonstrate their connectivity. Second, it is a call for organizing data collection processes so that they enhance our ability to observe these connections between changes in environmental quality and behavior adjustments.

The two remaining papers motivated by models of economic behavior pose different questions. They consider whether the properties of market equilibria can be used to inform benefits transfers. Turner and Kuminoff argue in their separate papers that if we recognize what is observed in labor and housing markets as part of a spatial equilibria, that reflects the sorting of firms and households, what we assume about this adjustment process influences how analysts should interpret market information. How do the implied properties of the compensating differentials in wage rates or the capitalized value of spatial amenities in land markets affect what is done in benefits transfer? Both papers suggest it may be possible to modify benefits transfer using the theoretically expected properties from the spatial equilibrium framework. Turner emphasizes the joint role of productivity effects on firms and amenity effects on households. His analysis highlights when it is possible to gauge when general equilibrium effects are important for cases where a rule is easily recognized as large. Kuminoff reinforces this message in specific terms. The capitalization effects of spatially delineated changes in amenities in land values cannot be considered sufficient statistics as measures of marginal willingness to pay. The strong assumptions in many urban spatial equilibrium models are often violated. They do have, in Kuminoff’s terms, “first order effects” on the ability of these models to estimate marginal benefits. Thus, while Turner’s analysis implies that it may be possible to calibrate simple versions of these models to assist in judging the plausibility of benefits transfers, Kuminoff warns us the assumptions are important. So one of the tasks for future research needs to be judging how important they are.

2.4 Meta Regression Models Should Be Different When Used for Benefits Transfer

The last paper by Boyle and Wooldridge addresses the econometric issues in using meta-analysis for benefit transfer. It draws a very important distinction in the objectives of meta-regression models. Much of the meta-analysis literature in environmental economics is about summarizing what has been learned. This research implicitly asks which modeling assumptions matter and why? Boyle and Wooldridge argue that using a meta-regression model for benefit transfer is a different task! This use of meta-analysis should be interpreted as a special type of prediction. This distinction is important because it affects how analysts consider estimation of meta-regression equations to support value predictions to support benefit transfer. Practices such as adjusting for sample selection, sample weighting, and the use of study or geographical area fixed effects as controls, that would usually be regarded as among the best-practice modeling decisions, should be viewed differently when a meta-regression model is interpreted as a prediction equation. It seems clear that a specific type of prediction is the best way of interpreting the use of meta-regression expressions in benefit transfer. In addition, there are many other detailed results in this paper that have direct implications for the routine decisions used in meta-analysis from data-set construction to estimation, but the prediction “take away” message is especially important.

3 Next Steps

The research discussed in this special issue reflects the challenges in what might be described as the “benefits-transfer balancing act”. We will never observe the equivalent of market prices for the services of environmental resources. As a result, we face the dual challenge of using indirect methods (or stated preference methods) that are specific to the contexts where they are applied and then adapting the results to apply them more generally for policies. Most of the papers in this special issue are focused on the first task and, Boyle and Wooldridge begin the process of developing consistent protocols for using meta-regression models as part of the second task. The papers here offer a rich menu of important questions for future research. So, I will not repeat these suggestions here. Instead, I will conclude with five general observations that emerged from the conference. First, I contrast the process and the resources associated with updating the consumer price index (CPI) with research to support benefits analysis for environmental policy, including systematizing benefit-transfer methods. Second, I discuss the advantages and disadvantages of adopting convenient modeling platforms for benefits transfer. Third, I consider the difference in what we can best measure and the situations where those measures are used. My fourth issue is prompted by the recent EPA Regulatory Impact Analysis (June 2017) rescinding the Obama administration’s proposed rule defining the waters of the United States as part of implementing the regulations implied by the Clean Water Act. The new RIA treats some estimated benefits as unquantified because, among other things, they are based on “old” research.12 My question is when should the estimates be regarded as unreliable. Finally, this year’s Nobel Award in economics to Richard Thaler is one more indication of the great interest in behavioral economics. Some of these analyses argue consumers are easily mislead into making mistakes and analysts should focus on the “correct” choices, not the ones some people make. I close by considering how these issues should affect research in benefits transfer?

3.1 CPI and Nonmarket Benefits

Thirty-five years ago, in 1982, the Bureau of Labor Statistics first published its Handbook of Methods describing how the agency collects the economic data used in its indexes and the methods used to organize these data. Beginning in 2015, to address the delays and publishing updates to the handbook, the BLS distributed the information through the Handbook of Methods components of their webpage. Chapter 17 of the Handbook details the practices for the CPI and Exhibit 1 in the chapter provides a complete chronology of the changes in this index. Part two of the chapter describes the construction of the CPI, including the price measure adjustment for new goods, item replacement, adjustments for quality changes in commodities, as well as a host of other issues. The Handbook also describes the practices used in the consumer expenditure survey to assemble information for the weights used in constructing the CPI. Overall, the integration of: the research design, documentation of results, and then policy changes in the implementation of the CPI is very impressive. This structure offers a model for what might be possible in developing a systematic approach to benefit measurement to support regulatory impact analyses.

The financial resources for developing price related information are dramatically different from the modest funds used for benefit transfers in RIAs. For example, the Price and Cost of Living Section of the Bureau of Labor Statistics had a 2017 budget that was $210 million. By contrast, an informal upper-bound estimate for the annual research budget to develop benefits estimates by EPA was about $1 million.13 I have not attempted to uncover the cost of surveys to collect price information through the Consumer Expenditure Survey and so forth. On the EPA side, no doubt staff time and resources are included in developing benefit estimates and the $1 million is simply external grants. Nonetheless, it helps to pose the question how would a serious effort to collect and analyze the necessary data be designed and what would it cost?

3.2 Platforms for Benefit Analysis

I have used the term platform to describe a logical process for developing measure of what people would pay for changes in some dimension of environmental quality. In my opinion one platform has been widely adopted for benefits transfers associated with air quality improvements related to reductions in fatality risks. It has led to at least two efforts to systematize the process—Benmap for ozone and fine particles (see and what is now referred to as the AP2 model (see Muller 2011).14 These models embed partial equilibrium estimates of unit benefits within integrated assessment models. The benefit measures may be subjected to adjustments for cost of living changes and changes in measures of average household income, but aside from these adjustments, they are not recognized as endogenous outcomes of decisions of households and firms. If these unit benefit measures were to be treated as endogenous, then we must ask about how policy affects their values. That is, when does the size or time horizon for a change imply that it is no longer reasonable to treat them as local constants? When are the “feedback loops” we expect for these large changes important to the reliability of these models?

3.3 When Does the Research Shelf Need Cleaning?

My CPI and benefit platform discussion presents examples of how benefit transfers can be systematic and accommodate periodic updating. One advantage of developing such protocols is that they provide more credible economic estimates to support major decisions, but this consistency comes at substantial cost. These processes require continuous updating. Benefit transfers, as practiced, can customize implicit prices to specific decisions, but may also suffer from a lack of consistency across applications.

The research question I want to raise here is about the shelf life of benefit estimates. More specifically, we know research methods change. Over time the proposed changes are evaluated and the widely accepted practices may change. How should we use these developments to screen past results that are used for benefits transfers? The age of a study, alone, is not a criterion for converting a benefit estimate from “quantified” to “unquantified”. Most analysts engaged in applied benefit analysis would likely agree with this statement but what are the criteria—both with respect to the data and the methods? I think it is fair to say these questions are implicitly answered in updates to consensus measures of key unit benefit measures such as the VSL or the scaling factor to estimate the opportunity cost of time in relation to the wage rate for recreation demand models. However, they have not been subjected to systematic analysis. Benefits transfer has no protocol for what might be termed “spring cleaning of the research shelf”. More formally we don’t have criteria for judging when the timing of data used or the methods applied are sufficiently different from current understanding of consumer behavior or best practices to call for removal of the estimates from a data base.

Some similar questions may seem to arise with price indexes. But they take a generically different form, because we accept market prices as the best measures of marginal willingness to pay. So, the focus in the case of price indexes is on when some commodities “leave” the choice set—buggy whips and kerosene home lamps and when new ones enter—cell phone and internet services. What is the process for excluding or adjusting existing benefit estimates as we learn?

3.4 Prudent Use of Revealed Preference Implications for Nonmarket Resources

People’s responses to change in environmental quality can be subtle—they decide to drive to work instead of biking when there is an air pollution alert. Or, their children don’t play outside after school during an alert. They don’t move homes with the first alert. A consistent pattern of deteriorating air quality no doubt is one of several motivations. Even when it is present, households may wait until a strategic point is reached to take action. For example, their child would be changing schools in the next year so they wait until that point to minimize disruptions to her life. Such constraints don’t easily fit within the calculus of our simple models, but that does not mean they are not important to understanding real decisions. What they do mean is that it will be easiest to find people’s responses to large and sustained changes in environmental quality.15 The measures we can recover in these circumstances are increasingly reduced form estimates of the impact of a discrete exogenous change. In these cases we can make the most convincing arguments, without extensive maintained assumptions, that the effect we seek to estimate is in fact identified. However, in practice the changes we need to evaluate are often much smaller. Even when they are large changes their effects are not realized immediately for some of the types of reasons I gave in my examples above.

Rules have phased implementation which means environmental quality changes in small increments over time. Thus, we have measure that are best for large, discrete increments but must use them for small, marginal changes. And we have not identified the underlying structural relationship that describes how the discrete tradeoff relates to a person’s other choices or constraints. Some authors have argued we should seek to partially identify “some” of the structure. Sometimes this is labeled or interpreted as a sufficient statistic. In a policy context I would argue sufficient means—relevant to the policy context. We don’t need to estimate a full structural model to address this issue, but what do we need and how does it change with the policy question?

3.5 Who Decides?

Finally, calls to “correct” consumer sovereignty inevitably face the issue of who decides which choice is the “best” one? Stylized examples of mistakes when the analyst uncovering them does not know the information consumers had, constraints she faced, or the objectives underlying the choices that are observed must be regarded as conditional. That is, they are based on what is assumed by that analyst. Another observer could well conclude something else and ultimately the corrections depend on who is judged to be correct—the consumer who must face the consequences of her actions or the analysts who do not.

Both aspects of the balancing act in benefits transfer need research. The larger issues are associated with systems for benefits transfer and ongoing questions involving the methods used to recover and adapt individual tradeoff estimates for environmental services so they can be used in transfers. When current practices are translated into “modeling platforms”, such as Benmap or AP2, to facilitate policy responses, there needs to be a protocol for periodically evaluating the integrated framework and updating it. This is analogous in the CPI context to new approaches to construct index numbers. Currently research that develops estimates for the marginal willingness to pay for changes in the amount or quality of environmental resources focuses primarily on new methodologies, not replication or systematic efforts to develop tradeoff measures for specific classes of resources. The papers that follow suggest ways to enhance the complementarity between the needs of benefits transfer and the methodological focus of professional journals. That said, there is no escaping the need for a system to produce the basic information needed for the benefits transfers associated with well-defined classes of resources. Progress in both aspects of the balancing act will be limited until this happens.


  1. 1.

    The Planning Committee included, in addition to Elena and me, Allen Ashley, Julie Hewitt, Al McGartland, Steven Newbold, David Simpson, Michael Trombley, and William Wheeler.

  2. 2.

    “Appendix” section provides the program for the conference.

  3. 3.

    I summarized the comments of the two discussants for each paper and provided my own review and the composite of these suggestions lead to revisions in each of the papers. The papers by EPA authors and by Kuminoff were also reviewed and revised. All this process added time to the overall time span for completing the research effort.

  4. 4.

    The 2015 accidental spill of mine waste in the Animas River in Colorado is a more complex example resembling the generic features of the stylized example in the text. That spill affected waterways in other states and impacted other uses than recreation which would require additional benefits transfer methods. Kopp and Smith (1989) discuss an early natural resource damage case for the Eagle River (also in Colorado) that illustrates how the plaintiff and defendant’s experts estimated the losses from mine waste contamination thru some primary research and from benefits transfer methods.

  5. 5.

    Unless there is an entrance fee or some type of limit imposed on how users can access, information on patterns of use depend upon periodic surveys such as the U.S Fish and Wildlife Agency’s survey of hunting and fishing activity in the US. This survey does not identify specific sites. It reports participation and expenditures by type of activity. The lowest level of geographic disaggregation is usually the state.

  6. 6.

    The Costanza et al. (1997) infamous Nature paper can be seen as an extreme example of the dangers associated with adopting constant unit values for services available outside markets using arbitrary physical units to normalize estimates and to define the resources being valued.

  7. 7.

    This condition assumes non-satiation.

  8. 8.

    In a response to a comment on Bishop et al. (2017) published as a letter on July 28, 2017, the authors provide a simple example. Paraphrasing their example—suppose a person buys a piece of toast with jelly (A+B) at a price, and also is observed buying toast without jelly (A) at a different, lower price. If she refuses jelly alone (B) at the price difference between the two previous options, this decision tells us nothing about whether this was an irrational choice. The jelly is simply worthless to her without the toast.

  9. 9.

    This term appears to have been first used in Smith et al. (2002).

  10. 10.

    Their approach parallels Hausman’s (1981) approach using Roy’s identity and Marshallian demand functions to derive the quasi indirect utility function. See Larson (1992) for the case of nonmarket goods. This paper is the first to my knowledge to demonstrate how the logic can be used to advance benefits transfer.

  11. 11.

    The form of this condition is given by: \(U\left( {x_{10} ,X_k ,q_t } \right) =U\left( {x_{1t} ,X_{kt} } \right) +\alpha _t x_{1t} \), where U is the preference function, \(x_{10} \) is the initial level of the private good linked to the environmental quality, q. \(X_k \) is a vector of all other goods. The second subscript designates another value different from the base level (0).\(\alpha _t \) is the change in marginal utility and can be very general. This structure can be consistent with repacking restrictions to allow quality or other commonly used revealed preference restrictions, such as translating or cross product translating models. It may be possible to consider weak complementarity restrictions in this format as well. This issue remains for future research.

  12. 12.

    See Boyle et al. (2017) for further discussion.

  13. 13.

    Private correspondence with Dr. Albert McGartland, Director of National Center for Environmental Economics.

  14. 14.

    This was originally labeled the Air Pollution Emission Experiments and Policy analysis (APEEP) model and documented in Muller and Mendelsohn (2007).

  15. 15.

    This set of issues was developed in response to some penetrating questions raised by Kevin Boyle.


  1. Bishop RC, Boyle KJ, Carson RT, Chapman D, Hanemann WM, Kanninen B, Kopp RJ, Krosnick JA, List J, Meade N, Paterson R, Presser S, Kerry SV, Tourangeau R, Welsh M, Wooldridge JM, DeBell M, Donovan C, Konopka M, Scherer N (2017) Putting a value on injuries to natural assets: the BP oil spill. Science 356(6335):253–254CrossRefGoogle Scholar
  2. Blow L, Blundell R (2017) A nonparametric revealed preference approach to measuring the value of environmental quality. Environ Resour Econ.
  3. Boyle KJ, Kotchen MJ, Kerry Smith V (2017) Deciphering dueling analyses of clean water regulations. Science 358(6359):49–50CrossRefGoogle Scholar
  4. Boyle KJ, Wooldridge JM (2017) Understanding error structures and exploiting panel data in meta-analytic benefit transfers. Environ Resour Econ.
  5. Costanza R, d’Arge R, de Groot R, Farber S, Grasso M, Hannon B, Limburg K, Naeem S, O’Neill RV, Paruelo J, Raskin RG, Sutton P, van den Belt M (1997) The value of the world’s ecosystem services and natural capital. Nature 387(May):253–260CrossRefGoogle Scholar
  6. Dudley S et al (2017) Consumer’s guide to regulatory impact analyses: ten tips for being an informed policymaker. J Benef Cost Anal 8:1–18CrossRefGoogle Scholar
  7. Hausman JA (1981) Exact consumer surplus and deadweight loss. Am Econ Rev 17:662–676Google Scholar
  8. Kling CL, Phaneuf DJ (2017) How are scope and adding up relevant for benefits transfer? Environ Resour Econ.
  9. Kopp RJ, Smith VK (1989) Benefit estimation goes to court: the case of natural resource damage assessments. J Policy Anal Manag 8:593–612CrossRefGoogle Scholar
  10. Kuminoff NV (2017) Can understanding spatial equilibria enhance benefit transfers for environmental policy evaluation? Environ Resour Econ.
  11. Larson DM (1992) Further results on willingness to pay for nonmarket goods. J Environ Econ Manag 23:101–122CrossRefGoogle Scholar
  12. McConnell KE, Siikamäki JV (2017) Seeking external evidence to assess benefit transfers for environmental policy. Environ Resour Econ.
  13. Muller NZ, Mendelsohn RO (2007) Measuring the damages due to air pollution in the United States. J Environ Econ Manag 54:1–14CrossRefGoogle Scholar
  14. Muller NZ (2011) Linking policy to statistical uncertainty in air pollution damages. BE J Econ Anal Policy 11:32Google Scholar
  15. Newbold SC, Simpson D, Massey M, Heberling MT, Wheeler W, Corona J, Hewitt J (2017a) Benefit transfer challenges: perspectives from U.S. practitioners. Environ Resour Econ.
  16. Newbold SC, Walsh P, Massey DM, Hewitt J (2017b) Using structural restrictions to achieve theoretical consistency in benefit transfers. Environ Resour Econ.
  17. Smith VK, Van Houtven G, Pattanayak SK (2002) Benefit transfer via preference calibration: ‘prudential algebra’ for policy. Land Econ 78(1):132–152CrossRefGoogle Scholar
  18. Turner MA (2017) Benefit-transfer and spatial equilibrium. Environ Resour Econ.
  19. U.S. Environmental Protection Agency (2017) SAB review of EPA’s proposed methodology for updating mortality risk valuation estimates for policy analysis. Letter to E. Scott Pruitt, February 23Google Scholar

Copyright information

© Springer Science+Business Media B.V., part of Springer Nature 2018

Authors and Affiliations

  1. 1.Arizona State UniversityTempeUSA

Personalised recommendations