Valuing instream-related services of wastewater

Ecosystem Services 21 (2016) 59–71

Contents lists available at ScienceDirect

Ecosystem Services journal homepage: www.elsevier.com/locate/ecoser

Valuing instream-related services of wastewater Matthew A. Weber a,n, Thomas Meixner b, Juliet C. Stromberg c a

US Environmental Protection Agency, 200 SW 35th St, Western Ecology Division, Corvallis, OR 97333, United States University of Arizona, Department of Hydrology and Water Resources, 1133 E James E Rogers Way, J W Harshbarger Bldg Rm 122, PO Box 210011, Tucson, AZ 85721, United States c Arizona State University, School of Life Sciences, 427 E Tyler Mall, Tempe, AZ 85281, United States b

art ic l e i nf o

a b s t r a c t

Article history: Received 8 April 2016 Received in revised form 21 July 2016 Accepted 25 July 2016 Available online 3 August 2016

In the southwestern US water resources are increasingly scarce, leaving perennial habitats and associated environmental amenities vulnerable to off-channel water demands. To provide management insight, the value of two instream flow related ecosystem services are estimated for two river reaches, for two separate population centers. The specific services are preservation of instream flow extent and accompanying Cottonwood-Willow riparian forest, and improving water quality to be safe for full body recreational contact. The case study is of a highly modified effluent-dominated waterway, yet strong support for maintaining wet river habitat was documented, apparently due mainly to ecological rather than recreational motivations. In general, the more distant river reach with more trees was more highly valued on a per mile basis, and the population center closest to both river reaches more highly valued their preservation. Support was mixed for increasing water treatment to allow safe full body contact. Well-known multinomial and mixed logit models are compared with a relatively new generalized mixed logit framework, with the latter performing best. Documentation of public values associated with the posed river management options assist decision-making for the case study and similar contexts lacking quantification of the value of instream flow related ecosystem services. & 2016 Elsevier B.V. All rights reserved.

Keywords: Instream flow Riparian area Swimmable water quality Generalized mixed logit Wastewater Effluent

1. Introduction In the southwestern United States, competing pressures on water resources are extreme. Multiple extractive water demands from the agricultural, municipal, and industrial sectors must be balanced with protection and restoration of environmental resources. The extent to which society values allocations for environmental public goods is typically poorly characterized, since there is often no direct market to “sell” such ecosystem services. Without information on public values, it is difficult to assess whether environmental resources are being appropriately managed. In many instances people have both recreational use values and nonuse values (e.g. existence value) for environmental resources, complicating management tradeoffs. The only known technique of capturing total economic value inclusive of nonuse values are stated preference valuation surveys, also known as “willingness to pay” (WTP) studies. Stated preference surveys concerned with freshwater ecosystem services are an active area of research, with numerous studies at n

Corresponding author. E-mail addresses: [email protected], [email protected] (M.A. Weber), [email protected] (T. Meixner), [email protected] (J.C. Stromberg). http://dx.doi.org/10.1016/j.ecoser.2016.07.016 2212-0416/& 2016 Elsevier B.V. All rights reserved.

regional and even national scales. In the US, a classic reference is the “boatable, fishable, swimmable” water quality ladder, and associated marginal values between rungs (Carson and Mitchell, 1993). A more recent work conducted by Viscusi et al. (2008) valued changes in US lake acreage and river miles with “good” water quality on three dimensions: aquatic life support; safe fish consumption; and primary contact recreation without illness. Innumerable regional case studies are also available, valuing a variety of freshwater ecosystem services. In addition to original data collection, there is growing interest in benefit transfer to glean valuation insights across studies (Johnston et al., 2005; Van Houtven et al., 2007). Despite the body of previous work, continued case study research remains important for two reasons. First, the river and stream attributes that people prefer may differ by region. Only by engaging in case studies can any variability and patterns in these attributes due to geography or other contextual factors be documented. For example, in one area the foremost issue may be water quality related, in another, water quantity. Second, the actual dollar value for the same ecosystem service change may differ radically depending on regional context. Additional targeted case studies not only inform local-scale decision-making, but also improve the robustness of benefit transfer techniques that rely on numerous empirically derived estimates under varying circumstances. Indeed, Boyd and Krupnick (2013) argue insufficient

60

M.A. Weber et al. / Ecosystem Services 21 (2016) 59–71

Fig. 1. Map of the Santa Cruz River watershed.

attention has been paid toward defining ecosystem service metrics for valuation, implying that existing studies may be of limited use if there has not been a defensible process for defining publicly relevant metrics, an issue also taken up by Jeanloz et al. (2016). Our case study focuses on the Santa Cruz River in southern

Arizona (see Fig. 1). The Santa Cruz River is a classic example of southwestern riparian area loss, with layers of contemporary water management challenges. The channel once naturally carried perennial flow in some locations between the Mexican border and Tucson (Logan, 2002). These flows were dewatered by the

Fig. 2. Detail map of wastewater treatment outfalls on the Santa Cruz River and the associated perennial flow.

M.A. Weber et al. / Ecosystem Services 21 (2016) 59–71

mid-twentieth century due to increased groundwater withdrawals for irrigation. As regional urban populations began to grow rapidly in the same time period, the river was reborn due to continuous releases of treated wastewater. Treatment outfalls create two separate reaches of perennial flow (see Fig. 2). The northern reach is approximately 20 miles long, starting in heavily urbanized northwest Tucson, AZ. The southern reach is approximately 17 miles long, starting in Rio Rico, a rural area just north of the Mexican border. To maintain mandated water quality standards, wastewater treatment has recently been upgraded to tertiary levels in both reaches. Odor is reduced and the river now supports small fish. Improvements in the resource have led to new river management questions: what was previously an obscure resource has potentially become a more valuable public good. Our study was designed to value potential Santa Cruz River ecosystem service changes to aid planning of various river management possibilities. This valuation is a long-planned component of a decision support tool for the Santa Cruz watershed developed by multiple institutions (Norman et al., 2010). While our study does not inform a specific project, multiple separate proposals have been made in the region which would affect the river in various ways. Local entities actively track river conditions for both reaches (e.g. Sonoran Institute, 2010, 2016), and officials at some public agencies are interested in how environmental values should affect river decision-making, given their authority and responsibility for managing public resources (e.g. City of Tucson and Pima County, 2009). Interest in the river has significantly increased in recent years, partially due to major wastewater treatment upgrades. Ambitious federal and county-level projects have been proposed for the Santa Cruz River (e.g. Pima County, 2013; United States Army Corps of Engineers, 2010). However, lack of economic information regarding the tradeoffs between instream flow and alternative water uses have impeded decision-making. Public support for preservation, or for bonding to fund different outcomes such as recreational development, has been poorly known. Most vitally, future releases of effluent into the Santa Cruz are uncertain for a number of reasons. In the northern reach, the City of Tucson delivers treated wastewater for landscaping purposes, and in the past has considered treating wastewater to meet potable municipal demand. In the southern reach, most of the effluent is actually owned by Mexico. Treatment and infrastructure costs historically prevented Mexico's utilization of much of their effluent. Thus Mexican effluent was delivered to the US for treatment and discharge north of the border. However, this situation changed in 2012 when the Los Alisos treatment facility began operation, allowing Mexico the potential to retain control of the majority of the water contributing to surface flow in the southern reach. A possible outcome would be for the US to pay Mexico to guarantee future releases into the US, however the value of the water would obviously need to be characterized. An added complication to the policy backdrop came in 2008 with designation of portions of both reaches as “traditional navigable waters” of the US, proffering protections under the Clean Water Act. While some stakeholder groups have criticized the decision as an unnecessary constraint on land development (Arizona Daily Star, 2013), an overarching question is the extent to which people in the region value Santa Cruz River ecosystem services. As water resources become increasingly scarce, tradeoffs regarding how to best manage rivers, including effluent-dominated systems, become more important. Our study was intended to investigate selected ecosystem service values to inform the divided policy debate.

61

2. Methods 2.1. Qualitative study The survey was reviewed and approved by the US Office of Management and Budget as well as US EPA human subjects research review. Survey design was an iterative process of four qualitative research phases engaging members of the general public in southern Arizona, see flow chart and additional details in Fig. 3. The first two phases did not present participants with a survey, and instead were wholly devoted to identifying river issues in general, as well as gauging public knowledge. A central goal of these interactions was to ensure the survey featured environmental changes of high public relevance, i.e. “final” ecosystem services important to people in and of themselves, rather than technical jargon likely to invite speculation and reinterpretation (Boyd and Banzhaf, 2007; Boyd and Krupnick, 2013; Johnston and Russell, 2011; Ringold et al., 2013). The second two phases iterated the survey instrument, testing the number of attributes that would

Define Relevant Aributes PHASE 1 Fall of 2010 Convenience sample of Tucson Neighborhood Presidents residing near the Santa Cruz River. n=12

PHASE 2 Spring of 2011 Tucson focus groups sampling broad sociodemographics. Two meengs in upper watershed. Two meengs in Spanish. n=70 (over 10 groups)

Define Relevant Aributes PHASE 3 Fall of 2011 Pretest survey with Tucson residents, sampling broad sociodemographics. n=17

PHASE 4 Spring of 2013 Pretest survey with Tucson and Phoenix residents, sampling broad sociodemographics. n=17

Fig. 3. Flow chart of qualitative research phases involved in survey design.

62

M.A. Weber et al. / Ecosystem Services 21 (2016) 59–71

be included, how the changes would be described, and the geographic scope of where the changes would occur. In the interviews conducted for phase one we gathered overall insights and tested initial ideas about the relationship between the public and the river. For example, it was found that few of the participants were aware of the perennial reaches of flow, underscoring the importance of visual aids and actual photos for describing current river status. Phase two focus groups documented numerous public preferences for river and streams, including the high importance placed on the mere presence of surface water. Also strongly appreciated were vegetation types associated with riparian areas, particularly tall trees. Cottonwood-Willow “gallery forests” are a well-known ecological feature of riparian areas in the region (Stromberg, 1993). Additional findings from the focus groups are summarized in Weber and Ringold (2015). Based on phases one and two, a draft survey was created in a choice experiment format, in order to allow estimation of the marginal value of separate resource changes (Bateman et al., 2002). Phase three then pretested this draft with additional interviews. Initially the survey ambitiously featured several attributes, yet processing these proved overly complex for participants and we soon recognized a need to simplify. The first draft also focused solely on the northern reach, under the rationale of it being closest to southern Arizona population centers of Tucson and Phoenix. However, to test geographic scope we asked participants to comment on the relative appeal of preserving instream flow in the northern reach versus the southern reach, based on a map of their respective locations, and representative photos of the two areas. Many participants preferred maintaining the southern reach despite it being further away, due to increased tree growth adjacent to the river. With evidence of substantial value for another effluent-dominated reach of the Santa Cruz River, a reach

also undergoing policy debate, the survey was expanded to include the southern reach as well. A revised survey incorporated change scenarios for both the north and south perennial reaches with just two environmental attributes: preserving water for varying instream flow lengths with accompanying acreages of Cottonwood-Willow trees; and improving wastewater treatment to allow safe full body recreational contact. A critical point is that instream flow lengths were “bundled” with their accompanying forest acreages, rather than separated for independent valuation. Public knowledge that flow and forest are correlated was highly evident from preceding qualitative research, thus scenarios in which flow varied independently of vegetation for a given reach would be questioned. Safe full body contact terminology was selected over swimmable water quality since the water tends to be shallow precluding swimming, and we did not want to mislead survey recipients. A stylized image of a person reclining in a shallow stream was shown in the survey to emphasize this point. The final attribute, cost, was phrased as a permanent increase in state taxes. A state tax was more realistic than a water utility charge, since the survey would be sent to households in both the Tucson and Phoenix metro areas. A perpetual tax was posed with the logic that the opportunity cost of water dedicated to environmental purposes, and/or the increased cost of water treatment for safe full body contact, would require continuous funding to be justified. Introductory text in the revised survey was shortened since there were now fewer attributes. A brief background was retained on Santa Cruz River history including two maps that showed the watershed location, as well as the location of the outfalls and perennial flow lengths (Figs. 1 and 2). Differences between the north and south reaches were plainly stated, with the south having “4–5 times” as much forest per mile as the north, something

Fig. 4. Photos of river condition.

M.A. Weber et al. / Ecosystem Services 21 (2016) 59–71

63

Fig. 5. Summary of north reach vs. south reach conditions.

also apparent in provided photos (Fig. 4). The reason for the difference was also described: in the south, effluent flows leverage the presence of shallow groundwater. In focus groups and in survey pretesting, a common debate was the appropriateness of ecological water use for instream flow or vegetation support, given scarce water conditions for humans. Thus a full page diagram showing how much water would be consumed (10%) versus infiltrating to replenish the groundwater table (90%) was included. This information was critical since people tended to think the consumptive use fraction would be much higher. Phase four, final pretesting, brought the total number of persons involved in the qualitative phase to 116. Higher survey comprehension was found with the narrowed scope of attributes. Pretests were purposefully staged, with several further minor edits made over the sequence of meetings. One such change was a conceptual bar chart to ensure that the difference between north and south flow lengths and forest acreages would be clear, illustrating the maximum (Current Condition), intermediate, and minimum (Expected Future) levels, see Fig. 5. Each choice question included 3 options, with options A and B changing one or more attribute levels from the “Expected Future” anticipated to occur within 10 years. The third Expected Future option was minimum flow and forest extents for the north and south reach, and water that remained unsafe for full body contact recreation. An example choice question is shown in Fig. 6. Throughout pretesting, participants demonstrated strong interest in maintaining instream flow, even in the case of treated wastewater. Different preferences for the various attributes were noted: some respondents looked for the greatest change in forest per dollar; others preferred preserving conditions closer to their residence. In addition, there seemed to be a split between those with high interest in water safe for full body contact, and others who indicated that it would be difficult for them to “trust” the water due to its source. Overall the survey was carefully written to contain neutral language to support an informed “vote”. It was noted that without instream flow the waterways would still support an ecosystem, just not a wet river type of ecosystem. As a reminder of substitutes, southern Arizona rivers having natural perennial flow were listed, namely the San Pedro River and the Upper Gila River. The consequentiality of respondent choices is emphasized in several places

Fig. 6. Sample choice experiment question.

in the survey, in keeping with the arguments of Carson and Groves (2007) and Vossler et al. (2012), e.g. on the inside front cover it states “This survey is designed to help managers select the best option”. The survey is available from the lead author upon request. 2.2. Biophysical modeling A highlight of the study is close coordination between the interdisciplinary author team, allowing the final choice experiment to be highly flexible to “final” ecosystem services defined by the public, with quantified changes estimated by expert biophysical modeling. Valuation studies do not always have this grounding, which may contribute to a phenomenon of vague attribute descriptions in stated preference studies (Boyd and Krupnick, 2013). Of particular note in this case was assessing the current extent of perennial flow and forest, predicting miles of flow and “tall tree” forest acreage for different flow scenarios, and assessing the envelope of future water release possibilities. While metrics besides those featured in the survey would have been easier to supply (for example streamflow in cubic feet per second), they would not have been as relevant to respondents. Moreover, relatively accurate forecasts were needed if results were to be useful for policy inference. Baseline instream flow lengths for the north and south reaches were assessed using orthophotos for low flow (June) conditions, when surface water in both reaches is essentially 100% effluent. Niraula et al. (2015) developed a model of the system utilizing the well-known Soil and Water Assessment Tool (SWAT) supported by the USDA Agricultural Research Service. This entailed estimating infiltration rates through the streambed of the Santa Cruz River. Modeling completed for this paper altered effluent discharge to the river and assumed bed infiltration rates did not vary. This resulted in the ability to translate different discharge rates into varying instream flow lengths for the north and south reaches. Partnering ecological research established a baseline estimate of Cottonwood-Willow or “tall tree” forest acreage along the Santa Cruz River, based on twelve field sites along the Santa Cruz River (White, 2011). Field sites were established in different portions of the river to sample areas with different vegetation types and densities. Data from individual sites were assumed to represent

64

M.A. Weber et al. / Ecosystem Services 21 (2016) 59–71

conditions throughout the corresponding reach. A condition class approach (Stromberg et al., 2006) was then applied to characterize similar reaches, and also model how tree acreage would decline with reduced stream flow lengths. Cottonwood and Willow trees are phreatophytes, requiring essentially continuous access to water, thus reductions in stream flow lengths translate directly into reduced tree acreage. Tree response modeling incorporated reachspecific factors such as depth to groundwater. For example, Cottonwood and Willow trees in the north reach are extremely sensitive to continued access to surface flows since the groundwater table is deeper than their roots, whereas in the south reach trees have access to relatively shallow groundwater. Despite the advantage of tailored hydrologic and vegetation modeling, it should be noted that impacts on wildlife dependent on instream flow and vegetation habitats was not available and could not be included in the survey. To finalize scenarios, consultations with county, state, and regulatory entities were made to verify current conditions and discuss future possibilities. Although exact forecasts could not be made, complete removal of effluent from either reach was deemed improbable even with increased use of effluent for off-channel purposes. Thus, some perennial streamflow remains in each reach for all scenarios, based on best professional judgment. The assumed minimum effluent flows were then translated into minimum streamflow lengths and Cottonwood-Willow forest acreages through the sequence of biophysical modeling described above. Ultimately each reach had three flow and forest possibilities, a maximum (Current Condition), a minimum (Expected Future), and an intermediate level. 2.3. Survey design A fractional factorial design was employed to make the most of survey respondent effort (Louviere et al., 2000). The statistical software package SAS (v 9.3) was used to develop an efficient and balanced choice experiment design, given a total number of design choice sets to manipulate, as well as a provisional beta vector obtained from pretests described above (Kuhfeld, 2010). A design size of 72 choice profiles, plus one constant, no cost, opt-out alternative (the Expected Future), was selected as the smallest number of profiles allowing both orthogonality and balance within a main effects model. These profiles were then optimally blocked into nine survey versions of four replications each. The computer generated design was manually adjusted for dominating choices within survey blocks. Four choice questions per respondent allows collecting more preference information per respondent, and also permits respondents to see and judge a variety of options rather than just one choice question, which can seem unrealistic given all the possibilities. A summary of the attributes and levels are shown in Table 1. The medium and maximum extents for north reach flow and forest are coded as NMIDD and NFULL, respectively. Corresponding extents for the south reach being SMIDD and SFULL, respectively. Safe full body contact for the north and south are coded as NCONT and SCONT, respectively. Preceding the four choice questions, question 1 is a “warmup” in which respondents were asked to rate the importance of the five attributes (including cost) on a scale of 1–5, with 5 being the most important, similar to Johnston et al. (2011). After the choice questions, several debriefing questions were included to gain further insight into respondent decision-making, such as protest of the tax payment vehicle, or otherwise rejecting the scenario. Briefly, recreational habits and preferences were collected, and the survey closed with several sociodemographic questions. A full page on the back of the survey as well as write-in fields within the debriefing and recreation questions facilitated collection of openended comments. The survey format was a booklet of letter-sized

Table 1 Summary of choice experiment attributes and levels. Attributes

Levels

North Reach Flow & Forest

Expected Future: 10 miles of flow and 45 forest acres NMIDD: 15 miles of flow and 65 forest acres NFULL (Current Condition): 20 miles of flow and 125 forest acres

North Reach Safe Full Body Contact

Expected Future (Current Condition): No NCONT: Yes

South Reach Flow & Forest

Expected Future: 9 miles of flow and 230 forest acres SMIDD: 13 miles of flow and 350 forest acres SFULL (Current Condition): 17 miles of flow and 460 forest acres

South Reach Safe Full Body Contact

Expected Future (Current Condition): No SCONT: Yes

Cost

Expected Future: $0 Intervention Options: $5; $15; $30; $60; $100

sized sheets folded on the short axis and saddle-stitched, a layout similar to Johnston et al. (2011). The sample frame was the metropolitan statistical areas (MSAs) of Phoenix and Tucson. Thus the sample does not represent all of the persons of the state, but instead focuses on the two major urban areas which together account for 79% of the population of Arizona (US Census). A mail survey was employed, a relatively low-cost technique not relying on internet connectivity and allowing use of visual aids (Champ, 2003). Sample sizes of 1000 household addresses for each MSA were purchased from a marketing firm. A pilot test of 10% of these addresses was used to update the provisional beta vector and cost levels for the full mailing. The most important difference between the pilot and full mailing was reducing the minimum bid to $5 and increasing the maximum bid to $100 to better constrain WTP. A full five-contact methodology was used, a best practices approach for achieving a high response rate (Dillman, 2000; Dillman et al., 2009). An email address and a toll-free number for the lead author were included in survey materials for transparency, which several respondents made use of. To make the survey more accessible to the Latino community, the cover letters, post card reminder, and final notices were all English/Spanish bilingual in a “swim-lane” layout (Rothhaas et al., 2011). These materials noted the availability of a Spanish version of the survey if desired, which could be requested by email or by phone, however only one such request was received. 2.4. Statistical analysis Choice experiment results were analyzed within a random utility framework. Respondents are assumed to choose the river management option that maximizes their utility. Utility for a given respondent is modeled as dependent upon the array of river attributes, sociodemographics, and an error (or “random”) term. Until recently the standard in environmental choice modeling analysis was the basic multinomial logit (MNL), e.g. as described by Ben-Akiva and Lerman (1985). The MNL requires relatively restrictive assumptions, and in recent years increases in computing power have made more advanced and flexible models possible. These models include the mixed logit (MXL), and most recently the generalized mixed logit (GMX). The MXL allows preference parameters to vary across individuals to account for taste heterogeneity, and can also account for correlation when multiple choice questions are being answered by a single respondent (Train, 2009). By allowing correlation of unobserved factors influencing choice,

M.A. Weber et al. / Ecosystem Services 21 (2016) 59–71

the MXL relaxes the so-called independence of irrelevant attributes (IIA) assumption. The GMX allows incorporation of both taste heterogeneity and scale heterogeneity (Fiebig et al., 2010; Greene and Hensher, 2010). The corresponding utility models are mathematically linked as follows, summarizing from Fiebig et al. (2010), and neglecting sociodemographic regressors for simplicity:

Unjt = βx njt + εnjt

( MNL model)

Unjt = (β + ηn)x njt + εnjt

( MXL Model)

Unjt = ⎡⎣ σnβ + γηn + ( 1 − γ )σnηn⎤⎦x njt + εnjt

( GMX Model)

where U is the utility for person “n” choosing “j” alternative in choice occasion “t”. The vector of choice attributes is x, β is the vector of coefficients, and ε is the error. The MXL includes taste heterogeneity with η, a vector of person-specific deviations from the mean. The GMX includes η, as well as scale heterogeneity with s, a person-specific scaling of the error term. The GMX also includes two additional parameters; the γ term shown in the equation accounts for how taste heterogeneity changes with scale. Model estimation will also return tau (τ), representing the standard deviation of s, with interpretation of τ being an indicator of the extent of scale heterogeneity. As τ approaches zero, the GMX model approaches the MXL. Finally, GMX model estimation will report sigma, which is not a parameter but simply the sample average of the computed values of the individual scale parameters sn. While advancements from the MNL are generally preferred, no single extension is unilaterally “best”. Theoretic tradeoffs exist (see Hensher et al., 2012; Scarpa et al., 2012), and model fit criteria such as the Akaike Information Criterion (AIC) are commonly used to aid model selection. In hypothesizing utility functions, each change from the Expected Future was treated as a categoric variable (i.e. a “different betas” approach), although differences in forest acreages and flow lengths are numeric values. The categoric variable approach allows more flexibility to gauge respondent response to each discrete change separately, rather than imposing a functional form tied to (e.g.) miles of perennial flow (Johnson et al., 2007). Ultimately utility functions are used to calculate welfare linked to environmental changes. There are arguments both for and against including alternative-specific-constants (ASCs) in calculating such welfare changes. Note that ASCs can be either positive, which can indicate yea-saying bias, or negative, which can indicate status-quo bias. Regardless, the ASC captures unobserved factors. Here we provide WTP estimates without the inclusion of the influence of ASCs in order to focus on how the attributes themselves affect welfare. Willingness to pay for each change is then calculated as

WTP = − βcategoric change /βcost

3. Results and discussion Overall response rates were 24% in Phoenix and 31% in Tucson, after accounting for changed or undeliverable addresses. All survey responses were entered twice for data quality control. Debriefing questions and written comments were used to detect responses protesting the tax payment vehicle, or otherwise rejecting the scenario. Eleven surveys were dropped due to protesting the tax payment vehicle, and nineteen surveys were dropped due to scenario rejection. Three forms of scenario rejection were noted: overall distrust of the government; a ‘leave

65

nature alone’ ideal (even though the system is human controlled by the treatment outfalls); and wanting additional information. Selected summary statistics for the final sample size of 421 are given in Table 2. A single combined ASC was created for options A and B to represent affinity for selecting an option other than Expected Future, independent of attribute changes. Numerous provisional MNL models were run to narrow the list of potential sociodemographic variables to those significantly influencing Santa Cruz River management choices. Despite our efforts to make the survey broadly accessible, the sample is relatively enriched in higher educated, higher income, and white respondents as compared with US Census data for Phoenix and Tucson. However, corresponding variables showed little or no influence in provisional models. Thus these variables were dropped, and subsequent weighting on these characteristics for WTP predictions was not considered. Model estimation was done with NLOGIT5. Initially we hypothesized the Tucson and Phoenix subsamples would have markedly different values. While this is true for at least some attributes, through the course of modeling a separate factor besides geography of respondent also became clear. Initial models returned relatively low importance and significance to the NCONT and SCONT variables; this led to further inquiry as to why a variable with such a strong link to the classic attribute of swimmable water quality would be disregarded. It turned out that respondents are actually divided into two groups with preferences that tend to cancel, one strongly preferring safe full body contact, and one with a strong aversion to the same. Latent class models were run which further evidenced this split. A method of sorting the sample was found based on responses to question 1. Recall this was the warmup asking respondents to rate each attribute on a five-point scale. Group 1 was defined as those persons selecting either NCONT or SCONT (or both) as having a low importance score of 1 or 2. Group 2 was defined as the remainder of the sample. The GMX results for the Pooled sample, the Phoenix subsample, the Tucson subsample, and Groups 1 and 2 are shown in Table 3. The MNL and MXL results, as well as AIC for all three model frameworks, are placed in Appendix A. For all model frameworks the Pooled, Group 1, and Group 2 samples are weighted based on US Census 2010 data to account for the greater number of households in the Phoenix MSA as compared with the Tucson MSA. For the GMX, WTP results for each categoric change for each sample are shown in Table 4, with standard errors calculated by the delta method. Corresponding WTP results for the MNL and MXL are found in Appendix B. In comparing extensions to the MNL, the MXL and GMX had progressively superior fit in terms of AIC. Normal and triangular distributions were tested for all choice attributes besides cost (which was fixed), with the normal performing best. It should also be noted that GMX results show strong evidence that scale heterogeneity is present with nonzero and statistically significant estimates for tau, indicating a preference phenomenon beyond the scope of the other model frameworks. Use of the GMX appears to still be scarce, and we do not know of another environmental study using it for final model runs. Instead, recent environmental choice experiments favor the MXL (e.g. Baulcomb et al., 2015; Christie et al., 2015; Remoundou et al., 2015). In addition to the above results, the GMX performed best in terms of matching expectations. Examining first the flow and forest variables (NMIDD, NFULL, SMIDD, SFULL), all are highly significant with the expected sign for all samples. The MNL and MXL show some exceptions to this foundational test. We can also check for the expected transitivity, i.e. coefficients on NFULL and SFULL exceeding those of NMIDD and SMIDD, respectively. This occurs in every case for the MNL, but is also typical for the MXL and GMX. Subsample results allow us to compare the preferences of Tucson, the relatively

66

M.A. Weber et al. / Ecosystem Services 21 (2016) 59–71

Table 2 Sample descriptive statistics, n¼ 421. Parameter

Explanation

Mean (or Percentage) Std. Dev. Min.

Max.

Q1-A Q1-B Q1-C Q1-D Q1-E YRBORN YRSAZ KIDS DONOR

Importance of North reach flow and forest (likert scale) Importance of North reach full body contact (likert scale) Importance of South reach flow and forest (likert scale) Importance of South reach full body contact (likert scale) Importance of Cost to household per year (likert scale) What year were you born? How many years have you lived in Arizona? How many people under 18 yrs of age live in your household? Has anyone in your household donated money to an environmental organization within the past 5 years? Are you male or female? (male coded as 1; female as 0) Composite index of number of times household members engaged in any of several listed river-related activities within the past 12 months Composite index of importance the household places on several listed river-related activities for the Santa Cruz River Distance of household from first North reach outfall (calculated in GIS) Distance of household from South reach outfall (calculated in GIS)

3.44 2.34 3.40 2.30 3.82 1955.34 28.96 0.24 0.25

1.48 1.48 1.49 1.45 1.23 13.37 18.76 0.43 0.44

1 1 1 1 1 1916 0.5 0 0

5 5 5 5 5 1991 87 1 1

0.68 13.44

0.47 5.71

0 0

1 42

17.78

10.05

0

50

78.13 150.68

70.11 79.71

1.79 245.29 31.65 324.91

2.0% 7.8% 11.3% 24.5% 28.7% 25.7%

N/A

N/A

N/A

1.2% 4.2% 2.2% 9.6% 78.4% 4.4%

N/A

N/A

N/A

2.4% 5.6% 20.6% 19.3% 14.6% 12.7% 12.7% 7.9% 1.6% 2.6%

N/A

N/A

N/A

MALE RECTIMES REC-SCR N-DIST S-DIST

EDUCATION What is the highest level of education that you have completed? 1. 2. 3. 4. 5. 6. RACE

From the following options, do you consider yourself to be: 1. 2. 3. 4. 5. 6.

HH-INC

Less than high school High school or equivalent High school þ tech school One or more years of college Bachelor's Degree Graduate Degree

American Indian Asian African American Latino/Hispanic White Other

What category comes closest to your total household income for 2013? 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

Less than $10k $10k to $19.9k $20k to $39.9k $40k to $59.9k $60k to $79.9k $80k to $99.9k $100k to $149.9k $150k to $199.9k $200k to $249.9k $250k or more

nearby urban area, with those of Phoenix, which is situated relatively further away from the featured river reaches. Limiting ourselves to GMX results for brevity, we find higher values in Table 4 for Tucson, the more proximate residents. An additional question is whether preserving the southern reach appears to be preferred to the northern reach, given the increased tree acreage per mile in the south, with trees tending to be larger in the south besides. The GMX results show a premium for the south reach despite greater distance from urban areas. This premium is accentuated on a per mile basis if the values for categoric changes in Table 4 are divided by their respective numbers of flow miles preserved. Moving on to results for NCONT and SCONT, these are more varied than flow and forest preferences. Appendix A shows that in the MNL, significance is lacking for all subgroups except negative significance for Group 2. Looking at MXL results, extracting Group 2 also yields negative significance, although significance is also found in other subsamples. The GMX results in Table 3 are never significant, but Group 2 is negative while Group 1 is positive in conformance with the other models. Overall, the results on full body contact are somewhat unstable across models for different samples. We believe different factors coalesce to produce this. First, the statistical design rested on an assumption that NCONT and SCONT would be amenities rather than disamenities for some subgroups.

That different subgroups view these attributes so differently makes efficient capture of those preferences challenging, unless different subgroups can be given different survey designs a priori, which would invoke concerns of “starting point bias”. It is notable that the simple MNL model can capture the Group 1 vs. statistically significant negative Group 2 preferences, while the more complex MXL and GMX models cannot. With a challenging statistical design scenario, combined with a relatively modest sample size, we appear to be at the limit of what the more complex models can discern. That two such dramatically different groups would exist, almost evenly split (Group 2 membership is only slightly smaller than Group 1), is remarkable. The negative coefficients in particular invite further investigation. A small coefficient or statistical insignificance would be easier to explain as indifference. Examination of survey responses including write-in comments provides insight on this unusual outcome. A relatively small proportion of the sample recreates on any portion of the Santa Cruz River – for example approximately one third of respondents reported no river-related recreation in the past year at all. Of the households that did report some recreation, more than half reported zero trips for the Santa Cruz River. For the few that did recreate on the Santa Cruz, the main destination was the dry reach in the downtown Tucson area. Only about 10% of the sample listed the perennial

M.A. Weber et al. / Ecosystem Services 21 (2016) 59–71

Table 3 Generalized Mixed Logit Model Results: for a given variable first and second reported numbers are coefficient and standard error respectively; third and fourth numbers are standard deviation and standard error of the individual-specific coefficient estimates respectively.

67

Table 4 Willingness to pay for categoric changes per household per year in 2014$, based on the Generalized Mixed Logit (GMX). First reported number is the estimate, second number is standard error. Variable

Pooled

Phoenix

Tucson

Group 1

Group 2

NMIDD

24.98*** 6.09

29.83*** 5.43

30.07*** 5.49

21.47*** 6.31

24.13*** 8.62

1.94*** 0.591 1.70* 0.990

NFULL

56.06*** 10.50

46.10*** 7.15

54.07*** 7.35

54.90*** 9.56

51.92*** 18.55

NCONT

 0.03 5.08

10.71*** 2.85

2.32 4.05

18.49*** 5.12

 9.29 8.85

4.19*** 0.910 2.76*** 1.04

4.18*** 1.34 0.402 1.26

SMIDD

67.09*** 10.48

41.94*** 6.46

66.99*** 9.27

73.33*** 14.73

43.97*** 16.89

SFULL

71.76*** 10.53

49.16*** 6.60

61.48*** 7.15

90.66*** 15.64

44.46*** 13.84

0.208 0.365 3.92*** 0.849

1.41*** 0.452 3.12*** 0.828

 0.748 0.684 1.55** 0.714

SCONT

 5.63 4.97

3.62 3.35

0.19 3.65

4.35 5.42

 12.55 9.59

5.75*** 1.18 1.43* 0.798

6.00*** 1.04 0.708 0.513

5.59*** 1.35 1.73* 0.937

3.54*** 1.29 0.0149 1.33

5.62*** 0.768 3.13*** 0.602

6.75*** 1.25 1.54*** 0.550

5.51*** 0.857 6.16*** 1.21

6.92*** 1.44 2.74*** 0.999

3.58*** 1.02 1.72* 0.960

SCONT

 0.441 0.378 2.86*** 0.462

0.497 0.468 4.38*** 0.924

0.0171 0.327 3.68*** 0.808

0.332 0.425 2.78*** 0.748

 1.01 0.750 2.90** 1.24

COST

 0.0784***  0.137*** 0.0106 0.0226

 0.0896***  0.0763***  0.0806*** 0.0110 0.0170 0.0149

ASC

 2.40*** 0.847

 1.88*** 0.709

YRSAZ

 0.0561***  0.0630***  0.0480***  0.0612***  0.0386** 0.0138 0.0169 0.0112 0.0233 0.0177

DONOR

2.99*** 0.620

3.12*** 0.793

2.63*** 0.441

3.30*** 1.04

1.97*** 0.687

MALE

 1.02** 0.469

 2.12*** 0.677

 0.951*** 0.358

0.893 0.699

 1.86*** 0.693

RECTIMES

0.0508 0.0440

0.244*** 0.0818

 0.141*** 0.0405

0.0573 0.0685

0.00403 0.0665

REC-SCR

0.132*** 0.0234

0.0877*** 0.0247

0.210*** 0.0255

0.0877** 0.0366

0.0690* 0.0376

Tau

1.33*** 0.207

1.28*** 0.169

1.14*** 0.148

1.39*** 0.176

1.53*** 0.445

Gamma

0.774*** 0.0945

0.408*** 0.109

0.139 0.103

0.790*** 0.115

0.303 0.264

Sigma: Sample Mean & Sample St Dev

0.904 1.47

0.898 1.27

0.963 1.31

0.951 1.68

0.850 1.53

Variable or Output n

Pooled

Phoenix

Tucson

Group 1

Group 2

1668

712

956

938

730

NMIDD

1.96*** 0.438 2.11*** 0.586

4.09*** 0.835 3.53*** 0.918

2.69*** 0.537 3.99*** 0.947

1.64*** 0.516 2.00** 0.941

4.39*** 0.706 1.41** 0.644

6.32*** 1.18 1.61* 0.962

4.84*** 0.808 0.448 0.320

NCONT

 0.00264 0.398 3.43*** 0.619

1.47*** 0.440 3.52*** 0.815

SMIDD

5.26*** 0.762 1.10** 0.559

SFULL

NFULL

 3.73*** 1.13

 1.34 1.19

 0.774 1.05

*

Significant at the 10% level. Significant at the 5% level. *** Significant at the 1% level. **

north or south reaches (where one might contact water) as the primary Santa Cruz River destination. Thus, recreation is not a strong explanation for the WTP respondents exhibited to preserve flow and forest. Furthermore, a debriefing question asked why people voted for safe full body contact (if they did). About 6% of the sample checked “My household is interested in recreation contact with the treated water, and 19% checked ”I want other people to be able to recreationally contact the treated water”. However, the most common multiple choice response was “I was

***

Significant at the 1% level.

forced to choose Full Body Contact in order to get more Flow & Forest” (23% of sample). Several write-in comments expressed lack of support for full body contact, including distrust of “toilet water”. There may have been some aversion on principle for options containing safe full body contact when that feature was not valued (yet appears to inflate the cost). In addition, many persons made it a point to provide written comments at various points in the survey that their main motivation was preserving ecological characteristics, as opposed to recreational motivations. A few respondents went further, speculating that increasing the safety of recreation would lead to increased use and impact ecological resources, adversely affecting their overall objective in regards to Santa Cruz River management, e.g.: “I think that not having the highest water treatment may limit people going in the water and be more beneficial for wildlife” (emphasis in the original). Besides ecosystem service attributes and cost, several additional regressors were also seen to influence choices. The ASC associated with selecting anything besides the expected future is typically negative, a common finding which can be interpreted as status quo bias. Interestingly the significance of the ASC tends to diminish with the separation of the sample into Group 1 and Group 2, apparently the ASC proxies for the Group 2 aversion to full contact recreation. Sociodemographic variables are interacted with the ASC since they do not vary by respondent. The interaction for DONOR is positive as expected and MALE is negative which is also common. Interestingly, YRSAZ is strongly negative. Perhaps longer-term residents have developed more affinity for rivers besides urban resources such as the Santa Cruz. The recreation variables RECTIMES and REC-SCR are typically positive but not unilaterally. Additional sociodemographic variables were tested at some effort, but ultimately dropped. Distance from respondents’ addresses to river reaches were calculated in GIS, but the associated variables were unstable, and of miniscule impact besides. Finally, the response time for each survey was painstakingly tracked with the hypothesis that early responders might be more supportive of management changes, however this variable was not significant. Our results can be placed in context of the larger literature valuing ecosystem services of rivers in the western US. A major factor in comparing studies is the variety in how ecosystem services are defined. To summarize, our study is designed to estimate total economic value (use and nonuse value) and encompass two types of ecosystem service changes: preserving miles of streamflow and associated acres of Cottonwood-Willow riparian vegetation; and making water safe for full body contact recreation. Whereas a

68

M.A. Weber et al. / Ecosystem Services 21 (2016) 59–71

classic valuation topic is quantifying the recreational benefit of instream flow from perspectives such as boating and fishing (e.g. Brown et al., 1991; Duffield et al., 1992), this excludes nonuse values, and the typical issue is varying flow rate rather than wet vs. dry conditions. A study more similar to ours is found in Berrens et al. (2000), who value maintaining instream flow sufficient for survival of endangered fish in New Mexico. Their study design does encompass nonuse values, and the authors estimate an annual per household WTP of $25 for the Middle Rio Grande, and $55 when the survey question was expanded to four major New Mexico Rivers. Our preservation values would seem to be on the high side in comparison, given that the Middle Rio Grande alone extends 170 river miles. Yet the difference may be explainable by the relatively extreme change of wet vs. dry conditions that we posed. A recent paper by Broadbent et al. (2015) most closely resembles ours in scope of ecosystem services and geography. The study examines the San Pedro River, which neighbors the Santa Cruz watershed. Although their survey has great detail on birds (with variables such as canopy bird numbers and water bird numbers for a reach), they also include wet vs. dry river conditions, as well as riparian vegetation of different types. They estimate WTP at $52 for preserving four miles of wet river and 231 “type 1″ vegetation acres (a type analogous to our CottonwoodWillow forest). They also provide similarly framed results for the Middle Rio Grande, but the preservation value is insignificant and negative. Our scenarios of NMIDD and SMIDD are similar to their San Pedro results in terms of miles of streamflow preserved, being five and four miles of flow preserved respectively. However, our respective acreages of Cottonwood-Willow forest preserved are far less, at 80 and 120 acres. Referencing just our pooled GMX results, our estimate for NMIDD preservation is about $25, and SMIDD preservation is about $67. The two studies are in a similar range particularly when considering confidence intervals, yet it should be noted that the San Pedro result is associated with a one-time payment rather than the annual payment mechanism we pose. Moving on to water safe for full contact recreation, as noted earlier, the issue of swimmable water quality is also a key topic in valuation research. The seminal paper by Carson and Mitchell (1993) cites large values for achieving swimmable water quality, values which have been routinely used by the United States Environmental Protection Agency in calculating costs and benefits of water quality regulations (Weber, 2010). We do not know of studies besides ours investigating swimmable water quality values for an effluent-dominated system, nor do we know of any other study showing mixed results. While our finding is attributable to the streamflow being effluent, it should be noted that many waterways have effluent as a significant portion of their flow, even if the public is unaware of it. However, the more important point is that our study shows substantial value for protecting ecosystem services of a waterway despite apparent lack of traditional contact recreation.

4. Conclusions This study provides estimates of the value of ecosystem service changes for an effluent-dominated river in the southwestern US. Two types of ecosystem services are valued, for two separate reaches of river, querying two separate metro areas. Moreover, the GMX is utilized, which in this case outperforms other tested ways of representing environmental preferences. A particular strength of the study design was extensive pretesting and design flexibility to incorporate public feedback. The interdisciplinary author team developed the survey through multiple iterations, finally presenting an instrument highly relevant to the public while having a strong foundation of case-specific biophysical modeling. Preserving instream flow and associated Cottonwood-Willow

riparian forest was strongly supported. For both northern and southern reaches, model results typically show preference for increased flow and forest as expected. Of the two reaches, the southern reach with increased forest growth was preferred, indicating that quality is more important than proximity. The more proximate Tucson subsample showed slightly larger support for flow and forest preservation for both reaches than the Phoenix subsample. Subsample modeling uncovered factions both for and against full body contact cutting across the two populations surveyed. Low river-related recreation rates raise a substantial possibility that those supporting safe full body contact are concerned with users outside of their own families. The substantial minority opposed to safe full body contact appears to be concerned with resource damage due to recreational use, use of public funds for a low priority, and stigma associated with wastewater. Mixed results on safe full body contact are interesting since similar “swimmable” water quality attributes are omnipresent in water quality valuation, and the authors do not know of any other model results showing neutral findings, let alone statistically significant negative coefficients. While this is undoubtedly partially due to the river carrying treated wastewater, a larger point is that in this case, recreation is secondary to ecological motivations. This is an extraordinary result given the highly impacted nature of the Santa Cruz River, part of which is situated within a heavily urbanized area. Normally nonuse values are associated with more pristine resources. We supply public values for changes in river ecosystem services in order to inform water management, in a region with numerous competing water demands. Benefit estimates provided by this paper could be combined with cost information in order to conduct benefit-cost analysis or net benefits analysis of different management options. The pejorative term “novel ecosystem” is sometimes applied to resources such as our case study that are heavily modified and/or rely strongly on human inputs. However, the Santa Cruz River and similar areas also contain natural elements significantly contributing to human welfare, thus their management should be carefully considered.

Disclaimer This manuscript has been subjected to Agency review and has been approved for publication. The views expressed in this paper are those of the authors and do not necessarily reflect the views or policies of the United States Environmental Protection Agency. Mention of trade names or commercial products does not constitute endorsement or recommendation for use.

Acknowledgements The lead author would like to thank postdoctoral funding through the US Environmental Protection Agency for supporting this research. The second and third authors would like to thank the National Science Foundation, DEB Grant #1038938. The authors are grateful for the comments and/or assistance of three anonymous reviewers, as well as the following persons: E. Brott; E. Canfield; E. Curley; C. Cvitanovich; J. Fonseca; J. Hoehn; E. Holler; H. Huth; R. Johnston; M. Massey; R. Niraula; P. Ringold; V.K. Smith; M.H. Weber, M. White; C. Zugmeyer; and of course all of the survey pretesters and respondents who took the time to submit their opinion and preference. Persons named above do not necessarily endorse the manuscript. All errors and faults remain with the authors. Appendix A See Table A1.

M.A. Weber et al. / Ecosystem Services 21 (2016) 59–71

69

Table A1 Supplemental Model Results: Akaike Information Criterion (AIC) for Multinomial Logit (MNL), Mixed Logit (MXL), and Generalized Mixed Logit (GMX), as well as coefficient estimates for the MNL and MXL. For a given variable first and second reported numbers are the coefficient and standard error respectively; third and fourth numbers (MXL model) are standard deviation and standard error of the individual-specific coefficient estimates, respectively. Variable or Output

Model

Pooled

Phoenix

Tucson

Group 1

Group 2

AIC

MNL MXL GMX

1156.6 995.5 944.5

1350.1 1182.3 1059.2

1873.2 1597.7 1492.8

629.1 579.3 542.7

457.8 402.0 387.8

NMIDD

MNL

0.362** 0.151

0.401*** 0.141

0.334*** 0.117

0.363* 0.186

0.469 0.286

MXL

1.09** 0.541 3.12*** 0.824

1.23** 0.492 4.28*** 1.03

0.447 0.380 2.98*** 0.504

0.874 0.581 2.79*** 0.884

0.950 0.997 3.97*** 1.39

MNL

0.772*** 0.177

1.01*** 0.167

0.609*** 0.137

0.764*** 0.220

0.986*** 0.333

MXL

1.73*** 0.558 3.02*** 0.857

3.06*** 0.815 3.93*** 0.928

1.42*** 0.414 3.15*** 0.625

1.75*** 0.657 2.74*** 0.880

2.53** 1.21 3.63*** 1.30

MNL

 0.00435 0.140

0.156 0.128

 0.136 0.109

0.277 0.175

 0.815*** 0.274

MXL

 1.00** 0.500 3.41*** 0.841

 0.443 0.482 2.89*** 0.835

 1.27*** 0.408 3.02*** 0.500

0.404 0.507 2.50*** 0.875

 3.20*** 1.14 3.77** 1.60

MNL

0.616*** 0.213

0.756*** 0.193

0.505*** 0.168

0.803*** 0.283

0.437 0.360

MXL

1.10* 0.585 2.27*** 0.580

1.83*** 0.665 3.68*** 0.839

0.992** 0.456 2.91*** 0.656

1.76** 0.706 1.56** 0.775

1.09 1.09 1.90* 1.08

MNL

0.826*** 0.223

0.937*** 0.201

0.731*** 0.177

1.06*** 0.295

0.484 0.382

MXL

1.68** 0.674 3.69*** 0.865

2.46*** 0.783 4.01*** 0.932

0.878* 0.504 3.50*** 0.684

1.96** 0.789 2.91*** 0.836

1.27 1.24 4.75*** 1.66

MNL

 0.0614 0.116

0.0306 0.111

 0.130 0.0894

0.128 0.143

 0.651*** 0.229

MXL

 0.997** 0.445 2.83*** 0.733

 0.287 0.411 3.45*** 0.737

 0.962*** 0.367 2.64*** 0.433

2.50*** 0.875 2.06*** 0.700

 2.86*** 1.09 4.33*** 1.35

MNL

 0.0176*** 0.00322

 0.0224*** 0.00304

 0.0143*** 0.00250

 0.0169*** 0.00415

 0.0239*** 0.00600

MXL

 0.0485*** 0.0102

 0.0711*** 0.0134

 0.0425*** 0.00736

 0.0427*** 0.0122

 0.0783*** 0.0246

MNL

 1.42*** 0.390

 1.75*** 0.365

 1.35*** 0.312

 1.26** 0.561

 0.510 0.628

MXL

 2.01** 0.996

 4.58*** 1.26

 1.81** 0.825

 2.22* 1.35

0.726 1.92

MNL

 0.0140*** 0.00505

 0.0227*** 0.00496

 0.00663* 0.00387

 0.0202*** 0.00759

 0.0131* 0.00763

MXL

 0.0282** 0.0142

 0.0554*** 0.0157

 0.0105 0.0105

 0.0290 0.0178

 0.0444* 0.0244

MNL

1.27*** 0.252

1.22*** 0.244

1.35*** 0.193

2.14*** 0.544

0.965*** 0.330

MXL

2.83*** 0.849

3.33*** 1.03

2.60*** 0.556

3.25*** 1.10

2.28* 1.18

MNL

 0.398** 0.203

 0.397** 0.181

 0.383** 0.167

0.170 0.308

 0.715** 0.298

MXL

 0.998* 0.591

 0.755 0.617

 0.834* 0.474

0.461 0.753

 2.03** 0.985

MNL

0.0185 0.0190

0.0667*** 0.0190

 0.0192 0.0150

0.0472* 0.0273

 0.0334 0.0330

NFULL

NCONT

SMIDD

SFULL

SCONT

COST

ASC

YRSAZ

DONOR

MALE

RECTIMES

70

M.A. Weber et al. / Ecosystem Services 21 (2016) 59–71

Table A1 (continued ) Variable or Output

REC-SCR

* **

Model

Pooled

Phoenix

Tucson

Group 1

Group 2

MXL

0.0458 0.0517

0.223*** 0.0724

 0.0310 0.0401

0.0693 0.0660

0.00734 0.105

MNL

0.0660*** 0.0112

0.0540*** 0.00948

0.0833*** 0.00999

0.0351** 0.0144

0.0750*** 0.0216

MXL

0.133*** 0.0305

0.128*** 0.0343

0.161*** 0.0284

0.0629** 0.0321

0.0716 0.0565

Significant at the 10% level. Significant at the 5% level. Significant at the 1% level.

***

Appendix B See Table B1.

Table B1 Supplemental willingness to pay results for Multinomial Logit (MNL) and Mixed Logit (MXL), with values for each categoric change per household per year in 2014$. First reported number is the estimate, second number is standard error. Variable NMIDD

NFULL

NCONT

SMIDD

SFULL

SCONT

Pooled **

Phoenix

Tucson ***

Group 1 **

Group 2

***

17.89 6.23

23.45 8.32

21.53 10.97

19.66 12.21

MNL

20.54 8.57

MXL

22.49** 10.14

17.32*** 6.05

10.52 8.62

20.48 12.50

12.12 12.15

MNL

43.81*** 9.59

45.18*** 7.10

42.68*** 9.13

45.31*** 12.51

41.31*** 13.57

MXL

35.61*** 9.84

43.09*** 7.47

33.44*** 8.61

40.95*** 12.62

32.24*** 11.64

MNL

 0.25 7.95

6.96 5.29

 9.56 8.72

16.43* 8.66

 34.14** 16.80

MXL

 20.64* 11.88

25.69*** 7.64

 29.85*** 11.27

9.46 10.83

 40.81** 16.21

MNL

34.95*** 10.23

33.72*** 7.40

35.44*** 9.88

47.63*** 13.75

18.32 13.61

MXL

22.59** 10.51

34.56*** 7.85

23.33** 9.40

41.14*** 13.06

13.96 12.58

MNL

46.88*** 9.87

41.80*** 7.14

51.28*** 9.53

62.74*** 13.39

20.30 14.03

MXL

34.62*** 10.98

 6.23 6.73

20.66* 10.66

45.93*** 14.17

16.15 13.14

MNL

 3.49 6.80

1.37 4.88

 9.11 6.82

7.62 8.08

 27.26** 12.54

MXL

 20.54** 10.01

 4.04 5.92

 22.62** 9.57

6.59 9.68

 36.47*** 12.80

*

Significant at the 10% level. Significant at the 5% level. *** Significant at the 1% level. **

References Arizona Daily Star, 2013. Builders Lose a Round in Dispute Over ‘Navigable’ Santa Cruz River. Howard Fischer Capitol Media Services. Bateman, I.J., Carson, R.T., Day, B., Hanemann, M., Hanley, N., Hett, T., Jones-Lee, M., Loomes, G., Mourato, S., Ozdemiroglu, E., Pierce, D.W., Sugden, R., Swanson, J., 2002. Economic Valuation with Stated Preference Surveys: A Manual. Edward Elgar,, Northampton, MA. Baulcomb, C., Fletcher, R., Lewis, A., Akoglu, E., Robinson, L., von Almen, A., Hussain, S., Glenk, K., 2015. A pathway to identifying and valuing cultural ecosystem

services: an application to marine food webs. Ecosyst. Serv. 11, 128–139. Ben-Akiva, M., Lerman, S.R., 1985. Discrete Choice Analysis. MIT Press, Cambridge, Massachusetts. Berrens, R.P., Bohara, A.K., Silva, C.L., Brookshire, D., Mckee, M., 2000. Contingent values for New Mexico instream flows: With tests of scope, group-size reminder and temporal reliability. Journal of Environmental Management 58 (1), 73–90. Boyd, J., Banzhaf, S., 2007. What are ecosystem services? The need for standardized environmental accounting units. Ecol. Econ. 63 (2), 616–626. Boyd, J., Krupnick, A., 2013. Using ecological production theory to define and select environmental commodities for nonmarket valuation. Agric. Resour. Econ. Rev. 42 (1), 1–32. Brown, T.C., Taylor, J.G., Shelby, B., 1991. Assessing the direct effects of streamflow on recreation: a literature review. Water Resour. Bull. Am. Water Resour. Assoc. 27 (6), 979–989. Broadbent, C.D., Brookshire, D.S., Goodrich, D., Dixon, M.D., Brand, L.A., Thacher, J., Stewart, S., 2015. Valuing preservation and restoration alternatives for ecosystem services in the southwestern USA. Ecohydrology 8, 851–862. Carson, R.T., Mitchell, R.C., 1993. The value of clean water: the public's willingness to pay for boatable, fishable, and swimmable quality water. Water Resour. Res. 29 (7), 2445–2454. Carson, R.T., Groves, T., 2007. Incentive and informational properties of preference questions. Environ. Resour. Econ. 37 (1), 181–210. Champ, P.A., 2003. Collecting survey data for nonmarket valuation. Chp 3. In: Champ, P.A., Boyle, K.J., Brown, T.C. (Eds.), A Primer on Nonmarket Valuation. Kluwer, Dordrecht, The Netherlands, 576 p. Christie, M., Remoundou, K., Siwicka, E., Wainwright, W., 2015. Valuing marine and coastal ecosystem service benefits: case study of St Vincent and the Grenadines’ proposed marine protected areas. Ecosyst. Serv. 11, 115–127. City of Tucson, Pima County, 2009. Phase II Final Report, Water & Wastewater Infrastructure, Supply, and Planning Study. 〈http://webcms.pima.gov/UserFiles/ Servers/Server_6/File/Government/Wastewater%20Reclamation/Water%20Re sources/WISP/Final-Report_LRG.pdf〉 (accessed 20.07.16). Dillman, D.A., 2000. Mail and Internet Surveys: The Tailored Design Method vol. 2. Wiley, New York. Dillman, D.A., Smyth, J.D., Christian, L.M., 2009. Internet, Mail and Mixed-Mode Surveys vol. 3. John Wiley and Sons,, NY, NY. Duffield, J.W., Neher, C.J., Brown, T.C., 1992. Recreation benefits of instream flow: application to montana’s big hole and bitterroot rivers. Water Resour. Res. 28 (9), 2169–2181. Fiebig, D.G., Keane, M.P., Louviere, J.J., Wasi, N., 2010. The generalized multinomial logit model: accounting for scale and coefficient heterogeneity. Mark. Sci. 29 (3), 393–421. Greene, W.H., Hensher, D.A., 2010. Does scale heterogeneity across individuals matter? An empirical assessment of alternative logit models. Transportation 37, 413–428. Hensher, D.A., Rose, J.M., Zheng, L., 2012. Does the choice model method and/or the data matter? Transportation 39, 351–385. Jeanloz, S., Lizin, S., Beenaerts, N., Brouwer, R., Passel, S.V., 2016. Towards a more structured selection process for attributes and levels in choice experiments: a study in a Belgian protected area. Ecosyst. Serv. 18, 45–57. Johnson, F.R., Kanninen, B., Bingham, M., Özdemir, S., 2007. Experimental design for stated-choice studies, Valuing environmental amenities using stated choice studies. Springer, Netherlands, pp. 159–202. Johnston, R.J., Besedin, E.Y., Iovanna, R., Miller, C.J., Wardwell, R.F., Ranson, M.H., 2005. Systematic variation in willingness to pay for aquatic resource improvements and implications for benefit transfer: a meta-analysis. Can. J. Agric. Econ./Rev. Can. Agroecon. 53 (2–3), 221–248. Johnston, R.J., Russell, M., 2011. An operational structure for clarity in ecosystem service values. Ecol. Econ. 70 (12), 2243–2249. Johnston, R.J., Schultz, E.T., Segerson, K., Besedin, E.Y., 2011. Bioindicator-based stated preference valuation for aquatic habitat and ecosystem service restoration. International Handbook on Non-Marketed Environmental Valuation. Edward Elgar, Cheltenham, UK, pp. 159–186. Kuhfeld, W.F., 2010. Marketing Research Methods in SAS. SAS 9.2 Edition, MR-2010. 〈http://support.sas.com/techsup/technote/mr2010.pdf〉 (accessed 20.07.16). Logan, M.F., 2002. The Lessening Stream: An Environmental History of the Santa

M.A. Weber et al. / Ecosystem Services 21 (2016) 59–71

Cruz River. University of Arizona Press, Tucson, AZ. Louviere, J., Hensher, D., Swait, J., 2000. Stated Choice Methods: Analysis and Applications. Cambridge University Press, New York. Niraula, R., Meixner, T., Norman, L.M., 2015. Determining the importance of model calibration for forecasting absolute/relative changes in streamflow from LULC and climate changes. J. Hydrol. 522, 439–451. Norman, L., Tallent-Halsell, N., Labiosa, W., Weber, M., McCoy, A., Hirschboeck, K., Callegary, J., Van Riper, C., Gray, F., 2010. Developing an ecosystem services online decision support tool to assess the impacts of climate change and urban growth in the Santa Cruz watershed; where we live, work, and play. Sustainability 2 (7), 2044–2069. Pima County, c. 2013. Trés Rios, Corazon, the Heart of Trés Rios del Norte at the confluence of the Santa Cruz, Rillito, and Cañada del Oro Rivers. 〈http://www. webcms.pima.gov/UserFiles/Servers/Server_6/File/Government/ Communications%20Office/corazon%20tres%20rios%20flyer%2004-08–13%20hires.pdf〉 (accessed 20.07.16). Remoundou, K., Diaz-Simal, P., Koundouri, P., Rulleau, B., 2015. Valuing climate change mitigation: a choice experiment on a coastal and marine ecosystem. Ecosyst. Serv. 11, 87–94. Ringold, P.L., Boyd, J., Landers, D., Weber, M., 2013. What data should we collect? A framework for identifying indicators of ecosystem contributions to human well-being. Front. Ecol. Environ. 112, 98–105. Rothhaas, C.A., Bentley, M., Hill, J.M., Lestina, F., 2011. Bilingual Questionnaire Assessment Report. US Census, Decennial Statistical Studies Division. 〈https:// www.census.gov/2010census/pdf/2010_Census_Bilingual_Questionnaire_As sessment.pdf〉 (accessed 20.07.16). Scarpa, R., Thiene, M., Hensher, D.A., 2012. Preferences for tap water attributes within couples: an exploration of alternative mixed logit parameterizations. Water Resour. Res. 48, W01520. Sonoran Institute, 2010. A Living River: Charting the Health of the Upper Santa Cruz River: 2010 Water Year. 〈https://sonoraninstitute.org/resource/living-river-

71

charting-the-health-of-the-upper-santa-cruz-river-2010-water-year-08-012011/〉 (accessed 20.07.16). Sonoran Institute, 2016. A Living River: Charting Wetland Conditions of the Lower Santa Cruz River: 2010 Water Year. 〈https://sonoraninstitute.org/resource/liv ing-river-lower-santa-cruz-2015-water-year/〉 (accessed 20.07.16). Stromberg, J.C., 1993. Frémont Cottonwood-Goodding Willow Riparian Forests: a review of their ecology, threats, and recovery potential. J. Ariz.—Nev. Acad. Sci. 27 (1), 97–110. Stromberg, J.C., Lite, S.J., Rychener, T.J., Levick, L.R., Dixon, M.D., Watts, J.M., 2006. Status of the Riparian ecosystem in the upper San Pedro River, Arizona: application of an assessment model. Environ. Monit. Assess. 115, 145–173. Train, K.E., 2009. Discrete Choice Methods with Simulation, 2nd edition. Cambridge University Press, Cambridge, UK. United States Army Corps of Engineers, May 2010. Tres Rios Del Norte – Pima County, AZ Ecosystem Restoration Feasibility Study (Review Draft: Subject to Edits). Van Houtven, G., Powers, J., Pattanayak, S.K., 2007. Valuing water quality improvements in the United States using meta-analysis: Is the glass half-full or half-empty for national policy analysis? Resour. Energy Econ. 29 (3), 206–228. Viscusi, W.K., Huber, J., Bell, J., 2008. The economic value of water quality. Environ. Resour. Econ. 41, 169–187. Vossler, C.A., Doyon, M., Rondeau, D., 2012. Truth in consequentiality: theory and field evidence on discrete choice experiments. Am. Econ. J.: Microecon. 4 (4), 145–171. Weber, M.A., 2010. EPA use of ecological nonmarket valuation. Assoc. Environ. Resour. Econ. Newsl. 30, 26–35. Weber, M.A., Ringold, P.L., 2015. River metrics for residents of an urbanized arid watershed. Landsc. Urban Plan. 133, 37–52. White, M.S., 2011. Effluent-dominated waterways in the southwestern United States: advancing water policy through ecological analysis. Arizona State University.