The Impacts of Harmful Algal Blooms and E. coli on Recreational Behavior in Lake Erie

1. Introduction

Harmful algal blooms (HABs) create freshwater toxins that are dangerous to humans and animals, raising public concern due to increasing occurrences both in the United States and globally (Morse et al. 2011; Duan et al. 2009; Bowling and Baker 1996). Apart from the direct adverse impacts on human health via drinking water provision, HABs also have the potential to negatively affect the well-being of recreational visitors and near-lake homeowners. For example, water quality degradation in lakes has been shown to significantly reduce the value of nearby properties (Liu et al. 2019; Walsh, Milon, and Scrogin 2011; Leggett and Bockstael 2000), with reductions in property values exceeding 30% for lake-adjacent homeowners when an algal bloom forms (Wolf and Klaiber 2017). Similarly, recreational visitors are known to actively avoid visiting lakes with poor water clarity (Wolf, Georgic, and Klaiber 2017; Keeler et al. 2015; Hanley, Bell, and Alvarez-Farizo 2003; Bockstael, Hanemann, and Kling 1987) and have stated concerns about areas that are persistently plagued by HABs (Zhang and Sohngen 2018).

The impact of water quality degradation is heterogeneous across different user groups including recreational visitors and households. In a survey-based choice experiment that identified five classes of freshwater anglers, only two of the five resource user groups were found to face significant damage from site closures caused by the spread of Didymosphenia geminata, a freshwater alga (Beville, Kerr, and Hughey 2012). Similarly, an empirical study in the Gulf of Finland showed that the willingness to pay (WTP) for a nutrient reduction policy that would decrease algae by up to 35% varied significantly among household groups. Some households were found to be indifferent toward water quality improvements, while others had a WTP of up to €457 per year (Kosenius 2010). Given the wide range of services provided by freshwater lakes and the preference heterogeneity across user groups, it is vital to examine trade-offs across amenities and water quality for water resource user groups.

Previous studies that examine the impact of water quality changes have focused on how pathogens, such as E. coli (Murray, Sohngen, and Pendleton 2001), beach closures (Palm-Forster, Lupi, and Chen 2016; Parsons et al. 2009), or perceived water quality (Keeler et al. 2015; Hynes, Hanley, and Scarpa 2008; Hanley, Bell, and Alvarez-Farizo 2003) influence recreation behavior. HABs have received less attention in the literature, however, likely due to the difficulty in observing varying concentrations of HABs. One notable exception is the recent analysis by Zhang and Sohngen (2018), which finds that Ohio anglers actively avoid HABs not only because they reduce angler’s chances of success, but also because of their impact on the overall fishing experience (reduction in water clarity, unpleasant odors produced from blooms, etc.).

In addition, surprisingly little attention has been given to exploring trade-offs across different sources of water pollution. Von Haefen (2003) includes measures of trophic state and dissolved oxygen to better understand recreational patterns in the Lower Susquehanna River basin; however, both of these measures of water quality are aggregated to a watershed or river reach level. Egan et al. (2009) combine 13 measures of water quality from 129 lakes in Iowa with recreational trip information to determine how responsive recreationalists are to each measure of water quality. They find that water clarity and nutrient levels are the two most important indicators of water quality for recreationalists. However, water-borne pathogens, such E. coli or fecal coliform bacteria, were not included within their analysis, and latent preference heterogeneity is largely left unexplored. In this paper, we extend the literature by including multiple sources of water pollution—E. coli and HABs—into a model of recreational demand, while simultaneously identifying multiple recreator groups. To our knowledge only one other paper has done both (Kosenius 2010), using data collected from a contingent valuation survey rather than the revealed preference approach we employ. By identifying various subsets of the recreational visitor populations and their WTP to avoid multiple sources of water pollution, we can better understand the distributional impacts of water quality changes stemming from potential policy scenarios.

Discrete choice modeling methods, combined with either survey responses of recreation behavior or site intercept data, are commonly used to estimate the impact of environmental change on recreation demand, and to reveal the preferences of recreational visitors. For this paper, we conducted a webbased survey of randomly selected households across 18 Ohio counties near the Lake Erie shoreline and collected information on trips to Lake Erie during the summer of 2016.¹ We focus on the western basin of Lake Erie as it has suffered from repeated exposure to HABs for nearly 20 years.² From this survey we obtained responses from 749 individuals³ with 549 of those individuals taking at least one day trip to one of 106 access locations along the Lake Erie shoreline. By combining these survey responses on recreation behavior with detailed geospatial data on HABs gathered from the National Oceanic Atmospheric Administration (NOAA), and site-specific characteristics provided by the Ohio Department of Natural Resources (ODNR), we created a novel dataset that allows us to estimate the welfare costs of blue-green algal blooms and E. coli contamination on recreational beach users and anglers along the Lake Erie shoreline in Ohio.

We make three contributions to the literature: (1) we use information on recreational behavior to provide new valuation estimates of the negative impact of HABs, (2) we show that heterogeneity in individual preferences for lakeshore amenities and disamenities–WTP to avoid HABs and E. coli–will lead to strong distributional effects from remediation policies, and (3) we predict the value of welfare losses from algal blooms and benefits from either algae or E. coli control under five hypothetical policy scenarios.

2. Survey Design and Data

The sample frame consists of 20,000 residential mailing addresses for single-family homes located within 50 miles of the Lake Erie shoreline, randomly drawn from 18 counties across the state of Ohio (Figure 1). Addresses were collected from county tax auditors’ databases. For each address selected, two postcards were sent out: one in the first week of February 2017 and a second in the last week of February of 2017. Each postcard included a brief description of the study along with a website address, directing the recipients to an online survey. A unique identification number was also included on each postcard and was subsequently entered into the online survey by respondents, allowing us to link survey responses with residential location information from the auditor data.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Sampled Counties and Mailing Addresses

The online survey contained three sections. The first section asked recipients questions about their typical day trip to Lake Erie between Memorial Day (May 30, 2016) and Labor Day (September 5, 2016), total trips taken to Lake Erie in the summer of 2016, travel mode to Lake Erie, ranking of site characteristics that were considered most important when deciding where to go, and if the respondent was aware of algae in Lake Erie or of any water quality advisories. In the second section respondents were guided through an interactive map of 185 access points monitored and managed by the ODNR, Office of Coastal Management to determine time and location of the most recent day trip to Lake Erie. Survey respondents then reported expenditures on gas, food, clothing, parking, and so forth, the primary purpose of their trip (boating, fishing, swimming, etc.), and perceived congestion at the chosen site. The final section contained questions on demographic characteristics (income, race, education, etc.). After removing respondents who took more than three hours to finish the survey, the final sample contains 749 responses. Of these respondents, 549 took at least one day trip to one of 106 different locations along the Lake Erie shoreline.

To understand and model motivations for recreation trips, we asked individuals to rate the importance of 11 site characteristics using a five-point Likert scale (Appendix A Figure A1). Survey respondents indicated greater concern for the quality/availability of natural amenities than man-made amenities. The overall beauty, health, clarity, and odor of the water were important, with 90% of respondents indicating these attributes were either very important or extremely important in deciding where to recreate along Lake Erie. Available parking, boating opportunities, and convenient facilities, on the other hand, were less important. Given the importance placed on water quality, we expect that differences in algal concentrations across sites likely influenced recreational location decisions.

Table 1 shows the demographics of our survey respondents compared to those of the general population living within the 18 counties either bordering or located near Lake Erie.⁴ We note a number of differences between our sample of recreational visitors and the general population. The surveyed sample tends to be more educated and wealthier, have a larger household, and live in more rural areas than the general population. Similar or even larger differences are also observed when comparing the general population with samples of Ohio recreational anglers collected by the ODNR (N = 425) (Zajac et al. 2011) and the U.S. Fish and Wildlife Service (N = 1,435) (USFWS-USCB 2011) between 2007 and 2011.

The ODNR’s targeted population were resident and nonresident anglers who purchased a fishing license from the state anytime between 2007 and 2010. The ODNR demographics shown in Table 1 are from a subsample of this survey who indicated Lake Erie as their preferred recreation destination. The U.S. Fish and Wildlife Service sample, on the other hand, comes from a nationwide census of 48,600 households—1,435 of which were identified as Ohio anglers. The U.S. Fish and Wildlife Service did not make a distinction between Lake Erie and non–Lake Erie anglers and therefore provides a representation of all the anglers statewide. When comparing these two external samples with the general population in Ohio, we find that the average ODNR respondent is more likely to be male, have higher educational attainment status, and be employed. Likewise, the U.S. Fish and Wildlife Service sample is disproportionately white, male, poor, and rural and has a lower level of education than the overall population. While not representative of the general public in Ohio, our surveyed sample provides a representative snapshot of Lake Erie visitors, compared to the ODNR and U.S. Fish and Wildlife Service samples, which focus solely on a subset of recreational anglers.

We obtained water quality data from NOAA (NOAA 2018) and Stumpf et al. (2012) and merged this information with site-specific characteristics provided by the ODNR. NOAA publishes 10-day HAB composites for Lake Erie, measured in Microcystis cells (10,000s per milliliter), between the months of June and October. Given the lack of blooms during June of 2016 however, NOAA only published composites between July 1 and October 30 for 2016. Using this dataset, we create a site-specific, summer-long mean algae measure using readings collected from the three closest remote sensing locations. Similarly, maximum Ohio Department of Health E. coli readings from the closest monitoring station⁵ were attached to each access point. A total of 274 algae and 60 E. coli monitoring locations were used for this aggregation strategy. Finally, we include proximity to other non–Lake Erie amenities as measured by the distance between each access point and Cleveland’s and Toledo’s central business district.

Table 2 reports summary statistics for site-specific characteristics for the 185 access points reported by the ODNR and the subset of 106 that were visited by our sample. On average, the visited access points have a beach, accessible parking nearby, restrooms, a picnic area, and walkable trails. Sites with algae conditions above 100,000 cells/mL are above the World Health Organization’s high advisory threshold,⁶ while the Environmental Protection Agency issues E. coli health advisories when conditions surpass 410 CFU/100 mL (EPA 2012). Both algae and E. coli levels varied substantially across sites during the study period in 2016.

3. Demand for Recreation

Following a standard discrete-choice random utility maximization framework (Haab and McConnell 2002; McFadden 1974), on each choice occasion, each recreator chooses where to access Lake Erie from among 106 locations. The utility of a trip to a chosen site is a function of the cost of traveling to the site, and other site-specific attributes including environmental quality.

Following Dundas, von Haefen, and Mansfield (2018), we use income data and a spatial matrix of distances between each site and mailing address to calculate the travel cost to each site for each respondent: where Travel Cost_ij measures the cost of traveling to site j for individual i. The first component of Travel Cost_ij represents the opportunity cost of travel time and is defined as the distance traveled at 55 mph (dist_ij /55 mph) multiplied by one-third of the wage rate (wages_i /3) (Haab and McConnell 2002; Cesario 1976), where wages_i represents individual i’s hourly wage rate calculated by dividing annual income by 2,000 hours,⁷ and dist_ij is the one-way distance in miles from respondent i’s residence to access point j. The second component of Travel Cost_ij approximates the out-of-pocket cost of driving (dist_ij·0.476). According to the American Automobile Association (2016), the average cost per mile of operating a vehicle in 2016 was $0.476. This includes the cost of gas, maintenance, and vehicle depreciation. One-way travel costs are doubled to give a measure of round-trip travel costs.

Finally, respondents were given a choice between seven different activities at the chosen recreation site, some of which included direct interaction with the water (boating, swimming, fishing, etc.), while other activities were less water focused (wildlife watching, walking/running, etc.). Participants were also given the chance to write in their own activity if none of seven provided categories accurately described their recreational activity. Summary statistics for the travel cost estimate and the visitor’s primary purpose of recreating are given in Table 3.

We estimate three specifications of the discrete choice model for recreational site choice. In all specifications, we assume recreators determine their optimal location by selecting the site that provides the highest level of utility. The utility associated with site j for household i is given by [1] where V_ij is the observed component of utility for household i choosing to recreate at site j, and ε_ij is an idiosyncratic component of utility unobserved to the researcher. The representative utility, V_ij, captures the effects of observed attributes that vary among individuals, sites, or both. The idiosyncratic error term is assumed to be distributed i.i.d. type I extreme value giving rise to the well-known logit family of models with probability of individual i choosing site l given by [2] Modeling variation and preference heterogeneity is introduced through treatment of covariates in the specification of the indirect utility function and its parameterization. Consider the standard specification of representative utility: [4] where Z_ij denotes the individual and site-varying travel cost for person i visiting site j, X′_j is a vector of site-specific attributes including water quality measures, γ is a scalar representing the marginal utility of income, and β is a vector of taste parameters to be estimated.

First, we estimate a conditional logit model using a standard fixed parameter logit maximum likelihood estimation routine. We introduce individual preference heterogeneity in the model by following a classical approach (Morey and Greer Rossmann 2003; Swallow et al. 1994) and interacting site-specific attributes with observable characteristics. In the second specification, we change the manner in which preference heterogeneity is introduced by assuming that the population is categorized into S unobservable (latent) classes (s = 1,…, S). For each group a unique set of parameters is estimated that allows consumers’ tastes for site attributes to vary across segments of the population. With preference heterogeneity, equation [2] becomes [4] where the probability of individual i visiting site l, conditional on being in segment s, is equal to P_il|s. As a part of the latent class model, a membership function must also be specified that assigns individuals into a specific segment of the population. Following Swait (1994) and Boxall and Adamowicz (2002), we specify this membership function using the following set of equations: [5] [6] where M_is is the membership likelihood function for individual i and segment s; D_i is a vector of observed sociodemographic characteristics of recreator i; is a vector of latent psychometric constructs held by recreator i; Z_i is a vector of observed indicators of latent constructs held by respondent i; ψ and θ are parameter vectors to be estimated; and μ and ξ represent error terms. Assuming ξ is an i.i.d. error term with a type I extreme value distribution, the probability of membership into group s for individual i can then be characterized by the following:⁸ [7] The product of equations [4] and [7] then reveals individual i’s unconditional probability of visiting site l within a latent class framework: [8] Finally, we estimate a random parameters specification, which allows even greater flexibility in preference heterogeneity compared to the latent class model. Specifically, preference parameters are allowed to vary across individuals rather than segments. This additional flexibility can be observed in equation [9]: [9] where β now varies across individuals (i) rather than just segments (s). One restriction with this approach, however, is the need to specify how the random parameters are distributed. We assume the coefficients on site-varying attributes are normally distributed. Estimation of the mixed logit model proceeds by simulating the choice probabilities given as [10] where f(β|μ, σ) is the probability density function of β. Using simulated maximum likelihood, we recover estimates of the parameters γ, μ, and σ. With estimates in hand, we can use the familiar log-sum rule to estimate WTP for nonmarginal changes in covariates and the associated changes in utility (Small and Rosen 1981).

The three model specifications discussed above differ in the analysis of preference heterogeneity. The first specification of a conditional logit model in equation [3] allows consumer preferences to be heterogeneous only through interaction terms between site-specific attributes and observed demographics. The latent class and mixed logit specification, on the other hand, introduces additional heterogeneity by allowing preferences to vary at either the group or individual level. It is important to allow for this additional heterogeneity in our study because there are different reasons why one might want to visit Lake Erie (boating, fishing, swimming, etc.), and depending on the purpose of the trip, the value of each observable site attribute is expected to differ across individuals or groups. In the case of the mixed logit, the distribution of the taste parameters must be specified in order to derive choice probability estimates. Latent class models require the specification of a membership function to not only account for heterogeneity in consumer tastes, but also provide an explanation for the resulting heterogeneity (Boxall and Adamowicz 2002).

4. Results

Table 4 reports estimates from the conditional logit, mixed logit, and latent class model of recreation choice. Examining results from the conditional logit model first, we find the travel cost estimate to be, as expected, negative and significant with a value of –0.049. Site-specific attributes also have the expected signs with positive and significant coefficients associated with picnic shelters, food services, showers, restrooms, and beach access. Controlling for population centers, we find that households prefer recreation sites that are farther away from the densely developed Cleveland downtown, whereas proximity to the less developed waterfront near Toledo was viewed positively.

Turning attention to the key water quality measures of interest—algae and E. coli—we find the expected negative and significant coefficients. The negative and statistically significant coefficient (–0.0121) on algae confirms our hypothesis that HABs influence visitor decisions when they select recreation sites along Lake Erie. Similarly, the concentration of waterborne E. coli has a negative and significant (–0.0620) impact on the decision-maker’s site choice, consistent with the previous literature (Awondo, Egan, and Dwyer 2011; Murray, Sohngen, and Pendleton 2001). Finally, following Timmins and Murdock (2007), we interact our two measures of water quality with an indicator for boat ownership and an indicator for whether the survey respondent has children.⁹ We find that recreators with children (Kid (0/1) = 1) avoid E. coli contaminated water to a greater degree than the average recreator without children.

We allow for increased heterogeneity in preferences for site attributes in our mixed logit model, which is shown columns (2) and (3) of Table 4. Consistent with the discrete choice literature, we estimate a fixed preference parameter associated with travel cost to aid in welfare calculations. Across all covariates we find that results are similar to those of the conditional logit model, with three notable exceptions. First, the mean coefficient on restrooms and its standard deviation parameter is statistically significant. This suggests that while, on average, recreators prefer access points with restrooms, there is significant heterogeneity across individuals. This is consistent with prior literature showing that particular segments of the population, generally families with children, prefer recreation sites with restrooms (Parsons et al. 2009; Timmins and Murdock 2007). Second, algae continues to have a negative and significant coefficient for the mean parameter. However, its standard deviation term is also statistically significant, indicating heterogeneity in preferences to avoid HABs within our sample of Lake Erie recreators. Third, the interaction between Kid (0/1) and E. coli is negative but no longer significant at the 10% level.

To further examine this underlying heterogeneity, we estimate a latent class model. A membership function, which categorizes different groups of Lake Erie visitors, is required for estimation. Following Boxall and Adamowicz (2002), we first create a set of latent motivational constructs using responses from the 11 psychometric questions displayed in Appendix A Figure A1 and factor analysis. Factor analysis allows us to distill the information gathered from these responses into a smaller set of determinants that can explain the underlying motivation for recreational site choice decisions on Lake Erie.

We recover four principal components from this analysis¹⁰ that summarize and account for the majority of variation present within these responses (Appendix A Table A1). We label the first component as “Amenity Preferences,” given the positive relationship between all the statements relating to site attributes and this factor; individuals in this segment tended to be more sensitive to differences in site-varying amenities than other lake visitors. The second component is called “Recreational Fishing Quality,” as the three primary determinants of this factor were the presence of a boat ramp, fishing opportunities, and the health of aquatic life. The third component, “Shoreline Amenities,” differentiates recreators who were more interested in land-based amenities than water-related characteristics. Finally, our fourth factor, “Aesthetic Amenities,” is characterized by people seeking nearby natural areas that are peaceful and isolated from crowded areas.

Factor scores were calculated for each individual and included, along with a vector of observable sociodemographic information,¹¹ in the membership function (equation [5]). In order to determine the optimal number of population segments included within the latent class model, we estimated latent class models with one to five segments. We selected the model with the lowest Bayesian information criterion score (3,856.0344), which was the model with two segments, and present the results from its membership function in Appendix A Table A2. Notice the coefficients for the first group, “Beachgoers,” have been normalized to 0. Consequently, the coefficients from the other segment, “Anglers,” are described in relation to the beachgoer segment. As is evident from Appendix A Table A2, recreational anglers are more likely to own a boating license, have a higher household income, and have a lower education level than the average beachgoer. Anglers are significantly more concerned about the availability and quality of fishing at access points than beachgoers, which is depicted by the positive and statistically significant coefficient on factor 2. The coefficients on the other principal components are not significant, indicating that the other three factors are not essential in differentiating beachgoers from the fishing group.

Two sets of utility parameters, one for each segment, are also estimated within the latent class model. We present these results in the final two columns of Table 4.¹² The sign on both E. coli and algae continue to have the expected negative sign for both beachgoers and fishing enthusiasts. However, there is significant heterogeneity in willingness to avoid water pollution not only across groups but also across the type of water pollution present. In particular beachgoers tend to avoid areas where there are high concentrations of E. coli but are less concerned by the presence of HABs. Recreational anglers, however, tend to avoid algal-infested waters but are indifferent toward E. coli. This divergence in water quality preferences is likely due to the manner in which each population segment benefits from recreational amenities in freshwater lakes. Anglers may be more concerned about visibility issues and reductions in game fish (Paerl and Otten 2013; Rensel, Haigh, and Tynan 2010; Richlen et al. 2010) that arise from HABs¹³ but are less concerned about the health side effects associated with E. coli as they are not often directly interacting with the water. Beachgoers, on the other hand, are more likely to be wading and swimming in the water and therefore are warier of E. coli, a well-known toxicogenic. The results from Table 4 support this hypothesis, with anglers avoiding sites with heightened levels of HABs while beachgoers avoid access points with high concentrations of E. coli. Despite there being no statistical relationship between site choice and HABs for the latent beachgoers group (t-value of 0.70), this coefficient is likely underestimated, as we are unable to observe the averting behavior of beachgoers who forego swimming and instead do something else at the beach (take a walk, sunbathe, etc.) due to HABs. In fact, only 13% of the latent beachgoer group indicated the primary purpose of their trip was to go swimming (Table 3), while 55% said they went to the beach to either walk, run, sunbathe, or participate in an organized activity. It is possible that many of our recreators initially went to the beach to swim but instead decided not to due to presence of HABs.

Across the two groups of visitors, there is significant heterogeneity in preferences for nonwater amenities as well. Beachgoers were more likely to visit access points that have a beach, picnic shelters, and food stalls. Meanwhile anglers tended to avoid areas with picnic shelters but sought out access points with boat ramps and restrooms. Similar to the parameters recovered from the conditional and mixed logit models, both groups also preferred to recreate away from densely developed areas (i.e., Cleveland). Finally, there is significant heterogeneity in the travel cost parameter, with beachgoers being more averse to long distance locations than anglers. This difference is also evident from the summary characteristics of these two latent classes (Table 3). Specifically, the average angler incurred an additional $21 in travel cost to reach his or her preferred location compared to the average beachgoer. This difference is likely driven by prominence of the walleye fishery in Lake Erie and the premium that anglers place (i.e., by driving further/spending more time to get to their recreation location) on locations with high walleye catch rates (Melstrom et al. 2015).

5. Welfare

We next examine the welfare implications of changes in water conditions arising from our estimates of the conditional logit, mixed logit, and latent class models in Table 5. We report marginal WTP measures (MWTP), which are calculated by dividing the estimated coefficients by the travel cost parameter, and use these estimates to recover welfare measures.¹⁴ According to the estimates from the conditional logit and mixed logit models, the welfare loss associated with an increase in algae is –$0.25 per 10,000 cells/mL. In other words, an increase of algae by 10,000 cells/mL will lead to a welfare loss of 25 cents per visitor per trip. Considering the mean algae level at the recreational sites in our sample is 127,900 cells/mL, this reflects a cost of $3.20 per visitor per trip associated with the presence of the mean level of algae at a recreation site relative to a site with no algae. A similar pattern emerges when we estimate the impact of E. coli on recreators’ well-being. A one-unit increase in E. coli (1,000 CFU/100 mL) reduces consumer welfare between $1.27 and $0.68, with the higher loss estimate derived from the conditional logit model. Evaluated at the average observed E. coli level, this represents a total cost of between $1.22 and $2.29 per visitor per trip.

In our preferred latent class model specification, we find significant heterogeneity when we estimate separate water quality parameters for recreational anglers and beachgoers. Anglers are the most affected by algae-contaminated water, losing 52 cents per 10,000 cells/mL, while beachgoers lose only 5 cents per 10,000 cells/mL. E. coli, on the other hand, is an important consideration for beachgoers, costing $1.29 per 1,000 CFU/100 mL. Recreational anglers also appear to be sensitive to increasing concentrations of E. coli with a MWTP of –$2.61 per 1,000 CFU/100 mL; however, this estimate is not statistically significant at the 10% level. Finally, we assess the sensitivity of our latent class model by using a lower vehicle cost per mile estimate—$0.25—which was used by Awondo, Egan, and Dwyer (2011) to recover estimates of Lake Erie beachgoers’ MWTP for wetland construction near Maumee Bay State Park. The MWTP estimates from this new vehicle per mile estimate (Appendix A Table A4) are lower than our original values and likely represent a scenario where recreators do not consider the depreciation cost of traveling. We continue to use the higher vehicle cost per mile estimate ($0.476), as many of our respondents were likely driving their boat to their preferred access point (24% of the sample indicated they owned a boat) and because of similar, or even higher, estimates used by Melstrom and Lupi (2013) and Zhang and Sohngen (2018).

In addition to estimating measures of MWTP, it is also possible to use the structure of the discrete choice utility framework and the estimated coefficients to calculate consumer surplus measures associated with nonmarginal changes in lake quality. As climate change and increased urbanization will likely exacerbate water quality conditions over time, we develop several scenarios that focus on the potential welfare losses associated with worsening water quality. We also conduct two policy counterfactuals that evaluate the welfare gains from a reduction in HABs and a reduction in E. coli. The first three scenarios involve changes to water quality for beach locations along the western shoreline of Lake Erie in Ohio,¹⁵ which is an area that suffers from repeated exposure to HABs. The fourth scenario evaluates the implications associated with a lakewide reduction in phosphorus as suggested by the 2012 Great Lakes Water Quality Agreement (GLWQA), while the fifth scenario reduces E. coli levels to just below the U.S. Environmental Protection Agency’s (EPA) advisory threshold for water recreational users for sites where this threshold is currently met or exceeded.

The five policies we evaluate include (1) blue-green algae causing severe blooms across the 31 access points in the western basin that were visited by our sample, with algae levels reaching the sample maximum (994,100 cells/ mL) at each site; (2) closure of the 31 western sites due to algal blooms; (3) a prevention scenario in which HABs are eliminated in the western basin; (4) a targeted 40% reduction in lakewide phosphorus loadings; and (5) a reduction in E. coli levels to 410 CFU/100 mL for the 79 sites that currently meet or exceed the EPA’s recommended advisory threshold for water recreators. These hypothetical scenarios provide insight into the potential welfare changes that would accrue to a typical day-visitor to Lake Erie.

Welfare losses/gains associated with these simulations are presented in Table 6 for the conditional logit, mixed logit, and latent class models. Based on the conditional logit estimates, the average per-trip cost to a visitor would be $3.44 when there are massive algae blooms in Lake Erie’s western basin. The cost is a little lower in the mixed logit model, with visitors losing $2.60 per trip. The latent class specification reveals that fishing enthusiasts would be the most impacted by increases in algae, losing $11.48 per trip, while beachgoers’ welfare would decrease by a modest $0.46 per trip. If these western sites were closed due to algal blooms, the closures would cost $2.32 per trip for beachgoers and $23.51 per trip for recreational anglers. In contrast, if HABs were completely eliminated from Lake Erie’s western basin, the benefit would range between $0.43 and $4.79 per visitor per trip. We note that these estimates are likely lower bounds as they do not account for averting behaviors of beachgoers who avoid swimming or any direct contact with the water but choose a different recreational activity at the beach due to HABs. Water quality improvements through E. coli reductions would also improve recreator well-being, with Lake Erie visitors willing to pay between $2.00 and $3.43 per trip for a reduction in E. coli to levels just below the EPA’s advisory threshold.¹⁶

Aggregating these per-trip losses/gains across all Lake Erie trip-goers results in meaningful welfare measures for Ohio recreators. Following prior work (Murray, Sohngen, and Pendleton 2001; Palm-Forster, Lupi, and Chen 2016), we estimate the annual number of trips to Lake Erie beaches in Ohio (6.25 million) by merging results from our survey with aggregate visitation data collected by the ODNR.¹⁷ Combining our valuation measures with this total day-trip estimate, and predicted membership shares from our latent class model,¹⁸ then allows us to recover group-specific welfare measures associated with nonmarginal improvements or degradations in water quality. We find from this analysis that recreational anglers and beachgoers would lose $69.1 million and $7.7 million, respectively, each year if all the access points within Lake Erie’s western basin were closed due to unsafe algae levels. Assuming there are approximately 90 days during the summer season, our aggregate losses closely resemble those recovered by Palm-Forster, Lupi, and Chen (2016) when evaluating a 33-site closure along the western basin of Lake Erie. Specifically, they find aggregate daily loses to be between $573,730 and $744,320, which would be equivalent to $51,635,700 and $66,988,800 for a 90-day summer season.

Finally, we examine the gains attributed to a 40% reduction in lakewide phosphorus loadings as suggested by the GLWQA. Using Stumpf et al.’s (2012) algae forecasting model¹⁹ we find this policy would reduce lakewide algae by approximately 32%, with western basin access points benefiting the most.²⁰ On average, beach visitors would gain $0.17 per trip taken; however, anglers would benefit the most, gaining $1.99 per trip. Aggregating these values to an annual basis would result in gains of $5.8 million and $563,000 for recreational anglers and beachgoers, respectively. Zhang and Sohngen (2018) estimate similar gains for recreational anglers when upstream phosphorus levels are reduced by 40%, with aggregate consumer surplus increasing $4 to $6 million per year.²¹

6. Conclusions

Freshwater lakes provide important ecosystem services and recreational amenities that impact a variety of resource users. Poor water quality due to the emergence and growth of HABs has been a growing concern in the Lake Erie region over the past two decades (Smith, King, and Williams 2015; Michalak et al. 2013; Rinta-Kanto et al. 2005). A number of costly management solutions have been proposed in response to this environmental concern (Scavia, DePinto, and Bertani 2016; Sohngen et al. 2015), but to evaluate potential policies we need reliable estimates of the benefits from water quality improvements along different margins of services. In this paper, we combine survey data with detailed geospatial information on water quality indicators—including algae and E. coli concentration—to estimate Lake Erie visitors’ recreational amenity benefits from water quality improvements. Using a random utility model we find Lake Erie recreational anglers and beachgoers would lose $69.1 million and $7.7 million, respectively, each year if Lake Erie’s western basin were closed due to severe algal blooms. The gains attributed to a 40% reduction in lakewide phosphorus, while substantial, have a smaller impact, with beachgoers and anglers gaining $563,000 and $5.8 million, respectively, each year. While these benefits are considerable, they represent only a fraction of the true gains associated with a reduction in HABs. Improvements in water quality are also likely to be capitalized inhousing markets and recreational fishing patterns (Wolf and Klaiber 2017; Wolf, Georgic, and Klaiber 2017) and improve the well-being of existing recreators who go to Lake Erie but avoid recreating in the water. We further find heterogeneity in individual preferences toward lakeshore amenities and disamenities including HABs. Lakeshore visitors’ WTP to avoid HABs is significantly heterogeneous across our sample, with anglers greatly benefiting from a reduction in algae concentration, while the welfare effects for beachgoers depend largely on reduction in E. coli.

Our analysis provides insights that are relevant for both local and regional policy decisions. Freshwater lakes provide a wide range of ecosystem services and recreational amenities that make regulations to address water quality a multijurisdictional concern. Whereas the provision of clean drinking water is a regional concern, amenities for beach visitors are local public goods. Understanding the heterogeneity in benefits from water quality changes across different user groups can better inform policy makers about the distributional implications of potential policies. Because water quality changes have implications for different uses and services at different scales, policy response necessarily requires a cross-scale approach. As policy makers consider a suite of potential solutions to address water quality concerns in the Great Lakes region in the United States, this paper provides estimates of the potential welfare implications of reductions in E. coli and HABs that can be used to determine the value of services provided and evaluate trade-offs under different policy scenarios.