What Can Cell Phone Data Tell Us about Outdoor Recreation?, with Daniel Phaneuf

Notable Quotes

• Environmental goods lack a market price: “There’s lots of goods and services that can generate happiness and well-being for us as human beings, and they’re not limited to just things that we buy in the market. We can think about nonmarket environmental goods like water quality generating well-being or happiness in the same way that we can imagine a bottle of wine or a six-pack of beer. The main difference, of course, is that we go to the store to buy a bottle of wine. We don’t go to the store to buy water quality in the lake we want to swim in.” (5:17)
• Estimating the value of recreational spaces: “We use statistical methods to compare the number of trips that people have taken and the places that they’ve gone to their travel cost, which is a little bit like putting a price on environmental conditions.” (10:12)
• Advantages of using cell phone data: “One of the really exciting things about these mobility data, though … is that some environmental problems, like harmful algal blooms or beach closures, are on-a-day types of things. With these data, we may ultimately be able to understand how people respond to, say, a warning about a harmful algal bloom or an announcement that a beach has closed.” (18:32)

Top of the Stack

• Arthur Brooks in the Atlantic

The Full Transcript

Margaret Walls: Hello, and welcome to Resources Radio, a weekly podcast from Resources for the Future. I'm your host, Margaret Walls. Today, we are recording this episode from the Annual Conference of the Association of Environmental and Resource Economists, or AERE, as it's called. AERE is the main professional membership organization for people like me and my guests today: environmental and resource economists in the United States. Every year, AERE holds its flagship conference around this time, and we're lucky enough today to have it in Washington, DC, in our backyard here at Resources for the Future (RFF).

I want to take just a minute, because there's some strong connections between AERE and RFF. For our Resources Radio audience, I think it's good to tell folks a little bit about that.

AERE was founded in the mid-1970s when environmental economics was a brand-new field, and some folks that were doing that kind of work said, "We need to have our own association." A lot of those people were either affiliated with Resources for the Future or at Resources for the Future. People like John Krutilla; Allen Kneese; Emery Castle, who was vice president of RFF at the time and then became president; and Kerry Smith were all affiliated with RFF in one way or another.

I've also learned that RFF gave a grant to get AERE started and help to do the work necessary to establish the organization as a 501(c)(3) nonprofit. There were a lot of connections in the early days and over the years. Past AERE presidents include RFF’s own John Krutilla—who was the first one, I think—and Senior Fellows Alan Krupnick and Karen Palmer. This year, AERE reached out to us at Resources Radio and said, “Would you all like to record at the conference?” So, here we are today. We eagerly agreed to that.

I'm really excited today to have my guest as Daniel Phaneuf. Dan is the Henry C. Taylor Professor of Agricultural and Applied Economics at the University of Wisconsin–Madison. Dan has published on a wide range of topics, but I'd say his work mainly focuses on nonmarket valuation methods—ways to attach values and estimate values of the environment when things aren't traded in markets. He's done a lot of work, for example, on water-quality issues. Dan has a very long bio. I'm not going to get into all of that. It's really impressive. Because we're at AERE, I want to mention that Dan himself was president of AERE in 2019 and 2020, when I was on the board, too. He was the inaugural editor of the Journal of the Association of Environmental and Resource Economists (JAERE), and, also in 2022, he was named as an AERE Fellow.

An AERE Fellow is the highest honor that the organization has for its members, so I'm really thrilled to have Dan here today. What we're going to talk to Dan about is one of the organized sessions here at the conference that he was in, which is entitled “Frontier Economic Applications Using Mobility Data.” Dan had a paper with some coauthors in that session. I'm going to ask him some questions about that research.

As background, we're going to start first by talking about nonmarket valuation methods more broadly and then work into talking about this new data, what these data are providing, the challenges associated with using them, and how they might improve the methods we use or not. Stay with us.

Hello, Dan. Welcome to Resources Radio. It's really great to have you on the show.

Daniel Phaneuf: Yeah, thanks. It's really great to be here.

Margaret Walls: It's fun to be in person doing this.

Daniel Phaneuf: It's nice.

Margaret Walls: We always start the podcast by learning a little bit about our guests. I want to ask you a little bit about your background. What inspired you to become an environmental economist and focus on the topics that you focus on?

Daniel Phaneuf: For me, it was pretty personal. I grew up in Minnesota. My family does a lot of hunting and fishing and a lot of outdoor activities, so that was always an important part of my life. When I was deciding to go to graduate school, I went to Iowa State, in part because I had a good friend in Des Moines who had a great hunting dog, and I thought, “If I go there, I can do a lot of pheasant hunting.” Much to my surprise, in my second year, my advisor, Cathy Kling, shows up and tells me that, in fact, you can study outdoor recreation as part of your dissertation. That was how it all started, and it seemed like a great fit.

Margaret Walls: That's awesome. I love personal stories like that. We're going to get into recreation-demand modeling, but let's start with an even bigger picture than that. I mentioned at the beginning—nonmarket valuation. Can you give our listeners a brief primer on what we mean by that and why it's important? What are some of the challenges? I think of that kind of work as fundamental to environmental economists and what we do, but give our audience a sense of what we mean by that.

Daniel Phaneuf: Sure. I think the starting point is to recognize that there's lots of goods and services that can generate happiness and well-being for us as human beings, and they're not limited to just things that we buy in the market. We can think about nonmarket environmental goods like water quality generating well-being or happiness in the same way that we can imagine a bottle of wine or a six-pack of beer. The main difference, of course, is that we go to the store to buy a bottle of wine. We don't go to the store to buy water quality in the lake we want to swim in, so goods with prices tell us something about the economic value.

When I buy that bottle of wine, I'm basically telling the world that I was willing to pay at least the price, $15, for that wine. I never show you through my behavior how much I'm willing to pay or how much I value water quality, so that's the difference.

The idea is important, because a lot of decisions in society involve trading off environmental preservation: keeping things in good environmental shape versus using them for other purposes. To understand how we should wisely make these trade-offs, we need a common metric. We need to be able to say that environmental quality has this value that can be compared to something in the market.

Margaret Walls: That is super important for all the policies we set in this country around environmental policy. You mentioned doing recreation-demand modeling. I want to ask you about that. It's often used as a method for getting a handle on these benefits and eliciting values for, as you say, water quality or things like that. Tell us about recreation-demand modeling and how it does that.

Daniel Phaneuf: I think the best way to start understanding that is to back up a little bit to the science of this nonmarket valuation thing. We usually define two ways to think about measuring these values that people hold. One is known as stated preference, and this relies on surveys where we would give a person a survey and put them into a situation where they have to respond to some environmental context.

A good example is a simulated referendum. I might ask somebody in a survey, “Suppose we are going to improve lake water quality by 10 percent, but it's going to increase your utility bill by $8.” They can vote yes or no, and by observing that simulated vote, they tell us something about their value. The recreation idea that you mentioned is a type of revealed-preference analysis where we use people's interactions with environmental resources to learn something about their values.

Recreation is really interesting, because it's one of the main ways that people interact with the environment. Recreation analysis basically uses this idea that people are willing to travel to a recreation site even though travel is costly. They're showing us through that behavior that, in fact, they value access to that recreation site at least as much as it costs for them to get there.

The additional part is that, if different recreation sites have different environmental conditions, so if different lakes have different water-quality conditions, and we see that people are willing to travel further to get to a lake that has better water quality, they've told us something. They've told us that they're willing to spend more in travel to get to the place that has better water quality, and that can allow us to get a sense of how important water quality is to people; how much they're willing to give up to have it in terms of traveling further and paying more.

Margaret Walls: That's a great explanation, Dan. What are the challenges, then? We have to have that data on people visiting those different kinds of lakes. What are the data challenges, methods challenges, and so forth in doing this empirically?

Daniel Phaneuf: Great question. Empirically, the way we think about this is that we usually use surveys to assemble data from a sample of people that we would ask. They would tell us about the trips they've made to different recreation sites, presumably in their area. We would ask them where they live so that we can compute how far from their home they would have to travel to these various recreation sites. Then, we would assemble information on the different levels of environmental quality at the recreation sites. So again, in the case of lakes, we would assemble information on water quality in the lakes.

The way these models work is we use statistical methods to compare the number of trips that people have taken and the places that they've gone to their travel cost, which is a little bit like putting a price on environmental conditions, which is this thing that we're interested in. We can then effectively estimate what economists would refer to as demand curves for these recreation trips as a function of travel cost and environmental quality. That's the basic idea of it.

You've probably gathered from my explanation that it's super data intense, right? We have to field a survey. We have to find a bunch of people to ask about their behavior. Surveys are expensive. They're difficult to often get people to respond to them. So, the method is, maybe we should say, expensive to implement in terms of the data that we need to assemble.

Margaret Walls: Yeah, good point. That brings us to the topic we really wanted to focus on, which is the new types of data that are becoming available. In the AERE session at the conference that I mentioned, all the papers are using so-called “mobility data,” or cell phone data. Tell us about those data.

Daniel Phaneuf: The mobility data is a really exciting new development, and it starts with something I think we're all familiar with. We all carry around our devices, and these devices are interacting with the world as we carry them. What they're doing is recording pings that are recording a place where the device is at a certain time. By assembling data on these pings, we can trace out how people have moved in the landscape, so to speak. That's why these are called mobility data.

We can see places that individual devices, presumably attached to people, have traveled; say, from their home to recreation sites. Then, through a collection of these pings, we can see how long they stayed at a particular place. Marketing folks are really excited about this data, because they can get a sense of traffic-to-retail operations. They can see sales or special offers drawing more people into stores. For environmental economists, it can be very useful, because we can see how people are interacting with environmental resources; for example, in the form of recreation behavior.

Margaret Walls: Let me turn to your paper in particular—and you have several coauthors, so I just want to be sure to name them. One is your PhD student, Dimitris Friesen, and the others are Yoojin Cha; Daniel Douglas; Yusuke Kuwayama, a fellow at RFF; and Sheila Olmstead, an RFF university fellow; and Jiameng Zheng. It's focused on using the mobility data you described to estimate water-quality benefits, which we've also been talking about. I know this is work in progress, Dan, so that's fine. Just tell us about the research project and what you're setting out to do.

Daniel Phaneuf: Thanks for mentioning my fantastic group of collaborators. Two of the graduate students on this are among the best we've seen, and, obviously, working with Sheila and Yusuke is really super.

This is a project that's funded by the US Environmental Protection Agency, and the objective is to understand the economic value of water-quality resources in a few different places, including the Long Island Sound. The Long Island Sound is interesting, because it's an estuary-type environment where, because of the closed-in nature of the sound, nutrient-pollution runoff from New York City, and other areas nearby can lead to algal blooms, low dissolved oxygen, and degraded ecology.

There's all kinds of debate about how much we should spend to try to reduce these runoffs into the Sound. If we get improvements in environmental quality, what are they going to be worth? Is it worth the cost? The Environmental Protection Agency was interested in helping to understand what kind of economic benefits could be found by improving conditions in the Sound. We have many activities that we're doing, but one of them is to model recreation behavior to coastal areas in the Sound. We're doing this in a few different ways, but one of them was the focus of the session where we're trying to use this new mobility data to estimate the demand for these recreation trips to places along the Long Island Sound shoreline as a function of water quality.

Margaret Walls: Can I just interrupt you for a second?

Daniel Phaneuf: Of course.

Margaret Walls: I'm not so familiar with the setting. Are there a lot of recreation sites along the water there?

Daniel Phaneuf: It is considered an area that's useful for local recreation. There's surfing on the south shore. There's a lot of water resources there. There's I think 70, 80-odd beaches that are on the island, so it is a recreation resource. It's not an iconic recreation resource, but it's locally important and close to population centers, so potentially very heavily used.

Margaret Walls: Okay. Sorry, I didn't mean to interrupt.

Daniel Phaneuf: No worries. I can continue talking a little bit about what we're trying to do. The connection to the mobility data is that, at the end of the day, we're trying to see if we can use the mobility data to get a similar answer to what we would get if we used a traditional method. With the mobility data, what we're able to do is see what I'm going to refer to as a sample of 10 million device holders—enormous amounts of data. For each of these device holders, we know what I'll refer to as the neighborhood they live in. This is actually a census block group, which would be the technical term, but let's just think of it as a neighborhood. We know the neighborhood that they live in.

Through this mobility data, we can count the number of trips that we see from that neighborhood to each of the 128 destinations that we've defined around Long Island Sound. With that, we can compute the travel distance to each of these places from each of these neighborhoods and basically back into the same kind of data that we would get from a recreation survey. But we get 10 million device holders and 33,000 different neighborhoods. We observe over 3 million trips in a given year. We have three years of data: 2019, 2020, and 2021 throughout the Long Island Sound. It's an enormous amount of data that we could never hope to gather from a survey where we might get 500 observations telling us about a thousand trips.

Margaret Walls: The temporal resolution, or throughout a day, you have all these different things. I don't know how you go about aggregating. Do you aggregate to the daily level or something like that?

Daniel Phaneuf: Great question. A huge advantage of these data is that, as us nerdy economists would say, we get great temporal resolution. Regular human beings would say we get to see behavior day to day, so there's a lot of information that we can see as time passes. For our particular application, we've aggregated it to trips in a year, because we want to compare to traditional recreation data sources, which usually will assemble data at the scale of a season or a year, simply because we can't gather it day over day. The information that we're going to use is going to be aggregated up to trips in a year.

Margaret Walls: If you didn't do that, do your water-quality variables vary much from day to day or month to month for various reasons?

Daniel Phaneuf: Great question. The water-quality variable we're using is something called dissolved oxygen, which is a good indicator of overall ecological health. The monitoring of dissolved oxygen is not as frequent as you would need to do day over day, so we're effectively averaging over months and even over the year. One of the really exciting things about these mobility data, though—and this gets a little ahead of us—is that some environmental problems, like harmful algal blooms or beach closures, are on-a-day types of things. With these data, we may ultimately be able to understand how people respond to, say, a warning about a harmful algal bloom or an announcement that a beach has closed. That's part of the potential of this data. It's not what we're doing in this instance.

Margaret Walls: Okay. Tell us a little bit more about some of the challenges of the data, because what you've been describing makes it sound great. Can you also clarify where you get these data from?

Daniel Phaneuf: Your point about these data sounding great—it's very seducing, right?

Margaret Walls: Right.

Daniel Phaneuf: For somebody who's done models with 500 observations to be subtly handed 10 million observations … You get giddy, and Minnesotans don't get giddy. It has this huge potential.

As you dig into it, though, like with any tool, you find that it's no panacea. In terms of where we get these data, these are assembled by private vendors. Because we have this grant from the Environmental Protection Agency, we have decent resources to purchase a subscription to use this data, but it's not inexpensive, so not any researcher right now would have the resources to do this. I think the more serious challenges and limitations, though, have to do with the fact that we just don't know very much about how these observations are generated. We don't know anything about the people. If we were taking a survey, we would ask about individual characteristics.

We don't know if, when we see somebody in the mobility data at a beach, if they're actually going there to visit the beach or if they just happen to be shopping in a store on the beach. We're not able to say what trips are for.

Probably, though, the most important challenge is that we don't understand what statisticians would refer to as the sampling properties. If you listen to the news, and they tell you about polls about preferences, the announcer will always explain the properties of the sample: “It was a random sample derived from a high-quality source.” With these data, we're not able to say what the properties of the sample are, so we're not sure what is being represented. We don't know what's being collected and what's being missed. A big challenge is coming to terms with the fact that these data are just different than randomly sampled observations, and we have to figure out how to learn something from the data without knowing all of those very specific properties that statisticians love.

Margaret Walls: That's well put. I'm glad you're the person working on this, Dan, because I think that's a challenge, but I think you're up to it.

Maybe it's too early in your work with the data to say, but what other kinds of questions do you think these sort of data are useful for? Especially since people are interested in these topics and they want to—and a lot of the young people in our profession are very tooled up to—handle this kind of data. What are the specific questions you think these might be? Is that a fair question to go ahead and ask you?

Daniel Phaneuf: Yeah, that's a really great question. Maybe we can start with what we've historically done in recreation-demand analysis, and that's what I call single-year case studies. I might gather data on people's behavior in Wisconsin during 2019, and we can do some really excellent analysis about Wisconsin in 2019 with those data. But if we want to do anything more, we have to rinse and repeat. We have to gather the data again. I think what the real promise of these data is, is that we can fill in over space and time; maybe we can examine water quality in lakes across the entire country and repeat that exercise year over year to see if behavior and people's economic values for water quality is evolving and to see what happens when conditions change. I think the real advantage is, with space and time, we can get big. We can get temporally precise. We can fill in places that have never been studied.

I reference these data as being expensive to subscribe to, but that aside, they're pretty inexpensive to access. They're much less expensive than fielding an original survey, and it's almost infeasible to field a survey at the scale that I mentioned and in the frequency that I mentioned. They allow us to do things that are just simply not possible without the current technology.

I mentioned a moment ago that, looking at these transient-type environmental conditions, this could be a real opportunity to do that. Your listeners may know that you've worked on some of these ideas of recreation and wildfire and how this is a very transient thing. It could affect people's behavior on the day when smoke is around, but maybe not in a week. Other examples of that are harmful algal blooms, beach closures, and one-time warnings that have to do with spills. These types of here-today-gone-in-a-week type of environmental conditions can be costly for people to deal with, but they are short in time. These types of high-frequency data are needed for that kind of analysis.

Margaret Walls: I meant to ask this a little bit earlier, but what about privacy issues with these data?

Daniel Phaneuf: Yeah, that's something we wrestle with. Even in our session, people were like, "Wow, people can observe my cell phone moving around in space like that." It's a real challenge both in terms of designing protocols for protecting privacy, as well as simply the ethics of using these kinds of data. People opt in in the sense that you can turn off your indicator of space and time on your phone, but probably a better way to think of it is they fail to opt out. There's some debate about the extent to which passively assembled mobility data of this type from people who failed to opt out is consent to assemble data in the same way that university boards that oversee these things would think about. I think that's an open question.

With applications like ours, where you're simply looking at people going to the beach, the privacy concerns are probably not going to get us too worried. But other applications of mobility data where maybe some of the places are more sensitive, the privacy concerns are real. The vendors have been working on ways to mask places that are considered sensitive. For example, we can't observe where people live; we only know their neighborhood. Other types of more sensitive destinations are also masked. People are asking the right questions and trying to take the right kind of steps, but I think more work is needed.

Margaret Walls: That's good to know. We are running out of time here, Dan, but this has been a great conversation.

Daniel Phaneuf: Yeah, this has been fun.

Margaret Walls: We need to close our podcast with our regular feature, which we call Top of the Stack. What books, articles, podcasts, or any kind of content would you like to recommend to our listeners?

Daniel Phaneuf: I was thinking about what I wanted to share with this, and what I ended up on is that I read the Atlantic, and one of the regular columnists for the Atlantic is a person named Arthur Brooks. He writes about ways to find happiness and fulfillment in life. He himself is an academic that works in this area, looking at things that lead to the “good life.” What is interesting to me about him is that, as an academic, he shares many of the personality traits that many of us do. We're very intense. We can't turn off our minds. We're always thinking, and he has some really nice ways of explaining how we can tame back our neuroses and try to enjoy life a little bit more. I really like his perspective as an academic and a researcher in this area—distilling down basic insights from his research on how to have a balanced, happy life.

Margaret Walls: Is he a regular columnist there?

Daniel Phaneuf: He is. If you read the Atlantic, you can search “Arthur Brooks,” and he'll have several articles, one of which was written essentially for academics who have all these neuroses that I just explained, and it was like looking in the mirror. It was fun to read.

Margaret Walls: That's great. I like that. It's been a pleasure having you here in our little recording studio at AERE and having you on Resources Radio. Thanks so much for taking the time from the conference to come and talk to us, and I'm a big admirer of your work, so it's really great to get you on the show, finally. We're going to be looking forward to where this research goes, so thanks very much.

Daniel Phaneuf: Thanks, Margaret. It was really fun to be here.

Margaret Walls: You've been listening to Resources Radio, a podcast from Resources for the Future, or RFF. If you have a minute, we'd really appreciate you leaving us a rating or a comment on your podcast platform of choice. Also, feel free to send us your suggestions for future episodes.

This podcast is made possible with the generous financial support of our listeners. You can help us continue producing these kinds of discussions on the topics that you care about by making a donation to Resources for the Future online at rff.org/donate.

RFF is an independent, nonprofit research institution in Washington, DC. Our mission is to improve environmental, energy, and natural resource decisions through impartial economic research and policy engagement. The views expressed on this podcast are solely those of the podcast guests and may differ from those of RFF experts, its officers, or its directors. RFF does not take positions on specific legislative proposals.

Resources Radio is produced by Elizabeth Wason, with music by Daniel Raimi. Join us next week for another episode.