Harold Pollack Interview w/Marynia Kolak Edited for length and clarity of translation from spoken to written word.

Harold Pollack. How did you get into GIS? 

Marynia Kolak. I got into GIS during my undergrad studies, I was actually trained as a geologist. I didn’t even take a class in GIS. One of my first college jobs was at the Illinois State Geological Survey. As a student worker, I was helping geologists with mapping. I learned GIS through that.   

So in my undergrad, I was a geology student. Well, we didn’t do GIS really. It was more, if we developed maps, they would be by hand. So I was taught how to develop maps… trained as a field geologist. But then at some point, I was really interested in delving in deeper in different ways, and the job that I got was at the Illinois State Geological Survey…. After the field geology assessment, and especially if you’re interested in what’s underneath the ground, not just what’s at the top, you really need to put that into a map.  

My first job there was considered “big data,” at that time, where there were tens of thousands of drill logs. Every time that you need to drill for water, you need to report that. Or anytime a company drills for anything, for developing a building or just for engineering information, they have to record all the different types of rock that they drill through at different stages. It’s a requirement. So my job was to take all of that and put it into a database. I learned SQL. I didn’t realize it was SQL at the time, but SQL is really fun. And again, it’s a puzzle. You’re trying to organize this big mess, and everyone records data in different ways. But even if it’s in the right format, they’re all recording it in different ways. After graduation, I put some of that GIS to use at the United States Geological Survey doing digital recovery — a historical project converting maps from the 1960s (from the US Army Corps of Engineers) to a new digital record. 

So I got to bring some of my GIS skills back into use, then, and that was really interesting and exciting to me. But it still wasn’t a focus on that stage, I was really interested in how the science and where people were feeling about the science was so disconnected, and especially with climate change and so on. 

I actually went back to school for writing better my writing skills. That was initially meant to be a short tangent before I went back to school for geology, but in my writing courses I learned a lot about critical theory, and I was introduced to a whole new range of research that really interested me. And at that point, I saw that the geology I was interested would require me to be out at sea for half of the year… I was starting a family, and that would have been difficult. So I made a decision to stay in Chicago, and to put my skills to different use.  

Fortunately, I had a chance to put my data and writing skills to work with community engagement I was doing as a volunteer at Engineers Without Borders. I was also present at Northwestern during the formation of the Institute of Public Health, where I worked at the time. I was exposed to public health at Northwestern, for the first time, in this big way. It was just so interesting because it connected science with research, and translated to community. There was a sense of urgency: How is the environment changing? How will that impact people? 

Thinking I knew better than others, as all 20 somethings do, I was convinced that health researchers (where I worked) were using maps all wrong, just as a visualization products or communication tools. This didn’t click with my geology training, which viewed maps as the final stage of after a long, robust research analysis. This still happens in public health today, where there’s a focus on one or two techniques that people use over and over, but that may not actually be the best technique or analysis. Of course, this happens in statistics, too.  

I did a Masters in GIS through Johns Hopkins to build my skills further. Community health assessments were becoming a top priority for hospital systems. I was really interested in how you measure and model space in such efforts… My masters’ project was my first big effort to connect the different pieces, modeling place from a public health lens: What are the different things that constitute a healthy place?  

There’s a big literature on this, but the complexities emerge when you try to operationalize that with data. How do you operationalize walkable places? There are dozens of different ways to do that. You know, you might use intersection density as one option. You might develop a massive access resource analysis. So it’s not necessarily walkability in general, it’s walkability to resources. 

And walkability matters differently in different places. It’s spatially heterogeneous. It means something different in rural and urban settings. I’ve become fascinated by the measurements of place. And also if you use one measure of walkability that was designed for urban spaces in rural areas, you will have a result, but it’ll be kind of meaningless to the people who live there– and vice versa. So I think that aspect is interesting.  

My dissertation focused on extending the potential outcomes framework in epidemiology and other disciplines with a spatial perspective. Much of what I did there was incorporating spatial econometrics. Different disciplines—epidemiology, econometrics, and others–have different jargon for the same phenomena. In public health we can’t do randomized control trials for everything, because they’re ethically bound, or they don’t make sense if we know that poverty is associated with many outcomes. Attention must be paid to causal inference. 

Harold Pollack. That’s a great point because GIS methods often produce awesome graphics, with less clarity about the specific question to which this awesome graphic is the answer: Is this documenting an important association for resource allocation? Is this supporting a causal story for policy analysis?  

Marynia Kolak. That’s so true. Coming from different disciplines myself, I learned that each approach has something to learn from others, especially for complex problems like the opioid overdose epidemic that we seek to address through JCOIN and the MAARC. I think they’re called “wicked” problems officially in the literature. You need a kind of Bell Laboratory mindset, where we mobilize different people from diverse backgrounds coming to address these complex problems.  

Harold Pollack. The maps you’ve produced show include some striking details. Opioids are a really challenging problem in places we don’t always expect. If you put up the map of where people are experiencing opiate overdoses and you put up another map of where the facilities are to help people with opioid use disorders, there’s less overlap in these maps than there needs to be. 

Marynia Kolak. Exactly. There is a big mismatch. It exists alongside stigma and other challenges. I think a lot of it also comes to science translation and research translation, not so different from the issues climate change scientists face in communicating the urgency of what they know. 

Harold Pollack. Very few of us in public policy, economics, and public health are actually trained in spatial econometrics even those of us with lots of other econometrics training. The idea that you should think about maps as explicitly econometric spaces is not something most of us have been taught. If you examine standard curricula in policy schools, economics departments, and public health schools, you rarely find that.  

Marynia Kolak. It’s funny because that’s when I was applying for PhD programs, I got a lot of pushback from my mentors because I was specifically looking for programs where I could get that specialized training. I chose the path less traveled. I still don’t know if it’s paid off; there is a bit of a gamble in that.  

For my dissertation, I studied the period from the late 60s-70s where there was a lot of pressure to reduce the drinking age. You have this classic natural experiment of different states adopting different policies over you know decades, and then looking at the impact of motor vehicle accident deaths. I looked at this as a spatial challenge as well. I replicated the initial experiment and added additional factors. Spatial analysis isn’t really about maps; it’s also about thinking about spatial effects. So many different things can generate patterns. If my neighbor passes a law, I might be swayed to follow. Or I might act in competition to that, for example to lower taxes right when nearby states raise them. On top of that you’ll have states acting in similar ways because of cultural factors. Southern states tend to act more similarly than those in other parts of the country. And you have states like Utah, whose unique culture often makes it an outlier.  

We have special terms for that–spatial heterogeneity versus spatial dependence. You can’t always use traditional tools to study this or to test different hypotheses. In the case of drinking age, there was a spatial multiplier effect among nearby states. There was a global spatial lag, meaning that additional lives were saved over time because of these local influences and neighbor relationships…   

Harold Pollack. Let’s zoom in now on the MAARC, What are some of the ways that you’ve brought these GIS skills in collaboration with JCOIN partners?  

Marynia Kolak. I run the Healthy Regions and Policies Lab. Dr. Qinyun Lin is the postdoc scholar at our group. Susan Paykin is a researcher who also supports with management. Dylan Halpern is a software engineer. That’s the core group. We also have guests, like a visiting engineer this year, Stuart Lynn, and then we have an army of research assistants for different projects. And Aresha Martinez-Cardoso as a colleague and collaborator from Public Health Sciences; she’s a co-investigator on the Covid Atlas team.  

We also have worked with Paul Joudrey and Emily Wang at the Yale clinical hub. For one paper, we measured zip-code level opioid use disorder treatment access linked with social vulnerability.  

We have a particular focus on two things. One is focused on the service part — our center seeks to make research more accessible to the whole community. I’m proud of the work we did in making data accessible from different layers of the community environment in the Opioid Environment Policy Scan (OEPS) for research across the country. This was a conceptually driven model, developed from the opioid risk environment framework. I like the risk environment framework because it focuses on context, not on individual behavior. It shifts the burden to our society. What can our society do to improve outcomes for individuals? The framework also includes social context, policy context, and other drivers of structures within the environment. We got great feedback from the JCOIN community on additional variables to integrate. Initially, we were going to just use that for our own research, but then we heard that similar data were needed for JCOIN. That worked out really well and became something much bigger than I expected. It turned into its own interactive website with extensive documentation and exploration of individual variables.  

We’ve been capturing how others have been using this data in different work. It’s been used at the New York State Psychiatric Institute, Chestnut Health Systems, in different work with Yale and other partners. This framework was also adapted for projects outside of JCOIN. There’s another series of NIH-funded clinical projects called ETHIC that seek to address opioid use disorder in rural contexts. And they learned about OEPS, readapted it, and extended it for rural environment specifically. We built it to be easily replicated and extended. The theme has been collaboration versus competition. 

In our work, we’ve found that social vulnerability intersected with access in totally different ways, depending on if you’re in urban, suburban, or rural areas. Some of the greatest mismatches between access and vulnerability arise in suburban areas which are often overlooked, whereas in rural areas access was so poor overall that it there wasn’t enough variation for there to be an outcome. There are really three “spatial regimes,” in the U.S.–urban, suburban, and rural—that are having distinctly different experiences right now.  

Some of our future research is trying to delve more deeply into the rural landscape. We understand these areas the least. They comprise so much of the U.S. landscape, and whose experiences in the opioid epidemic has weighed so heavily in national politics.  

Harold Pollack. Rural areas have been distinctively harmed by opioids. That’s apparent when we consider media coverage of what’s happening in parts of Kentucky, West Virginia, southern Ohio. As with all social problems, the public and policymakers can embrace simplified or stereotyped view of what’s going on. Many of us tend to view those the rural opioid epidemic as a nearly-exclusively non-Hispanic White phenomenon—which it’s not.  

Marynia Kolak. And there’s been a huge suburbanization of poverty. Researchers have been talking about this over the past decade. That isn’t necessarily communicated to the public. 

Marynia Kolak. Some of what you’re talking about now links to my past work on food deserts and food access analysis. A lot of that matches the same issues here. On Chicago’s North Side, you don’t have access to just one supermarket. There are dozens. I think we sometimes just focus on deprivation. It’s no less important to focus on the opposite—how more-privileged folks get access to a vast range of options. Taking a step back, what are the processes that drive that kind of inequality in the first place.  

That’s something I was trained to study. Spatial econometrics comes back to that. What is the data generating process that created these patterns? How can we capture that within a spatial analysis framework that allows robust causal inference?  That’s difficult to do when you usually only have natural experiments.  You know the data is going to be imprecise. There’s so many limitations. But I feel strongly that just because something’s difficult doesn’t mean we shouldn’t do it, or at least develop a baseline for that sort of rigorous analysis. 

Harold Pollack. It seems like your group is creating public goods, working with research partners to combine their local knowledge with your expertise in GIS and social determinants of health. 

Marynia Kolak. Yes, definitely. And then on the research side, there’s a lot of really exciting work coming out. It’s been great to be connected with others in JCOIN and beyond to take deep dives into access to medications for opioid use disorder (MOUD) and really focus on accessibility: looking at this historically across different types of access, not just spatial access. Many of the research articles with different partners highlight different aspects of that.  

We found that access can be associated at a structural level with stigma, which was really interesting. In one study [Lin et al 2022] we found that if communities tended to have more providers or better MOUD access, they tended to have slightly less stigma, controlling for other factors. There’s a bit of a chicken-and-egg challenge, what came first? We need to understand more about access to MOUD before we can choose the right policies to improve that. 

In term of geographic access, we’re continuing to research different travel metrics. How do people actually access MOUDs? And then it also depends on which medications people prefer. Not everyone wants buprenorphine. Some folks prefer methadone. Within the justice-involved community, when you’re working with drug courts, there’s high stakes for falling out of treatment. If you don’t have access to MOUD in the community, it’s a recipe for disaster. I think that our team is really becoming a leader in this space. 

We bring methodological expertise. There’s the Geographic Information Science–which empirical methods do we use? There are broader questions, too: What processes create the population patterns that we see?  

We’re thinking about problems of scale. Much of my work concerns inequity, and structural racism. What is this scale of structural racism? Well, it exists at many levels, but policies exist at larger levels. So we’ll see a lot of inequity at those scales that reflects that. In the case of MOUD, we see structural inequities with key differences in policy and social context between more-punitive and more-public-health-friendly jurisdictions. One of my goals is to think about that wider context of health geography. That’s a really old field, but just we’re trying to catch up with the rest of the world. Many countries outside of the US are leaders in this, and so we’re playing catch-up in some ways. 

Harold Pollack. In the context of the opioid epidemic, the numbers speak for themselves. And aspects brought public attention to long- neglected issues. The epidemic forced many of us to look more seriously at the challenges facing rural America, which was so heavily affected.  

Marynia Kolak. Job opportunities were lost- but with no change in infrastructure. We have huge economic changes that result in job loss. But we don’t have any change in public or transit infrastructure that might make it easier for folks. And if anything, it gets worse. We see analogous challenges in other places, including Chicago’s South Side. I was just giving an interview with a reporter about food access. These issues bear similarity to rural spaces. You have wide geographic places with not as many resources. It’s coupled with worse transit; those combine to worsen access to resources. 

Harold Pollack. These spatial challenges are intermingled with the sociological challenges in these same areas. 

Marynia Kolak. I would say the sociological processes are driving the geographic patterns we see at a larger scale. You can’t model the geographic processes without understanding the sociological ones. They’re really connected. It’s a social science. I think that’s what’s so fun about it. You get all the methods in computational GI science, but then you also have to really think about those processes and get into the nitty gritty arguments of, how do we construct our societies, and what are the consequences of these arrangements? 

Harold Pollack. One of the things that interest me in what you just said is to the outsider, you might think, boy, people that do very methodologically sophisticated GIS work or whatever, there must be… It’s easy for us to stereotype that comes with a kind of sociological naivete. 

Marynia Kolak. Oh yeah. I get that a lot. 

Harold Pollack. Of course, you’re saying the opposite. The methodological richness is needed and is there because you’re trying to understand the social processes and bring them into geospatial understanding. 

Marynia Kolak. That’s where some of the most exciting work will happen: tying causal inference to GI science. Sometimes you have to test different theories or different hypotheses, which it could be one, two, three, four different sociological processes, or a combination of those. 

As part of MAARC we’re not going to just do that analysis and develop that database. We’re also going to share how we’re doing it. We’ll share the products of that with the JCOIN Community, and we’ll help train others.  

So, for example, I’m just developing that OEPS database. It’s one of the main ways that we’ve connected with different JCOIN hubs. We started with an initial list, based on the risk environment framework, of what data to include that might influence opioid overdose outcomes, with a focus on MOUD treatment. From the beginning we were also interested in developing a database of community contextual variables, so that each clinical site would have an ability to prepare for what happens after jail or prison release. When folks go back home to their communities, there needs to be additional information to account for these contextual factors.  

We held workshops with different clinical sites, to talk about what community context variables to include. Bruce Taylor’s NORC team was a part of that. As we develop those data and share that with different groups, the work continued to mature. When we developed or created access metrics, we didn’t just add them to the database. We also share code for how to reproduce this yourself with your own data. For example, the Columbia site developed their own validated list of resources. This was a step beyond what we could do, because they have specific and validated local data. They were also the first to pilot the tutorials we generated.  That paper is in final stages of journal submission. So there have been some early of both the resource tutorials, but then also on the available data. 

Harold Pollack. Over the past two years, we’ve had the twin epidemics of opioids and covid. Can you talk about some of your work at the intersection of these epidemics?  

Marynia Kolak. The Center, and thus our spatial MAARC team, kicked off not long before the pandemic did. We launched the US Covid Atlas around the same time we began the spatial work of the MAARC. The Covid Atlas [https://uscovidatlas.org] aggregates data from multiple sources, and then integrates this data with community contextual data. Then we make that all accessible in a way that people can explore data collected throughout the whole pandemic. It’s a spatial temporal resource, and it adds classic tools of spatial analysis people need to interpret the data in a meaningful way. This was a valuable experience to improve and refine our MAARC work.  We’ve had an ongoing project funded by RWJF, and a recent JAMA Network Open publication, led by Dr Lin, that examines the relationship between structural barriers and disparities in Covid mortality across the country. 

Harold Pollack. What do you mean by “structural barriers? 

Marynia Kolak. This is really the Community context. Our group specializes in understanding social determinants of health, structural determinants of health, and ways spatial data often signals structural barriers. That’s technically what we’re trying to achieve by measuring the environment. 

Harold Pollack. It might be valuable to think in a more intentional way about spatial experimentation. There’s a great paper in the Journal of Economic Perspectives by Ludwig and colleagues on a mechanism experiments. Among other things, they note that we don’t know the actual public health impact of food deserts. They note one thing we could actually do. We could go to some potential food deserts. where there are not very many stores. We could conduct a randomized trial, in which we gave 500 people generous Pea Pod subscriptions. We say: “For you there’s no food desert, you know you, we’ll just bring groceries to your house, whatever you want.” And then we compare their health and nutrition outcomes to 500 other people in a control group.  

Marynia Kolak. There were a few studies like that, maybe more natural experiments, where stores were opening for the first time. But then, from my perspective, there’s also this temporal dimension. There is a component of complex PTSD from multi-generational trauma. It will take more than a year or two to shift eating patterns. What’s the right scale of analysis from that perspective? I think that spatial analysis is powerful in creating ways to create more matching possibilities. Alberto Abadie’s work in matching controls provides an approach to that. That’s the sort of direction some of this needs to go. However, even with this work, there’s this assumption that matching happens within administrative boundaries, without considering interaction effects with nearby places.  

Harold Pollack. Maybe that’s our next project. Thanks so much for speaking with me.