Agent-based modeling simulates the interaction between people and environment–and may offer valuable insights about preventing substance use-related harms
One of the first things Dr. Christina Mair did during our conversation was disabuse me of my whack-a-mole theories. “Like when bars get serious about carding, aren’t there just more house parties for minors to drink at?” I asked.
“You’d be surprised,” she said. “That’s a very commonly-held idea, that if you get rid of it here, it turns up there. And yes, there is some displacement, but it’s definitely not a 1:1 ratio.”
“Another example of the whack-a-mole theory is when you hear parents say, ‘Oh, well, I know my kid is going to drink anyway, so if I let him drink responsibly in the home, he won’t have problems.’ Which, of course, is not true. Studies show over and over again that those kids are drinking more,” Mair said.
Human behavior is context-dependent. No one lives in a vacuum. And according to Mair, when you make something harder to do, people do it less.
It’s an encouraging premise: making changes to the context of our lives can reduce substance use related harms. And nowadays, when the phrase “addiction should be treated as a public health issue” is being tossed around willy-nilly, this premise is an example of what that catchphrase actually looks like.
Mair is an assistant professor at the University of Pittsburgh Graduate School of Public Health. She came to Pittsburgh from the Prevention Research Center (PRC) in Berkeley, CA, where she researched environmental prevention strategies—well, much more than that. Her research examines the “social ecology” of substance use, which means the messy fabric of life in which we make decisions about drugs and alcohol.
“Social ecology is about social mechanisms. How are people acting in different contexts? How do personal characteristics and environmental characteristics combine to affect the way people are behaving?” Mair explained.
Social ecological theories focus on integrating what we know about individual-level behavior—a person who tends to drink after an unusually stressful day, for example—and population-level behavior—like a neighborhood with high bar density that has higher levels of alcohol-related violence.
“The explicit integration of those different levels is a wide open field with all sorts of questions,” Mair said. “This work is just beginning.”
Historically, many prevention efforts have focused on school-based approaches to educate kids about the danger of drugs, including the now-debunked D.A.R.E. program. Mair’s investigation of social mechanisms that promote or discourage substance use and its harms can be used to develop a much wider range of interventions, from enforcing loitering bans outside of bars to publicizing roadside driver checks in target areas.
Unpacking what “socioeconomic status” really means
At the PRC, Mair researched the social ecology of substance use under an NIAAA grant–an area that the NIAAA doesn’t often fund. “Most other NIAAA grantees are clinically-oriented or focus on biological or genetic issues related to alcohol. And we know that stuff is important,” Mair said. ”But this really spoke to me.”
Why? “As a public health person who’s always thinking about populations, the question for me about substance use is what can we do about prevention–specifically, environmental prevention. Because those efforts can reach so many people,” she said.
At the University of Washington, where she received her Master’s of Public Health in epidemiology, Mair’s interest was piqued when she took a class on the social determinants of health.
“There, I realized that a lot of people were simply plugging one big factor called ‘socioeconomic status’ into their models,” she said.
Mair wanted to take a more nuanced approach to understanding the social factors affecting disease.
“I realized that so much interesting work thinking about real world problems was happening in the area of social epidemiology.”
With that in mind, Mair studied for her PhD at the University of Michigan, examining how neighborhoods—their social and environmental composition—affect depression in men and women.
Her research showed that a better social neighborhood environment lowered depression in women but had no impact on men. In contrast, a negative physical environment (physical decay, litter, graffiti) was associated with higher levels of depression in both men and women.
After finishing her doctorate, she accepted a postdoc at the PRC, where she began specifically looking at behavior related to alcohol and drugs. “I came to substance use from mental health,” Mair said.
Agent-based modeling: understanding the complexity of human substance use
One of the exciting methods that Mair uses in her research is called agent-based modeling, an approach that biologists have been using for decades but has only recently been adopted by social scientists.
How do you make a model of the way people use substances? You start by doing descriptive research, looking at individual-level behavior and then population-level data. And from there, you generate a theory.
“What is one social mechanism that might be important? You need an idea about how the environment interacts with individual behaviors. Then you can program a world with certain environments where people have certain traits. And you can play with these characteristics–we say you can ‘parameterize them’–until you see that the models resemble what you see happening in the real world.”
And models are cool because you can play them forward in time. “You can see if it looks like what over time is actually happening in the world. You can see if the social mechanism works the way you think it does,” said Mair. “You can also model potential interventions. There’s a lot you can do with this method.”
If this sounds complicated, it’s because it is. Building good models for human behavior requires an interdisciplinary team knowledgeable about the many aspects of individual- and population-level substance use, including fields like epidemiology, public health, psychology, and neuroscience.
“You need the people who know the right questions to ask and you also really need a computer programmer. This approach explicitly forces cross-disciplinary work,” said Mair.
At Pitt’s School of Public Health, Mair joins Dr. Christopher Keane, another young researcher who favors agent-based modeling as a research methodology. Keane’s recently published book, Modeling Behavior in Complex Public Health Systems, is summarized chapter-by-chapter online. Its underlying premise, according to Keane: “Public health behavior makes the most sense in light of evolutionary games.”
Sprinkled throughout the book are psychedelic videos that illustrate Keane’s models.
I like the optimism that underlies the use of agent-based models. What a tall order, simulating the many internal and external factors that combine to produce human behavior at any given moment. It’s sure to be slow-going and to miss the mark. But as new research is conducted, new theories are developed, and we watch the epidemiology of substance use-related phenomena unfold, these models can be continually tweaked. They can become increasingly useful.
In that sense, the decision to incorporate models into public health research represents a commitment to looking more and more deeply at the factors that affect human behavior, however complex those turn out to be.
“People are really confusing and messy and complicated,” Mair said. “It’s harder work to use these methods in the social world than in the biological world, which is one of the reasons that this work is just beginning.”
Agent-based models can also be used to anticipate the effects of a policy before it is created and to project the cost-benefit of various interventions–two challenging issues for the public health field.
Of course, learning which types of interventions are effective doesn’t mean they will be implemented. “What is the role of a prevention researcher,” I asked, “who knows that applying certain environmental strategies may be more effective than designing certain curricula?”
“Research and politics, of course, can’t be separated,” Mair said. “A prevention researcher can look at certain policies, like taxation and pricing, and make recommendations about implementing them. She can say, ‘There’s a lot of good evidence that reducing access to alcohol does make a difference so if you’re thinking of making a policy, this is something we would recommend.’”
“Of course, that’s not a politically popular thing to do,” I said.
Mair laughed. “No, it’s not. It’s definitely not.”