U.S. poverty has been a persistent problem. President Franklin D. Roosevelt, in 1937, suggested that the test of our progress as a country is “not whether we add more to the abundance of those who have much; it is whether we provide enough for those who have little.” President Lyndon B. Johnson declared his war on poverty in 1964.
And today? Though almost $1 trillion of federal spending goes to programs for reducing economic hardship and improving welfare — one quarter of the annual budget — more than 13% of the population lives on an income below the poverty level. “Poverty is wreaking havoc on a lot of families,” says Geoffrey Lawrence Cohen, a Stanford University professor with a courtesy appointment at Stanford Graduate School of Business.
In response to this havoc, Cohen and a multidisciplinary team of professors are launching a novel search for solutions to poverty. They will use machine learning to cull through huge data sets to understand the many variables that lead to, perpetuate, and potentially even prevent impoverishment. “We’re taking the tools of Silicon Valley to this historic problem,” says Cohen.
Not Poverty, but Poverties
Poverty, the researchers argue, is not one problem but a multitude disguised as one — each with its own etiology, its own individual experience, and its own mechanisms. The policies to address poverty must therefore be attuned to the specific conditions of each case.
Cohen and the team will combine multiple existing and new data sets, including U.S. demographics, social network data, census and Google Street View information, psychological data, and employment and earnings statistics. They will use machine learning to analyze these different variables in combination, to root out the various factors involved in poverty and some of its consequences, such as poor health and educational underachievement.
The project was inspired in part by recent advances in precision medicine. “We haven’t cured cancer,” says Cohen, “but almost everyone agrees that we’re making much more significant progress since we realized cancer is not a single entity, but many entities, and that we ought to tailor treatment to the underlying problem.”
Making the War on Poverty “Smart”
The research team includes computer scientists who will apply data analytics and machine learning to the vast web of information collected, ultimately creating a fine-grained classification of the different kinds of poverty — the many pathways in and potential trajectories out. How, for example, do addiction and housing insecurity interact? Do strong social programs tend to counteract weak social networks? This information will also form the bedrock of predictive models that suggest which interventions are likely to work best in a given context.
To create these models, the team will carry out randomized controlled trials that test the effect of small interventions on poverty-related outcomes. Cohen has designed a suite of psychological interventions on heads of household. For instance, one kind of intervention has people reflect on core values that give them sustenance and a sense of dignity. This reflection has been shown to buffer people against stress and, in education and health contexts, has led to lasting improvements.
Though not typically associated with solutions to poverty, this type of intervention is based on research pioneered in the Psychology Department at Stanford University. Cohen noted one study in which poor immigrant children were asked to affirm their deeply held values, which led to improved school outcomes years later and a higher likelihood of being on track for college.
Cohen and his colleagues don’t believe an intervention like this will work for everyone, but it’s part of the project’s vision. “We plan to use this as a proof point that precision treatment is possible,” he says. “The hope is not to find a one-size-fits-all solution to poverty, but to demonstrate how we might tailor policies and interventions to the idiosyncratic vulnerabilities and opportunities of the people we’re trying to reach.”
Beginning with the Problem
To Cohen and his team, this approach represents a radical shift in the way social science is conducted. “Most researchers begin with a solution,” Cohen says. They decide to study a job training program, for instance, or a school health initiative. They roll the study out, ideally as a randomized controlled trial, and then dissect the results. “We’re doing the opposite: We’re beginning with the problem and trying to map all of its contours so that we can target interventions precisely.”
The team says future research with these data could examine a wide range of poverty-related concerns, even individual genomics. Given that prolonged exposure to stress can activate genes that adversely affect health outcomes, might there be targeted medical interventions that could help?
In the long term, Cohen imagines converting this data bank into a diagnostic map to guide the nation’s efforts to address poverty. “With enough precision intervention studies, this could become like a recommendation engine that tells which kind of policy is most likely to work given specific circumstances,” he says. “There are a lot of people who are teetering on the cusp of a different destiny — and the right act of support at the right time could make all the difference.”