While hypothesis testing is a highly formalized activity, hypothesis generation remains largely informal. We propose a systematic procedure to generate novel hypotheses about human behavior, which uses the capacity of machine learning algorithms to notice patterns people might not. We illustrate the procedure with a concrete application: judge decisions about whom to jail. We begin with a striking fact: the defendant’s face alone matters greatly for the judge’s jailing decision. In fact, an algorithm given only the pixels in the defendant’s mug shot accounts for up to half of the predictable variation. We develop a procedure that allows human subjects to interact with this black-box algorithm to produce hypotheses about what in the face influences judge decisions. The procedure generates hypotheses that are both interpretable and novel: they are not explained by demographics (e.g., race) or existing psychology research, nor are they already known (even if tacitly) to people or experts. Though these results are specific, our procedure is general. It provides a way to produce novel, interpretable hypotheses from any high-dimensional data set (e.g., cell phones, satellites, online behavior, news headlines, corporate filings, and high-frequency time series). A central tenet of our article is that hypothesis generation is a valuable activity, and we hope this encourages future work in this largely “prescientific” stage of science.

The Moving to Opportunity (MTO) demonstration assigned housing vouchers via random lottery to public housing residents in five cities. We use the exogenous variation in residential locations generated by MTO to estimate neighborhood effects on youth crime and delinquency. The offer to relocate to lower-poverty areas reduces arrests among female youth for violent and property crimes, relative to a control group. For males the offer to relocate reduces arrests for violent crime, at least in the short run, but increases problem behaviors and property crime arrests. The gender difference in treatment effects seems to reflect differences in how male and female youths from disadvantaged backgrounds adapt and respond to similar new neighborhood environments.

The standard economic view suggests that people will commit an action if its expected benefits outweigh its costs. But before people weigh the costs and benefits of an action, what affects whether they think of the action in the first place? We argue that actions are more likely to enter into consideration when they are cognitively accessible. We describe three psychological parameters that influence accessibility: automatic assumptions, identity, and perceptions of privacy. These parameters make it possible to identify new interventions for behavior change.

This study estimates the effects of means-tested housing programs on labor supply using data from a randomized housing voucher wait-list lottery in Chicago. Economic theory is ambiguous about the expected sign of any labor supply response. We find that among working-age, able-bodied adults, housing voucher use reduces labor force participation by around 4 percentage points (6 percent) and quarterly earnings by $329 (10 percent), and increases Temporary Assistance for Needy Families program participation by around 2 percentage points (15 percent). We find no evidence that the housing-specific mechanisms hypothesized to promote work, such as neighborhood quality or residential stability, are important empirically.

One long-standing motivation for low-income housing programs is the possibility that housing affordability and housing conditions generate externalities, including on children’s behavior and long-term life outcomes. We take advantage of a randomized housing voucher lottery in Chicago in 1997 to examine the long-term impact of housing assistance on a wide variety of child outcomes, including schooling, health, and criminal involvement. In contrast to most prior work focusing on families in public housing, we focus on families living in unsubsidized private housing at baseline, for whom voucher receipt generates large changes in both housing and nonhousing consumption. We find that the receipt of housing assistance has little, if any, impact on neighborhood or school quality or on a wide range of important child outcomes.

Most empirical policy work focuses on causal inference. We argue an important class of policy problems does not require causal inference but instead requires predictive inference. Solving these “prediction policy problems” requires more than simple regression techniques, since these are tuned to generating unbiased estimates of coefficients rather than minimizing prediction error. We argue that new developments in the field of “machine learning” are particularly useful for addressing these prediction problems. We use an example from health policy to illustrate the large potential social welfare gains from improved prediction.

Can machine learning improve human decision making? Bail decisions provide a good test case. Millions of times each year, judges make jail-or-release decisions that hinge on a prediction of what a defendant would do if released. The concreteness of the prediction task combined with the volume of data available makes this a promising machine-learning application. Yet comparing the algorithm to judges proves complicated. First, the available data are generated by prior judge decisions. We only observe crime outcomes for released defendants, not for those judges detained. This makes it hard to evaluate counterfactual decision rules based on algorithmic predictions. Second, judges may have a broader set of preferences than the variable the algorithm predicts; for instance, judges may care specifically about violent crimes or about racial inequities. We deal with these problems using different econometric strategies, such as quasi-random assignment of cases to judges. Even accounting for these concerns, our results suggest potentially large welfare gains: one policy simulation shows crime reductions up to 24.7% with no change in jailing rates, or jailing rate reductions up to 41.9% with no increase in crime rates. Moreover, all categories of crime, including violent crimes, show reductions; these gains can be achieved while simultaneously reducing racial disparities. These results suggest that while machine learning can be valuable, realizing this value requires integrating these tools into an economic framework: being clear about the link between predictions and decisions; specifying the scope of payoff functions; and constructing unbiased decision counterfactuals.

We present the results of three large-scale randomized controlled trials (RCTs) carried out in Chicago, testing interventions to reduce crime and dropout by changing the decision making of economically disadvantaged youth. We study a program called Becoming a Man (BAM), developed by the nonprofit Youth Guidance, in two RCTs implemented in 2009–2010 and 2013–2015. In the two studies participation in the program reduced total arrests during the intervention period by 28–35%, reduced violent-crime arrests by 45–50%, improved school engagement, and in the first study where we have follow-up data, increased graduation rates by 12–19%. The third RCT tested a program with partially overlapping components carried out in the Cook County Juvenile Temporary Detention Center (JTDC), which reduced readmission rates to the facility by 21%. These large behavioral responses combined with modest program costs imply benefit-cost ratios for these interventions from 5-to-1 up to 30-to-1 or more. Our data on mechanisms are not ideal, but we find no positive evidence that these effects are due to changes in emotional intelligence or social skills, self-control or “grit,” or a generic mentoring effect. We find suggestive support for the hypothesis that the programs work by helping youth slow down and reflect on whether their automatic thoughts and behaviors are well suited to the situation they are in, or whether the situation could be construed differently.