Knowledge that Transforms

Abstract Many school districts with centralized school choice adopt strategy-proof assignment mechanisms to relieve applicants from needing to strategize based on beliefs about their own admissions chances. This article shows that beliefs about admissions chances shape choice outcomes even when the assignment mechanism is strategy-proof by influencing how applicants search for schools and that “smart matching platforms” that provide live feedback on admissions chances help applicants search more effectively. Motivated by a model in which applicants engage in costly search for schools and overoptimism can lead to undersearch, we use data from a large-scale survey of choice participants in Chile to show that learning about schools is hard, beliefs about admissions chances guide the decision to stop searching, and applicants systematically underestimate nonplacement risk. We use RCT and RD research designs to evaluate scaled live feedback policies in the Chilean and New Haven choice systems. Twenty-two percent of applicants submitting applications where risks of nonplacement are high respond to warnings by adding schools to their lists, reducing nonplacement risk by 58% and increasing test score value added at the schools where they enroll by 0.10 standard deviations. Reducing the burden of school choice requires not just strategy-proofness inside the centralized system but also choice supports for the strategic decisions that inevitably remain outside of it.

Abstract We use admissions lotteries to estimate the effects of large-scale public preschool in Boston on college-going, college preparation, standardized test scores, and behavioral outcomes. Preschool enrollment boosts college attendance as well as SAT test taking and high school graduation. Preschool also decreases high school disciplinary measures including juvenile incarceration, but has no detectable effect on state achievement test scores. An analysis of subgroups shows that effects on college enrollment, SAT-taking, and disciplinary outcomes are larger for boys than for girls. Our findings illustrate possibilities for large-scale modern, public preschool and highlight the importance of measuring long-term and non–test score outcomes in evaluating the effectiveness of education programs.

Abstract A large literature in cognitive science studies the puzzling “Flynn effect” of rising fluid intelligence (reasoning skill) in rich countries. We develop an economic model in which a cohort’s mix of skills is determined by different skills’ relative returns in the labor market and by the technology for producing skills. We estimate the model using administrative data from Sweden. Combining data from exams taken at military enlistment with earnings records from the tax register, we document an increase in the relative labor market return to logical reasoning skill as compared to vocabulary knowledge. The estimated model implies that changes in labor market returns explain 37% of the measured increase in reasoning skill, and can also explain the decline in knowledge. An original survey of parents, an analysis of trends in school curricula, and an analysis of occupational characteristics show evidence of increasing emphasis on reasoning as compared to knowledge.

Abstract Using a randomized experiment with an automobile manufacturing firm in China, we measure the effects of letting workers evaluate their managers on worker and firm outcomes. In the treatment teams, workers evaluate their managers monthly. We find that providing feedback leads to significant reductions in worker turnover and increases in team-level productivity. In addition, workers report higher levels of happiness and well-being. The evidence suggests that these results are driven by learning by managers, leading to changes in their behavior and an overall better relationship between managers and workers.

Abstract Different agents need to make a prediction. They observe identical data, but have different models: they predict using different explanatory variables. We study which agent believes they have the best predictive ability—as measured by the smallest subjective posterior mean squared prediction error—and show how it depends on the sample size. With small samples, we present results suggesting it is an agent using a low-dimensional model. With large samples, it is generally an agent with a high-dimensional model, possibly including irrelevant variables, but never excluding relevant ones. We apply our results to characterize the winning model in an auction of productive assets, to argue that entrepreneurs and investors with simple models will be overrepresented in new sectors, and to understand the proliferation of “factors” that explain the cross-sectional variation of expected stock returns in the asset-pricing literature.

Abstract This article studies external sovereign bonds as an asset class. We compile a new database of 266,000 monthly prices of foreign-currency government bonds traded in London and New York between 1815 (the Battle of Waterloo) and 2016, covering up to 91 countries. Our main insight is that, as in equity markets, the returns on external sovereign bonds have been sufficiently high to compensate for risk. Real ex post returns average more than 6% annually across two centuries, including default episodes, major wars, and global crises. This represents an excess return of 3%–4% above US or UK government bonds, which is comparable to stocks and outperforms corporate bonds. Central to this finding are the high average coupons offered on external sovereign bonds. The observed returns are hard to reconcile with canonical theoretical models and the degree of credit risk in this market, as measured by historical default and recovery rates. Based on our archive of more than 300 sovereign debt restructurings since 1815, we show that full repudiation is rare; the median creditor loss (haircut) is below 50%.

How effective are physicians at diagnosing heart attacks? To answer this question, we contrast physician testing decisions with a machine learning model of risk. When the two deviate, we use actual health outcome data to judge whether the algorithm or the physician was right. We find physicians over-test: tests that are predictably useless are still performed. At the same time, physicians also under-test: many predicted high-risk patients are untested and then suffer adverse health events (including death) at high rates. A natural experiment using shift-to-shift testing variation confirms these findings: increasing testing improves health and reduces mortality, but only for patients flagged as high-risk by the algorithm. The simultaneous existence of over- and under-testing cannot easily be explained by incentives alone, and instead suggests errors. We provide suggestive evidence on the psychology behind these errors:(i) physicians use too simple a model of risk, suggesting bounded rationality; (ii) they over-weight salient information; and (iii) they over-weight symptoms that are representative or stereotypical of heart attack. Together, these results suggest the need for health care models and policies to incorporate not just physician incentives, but also physician mistakes.

Abstract We document two robust features of the cross-sectional distribution of usual weekly hours and hourly wages. First, usual weekly hours are heavily concentrated around 40 hours, while at the same time a substantial share of total hours come from individuals who work more than 50 hours. Second, mean hourly wages are nonmonotonic across the usual hours distribution, with a peak at 50 hours. We develop and estimate a model of labor supply to account for these features. The novel feature of our model is that earnings are nonlinear in hours, with the extent of nonlinearity varying over the hours distribution. Our estimates imply significant wage penalties for people who deviate from 40 hours in either direction, leading to a large mass of people who work 40 hours and are not very responsive to shocks. This has important implications for the role of labor supply as a mechanism for self-insurance in a standard heterogeneous-agent incomplete-markets model and for empirical strategies designed to estimate labor supply parameters.

Abstract This article uses administrative tax data to estimate top wealth in the United States. We assemble new data that link people to their sources of capital income and develop new methods to estimate the degree of return heterogeneity within asset classes. Disaggregated fixed-income data reveal that rich individuals earn much more of their interest income in higher-yielding forms and have much greater exposure to credit risk. Consequently, in recent years, the interest rate on fixed income at the top is approximately 3.5 times higher than the average. We value the population of U.S. firms using firm-level characteristics and apportion this wealth using firm-owner links. We combine this new data on fixed income and pass-through business returns with refined estimates of C-corporation equity, housing, and pension wealth to deliver new capitalized wealth estimates that build upon the methods of Saez and Zucman (2016a). From 1989 to 2016, the top 1%, 0.1%, and 0.01% wealth shares increased by 6.6, 4.6, and 2.9 percentage points, respectively, to 33.7%, 15.7%, and 7.1%. Overall, although we estimate a large degree of return heterogeneity, accounting for this heterogeneity does not change the fundamental story for top wealth shares and their growth—wealth inequality is high and has risen substantially over recent decades.

Abstract The economic analysis of racial discrimination in public accommodations is remarkably limited. To study this issue, we construct a national data set of nondiscriminatory establishments from the Negro Motorist Green Books, a travel guide published from 1936 to 1966 to aid Black Americans in finding nondiscriminatory retail and service establishments. We document patterns in the geographic spread and evolution of Green Book establishments, as well as the correlates of Green Book presence. We find that economic and social measures, as well as state laws relating to racial discrimination and antidiscrimination, were correlated with the provision of nondiscriminatory services. We then use the Green Book data to test whether market conditions and white consumer discrimination led businesses to bar Black customers prior to the Civil Rights Act of 1964. We use plausibly exogenous variation from white World War II casualties and Black migration patterns to isolate the effect of a change in the racial composition of consumers on the growth of nondiscriminatory businesses. We find that the share of nondiscriminatory establishments grew faster in locations with larger increases in the share of the Black population, but the magnitudes were small. These results highlight the importance of federal legislation in ending racial discrimination in public accommodations.