Knowledge that Transforms

Abstract This paper theoretically and empirically examines the role of information in pharmaceutical detailing (promotional interactions between drug representatives and physicians). We start with a theoretical framework in which pharmaceutical firms target detailing visits to physicians who learn about drug quality and prescribe accordingly. We derive several predictions about the role of information in these visits, which we then test empirically using prescriptions and pharmaceutical detailing visit data. We find limited empirical support of learning as the dominant mechanism, though cannot rule it out completely. We conclude with discussing alternative models that may be more consistent with the observed empirical patterns.

Abstract Standard estimators in information provision experiments place more weight on individuals who update their beliefs more in response to new information. This paper shows that, in practice, these individuals who update the most have the weakest causal e!ects of beliefs on outcomes. Standard estimators therefore understate these causal e!ects. I propose an alternative local least squares (LLS) estimator that recovers a representative unweighted average e!ect in a broad class of learning rate models that generalize Bayesian updating. In five of six recent studies, estimates of the e!ects of beliefs on outcomes increase. In two, they more than double.

Abstract We introduce a notion of money-metric social welfare for discrete choice under unrestricted heterogeneity and income effects. It is the maximized indirect utility under normalization of the outside option. It also equals the amount of income necessary to achieve a given level of utility, while certain choices are prohibited. We show that the distribution of this quantity is nonparametrically identified as a closed-form functional of average structural demand for the outside option, making it useful for cost-benefit analysis and optimal targeting. An illustration with private tuition subsidies in India shows that the income path of usage-maximizing subsidies differs significantly from welfare-maximizing ones.

Abstract Global phenomena, such as climate change, often have local impacts that are spatially correlated. We show that greater spatial correlation of productivities can increase international inequality by increasing the correlation between a country's productivity and its gains from trade. We confirm this prediction using a half-century of exogenous variation in the spatial correlation of agricultural productivities induced by a global climatic phenomenon. We introduce this general-equilibrium effect into projections of climate-change impacts that typically omit spatial linkages and therefore do not account for the global scope of climate change. We project greater international inequality, with higher welfare losses across Africa.

Abstract We examine the effect of investments by large technology firms on innovation by AI startups. Large technology firms have substantial advantages in data, a key non-rival input for developing AI technology. We assemble a unique dataset containing approximately 9,800 AI-inventing startups in China. Using a triple-differences strategy, we find that AI startups receiving investments from large technology firms file more AI patent applications and develop more software products following the investment, relative to their counterparts receiving investments from other firms without data advantages. Finally, we provide novel evidence that the innovation effect works mainly through sharing non-rival data.

Abstract We investigate the causal effects of financial incentives supporting geographical mobility among unemployed workers on their job search behavior and labor market outcomes. Exploiting regional variation in the promotion of mobility programs along administrative borders of German employment agency districts, we show that promoting mobility—as intended—causes job seekers to increase their search radius, apply for, and accept distant jobs. At the same time, local job search is reduced with adverse consequences for reemployment and earnings. A detailed analysis of the underlying mechanisms suggests spatial search frictions as the driver of the unintended adverse labor market effects.

Abstract I partially identify the marginal treatment effect (MTE) when the treatment is misclassified. I explore two restrictions, allowing for dependence between the instrument and the misclassification decision. If the signs of the propensity scores’ derivatives are equal, I identify the MTE sign. If those derivatives are similar, I bound the MTE. To illustrate, I analyze the impact of alternative sentences (fines and community service versus no punishment) on recidivism in Brazil, where court appeals processes generate misclassification. The estimated misclassification bias may be as large as 10% of the largest possible MTE, and the bounds contain the correctly estimated MTE.

Abstract In markets with asymmetric information between sellers and buyers, feedback mechanisms are important to increase market efficiency and reduce the informational disadvantage of buyers. Feedback mechanisms might work because of self-selection of more trustworthy sellers into markets with such mechanisms or because of reputational concerns of sellers. We show in a field experiment how to disentangle self-selection from reputation effects. Based on 476 taxi rides with four different types of taxis, we find strong evidence for reputation effects but little support for self-selection effects. We discuss policy implications of our findings.

Abstract We provide the first estimates of intergenerational income mobility using tax data for a large developing country, namely Brazil. We measure formal income from tax and payroll data, and we train machine learning models on census and survey data to predict informal income. We quantify the estimation bias resulting from income imputation and other sources of measurement error, and show that such bias remains negligible in our context. A 10 percentile increase in parental income rank is associated on average with a 5.5 percentile increase in child income rank, with considerable variation across sociodemographic groups and geographical areas.

Abstract Predictive algorithms inform consequential decisions in settings with selective labels: outcomes are observed only for units selected by past decision makers. This creates an identification problem under unobserved confounding — when selected and unselected units differ in unobserved ways that affect outcomes. We propose a framework for robust design and evaluation of predictive algorithms that bounds how much outcomes may differ between selected and unselected units with the same observed characteristics. These bounds formalize common empirical strategies including proxy outcomes and instrumental variables. Our estimators work across bounding strategies and performance measures such as conditional likelihoods, mean squared error, and true/false positive rates. Using administrative data from a large Australian financial institution, we show that varying confounding assumptions substantially affects credit risk predictions and fairness evaluations across income groups.