We analyze over 44,000 NBER and CEPR working papers from 1980 to 2023 using a custom language model to construct knowledge graphs that map economic concepts and their relationships. We distinguish between general claims and those documented via causal inference methods (e.g., DiD, IV, RDD, RCTs). We document a substantial rise in the share of causal claims-from roughly 4% in 1990 to nearly 28% in 2020-reflecting the growing influence of the "credibility revolution." We find that causal narrative complexity (e.g., the depth of causal chains) strongly predicts both publication in top-5 journals and higher citation counts, whereas non-causal complexity tends to be uncorrelated or negatively associated with these outcomes. Novelty is also pivotal for top-5 publication, but only when grounded in credible causal methods: introducing genuinely new causal edges or paths markedly increases both the likelihood of acceptance at leading outlets and long-run citations, while non-causal novelty exhibits weak or even negative effects. Papers engaging with central, widely recognized concepts tend to attract more citations, highlighting a divergence between factors driving publication success and long-term academic impact. Finally, bridging underexplored concept pairs is rewarded primarily when grounded in causal methods, yet such gap filling exhibits no consistent link with future citations. Overall, our findings suggest that methodological rigor and causal innovation are key drivers of academic recognition, but sustained impact may require balancing novel contributions with conceptual integration into established economic discourse.

Generative AI is transforming higher education, yet systematic evidence on student adoption remains limited. Using novel survey data from a selective U.S. college, we document over 80 percent of students using AI academically within two years of ChatGPT's release. Adoption varies across disciplines, demographics, and achievement levels, highlighting AI's potential to reshape educational inequalities. Students predominantly use AI for augmenting learning (e.g., explanations, feedback), but also to automate tasks (e.g., essay generation). Positive perceptions of AI's educational benefits strongly predict adoption. Institutional policies can influence usage patterns but risk creating unintended disparate impacts across student groups due to uneven compliance.

The COVID-19 pandemic caused widespread disruptions to education, with school closures affecting over one billion children. These closures, aimed at reducing virus transmission, resulted in significant learning losses, particularly in mathematics and science. Using data from TIMSS 2023, which assesses fourth and eighth-grade achievements across 71 education systems, this study analyzes the impact of school closure duration on learning outcomes. Mixed-effect models estimate deviations from pre-pandemic trends, adjusting for demographic factors. Results show a global average decline of 0.11 standard deviations (SD) in student achievement, with longer closures linked to more severe losses. The effects on low performers, girls, and linguistic minorities show effect sizes up to 0.22 SD. These findings highlight the lasting impact of school closures and emphasize the need for targeted recovery strategies and international cooperation to promote equitable educational outcomes post-pandemic.

Decision-making plays a pivotal role in shaping outcomes in various disciplines, such as medicine, economics, and business. This paper provides guidance to practitioners on how to implement a decision tree designed to address treatment assignment policies using an interpretable and non-parametric algorithm. Our Policy Tree is motivated on the method proposed by Zhou, Athey, and Wager (2023), distinguishing itself for the policy score calculation, incorporating constraints, and handling categorical and continuous variables. We demonstrate the usage of the Policy Tree for multiple, discrete treatments on data sets from different fields. The Policy Tree is available in Python's open-source package mcf (Modified Causal Forest).

Analysis of effect heterogeneity at the group level is standard practice in empirical treatment evaluation research. However, treatments analyzed are often aggregates of multiple underlying treatments which are themselves heterogeneous, e.g. different modules of a training program or varying exposures. In these settings, conventional approaches such as comparing (adjusted) differences-in-means across groups can produce misleading conclusions when underlying treatment propensities differ systematically between groups. This paper develops a novel decomposition framework that disentangles contributions of effect heterogeneity and qualitatively distinct components of treatment heterogeneity to observed group-level differences. We propose semiparametric debiased machine learning estimators that are robust to complex treatments and limited overlap. We revisit a widely documented gender gap in training returns of an active labor market policy. The decomposition reveals that it is almost entirely driven by women being treated differently than men and not by heterogeneous returns from identical treatments. In particular, women are disproportionately targeted towards vocational training tracks with lower unconditional returns.

Until the mid 1960s, the UK experienced regular measles epidemics, with the vast majority of children being infected in early childhood. The introduction of a measles vaccine substantially reduced its incidence. The first part of this paper examines the long-term human capital and health effects of this change in the early childhood disease environment. The second part investigates interactions between the vaccination campaign and individuals' endowments as captured using molecular genetic data, shedding light on complementarities between public health investments and individual endowments. We use two identification approaches, based on the nationwide introduction of the vaccine in 1968 and local vaccination trials in 1966. Our results show that exposure to the vaccination in early childhood positively affects adult height, but only among those with high genetic endowments for height. We find no effects on years of education; neither a direct effect, nor evidence of complementarities.

In this paper we develop a new machine learning estimator for ordered choice models based on the random forest. The proposed Ordered Forest flexibly estimates the conditional choice probabilities while taking the ordering information explicitly into account. In addition to common machine learning estimators, it enables the estimation of marginal effects as well as conducting inference and thus provides the same output as classical econometric estimators. An extensive simulation study reveals a good predictive performance, particularly in settings with non-linearities and near-multicollinearity. An empirical application contrasts the estimation of marginal effects and their standard errors with an ordered logit model. A software implementation of the Ordered Forest is provided both in R and Python in the package orf available on CRAN and PyPI, respectively.

This study evaluates the macroeconomic effects of active labour market policies (ALMP) in Germany over the period 2005 to 2018. We propose a novel identification strategy to overcome the simultaneity of ALMP and labour market outcomes at the regional level. It exploits the imperfect overlap of local labour markets and local employment agencies that decide on the local implementation of policies. Specifically, we instrument for the use of ALMP in a local labour market with the mix of ALMP implemented outside this market but in local employment agencies that partially overlap with this market. We find no effects of short-term activation measures and further vocational training on aggregate labour market outcomes. In contrast, wage subsidies substantially increase the share of workers in unsubsidised employment while lowering long-term unemployment and welfare dependency. Our results suggest that negative externalities of ALMP partially offset the effects for program participants and that some segments of the labour market benefit more than others.

Estimators that weight observed outcomes to form effect estimates have a long tradition. Their outcome weights are widely used in established procedures, such as checking covariate balance, characterizing target populations, or detecting and managing extreme weights. This paper introduces a general framework for deriving such outcome weights. It establishes when and how numerical equivalence between an original estimator representation as moment condition and a unique weighted representation can be obtained. The framework is applied to derive novel outcome weights for the six seminal instances of double machine learning and generalized random forests, while recovering existing results for other estimators as special cases. The analysis highlights that implementation choices determine (i) the availability of outcome weights and (ii) their properties. Notably, standard implementations of partially linear regression-based estimators, like causal forests, employ outcome weights that do not sum to (minus) one in the (un)treated group, not fulfilling a property often considered desirable.

Despite the widespread use of graphs in empirical research, little is known about readers' ability to process the statistical information they are meant to convey ("visual inference"). We study visual inference within the context of regression discontinuity (RD) designs by measuring how accurately readers identify discontinuities in graphs produced from data generating processes calibrated on 11 published papers from leading economics journals. First, we assess the effects of different graphical representation methods on visual inference using randomized experiments. We find that bin widths and fit lines have the largest impacts on whether participants correctly perceive the presence or absence of a discontinuity. Our experimental results allow us to make evidence-based recommendations to practitioners, and we suggest using small bins with no fit lines as a starting point to construct RD graphs. Second, we compare visual inference on graphs constructed using our preferred method with widely used econometric inference procedures. We find that visual inference achieves similar or lower type I error (false positive) rates and complements econometric inference.

Using administrative data on all induced abortions recorded in Spain in 2019, we analyze the characteristics of women undergoing repeat abortions and the spacing between these procedures. Our findings indicate that compared to women experiencing their first abortion, those who undergo repeat abortions are more likely to have lower education levels, have dependent children, live alone, or be foreign-born, with a non-monotonic relationship with age. We also report that being less educated, not employed, having dependent children, or being foreign-born are all strongly related to a higher number of repeat abortions. Lastly, we find that being less educated, foreign-born, or not employed is correlated with a shorter time interval between the last two abortions.

Fairness and interpretability play an important role in the adoption of decision-making algorithms across many application domains. These requirements are intended to avoid undesirable group differences and to alleviate concerns related to transparency. This paper proposes a framework that integrates fairness and interpretability into algorithmic decision making by combining data transformation with policy trees, a class of interpretable policy functions. The approach is based on pre-processing the data to remove dependencies between sensitive attributes and decision-relevant features, followed by a tree-based optimization to obtain the policy. Since data pre-processing compromises interpretability, an additional transformation maps the parameters of the resulting tree back to the original feature space. This procedure enhances fairness by yielding policy allocations that are pairwise independent of sensitive attributes, without sacrificing interpretability. Using administrative data from Switzerland to analyze the allocation of unemployed individuals to active labor market programs (ALMP), the framework is shown to perform well in a realistic policy setting. Effects of integrating fairness and interpretability constraints are measured through the change in expected employment outcomes. The results indicate that, for this particular application, fairness can be substantially improved at relatively low cost.

Understanding the processes that drive galaxy formation and shape the observed properties of galaxies is one of the most interesting and challenging frontier problems of modern astrophysics. We now know that the evolution of galaxies is critically shaped by the energy injection from accreting supermassive black holes (SMBHs). However, it is unclear how exactly the physics of this feedback process affects galaxy formation and evolution. In particular, a major challenge is unraveling how the energy released near the SMBHs is distributed over nine orders of magnitude in distance throughout galaxies and their immediate environments. The best place to study the impact of SMBH feedback is in the hot atmospheres of massive galaxies, groups, and galaxy clusters, which host the most massive black holes in the Universe, and where we can directly image the impact of black holes on their surroundings. We identify critical questions and potential measurements that will likely transform our understanding of the physics of SMBH feedback and how it shapes galaxies, through detailed measurements of (i) the thermodynamic and velocity fluctuations in the intracluster medium (ICM) as well as (ii) the composition of the bubbles inflated by SMBHs in the centers of galaxy clusters, and their influence on the cluster gas and galaxy growth, using the next generation of high spectral and spatial resolution X-ray and microwave telescopes.

The Coase Theorem has a central place in the theory of environmental economics and regulation. But its applicability for solving real-world externality problems remains debated. In this paper, we first place this seminal contribution in its historical context. We then survey the experimental literature that has tested the importance of the many, often tacit assumptions in the Coase Theorem in the laboratory. We discuss a selection of applications of the Coase Theorem to actual environmental problems, distinguishing between situations in which the polluter or the pollutee pays. While limited in scope, Coasian bargaining over externalities offers a pragmatic solution to problems that are difficult to solve in any other way.

In their IZA Discussion Paper 10247, Johansson and Lee claim that the main result (Proposition 3) in Abbring and Van den Berg (2003b) does not hold. We show that their claim is incorrect. At a certain point within their line of reasoning, they make a rather basic error while transforming one random variable into another random variable, and this leads them to draw incorrect conclusions. As a result, their paper can be discarded.

Binary treatments are often ex-post aggregates of multiple treatments or can be disaggregated into multiple treatment versions. Thus, effects can be heterogeneous due to either effect or treatment heterogeneity. We propose a decomposition method that uncovers masked heterogeneity, avoids spurious discoveries, and evaluates treatment assignment quality. The estimation and inference procedure based on double/debiased machine learning allows for high-dimensional confounding, many treatments and extreme propensity scores. Our applications suggest that heterogeneous effects of smoking on birthweight are partially due to different smoking intensities and that gender gaps in Job Corps effectiveness are largely explained by differential selection into vocational training.

Surveys are an indispensable source of data for applied economic research; however, their reliance on self-reported information can introduce bias, especially if core variables such as personal income are misreported. To assess the extent and impact of this misreporting bias, we compare self-reported wages from the German Socio-Economic Panel (SOEP) with administrative wages from social security records (IEB) for the same individuals. Using a novel and unique data linkage (SOEP-ADIAB), we identify a modest but economically significant reporting bias, with SOEP respondents underreporting their administrative wages by about 7.3%. This misreporting varies systematically with individual, household, and especially job and firm characteristics. In replicating common empirical analyses in which wages serve as either dependent or independent variables, we find that misreporting is consequential for some, but not all estimated relationships. It turns out to be inconsequential for examining the returns to education, but relevant for analyzing the gender wage gap. In addition we find that misreporting bias can significantly affect the results when wage is used as the independent variable. Specifically, estimates of the wage-satisfaction relationship are substantially overestimated when based on survey data, although this bias is mitigated when focusing on interpersonal changes. Our findings underscore that survey-based measures of individual wages can significantly bias commonly estimated empirical relationships. They also demonstrate the enormous research potential of linked administrative-survey data.

We study how the diffusion of broadband Internet affects social capital using two data sets from the UK. Our empirical strategy exploits the fact that broadband access has long depended on customers' position in the voice telecommunication infrastructure that was designed in the 1930s. The actual speed of an Internet connection, in fact, rapidly decays with the distance of the dwelling from the specific node of the network serving its area. Merging unique information about the topology of the voice network with geocoded longitudinal data about individual social capital, we show that access to broadband Internet caused a significant decline in forms of offline interaction and civic engagement. Overall, our results suggest that broadband penetration substantially crowded out several aspects of social capital.