Edward Conard

Top Ten New York Times Bestselling Author

  • “A full-throated defense of economic dynamism.” - The Wall Street Journal
  • “…a very valuable contribution.” - Larry Summers, former Secretary of the Treasury and director of the National Economic Council, president emeritus, Harvard University
  • “…challenges misconceptions that distort our economic debates.” - Arthur Brooks, President of the American Enterprise Institute
  • “…a fresh argument for the productive value of inequality.” - David Autor, Professor of Economics, Massachusetts Institute of Technology
  • “…a fresh argument for the productive value of inequality.” - David Autor, Professor of Economics, Massachusetts Institute of Technology
  • “…reminds us that inequality sends a signal of what society lacks most, in America’s case, entrepreneurship and risk taking.” - Lawrence Lindsey, CEO, The Lindsey Group, former Director of the National Economic Council
  • “…reminds us that inequality sends a signal of what society lacks most, in America’s case, entrepreneurship and risk taking.” - Lawrence Lindsey, CEO, The Lindsey Group, former Director of the National Economic Council
  • “…serious thinking for serious thinkers. …a thought-provoking blueprint for growing middle- and working-class incomes.” - Mitt Romney, former Governor of Massachusetts
  • “There are an amazing number of good ideas and interesting points made in Unintended Consequences. The thinking underlying it, and the obvious depth of understanding of the author, are very impressive.” - Steven Levitt, coauthor of Freakonomics; 2004 John Bates Clark Medal
  • “Unintended Consequences should be read by anyone who takes for granted the superiority of progressive taxation and has not thought carefully about the trade-offs involved.” - The New Republic
  • “Unintended Consequences is far smarter and more thought-provoking than most economics written for the general public” - Greg Mankiw, Harvard University, Former Chairman of the Council of Economic Advisors
  • “…a comprehensive explanation of the modern economy.” - Julian Robertson, Founder, Tiger Management
Upside of Inequality Oxford Unintended Consequences

AI Summary. Using anonymized federal tax records linked to workforce training participants, the authors build a machine-learning estimator whose earnings-impact estimates correlate 0.93 with those from 37 randomized trials, with error comparable to the trials themselves. They conclude such observational estimators can complement, but not replace, randomized experiments, which may serve as calibration tools.

Raj Chetty, Jamie Fogel, Lawrence Katz, Kadeem Noray, et al. Harvard University
Date Posted:

Chetty et al. propose an alternative to expensive randomized controlled trials (RCTs). Their machine-learning model uses nonparticipant data to predict participant earnings in the absence of the program. Its treatment-effect estimates correlate 0.93 with those from 37 RCTs. This method may facilitate more rapid and extensive policy evaluation.

Can machine learning on tax records match randomized trials' accuracy?

Core argument: Chetty, Katz, and co-authors report a double/debiased machine learning estimator, built on tax records for 327 million U.S. adults, correlates 0.93 with estimates from 37 RCTs and has an RMSE of $2,073 versus $2,126 for the RCT estimates themselves.

We train gradient-boosted trees [a sequence of branching prediction models, each correcting errors left by the earlier model] on the population of non-participants; [e.g., were prior earnings below $20,000? If yes, follow one branch; if no, follow another] to predict future earnings based on a rich set of pre-program individual characteristics: work histories, demographics, and residential locations. We then estimate treatment effects by comparing participants’ realized outcomes to the predicted counterfactuals. Finally, we combine the outcome and propensity score models using the double/debiased machine learning (DML) estimator. We use 37 randomized controlled trials (RCTs) to estimate the distribution of their bias. We use the term “digital twins” for an estimator because it yields counterfactuals — statistical twins constructed from data on non-participants — as accurate as those one would obtain from random assignment. Figure 2 plots the RCT estimates against the DML estimates. The estimates are clustered around the 45-degree line, with an unweighted correlation of 0.93. Our method reduces RMSE by 60% relative to parametric estimators used in prior work.

Takeaways by Macro Roundup® AI

  1. Chetty, Katz, and co-authors report a double/debiased machine learning estimator, built on tax records for 327 million U.S. adults, correlates 0.93 with estimates from 37 RCTs and has an RMSE of $2,073 versus $2,126 for the RCT estimates themselves.
  2. The estimator’s noise-adjusted RMSE attributable to bias is $1,717, against $4,485 for LaLonde (1986) and $3,960 for Dehejia and Wahba (1999), with gains attributed to richer covariates, larger comparison samples, and more flexible gradient-boosted tree models.
  3. For Year Up United, estimated average annual earnings impacts rose from $5,198 for the 2007 enrollment cohort to $12,000 for 2017 before falling to $9,077 for 2022, with equal 2007 and 2017 effects rejected at p = 0.01.

Related Articles:

  • Income Shocks and Intrahousehold Dynamics: Evidence from a Guaranteed Income Experiment — In a guaranteed income experiment, the gap between participant income and total household income fell by roughly $1,700 per year, which the authors attribute to lower earnings among other household members. Net transfers to extended family rose by about $135 per year, while measures of household stability, decision-making, and division of labor clustered near zero.
  • Does Generative AI Narrow Education-Based Productivity Gaps? Evidence from a Randomized Experiment — In a randomized experiment on a realistic business analysis task, participants without an AI assistant showed an education gap, with high-education participants scoring 0.548 standard deviations higher overall. Access to the AI assistant raised scores by 1.242 standard deviations for low-education participants versus 0.834 for high-education participants.
  • Can Gifted Education Help Higher-Ability Boys from Disadvantaged Backgrounds? — A regression discontinuity study of a large urban school district, where disadvantaged students qualify for gifted status at a lower IQ threshold, estimates that being identified as gifted raises the college entry rate of marginally eligible boys by 28 percentage points, from 46% to 74%. That matches the 74% rate of marginally ineligible girls, closing the gender gap.
Showing 785 database articles primarily about either Workforce, Demographics, Education, Family/Marriage, Gender Pay Gap, Immigration, Inequality, Minimum Wage, Poverty/Crime, Unemployment/Participation, or Wages/Income
Currently filtering by: