The Stata Journal Editors’ Prize 2021: Mark E. Schaffer

1 Prize announcement

The editors of the Stata Journal are delighted to announce the award of the Editors’ Prize for 2021 to Mark E. Schaffer.

The aim of the prize is to reward contributions to the Stata community in respect of one or more outstanding articles published in the Journal in the previous three calendar years. For the original announcement of the prize and its precise terms of reference, see Newton and Cox (2012), which is accessible at the following website: http://www.stata-journal.com/sjpdf.html?articlenum=gn0052.

Mark Edwin Schaffer was born in 1959 and grew up in New Jersey and Arizona. He received an AB degree in social studies magna cum laude from Harvard University in 1982 and master’s and PhD degrees in economics from the London School of Economics (LSE), Stanford, and LSE again. After teaching and research posts at Sussex and LSE, Schaffer joined Heriot-Watt University in Edinburgh in 1995. He has since occupied many senior management, research, and consultancy roles at Heriot-Watt, across consortia of Scottish universities, and internationally. Honors include election as a Fellow of the Royal Society of Edinburgh in 2009 and as a Fellow of the Royal Society of Arts in 2017.

Schaffer’s research ranges widely, with interests in applied econometrics, labor markets, economic history, energy economics, and evolutionary biology. He has acted as an Associate Editor of the Stata Journal since 2003 and has given talks at various Stata meetings.

The award recognizes specifically an outstanding article by Schaffer and coauthors Achim Ahrens and Christian B. Hansen (2020)

lassopack: Model selection and prediction with regularized regression in Stata. Stata Journal 20: 176–235.

This is the most recent example of Schaffer’s output in what is a strong history of publication of major articles and Stata programs. In particular, a series of articles with a theme of instrumental variables estimation with coauthors Christopher F. Baum and Steven Stillman (2003, 2004, 2005, 2007) remain heavily cited and influential. According to Google Scholar, the 2003 and 2007 articles have been cited over 4,000 times. These Stata Journal publications understate Schaffer’s Stata productivity: he is author or coauthor of multiple programs on the Statistical Software Components (SSC) Archive, principally in the areas of regression analysis and panel-data analysis as well as instrumental variables estimation. Schaffer is one of the most downloaded authors of Stata work at the SSC Archive, with hits of the order of 50,000 per month. For sample results, type in Stata ssc hot, author(schaffer) n(30) to see the latest monthly numbers.

We turn now to detailed comments on the article behind this award.

The article introduces lassopack, a suite of programs for regularized regression in Stata. The programs implement lasso, square-root lasso, elastic net, ridge regression, adaptive lasso, and postestimation ordinary least squares. The methods are suitable for the high-dimensional setting, where the number of predictors, p, may be large and possibly greater than the number of observations, n. Three approaches are offered for selecting the penalization (or tuning) parameters: information criteria (implemented in lasso2); K-fold cross-validation and h-step-ahead rolling cross-validation for crosssection, panel, and time-series data (cvlasso); and theory-driven (“rigorous” or plugin) penalization for the lasso and square-root lasso for cross-section and panel data (rlasso). The article discusses the theoretical framework and practical considerations for each approach and also presents Monte Carlo results to compare the performances of the penalization approaches.

This article is an exemplary blend with several strands: high-quality exposition, original results, and empirical illustrations showing how to use the programs. At the same time lassopack was published on the SSC Archive, the authors released an accompanying package, pdslasso, that integrates these high-dimensional estimators with methods for causal inference using regression and instrumental variables methods. Since the article was submitted, the lassopack suite has been extended to support logistic lasso regression, through lassologit. In the same vein, the authors have recently released the pystacked package. pystacked implements “stacking regression” via Stata’s Python integration and scikit-learn’s (Pedregosa et al. 2011) sklearn.ensemble.StackingRegressor and sklearn.ensemble.StackingClassifier. Stacking is a way of combining predictions from multiple supervised machine learners into a final prediction to improve performance.

We note that the lassopack article was accepted for publication, and lassopack released on the SSC Archive, before the release of Stata 16, which introduced lasso support. But even though official Stata has now added lasso and related routines, lassopack and pdslasso offer much extra functionality and perhaps may serve as a stimulus for further work in the territory between mainstream statistics and machine learning.

In sum, we salute Mark Schaffer for outstanding contributions to the Stata community, particularly through his publications in the Stata Journal and his many programs on the SSC Archive.

As editors, we are indebted to the awardee for biographical material and to necessarily anonymous nominators for most helpful appreciations.

H. Joseph Newton and Nicholas J. Cox Editors, Stata Journal