MODELS OF LINKED EMPLOYER-EMPLOYEE DATA: CONFERENCE

Project Goal

In their 1999 paper, John Abowd, Francis Kramarz, and David Margolis introduced what is arguably the most influential approach to the analysis of linked employer-employee data, now commonly referred to as the AKM model. Estimates of their model describe how much of a worker’s wage is determined by their own characteristics (“worker effects”), and how much is determined by where they work (“firm effects”). This model has become influential both because it is conceptually simple, and because it yields empirical results inconsistent with existing theory. Empirical studies based on the AKM model consistently find that firms play a very large role in wage-setting, and that high-wage workers are not concentrated in high-paying firms. Both results are difficult to square with standard competitive models of the labor market.

Only recently has linked data become sufficiently available to foster a broad academic literature. The last five years have seen the rapid, widespread application of the AKM model to many of the fundamental topics in labor economics: labor market inequality (Card et al., 2013), gender disparities (Barth et al., 2017; Card et al., 2016), the value of workplace amenities (Lavetti, Kurt and Schmutte, 2018), the minimum wage (Engbom and Moser, 2017), immigration (Pendakur and Woodcock, 2010), the costs of job displacement (Lachowska et al., 2018), social networks (Schmutte, 2015), and outsourcing (Goldschmidt and Schmieder, 2017). The often surprising, but persuasive, results from the AKM model have had just as much influence outside labor economics in such diverse fields as macroeconomics (Eeckhout and Kircher, 2011; Shimer, 2005), finance (Babina et al., 2018), and management.

The twenty years since publication of the original AKM paper have brought many refinements of their original approach. Nevertheless, considerable questions remain regarding how the AKM model should be estimated, and how its results should be interpreted. First, the validity of the AKM model depends on strong assumptions about how employment relationships form, and how workers move between jobs. Second, it is not clear what economic models are and are not consistent with the evidence from AKM models. In particular, it is not clear whether estimated firm effects represent differences in unobserved amenities, or instead reflect the market power of firms over workers. Third, the worker and firm effects are inconsistently estimated in the AKM decomposition, reflecting broader econometric challenges to working with relational data. Fourth, different kinds of matched employer-employee data contain different information, creating some confusion about how to prepare data for analysis, and how to compare results across studies.

Purpose of Funding

The grant provided critical funding for a conference, together with funding from the Pierce Memorial Fund, the Dean, and the Labor Dynamics Institute at the Cornell ILR School, as well as from the Class of 1950 Professor of Economics Chair at the University of California at Berkeley. In-kind contributions were made by the Labor Dynamics Institute and the University of Georgia.

Expected Products

• a conference
• multiple papers submitted by conference participants to a special issue of the Journal of Econometrics
• a conference website as a record of the event
• student participation in the planning and execution of the conference
• student participation in the subsequent special issue of the journal

Outcomes

The conference was held over two days in the ILR Conference Center in New York City, with 69 participants from Europe and North America. 10 of the participants were graduate students from various universities (we sponsored travel for 8 of them). 15 papers were presented in 45 minute sessions, with assigned discussants providing comments. Each session had two graduate students assigned as note takers. A pre-conference dinner with all presenters and discussants, as well as lunch and a reception on the main conference day allowed for ample networking.

The conference website hosts the program and links to preliminary papers for those authors who wished to have their papers posted. Several authors made their papers available through a eCommons repository at Cornell (https://hdl.handle.net/1813/69482), where they will be preserved.

The call for papers at the Journal of Econometrics closed in June 2020, though some papers were accepted after the deadline due to COVID-19-related delays. We received 17 submissions, of which 3 were not from conference presenters. Most of the graduate students who had participated as note takers at the conference were among the referees of the submitted manuscripts.

The journal issue was published in April 2023 (Journal of Econometrics, Volume 233, Issue 2), with 15 articles, and an introduction by the guest editors David Card, Ian Schmutte, and Lars Vilhuber. The following are the articles published (also see on Zotero).