Comment from Thornton, Robert EEOC-2026-0034

Robert J. Thornton, Professor Emeritus of Economics, Lehigh University. Institutional affiliation is provided for identification; these views are my own. I study labor economics and have used the EEO-1 aggregate files in published research. I comment on a narrow question the NPRM does not take up: what happens to the data series if this collection ends. The short answer is that it ends too, and nothing replaces it. The EEO-1 file is a near-census of large-employer employment, disaggregated by establishment, occupational category, race and ethnicity, and sex, running continuously since 1966. The usual alternatives are not substitutes. CPS and ACS sample households and cannot be linked to the employer; LEHD and QWI are built from administrative wage records and carry no occupational detail of the sort the EEO-1 provides; industry surveys are voluntary, unrepresentative, and discontinuous. To my knowledge there is no other source that permits within-firm, within-occupation comparison at national scale, and I am not aware of any proposal to construct one. I want to flag a distinction the regulatory analysis collapses. Terminating a series and interrupting a series are different costs, and the second is larger than it looks. Panel methods depend on continuity. A multi-year gap does not merely remove the years in the gap; it degrades identification for periods on both sides of it, because researchers lose the ability to distinguish firm-level change from composition change across the break. If the Commission is weighing this at all, and Section V.A.5 indicates it is not, the discontinuity should be costed separately from the ongoing collection expense. On the substantive value: the EEO-1 aggregates underpin a sizable empirical literature on occupational segregation, on the effects of enforcement activity on workforce composition, and on how local labor markets absorb firm entry and exit. Kalev, Dobbin, and Kelly (2006) is the canonical example of work that would have been impossible without it, and there is a steady flow of newer work in the same vein. The Commission's own Job Patterns for Minorities and Women in Private Industry, published for over four decades, is a public statistical product with a user base well outside the agency. None of this appears in the analysis. One further observation, offered in the spirit of taking the Commission's stated concern seriously. The worry expressed in the preamble is that regulated parties may misread statistical imbalance as evidence of liability and respond badly. That is a concern about interpretation and dissemination. It is not a concern about whether the underlying data exists. If the Commission believes published aggregates invite bad inferences, the proportionate instruments are interpretive guidance, revised cell-suppression and disclosure rules, or a restricted-access arrangement under which the data remain available to qualified researchers under license while public release is curtailed. Any of these would address the stated problem at something closer to its actual scale. The NPRM considers none of them, because it considers no alternatives at all. I would ask the Commission to do one thing before finalizing, whatever it decides on the merits: consult the Bureau of Labor Statistics, the Census Bureau, and OMB's Statistical and Science Policy branch on whether any component of this series warrants preservation or transfer to another agency, and place the results of that consultation in the docket. A sixty-year federal statistical series should not be discontinued without the federal statistical system having been asked whether it should be. I also join the requests for a ninety-day comment period. Robert J. Thornton