Texas graduating cohorts: worse in-state first employment and wages for exposed majors

Samuel Dodini and Tucker Smith, “AI plays a role in weak labor market for college graduates,” Dallas Fed Economics, 22 September 2026. Authored analysis of Texas public-four-year-university enrollment and graduate records matched to Texas earnings files. They map college majors to pre-ChatGPT job advertisements and classify likely occupations using Anthropic’s task exposure measure, reducing reverse causation from job ads after AI’s release. Their event-study comparison says a 10-percentage-point increase in major exposure corresponds to a 1.7-percentage-point decline in probability of employment within Texas during the first year, and ~5% lower earnings conditional on in-state employment among recent graduates relative to less-exposed majors. May 2024 exposed graduates were more likely (+1.4 points per 10 points exposure) to enroll in graduate school, often in their original major; undergraduate enrollment later shifted away from exposed majors by fall 2025.

The authors conclude AI affected graduates but the research design compares majors, not verified adopters against nonadopters. In-state employer records cannot show a job found across the state line or outside unemployment insurance, and the pay series excludes those not observed working. Graduate school may be a response to numerous demand shocks, not a directly proven AI-mediated choice. Computer science, engineering and languages are among high-exposure majors and have other cyclical/sector shocks. These results reinforce the Census linked-record harm signal, but the different exposure contrast (10 points here, extreme decile there) means their 1.7 and 5 percentage-point employment estimates must not be treated as discrepant measures of the same effect. National CPS tests a different endpoint.