How to Read an Age–Earnings Profile Without Being Misled
College-major age–earnings profiles are the subject of this final synthesis, and the central question is whether the apparent decline of engineering and computer-science earnings with age describes a real within-career process. The evidence supports a direct but bounded answer: in the long linked data, technical majors do not show slower relative earnings growth than the broad comparison group in the paper’s main specifications. The useful judgment is not to replace one slogan with another, but to understand how the people observed, the cohort they belong to, the earnings being counted, and the model used together determine what an age curve means.
The profile is a map of comparisons, not a film of careers
The apparent decline begins with a legitimate observation. In the short American Community Survey cross-section, engineering and computer science have high relative earnings near labor-market entry, but their estimated advantage narrows with age. In the specification that reproduces the earlier result, the technical profile falls by 0.167 log points between age 23 and ages 49–50 relative to the broad other-major group. The finding is real as a description of that cross-sectional comparison. The mistake would be to treat it as a complete record of what happens inside a typical technical career.
The comparison changes when the data design changes. Restricting the ACS to people who can be matched to the Longitudinal Employer–Household Dynamics data makes the age-49–50 estimate more negative, at −0.216, while keeping the observation essentially a one-year snapshot. Replacing self-reported ACS earnings with LEHD earnings for the same survey-year comparison produces −0.165. Those two steps show that neither linkability and geography alone nor the switch from survey earnings to unemployment-insurance earnings explains the main reversal.
The decisive movement begins when the full earnings histories are used. The linked workers contribute available person-year earnings rather than only the year in which the ACS recorded them. The technical estimate at ages 49–50 shrinks to −0.032 without worker fixed effects and becomes +0.071 in the baseline worker-fixed-effects specification. With age-specific graduate-degree profiles and labor-supply controls added, the technical values at ages 29–30, 39–40, and 49–50 are −0.027, +0.004, and +0.022. These estimates are close to zero or positive relative profiles, consistent with technical wage growth being on par with the reference group rather than with a large technical-major decay.
The important result is the path across those estimates. Additional years per worker reduce the need to place different cohorts at different ages. Worker fixed effects then absorb a persistent part of each worker’s earnings level, making the remaining comparison more about how earnings change over time for workers whose majors are observed. The controls separate some education and work-attachment pathways from the annual total. Each change answers a different concern, so the final result should not be presented as though one statistical switch magically corrected everything that came before it.
The cohort mechanism ties the pieces together. In a short recent survey, younger technical graduates tend to come from later cohorts and older technical graduates from earlier cohorts. If later cohorts receive a larger technical premium at entry, the younger observations start higher on the chart. A line connecting them to older, lower-premium cohorts slopes downward even if each cohort’s own earnings profile is flat or rising relative to other majors. Age is not meaningless in that line; it is carrying information about who entered the labor market when and with what starting position.