The Technical Premium Rose—So What Changed?

The technical-major profile raises a second question after the measurement correction: if engineering and computer-science earnings do not grow more slowly across the observed lifecycle, why did newer technical cohorts receive a larger premium? This chapter weighs cohort change against calendar-time change, then tests whether the result survives different samples and major classifications. The goal is not to identify one winning story, but to distinguish what the evidence supports from what remains a plausible explanation.

A premium can rise because the population changes

The paper considers three ways the cross-sectional and panel profiles might diverge. Returns to a major may differ across birth cohorts. They may change by calendar year as technology, demand, or institutions change. Or the age slope itself may differ across cohorts, meaning that newer graduates accumulate earnings differently as they grow older. These possibilities can produce similar-looking data, but they imply different interpretations of the apparent decline in a short cross-section.

The cohort explanation fits the main pattern best. In the paper’s simplified model, a newer cohort can have a higher technical-major premium at the point of entry. Because newer cohorts are concentrated at younger ages in the short ACS window, their higher starting point is mistakenly placed on the age axis. Once the long panel allows the analysis to account for cohort-major differences, the engineering-and-computer-science profile moves away from the strongly negative cross-sectional result.

The reported magnitudes make the level shift visible. Relative to workers born by 1960, the cohort-by-major estimate for engineering and computer science is about 0.134 log points higher for people born in 1991 or later. The authors also find that recent technical cohorts have increased their earnings advantage, particularly among those born after 1981. This is evidence of a changing premium level. It is not direct evidence that those cohorts have a different earnings slope as they age.

The calendar-time test asks a different question. A major may become more valuable in later years because employers demand more of its associated skills, even if workers from different birth cohorts are otherwise comparable. If that time trend explained most of the cross-section/panel gap, adding major-by-year interactions to the long panel should pull the age profile toward the cross-sectional estimate. Instead, the paper finds that the estimates with those interactions are similar to the panel estimates without them and remain unlike the short-cross-section result. Technical returns did rise mildly over time, but the authors judge year-major changes insufficient to explain the large discrepancy.

The slope test supplies another boundary. When the authors split the worker-fixed-effects sample into people born by 1970 and people born after 1970, the technical coefficients at ages 47–48 are 0.039 and 0.040. The estimates across the two groups are generally similar. That offers little evidence that later technical cohorts have a dramatically different age slope.