Overview

This chapter traces the engineering-and-computer-science profile through nested comparisons: the ACS snapshot, matched survey-year earnings, the full LEHD history, worker fixed effects, and added education and labor-supply controls. At ages 49–50, the estimate moves from −0.167 in the replicated cross-section to about −0.032 in the full panel without fixed effects, then to +0.071 in the baseline worker-fixed-effects model and +0.022 with additional controls. The sequence shows why time depth and persistent worker differences matter, while leaving the result a relative profile rather than a causal individual return.

Learning outcomes

Adding full LEHD histories reduces the technical estimate at ages 49–50 from −0.167 in the ACS profile to about −0.032 before worker fixed effects.
The baseline worker-fixed-effects estimate is +0.071 at ages 49–50, while added graduate-degree and labor-supply controls make it +0.022; specification remains important.
The panel improves the lifecycle comparison but does not identify an individual counterfactual or occupation-level skill obsolescence.

Keywords

EngineeringComputer scienceLEHDFixed effectsEarningsControls

What Changes When the Same Workers Are Followed

The subject here is the engineering-and-computer-science age–earnings profile as it passes through a sequence of increasingly informative comparisons. The specific question is why the estimated technical-major penalty is large in the short ACS cross-section, much smaller in the full LEHD panel, and positive in the baseline worker-fixed-effects model. Tracing that sequence lets readers interpret the result as a relative profile finding—rather than as a promise about what any individual would have earned after choosing a different major.

The same profile under different comparisons

The comparison is easiest to understand when the outcome and reference group are held in view. The paper estimates log annual earnings by age and major, with engineering and computer science compared with the broad group of other majors. The coefficients at ages such as 29–30, 39–40, and 49–50 describe how the technical group’s relative earnings growth differs from the reference group’s growth from the baseline age. A negative coefficient at 49–50 means that the technical group’s estimated advantage has narrowed relative to the baseline—not that every technical worker’s pay has fallen.

The first estimate comes from the ACS sample designed to reproduce the earlier cross-sectional result. Between age 23 and ages 49–50, the engineering-and-computer-science profile falls by 0.167 log points relative to the other-major group. In this comparison, a younger technical graduate and an older technical graduate are different survey respondents, usually from different birth cohorts. The estimate is therefore consistent with slower relative growth, but it also carries the cohort-composition problem established earlier.

The second comparison keeps only ACS respondents who can be matched to the LEHD while still using the ACS survey-year earnings. The age-49–50 technical estimate is more negative, at −0.216 log points. That movement does not show that the panel made the technical profile worse. It is a diagnostic: the people and geography have changed, but the observation is still essentially a one-year cross-section. The difference between the first two estimates is therefore not the main correction the authors are seeking. It shows that restricting the sample to linkable workers does not by itself eliminate the apparent decline.

The third comparison uses LEHD earnings for that same matched, survey-year sample. The outcome has changed from self-reported ACS earnings to earnings derived from unemployment-insurance records, but the people, major labels, and survey-year structure remain closely tied to the previous comparison. The age-49–50 estimate is −0.165 log points. Its similarity to the ACS result suggests that the striking cross-sectional pattern is not primarily a consequence of switching between those two earnings measures.

The fourth comparison uses the full LEHD earnings history for the same linked workers. Instead of taking only the year in which a person appears in the ACS, the model uses their available person-year observations across the long panel. At ages 29–30, 39–40, and 49–50, the technical estimates are approximately −0.002, −0.009, and −0.032. The age-49–50 difference is now small and statistically insignificant compared with the much larger negative cross-sectional estimates. The addition of years, before any worker fixed effects are imposed, changes the profile substantially.

That change is informative because it isolates the value of time depth.

Published by DeepMox

FAQ

Why does the technical profile change when full LEHD histories are added?

Adding person-year histories changes the technical profile because the model no longer places each linked worker only at the age observed in the ACS, weakening short-window cohort substitution.

What do worker fixed effects remove?

They absorb a persistent component of each worker’s earnings level and compare changes within the worker’s record; they do not create a counterfactual major choice.

What does a positive technical coefficient mean here?

It means the engineering-and-computer-science group has a higher or steeper estimated age profile than the broad other-major group under the stated controls, not that it leads every major at every age.

Previous9 minMatched ACS–LEHD Earnings Histories

Next11 minBusiness Annual Earnings and Labor Supply

Engineering and Computer Science Worker Panel

What Changes When the Same Workers Are Followed

The same profile under different comparisons

FAQ