Why Mathematics Pulls Ahead in Earnings
The question in this chapter is why mathematics has the larger estimated relationship with earnings even though verbal skill has the stronger relationship with educational progression. The answer begins by separating two labour-market outcomes—whether someone is employed and how much they earn—and then following both across the early career. This lets readers interpret the mathematics premium as a conditional lifecycle pattern shaped by field choice, education timing, skill level, and combined capabilities, not as a universal price attached to one school subject.
By the time the English cohorts reach their early thirties, the education-versus-earnings reversal is clear. The study’s instrumental-variables estimates associate both mathematics and verbal skill with higher pay, but the mathematics coefficient is about twice as large. That result does not erase the verbal pathway. It shows that a skill can matter strongly by helping a person reach education while another skill is more directly connected to the destinations and tasks that the labour market rewards with higher pay.
Pay and employment are different labour-market outcomes
The baseline pay estimates use log earnings. “Log pay” is a technical way of expressing proportional differences: a coefficient describes how the logarithm of earnings changes with a skill rank, and a small rank change can then be translated approximately into a percentage difference in pay. The study reports an IV coefficient of 0.583 for mathematics and 0.287 for verbal skill for adults observed at ages 30–34. A ten-percentile-point increase in the mathematics rank is associated with roughly 6% higher earnings; the equivalent verbal increase is associated with roughly 3% higher earnings, using the paper’s interpretation.
The comparison is not between a mathematical person who earns a fixed amount and a verbal person who earns another fixed amount. It is a comparison of estimated changes in average log pay as the measured rank changes, under the study’s controls and identification strategy. The unit is especially important because percentile ranks run from relative position in a cohort distribution, not from zero skill to perfect skill. A ten-point move can represent different underlying score changes at different parts of the distribution, and the later nonlinear estimates show that the associated return is not constant.
Employment tells a related but distinct story. In the full-sample IV estimates, mathematics has a coefficient of 0.163 for being employed, compared with 0.085 for verbal skill. Both relationships are positive, but the gap is smaller than it is for log pay. The result suggests that the strongest difference between the two domains may concern the type of work and the pay attached to it, not simply whether a person works at all. A skill can help someone cross an employment threshold, while other features of the person’s field, occupation, and job determine how much that work pays.
This distinction prevents two common errors. First, the mathematics earnings advantage should not be described as proof that verbal skill has no labour-market value.