On-the-job training and the age-earnings profile

On-the-job training is human capital acquired after leaving school, through instruction and experience at work. General training raises a worker’s productivity at many firms; specific training raises it only at the firm that provides it.¹ In a competitive labour market workers pay for general training through lower wages while they train, firms and workers share the cost of specific training, and because investment falls with age, earnings rise steeply early in a career and then flatten.

Where wage differences come from

Workers earn different wages because jobs differ (risky or unpleasant jobs pay compensating differentials), because workers differ in human capital from schooling and training, because they are at different stages of the life cycle, and because different labour markets pay different prices for skills. Human capital starts accumulating before school: social capital, the quality of the environment a child grows up in, including the neighbourhood, schools, peers and community institutions, shapes what children learn. In the Moving to Opportunity experiment, children whose families were randomly offered housing vouchers to move to lower-poverty neighbourhoods before they turned 13 were more likely to attend college and earned more as adults.²

General and specific training

Most training is a mix, but the two pure cases are:

• General: learning spreadsheets, driving a truck, programming in a common language, a trade certification. Other employers value these skills as much as the trainer does.
• Specific: learning a firm’s own software and procedures, its organization and who to ask, or operating equipment used only by one employer, such as a military tank. Other employers do not value these skills.

Skills can also be specific to an occupation or an industry rather than to one firm. What matters for wages is how much other employers would pay for the skill, because that determines who can afford to pay for it.

Who pays for training: the two-period model

A firm employs a worker for two periods in a competitive labour market. Her value of marginal product is $VMP_1$ in period 1 and $VMP_2$ in period 2, and future amounts are discounted at rate $r$. The firm hires until the present value of its costs equals the present value of what she produces. With costs $C_1$ and $C_2$,

$$C_1 + \frac{C_2}{1+r} = VMP_1 + \frac{VMP_2}{1+r}.$$

Training costs $H$ in period 1 and none in period 2, so $C_1 = w_1 + H$ and $C_2 = w_2$:

$$H + w_1 + \frac{w_2}{1+r} = VMP_1 + \frac{VMP_2}{1+r} \quad\Longrightarrow\quad w_1 = VMP_1 - H + \frac{VMP_2 - w_2}{1+r}.$$

General training: the worker pays

General training raises her productivity at every firm, so rivals will offer her $VMP_2$ in period 2. If the firm paid less, she would leave; so $w_2 = VMP_2$. The last term vanishes:

$$w_1 = VMP_1 - H .$$

The worker pays the whole cost through a first-period wage $H$ below her value of marginal product, and she collects the whole return in period 2. The firm cannot pay for the training, because it has no way to recover the cost: any attempt to pay her less than $VMP_2$ after training sends her to a rival. Apprentices, medical residents and articling law students fit this pattern, accepting modest pay while they acquire skills valued across their profession.

Specific training: the cost is shared

Specific training raises productivity only at this firm, so her best outside offer stays at $w^*$, her productivity elsewhere, with $w^* < VMP_2$. The firm could pay for the training and keep the return, setting $w_2 = w^*$, which gives

$$w_1 = VMP_1 - H + \frac{VMP_2 - w^*}{1+r}.$$

But a worker paid only her outside wage has nothing to lose by quitting, and if she quits the firm loses its investment. If instead she paid for the training ($w_1 = VMP_1 - H$, $w_2 = VMP_2$), the firm would have nothing to lose by laying her off after a small fall in demand, and her investment, worthless elsewhere, would be lost. So the two share the cost and the return:

$$w^* < w_2 < VMP_2 .$$

She earns more than elsewhere, so she stays; the firm earns more from her than it pays, so it keeps her. The firm finances part of the training, and it invests only if its share of the return covers its share of the cost.

Second-period wage under specific training, w₂ ($ thousands)

$35.0k

First-period wage, specific training

$28.76k

First-period wage, general training (VMP₁ − H)

$24.00k

Worker's share of H, specific

$1.24k

Firm's share of H, specific

$4.76k

Firm's period-2 margin VMP₂ − w₂, specific

$5.0k

Open in Graph Lab

Figure 1. A worker producing $30,000 in period 1 and $40,000 in period 2, trained at a cost of $6,000 with a 5 percent discount rate: general training gives $w_1 = VMP_1 - H$, or $24,000, and $w_2 = VMP_2$. Specific training with a second-period wage of $35,000 gives a first-period wage of $28,762, so the firm finances $4,762 of the cost and the worker $1,238.

Specific training explains two facts. Last hired, first fired: a senior worker’s value of marginal product exceeds her wage by the firm’s share of the return, so the firm keeps her through a fall in demand that makes a new hire, paid about his value of marginal product, unprofitable. Falling separations: as specific capital builds up with tenure, both quits and layoffs become less likely. General capital, which is portable, cannot explain either. The probability that a job ends does fall with tenure, though part of that pattern reflects differences between workers, with frequent movers leaving early.³ Wages also rise with seniority: Topel estimates that ten years with the same employer raise a typical American man’s wage by more than 25 percent, a measure of what he would lose if the job ended.⁴

Doubts about the dichotomy

Firms often pay for training that looks general, which the competitive model rules out. Acemoglu and Pischke argue that an employer knows more about its own workers than rival firms do, giving it some monopsony power over them and a reason to fund transferable skills; data on German apprentices support parts of their model.⁵ Lazear treats every skill as general but notes that firms use different combinations of skills; in a thin market few other firms want a worker’s particular mix, so the employer can recoup an investment in skills that are each general.⁶ The general-versus-specific framework remains the starting point for thinking about who pays.

Human capital over the life cycle

Measure human capital in efficiency units. The marginal revenue of a unit is the present value of the extra earnings it yields until retirement. Each unit earns the same amount, so marginal revenue is a horizontal line, and it is lower at older ages because fewer working years remain. The marginal cost of producing units rises with the number produced in a year, because of diminishing returns in producing human capital. At each age the worker invests where marginal revenue equals marginal cost.⁷

Age

30

Marginal revenue of one more unit

$1,637

Units acquired this year (Q^*)

81.9

Years left to collect the return

35

Open in Graph Lab

Figure 2. An efficiency unit adds $100 a year to earnings until 65, discounted at 5 percent; the marginal cost of units rises with the number produced in a year. Marginal revenue falls from $1,777 at age 20 to $1,637 at 30 and $433 at 60, so optimal investment falls from 89 units to 82 and 22.

Investment is therefore heaviest when young and tapers off: full-time schooling first, then on-the-job training that declines with age.

The age-earnings profile

Falling investment produces an age-earnings profile that is upward sloping and concave. Earnings rise because workers keep adding human capital and because, as they invest less, they give up less current pay to training. Growth slows because each year adds fewer units. Near retirement earnings can level off or fall if human capital depreciates faster than it is replaced. Observed profiles share three features: more educated workers earn more, earnings rise at a decreasing rate, and the profiles of education groups diverge with age, suggesting that schooling and later training are complements.

The Mincer earnings function puts this into a regression:⁸

$$\ln w = a\,s + b\,t - c\,t^2 + \text{other variables},$$

where $s$ is years of schooling and $t$ years of experience. Human capital theory predicts $b > 0$ and $c > 0$. The slope $d\ln w/dt = b - 2ct$ falls with experience, and log earnings peak at $t^* = b/(2c)$.

Coefficient on experience squared (c)

0.00070

Experience at the peak, b/(2c) with b = 0.04

28.6 years

High school graduates peak at age

46.6

University graduates peak at age

50.6

Log wage growth over the first year of work, b − c

3.93%

Open in Graph Lab

Figure 3. Profiles implied by $\ln w = 0.08s + 0.04t - 0.0007t^2$ plus a constant, for high school graduates working from age 18 and university graduates from 22. Both peak after $b/(2c) \approx 28.6$ years of experience, at ages 46.6 and 50.6, and the dollar gap between them widens with age.

As a test of human capital theory, the regression is weak in two ways. Other theories also predict wages that rise with experience, such as deferred pay that discourages shirking or promotion tournaments, so finding $b > 0$ and $c > 0$ is consistent with human capital without proving it. And the schooling coefficient $a$ suffers from ability bias unless ability is controlled for.

Key idea

Who pays for training depends on who else values it: workers pay for general training, firms and workers share specific training. Investment that falls with age makes earnings rise at a decreasing rate.

Worked example

Use the numbers of Figure 1: $VMP_1 = 30{,}000$, $H = 6{,}000$, $VMP_2 = 40{,}000$, $r = 0.05$, and an outside offer $w^* = 30{,}000$ if the training is specific.

1. General training. Rivals pay $VMP_2$, so $w_2 = 40{,}000$ and $w_1 = 30{,}000 - 6{,}000 = 24{,}000$ dollars. The worker pays all of $H$.

2. Specific training, firm keeps the return. With $w_2 = w^* = 30{,}000$: $w_1 = 24{,}000 + 10{,}000/1.05 = 33{,}524$ dollars. The firm pays all of $H$, and because firms compete for trainees, it also hands the rest of the expected return to the worker up front. She has no stake in staying.

3. Specific training, shared. With $w_2 = 35{,}000$: $w_1 = 24{,}000 + 5{,}000/1.05 = 28{,}762$ dollars. The firm pays $28{,}762 - 24{,}000 = 4{,}762$ of the cost and keeps $5,000 in period 2, worth $4,762 today; the worker pays $30{,}000 - 28{,}762 = 1{,}238$ and gains $5,000 over her outside offer in period 2.

4. Mincer peak. If $\ln w = 0.09s + 0.05t - 0.0008t^2$, earnings peak at $t^* = 0.05/(2 \times 0.0008) = 31.25$ years of experience. Log earnings grow by about $0.05 - 2(0.0008)(10) = 0.034$, or 3.4 percent, per year after 10 years of experience.

Evidence: earnings by age in Canada

Median employment income by age group, Canada

2024, people with employment income, both genders

15 to 24$13,400

25 to 34$49,400

35 to 44$62,800

45 to 54$67,500

55 to 64$52,700

65 and over$13,200

Median employment income by age group, Canada, 2024. Income climbs steeply from the youngest group, peaks at 45 to 54, and falls after 55.Source: Statistics Canada, Table 11-10-0239-01, Income of individuals by age group, gender and income source, Canada, provinces and selected census metropolitan areas. Retrieved 2026-10-23.

Show the data and notes

Canadian Income Survey. Median employment income (wages, salaries, commissions and net self-employment income) among people with employment income (zeros excluded), in 2024 dollars. Includes part-time and part-year workers, which pulls down the medians of the youngest and oldest groups.

Category	Median employment income
15 to 24	$13,400
25 to 34	$49,400
35 to 44	$62,800
45 to 54	$67,500
55 to 64	$52,700
65 and over	$13,200

Median employment income was $13,400 at ages 15 to 24, $49,400 at 25 to 34, $62,800 at 35 to 44 and $67,500 at 45 to 54, before falling to $52,700 at 55 to 64. The gains shrink with age, as the theory predicts: $36,000 between the first two groups, then $13,400, then $4,700. The chart is a snapshot of different people, not one cohort followed over time, and it includes part-time and part-year workers, so the low figures for the youngest and oldest groups partly reflect fewer hours, schooling and retirement rather than lower pay per hour.

Common mistakes

Traps on exams

• Saying firms pay for general training. In a competitive market they cannot recover the cost, because trained workers can leave for a rival paying $VMP_2$.
• Saying the worker pays for specific training. She would bear the risk of a layoff that wipes out her investment; the cost and return are shared, with $w^* < w_2 < VMP_2$.
• Dropping the 2 in the peak. Earnings peak at $t^* = b/(2c)$, not $b/c$.
• Reading a cross-section of ages as one worker’s path. Differences between cohorts and in hours worked shape the chart too.

On the exam

For “who pays”, write the zero-profit condition $H + w_1 + \frac{w_2}{1+r} = VMP_1 + \frac{VMP_2}{1+r}$, solve for $w_1$, then substitute $w_2 = VMP_2$ (general) or argue $w^* < w_2 < VMP_2$ (specific). For the life cycle, draw horizontal $MR_{20}$ above $MR_{30}$ against a rising $MC$ and show investment falling.

Check yourself

1.

In a competitive labour market, who pays for general training, and how?

1The worker, through a first-period wage equal to the value of marginal product minus the training cost2The firm, which recovers the cost by paying less than the value of marginal product after training3The firm and the worker split the cost equally4Nobody, because training raises productivity enough to pay for itself

Check answer

2.

Why do the firm and the worker share the cost of specific training rather than one side paying all of it?

1If the worker paid it all, the firm could lay her off at no loss to itself; if the firm paid it all, she could quit at no loss to herself. Sharing gives both a stake in the match2The law requires employers to pay half the cost of training3Specific training also raises productivity at other firms, so both sides benefit4Workers cannot borrow, so the firm must pay at least half

Check answer

3.

A Mincer earnings function estimates $\ln w = 0.09\,s + 0.05\,t - 0.0008\,t^2 + \text{other variables}$, where $t$ is years of experience. After how many years of experience are predicted earnings highest?

Your answer (years)

Check answer

Practise more questions on this topic

Sources

1. Becker, G. S. (1962). Investment in human capital: A theoretical analysis. Journal of Political Economy, 70(5, part 2), 9 to 49. ↩

2. Chetty, R., N. Hendren and L. F. Katz (2016). The effects of exposure to better neighborhoods on children: New evidence from the Moving to Opportunity experiment. American Economic Review, 106(4), 855 to 902. ↩

3. Farber, H. S. (1999). Mobility and stability: The dynamics of job change in labor markets. In O. Ashenfelter and D. Card (eds.), Handbook of Labor Economics, vol. 3, chapter 37, 2439 to 2483. Elsevier. ↩

4. Topel, R. H. (1991). Specific capital, mobility, and wages: Wages rise with job seniority. Journal of Political Economy, 99(1), 145 to 176. ↩

5. Acemoglu, D. and J.-S. Pischke (1998). Why do firms train? Theory and evidence. Quarterly Journal of Economics, 113(1), 79 to 119. ↩

6. Lazear, E. P. (2009). Firm-specific human capital: A skill-weights approach. Journal of Political Economy, 117(5), 914 to 940. ↩

7. Ben-Porath, Y. (1967). The production of human capital and the life cycle of earnings. Journal of Political Economy, 75(4), 352 to 365. ↩

8. Mincer, J. (1974). Schooling, Experience, and Earnings. National Bureau of Economic Research. ↩