Mincer Earnings Function

Mincer Earnings Function is a single-equation semi-logarithmic regression model that expresses the natural logarithm of individual labor earnings as a linear function of years of completed schooling and a concave quadratic function of potential post-schooling labor market experience. Formulated by Jacob Mincer in his pioneering 1958 Journal of Political Economy article and expanded in his classic 1974 treatise, Schooling, Experience, and Earnings, the equation is the most widely estimated empirical relationship in labor economics and applied microeconometrics. By grounding its functional form directly in the theory of lifecycle human capital investment developed by Mincer, Gary S. Becker (1964), and Yoram Ben-Porath (1967), the Mincer earnings function provides a unified framework for estimating the private rate of return to education, lifecycle wage growth, gender and racial wage differentials, and cross-country income inequality.

In its standard cross-sectional econometric specification, the Mincer earnings function relates the natural logarithm of hourly wages or annual earnings w_i for worker i to years of completed formal schooling s_i and years of potential labor market experience x_i:

ln w_i = ln w₀ + ρ s_i + β₁ x_i + β₂ x_i² + ε_i

When direct longitudinal records of actual years worked are unavailable in census or household survey datasets, Mincer introduced potential experience x_i = Age_i − s_i − 6 (assuming children begin formal schooling at age six and enter the full-time labor force upon completing s_i years of education). Each parameter in the canonical Mincer equation carries a structural economic interpretation:

1. Baseline Log Wage (ln w₀): The intercept represents the predicted log earnings of an individual with zero years of schooling (s = 0) and zero labor market experience (x = 0).
2. Mincerian Return to Schooling (ρ = ∂ ln w / ∂s): The slope coefficient on schooling measures the proportional increase in earnings associated with one additional year of formal education, holding experience constant. Under the compensating-differentials assumptions described below, ρ equals the internal rate of return to schooling investment. Across hundreds of empirical studies spanning more than 100 countries (Psacharopoulos & Patrinos, 2018), OLS estimates of ρ typically fall between 0.06 and 0.12 (6% to 12% per additional year of schooling).
3. Lifecycle Experience Profile (β₁ > 0, β₂ < 0): The linear experience coefficient β₁ > 0 captures the initial rate of wage growth from post-schooling on-the-job training, while the negative quadratic coefficient β₂ < 0 captures the concavity of lifecycle age-earnings profiles as workers reduce human-capital investment rates and skills depreciate approaching retirement. Setting ∂ ln w / ∂x = β₁ + 2β₂x = 0 identifies the peak-earnings experience horizon x* = −β₁ / (2β₂), which typically occurs at 25 to 35 years of experience (roughly ages 48 to 55).

Rather than being an ad hoc curve-fitting exercise, the semi-logarithmic specification of schooling and quadratic specification of experience arise from two complementary microeconomic models developed by Mincer (1958, 1974).

1. The Compensating-Differentials Schooling Model (Mincer, 1958)

Suppose individuals are initially identical in innate ability, face a perfect capital market with continuous-time real interest rate r, and choose how many years s to spend in full-time schooling before working for a fixed career length of T years (or infinitely long as T → ∞). While in school, a student earns zero labor income and incurs zero direct tuition cost (so the opportunity cost of schooling is solely forgone earnings). Upon graduation after s years, the worker earns a constant annual wage w(s) from time t = s to t = s + T.

The present discounted value of lifetime earnings evaluated at age zero is:

V(s) = ∫_s^s+T w(s) e^−rt dt = [w(s) / r] e^−rs (1 − e^−rT)

In a competitive long-run labor market equilibrium where all workers have equal working lifespans T regardless of schooling level, the principle of compensating wage differentials requires that the present discounted value of lifetime earnings be equalized across all schooling choices (V(s) = V(0)):

[w(s) / r] e^−rs (1 − e^−rT) = [w(0) / r] (1 − e^−rT) ⇒ w(s) = w(0) e^rs

Taking the natural logarithm of both sides yields the linear log-schooling equation:

ln w(s) = ln w(0) + r s

Thus, if the only cost of education is forgone earnings and post-schooling earnings streams are flat, the semi-elasticity of wages with respect to schooling ρ equals the market discount rate (or internal rate of return) r.

2. The Post-Schooling On-the-Job Investment Model (Mincer, 1974)

To explain why earnings rise concavely with experience after graduation, Mincer (1974) incorporated the Ben-Porath (1967) model of post-schooling on-the-job training. Let E_t denote an individual's earnings capacity (potential gross earnings if zero time were spent in training) after t years in the labor force, and let k_t ∈ [0, 1) denote the fraction of potential earnings capacity invested in on-the-job training at experience t. If formal schooling invests 100% of potential capacity (k = 1 for s years at rate of return ρ_s) and post-schooling investments earn a rate of return ρ_p, potential log earnings capacity after x years of work experience is:

ln E_x = ln E₀ + ρ_s s + ρ_p ∫₀^x k_t dt

Because the remaining working horizon shrinks as a worker ages, the marginal incentive to invest in on-the-job training declines monotonically with experience. Mincer parameterized the post-schooling investment ratio as a linearly declining function of experience:

k_x = k₀ (1 − x / T*)

where k₀ ∈ (0, 1) is the fraction of time invested in training upon entering the labor force at x = 0, and T* is the total investment horizon. Integrating k_t from 0 to x and noting that observed net earnings w_x equal potential earnings capacity minus current training investments (w_x = E_x(1 − k_x)), using the Taylor approximation ln(1 − k_x) ≈ −k_x − k_x² / 2 yields the quadratic Mincer earnings function:

ln w_x ≈ [ln E₀ − k₀(1 + k₀/2)] + ρ_s s + [ρ_pk₀ + (k₀/T*)(1 + k₀)] x − [ρ_pk₀/(2T*) + k₀²/(2T*²)] x²

This derivation also yields Mincer's overtaking age x̂ ≤ 1 / ρ_p (typically 7 to 10 years of experience): the point in the lifecycle where the return on accumulated previous on-the-job investments ρ_p ∫₀^x̂ k_t dt equals current training outlays k_x̂, so that observed earnings w_x̂ match true underlying human capital capacity E₀e^ρ_ss.

A central debate in labor economics concerns whether the OLS coefficient ρ̂ from the Mincer earnings function measures the true causal effect of human capital accumulation on productivity or reflects omitted-variable bias and market signaling:

1. Upward Ability Bias (Griliches, 1977): Suppose true log earnings are determined by ln w_i = α + ρs_i + β₁x_i + β₂x_i² + γA_i + u_i, where A_i represents unobserved cognitive ability, motivation, or family connections (γ > 0). If higher-ability individuals find schooling easier or more rewarding (Cov(s_i, A_i) > 0), omitting A_i biases the OLS estimator upward: plim ρ̂_OLS = ρ + γ [Cov(s, A) / Var(s)].
2. Downward Measurement Error (Attenuation Bias): Survey reports of completed schooling s_i contain classical reporting noise, which biases OLS estimates toward zero and is exacerbated in within-family or twin-difference regressions unless corrected via multiple reports (Ashenfelter & Krueger, 1994).
3. Human Capital vs. Educational Signaling: In the Spence signaling model (Spence, 1973), schooling does not necessarily raise a worker's innate productivity; instead, because the psychic cost of completing degrees is lower for high-ability workers, education serves as a credible signal that sorts workers under adverse selection. Empirical studies distinguish human capital from pure signaling via "sheepskin effects" (extra wage jumps at diploma completion years 12 and 16) and by comparing returns in self-employed vs. salaried sectors.

During the "credibility revolution" in applied econometrics, researchers led by Joshua D. Angrist, Alan B. Krueger, and David Card—who shared the 2021 Nobel Memorial Prize in Economic Sciences—used instrumental variables (IV) and monozygotic twin registries to isolate exogenous variation in schooling s_i. Strikingly, IV estimates using compulsory attendance laws (Angrist & Krueger, 1991) and geographic proximity to four-year colleges (Card, 1995, 1999) consistently produced causal return estimates ρ̂_IV equal to or 10%–30% larger than baseline OLS estimates. Card (1999) explained this result through a heterogeneous-returns framework (a generalized Roy model): because marginal students affected by compulsory schooling laws or college proximity face higher credit constraints and higher marginal returns to schooling than the average student, the Local Average Treatment Effect (LATE) identified by IV exceeds the population average return, while upward ability bias in OLS is largely offset by downward measurement error attenuation.

Subsequent empirical work has refined the Mincer specification in two major directions. First, Thomas Lemieux (2006) demonstrated that while a quadratic in experience was adequate in 1960 and 1970 census data, estimating lifecycle profiles reliably in contemporary labor markets requires a quartic specification in experience (x, x², x³, x⁴) because quadratic approximations systematically overstate peak mid-career wages and understate early-career wage growth.

Second, James J. Heckman, Lance J. Lochner, and Petra E. Todd (2006) evaluated the Mincer model across six decades of U.S. Census data (1940–2000) and showed that the textbook equivalence between the Mincer regression slope ρ and the true internal rate of return to schooling requires four stringent conditions: (i) log earnings must be parallel across schooling levels across the lifecycle (multiplicative separability of schooling and experience), (ii) linear log schooling returns, (iii) zero direct tuition costs and zero income taxes, and (iv) certainty with absence of sequential option values. Since the 1980s, college-educated workers have experienced steeper experience-earnings profiles than high-school graduates (non-separability) alongside rising tuition costs and the option value of continuing from high school to college, meaning structural internal rate of return calculations must account explicitly for tuition, taxes, and lifecycle non-separability.

Why does the Mincer earnings function use the natural logarithm of wages rather than dollar wages?

Taking the natural logarithm of wages satisfies both theoretical and statistical requirements. Theoretically, human capital investments compound proportionally over time (w(s) = w₀e^ρs), so log transformation linearizes the multiplicative relationship and allows ρ to be interpreted directly as a percentage rate of return. Statistically, dollar earnings are heavily right-skewed and heteroskedastic, whereas log earnings are approximately normal with much more homoskedastic residuals.

Why can potential experience (x = Age − s − 6) bias gender wage gap estimates in a Mincer regression?

Potential experience x = Age − s − 6 assumes continuous, uninterrupted full-time employment after finishing school. For workers who take career breaks for child-rearing or family care, potential experience overstates actual labor market experience and accumulated on-the-job training, downward-biasing the estimated return to experience β₁ and attributing differences in actual accumulated work tenure to the unexplained gender intercept gap.

What is Mincer's overtaking age and why is it useful?

Mincer's overtaking age (x̂ ≤ 1 / ρ_p, roughly 7 to 10 years after labor force entry) is the experience level at which the positive return from past post-schooling training investments exactly offsets the current income forgone due to ongoing on-the-job training. At the overtaking age, residual variance in log earnings across cohorts with different post-schooling investment intensities k₀ is minimized, providing the cleanest cross-sectional estimate of the pure return to formal schooling ρ_s.

1. Mincer, J. (1958). Investment in human capital and personal income distribution. Journal of Political Economy, 66(4), 281–302. https://doi.org/10.1086/258055
2. Ben-Porath, Y. (1967). The production of human capital and the life cycle of earnings. Journal of Political Economy, 75(4, Part 1), 352–365. https://doi.org/10.1086/259291
3. Angrist, J. D., & Krueger, A. B. (1991). Does compulsory school attendance affect schooling and earnings? The Quarterly Journal of Economics, 106(4), 979–1014. https://doi.org/10.2307/2937954
4. Card, D. (1999). The causal effect of education on earnings. In O. Ashenfelter & D. Card (Eds.), Handbook of Labor Economics (Vol. 3A, pp. 1801–1863). Elsevier. https://doi.org/10.1016/S1573-4463(99)03011-403011-4)
5. Heckman, J. J., Lochner, L. J., & Todd, P. E. (2006). Earnings functions, rates of return and treatment effects: The Mincer equation and beyond. In E. A. Hanushek & F. Welch (Eds.), Handbook of the Economics of Education (Vol. 1, pp. 307–458). Elsevier. https://doi.org/10.1016/S1574-0692(06)01007-501007-5)