Estimating the return to schooling using the Mincer equation

Interpretation of results from an earnings function

The form of the earnings function—that is, which variables to include and how they enter the equation—is important [4]. Researchers typically use hourly earnings as the dependent variable (the outcome of interest) in these calculations. However, the time frame over which earnings are measured is often determined by necessity: some data sets report annual earnings whereas others report weekly or hourly wages. Because individuals with higher levels of schooling tend to work more, the resulting return to schooling will be greater when using weekly or annual earnings than hourly earnings [5]. This must be kept in mind when constructing and interpreting the results from earnings functions.

Next to education, work experience is another important (explanatory) variable that accounts for the post schooling accumulation of human capital and therefore consequently affects earnings. Age is commonly used to approximate experience, though one must account for each individual’s number of years spent in school when doing so as this has an impact on the individual’s starting qualifications. In developing countries where children may not start school on time due to, for example, illness, cost, or distance, or when there is a high incidence of serial repeaters, there may be a bit of measurement error in using years of schooling to approximate experience. However, omitting experience would lead to biased results due to the relationship between experience and other relevant characteristics. In particular, experience and schooling are negatively correlated, i.e. when examining people of the same age, those with more years of schooling have less work experience. And both schooling and experience are positively correlated with earnings. This means that omitting experience could lead to an underestimation of the effects of schooling on earnings [6].

Moreover, one of the limitations of the Mincer equation is the assumption that returns to experience are the same at all levels of education. The problem of heterogeneous experience premiums is noted in the literature, but no agreed solution has so far appeared. Furthermore, experience not only captures on-the-job learning and the deterioration of skills, but also picks up institutional or contractual issues, such as when firms offer earning profiles that are skewed toward older workers in order to prevent shirking. In labor economics one speaks of the efficiency wage hypothesis, which argues that some managers may have an incentive to pay employees a salary above market rates in order to reduce the costs associated with turnover, especially in industrial sectors where the costs of replacing labor are high. Part of this incentive wage may be captured by the experience coefficient.

The interpretation issues that arise when reading the results of an earnings function are illustrated in the compilation of returns to schooling from 139 economies (Figure 1). The study shown is typical, in the sense that it uses representative, structured surveys and estimates earnings profiles using cross-section data. The first column shows the results for the typical Mincerian function with years of schooling as a continuous variable. The results represent workers with positive earnings in the labor market, which is a large sample. Overall, the returns to schooling are highest in sub-Saharan Africa, at 12.5%. Returns are also high in high-income economies, Latin America and the Caribbean, and East Asia and the Pacific, at 10%, 9.3%, and 9%, respectively. The lowest returns are found in the Middle East and North Africa, at only 6.5%. Overall, the global average return is just under 10% [7].

Figure 1 also shows returns by gender. On average, returns to schooling are significantly higher for women than for men. The lowest returns to schooling are for men in the Middle East and North Africa, at only 6%. The highest returns to schooling are for women in sub-Saharan Africa, at almost 15%.

Global estimates of the returns to schooling

Most studies that present compilations of returns to schooling suffer from problems of comparability due to data sample coverage and methodology. One common problem is that surveys may not accurately reflect population averages. Researchers also typically include many different independent (explanatory) variables in their models, which may affect estimates of the returns to schooling, and different studies rarely use the same models. The typical Mincerian earnings function uses a limited number of variables—earnings, schooling, and experience. But, some researchers add many other variables including sector of employment, region, and so on. They may also specify the variables differently, all of which makes comparability difficult. However, recent estimates provide results for a large number of countries, with one new compilation presenting comparable estimates of the private returns to schooling using 911 household surveys from 139 economies [7]. The study maintains a constant definition for the outcome variable, the set of control (independent explanatory) variables, the sample definition with respect to age, and the estimation method for all surveys represented in the sample.

Previous compilations of the literature suggest certain conclusions [7]:

• The global average private rate of return to a year of schooling is about 10%. New evidence suggests that the return to another year of schooling (marginal return) is indeed positive, significant, and more concentrated around the mean than previously thought, at 9.7% overall [7].

• The returns to schooling are higher in high-income economies and in sub-Saharan Africa.

• The returns to schooling are higher for women than for men. Women receive a rate of return of 11.5%, compared to 9.6% for men [7]. (However, simply applying the Mincer earnings function for women may be biased, for instance, due to incorrect measurement of experience, selection due to the fact that women are less likely to be wage earners, and because women often have discontinuous labor force participation.)

• Though the returns to schooling decline as the supply of schooling in a country increases, they tend to decline only modestly over time, by no more than two percentage points in a given decade, and actually much less than that in recent years. Using global data, returns to schooling are shown to have declined significantly over time, but at a much slower rate than has occurred with the expansion in schooling, especially since the late 1990s [7]. While the supply of schooling has expanded by almost 50% worldwide since 1980, the returns to schooling have declined by only 3.5 percentage points, or 0.1% per year. At the same time, the average length of schooling has increased by more than 3 years, or 2% per year worldwide. On average, another year of schooling leads to a reduction of the returns to schooling by one percentage point.

Determining the most valuable return to educational investment

Individuals invest in a level of education or a degree program, rather than in years of education. The Mincerian equation is a flexible specification. Years of schooling can be replaced with other variables to represent the levels of schooling (e.g. specific types of degree programs). These become “dummy variables” (variables with two values: 0 and 1) for each level of schooling. Since levels of schooling are sequential—that is, you must complete primary to enter secondary, and secondary to enter tertiary—the returns to schooling for each level also incorporate the benefit of previous levels of schooling.

One of the long-standing conclusions from the literature is that returns to schooling are highest at the primary education level, but this is no longer considered true [7]. Globally, the returns to tertiary education are highest, followed by primary; the lowest returns are for secondary schooling (see Figure 2). On average, the returns to tertiary education are 17%, while they are 10% for primary education. The highest returns to tertiary are in Africa, at 22%, where one also finds the highest returns to primary schooling, at 13%. To some extent, the effect on primary education may be driven by an increase in compulsory education. That is, as countries develop and education systems expand, primary school becomes compulsory and universal, making it difficult or impossible to estimate returns to education, or to compare them to older workers who may have lower wages for other reasons.

Do skills gained from schooling or the acquisition of credentials lead to increased wages?

The earnings premium associated with level of education suggests that productivity increases as people acquire additional qualifications. An alternative view is that earnings increase with education due to credential effects. This refers to the idea that higher levels of schooling are associated with higher earnings, not because they directly raise productivity, but because they certify that the worker is likely to be productive. In this sense, education merely sorts workers according to their unobserved attributes; it does not necessarily augment their intrinsic productivity. For public policy reasons it is important to distinguish between the human capital (productivity) and screening hypotheses about returns to education. In very basic terms, these two hypotheses mean, respectively: schooling imparts skills that enhance productivity; hence, increases in earnings are due to the increased productivity brought about by investments in schooling (human capital); while the screening hypothesis maintains that employers select workers with higher qualifications to reduce their risk of hiring someone with a lower capacity to learn; in this case, higher earnings may not be due to productivity alone (screening). With these concepts in mind, if the only purpose of schooling is to sort prospective employees, then questions arise as to the appropriateness of public investment in the expansion or improvement of schooling.

The above debate is typically presented as an either/or argument. However, the observed correlation between education and earnings shows that, at an individual level, schooling does augment earnings. Some level of screening will likely exist, at least at the entry level, and perhaps also influence the decision to enroll. The earnings function can be used to examine the case for or against screening.

Most tests of the screening hypothesis use an earnings function. One test for screening examines returns for individuals with a completed education versus those who dropped out; if education serves as a signal (i.e. screening mechanism), then certification from a course should convey more information to prospective employers about an applicant’s ability than the number of years of schooling. Returns for completers would thus be higher than returns for non-completers. To examine the “weak versus the strong version of the screening hypothesis” one draws a distinction between employers paying irrational wages (i.e. wages that do not match the applicant’s qualifications or experience) at the initial hiring point (weak) versus those that continue paying such wages thereafter (strong). Most applications of the “weak versus the strong version of the screening hypothesis” do not find strong evidence of screening. A recent analysis that uses rigorous evaluation techniques to compare the earnings of workers who barely passed versus those who barely failed high school exit exams finds little evidence of diploma screening effects [8].

More rigorous tests of the screening hypothesis involve taking advantage of “natural experiments” such as changes in the school leaving age or college openings. By and large, while some evidence of weak screening is revealed, education is generally associated with higher earnings due to productivity rather than to screening. Thus, investment in schooling continues to be a worthwhile activity for individuals and societies to undertake [9].

The debate is ongoing about the external validity and causality of rate of return estimates. More conclusive evidence will emerge once data on the lifetime earnings profiles of beneficiaries of voucher programs can be obtained, as many of these programs use lotteries to assign places, thus giving researchers access to randomized data. In 1981, Chile introduced nationwide school choice by providing vouchers to any student wishing to attend a “voucher school” (essentially, a private school participating in the program, whereby the funding from the voucher would be used to pay the fees and tuition). Beneficiaries of the vouchers obtained more schooling and subsequently earned more than non-voucher students. Also, it is estimated that formal sector earnings are higher for lottery winners in Colombia’s large-scale government program which used a lottery to distribute scholarships for private secondary school to socially disadvantaged students.

Institutional factors have also been used to more precisely estimate the returns to schooling, including the effect of birth date. Those who are forced to remain in school because of their birth date and the school attendance law receive the same rate of return to education as those who voluntarily continue schooling.