Are Educational Mismatches Responsible for the ‘Inequality Increasing Effect’ of Education?

For evidence on the social and economic benefits of education see, for example, Ashenfelter and Rouse (2000), Wolfe and Haveman (2001) and Dolton et al. (2009).

Through the paper we abuse language somewhat and will refer to these workers as workers who ‘lack necessary skills’. We are aware, however, that there might be individuals who have not had formal education and training for unskilled jobs but who have acquired the necessary background through other sources, including peer observation, learning by doing and general work experience. Although these channels are typically less relevant, they might be important for a small fraction of uneducated individuals working in low level jobs. As most other measures of mismatch, a limitation of our definition is that it focuses on formal education and training and disregards other sources of skills acquisition.

In Portugal, only 27.6% of the adult population (25–64 years old) has completed upper secondary education, while in Europe as a whole (EU-25) this proportion rises to 69.7%. Similarly, training participation in Portugal is 3.8%, against 10.1% in EU-25 (Eurostat 2007).

Using the mean rather than the modal value produced only small changes in the estimates. The results are available upon request.

This exclusion restriction affected 3 occupations and 0.1% of the workers in the initial sample.

The estimates for the full set of controls are available upon request.

Through the paper we refer to the coefficients reported in the tables as ‘wage effects’ or ‘wage differentials’. To be precise, however, the percentage wage difference is given by e^β−1, rather than by β itself, especially when the estimates are large. We do not perform this transformation to facilitate the correspondence between the tables and the text. .

Following a human capital interpretation of the mismatch phenomenon, some authors have suggested that workers may accept mismatched work in exchange of training, to compensate for low tenure and experience, or to access higher level occupations (Sicherman 1991; Groot 1996; Sloane et al. 1999). The full set of controls used in our earnings equations is aimed to remove the impact of these and other variables from the mismatch effect. It may be argued, however, that most of these covariates are endogenous and that a more parsimonious specification would capture the ‘true’ penalty of mismatch more appropriately. Thus, for example, we may not include controls for training. As the acceptance of mismatched work may allow some individuals to participate in training activities that later on are rewarded in the labour market, we could interpret these wage gains as a return to mismatch rather than a return to training. A similar argument applies to other variables, such as occupation and tenure. Following this reasoning, the estimates reported in “Appendix 2” are obtained dropping from vector X _i all the controls except experience.

The more differentiated view provided in Panel 3 comes at the cost of reduced cell size in some groups. Thus, for example, the number of overqualified workers drops from 1,047 in the total sample to 231 when we consider only the group of workers with university education, and a more reduced group results when we consider skills mismatches and strong mismatch among university graduates. Even though the interaction coefficients reported in Panel 3 exhibit moderate standard errors and are not erratic across quantiles, the reduced cell size of specific groups in a quantile regression framework recommend us to interpret the results with some caution.

Across studies, the return to surplus, required and deficit schooling range from 3 to 5%, from 5 to 11%, and from −2 to −6%, respectively.