The Returns to Education: Evidence from the Labour Force Survey

This report details research on the relationship between education and wages. The work is largely based on the Labour Force Surveys 1992-2000 and the focus of the research is largely on academic education. The report contains microeconometric estimates of the relationship between (log) wages and years of education and allows: for this relationship to be non-linear, so as to separately identify the effect of higher levels of education from the effect of earlier years; for the relationship to shift over time so we provide estimates that show how returns to education vary over time; and for the relationship to vary across individuals according to their observable and unobservable differences. Separate results for men and women are presented. The LFS data is large and this enables separate analysis of particular groups of individuals. In particular, the report contains microeconometric estimates of the effects of a degree on wages and allows for: different degree subjects to have differential effects; “sheepskin” effects associated with years of education that yield a qualification; and different lengths of study. Separate results for men and women are presented. In addition to estimating the mean effect of education on wages we also estimate the variance in returns around this mean. There are two complementary ways in which we pursue this. In the first method, estimation is by “quantile regression” methods to estimate the effect on different parts of the wage distribution. We are particularly concerned to show the extent to which the returns to education differ across the wage distribution. If the average ability of graduates has fallen over time then we might see this reflected in the size of the returns across quartiles of the wage distribution. The second method estimates a “random coefficients” model. Instead of assuming that the effect of education is the same for all individuals this model assumes that the effect differs (randomly) across individuals. The model estimates the mean effect of education and the variance around this mean. Again, by estimating the models for each separate year it is possible to see if the variance is getting larger over time. The modelling controls for observable differences in returns across individuals.