An alternative interpretation of ‘average years of education’ in growth regressions