Can skill-biased technological change compress unemployment rate differentials across education groups?