Conclusions about changes in homogamy are sensitive to how educational groups are defined. In essence, if all college graduates are grouped together, homogamy increased. If we separate college graduates from those with a more advanced degree as is common in the wage structure literature (Acemoglu and Autor, 2011), then, if anything, homogamy appears to have declined.
The ideal statistic for addressing the degree of assortative mating would be a rank-order correlation coefficient that works well when there are a large number of ties, as there are in the education distribution. Unfortunately, none exists. However, if we use either a standard Pearson correlation coefficient or standard rank-order statistics that correct for ties, we find no increase in the correlation between husband’s and wife’s education regardless of whether we use 5, 6 or 12 categories of education. Only if we use a rank-order correlation coefficient that does not correct for ties and five categories do we reach the conclusion that the correlation has increased.
--Rania Gihleb and Kevin Lang, NBER Working Paper 22927, on a potential statistical mirage