A month or so ago, a new study from four respected economists argued that “positive assortative mating” — when men marry women of similar educational attainment — is a major driver of economic inequality in the U.S. Positive assortative mating could describe people of similar social status or income levels marrying, too, or any number of other things, but they examined education levels — a more stylized description of the larger dynamic is featured in Charles Murray’s book *Coming Apart*, in which residents of a prototypical low-education, low-income community called Fishtown now don’t mix with and marry the college-education, high-income residents of Belmont anymore.

In other words, because college-educated people are more likely to marry other college-educated people than they were in past decades, economic inequality is higher than it otherwise would be, when looking at household incomes (which is what measures of inequality do).

But does a rise in positive assortative mating explain higher inequality — something that’s especially jumped over the past three decades? (An increase that’s overstated by those on the left, but is real.) No, argues Larry Mishel of the liberal Economic Policy Institute. He raises a fair point: One of the statistics the paper presents seems to show that positive assortative mating increased from 1960 to 1980, but not in recent years:

Mishel argues, “the question that should be answered is why income inequality grew a great deal between 1979 and now and was fairly stable before then.” But: “Greenwood et al.’s data show that positive assortative mating declined(!) from 1980 to 2005, which directly tells us that this phenomenon did not cause any of the income inequality generated after 1980: in fact, positive assortative mating was a force that equalized incomes after 1980. It was in the period from 1960 to 1980 that positive assortative mating lead to more unequal incomes. Consequently, their research in no way lifts up the role of ‘like marrying like’ in generating inequality since 1980.”

Mishel’s right about that one statistic, but he makes two mistakes: He’s looking at the least accurate representation of the data Greenwood et al.’s consulted, and he’s misunderstanding their argument.

It’s not exactly simple to measure the degree to which positive assortative mating is going on. For one, marriages aren’t discrete, one-time events: Once a very equal marriage happens, it affects the makeup of married America until the marriage ends or a spouse passes away. Since marriages occurring before 1975 were less sorted than marriages occurring at any point afterward, more positively sorted marriages accumulated over time.

Which is why the paper relies most heavily on other statistics, showing an unambiguous increase:

By the other two statistics, the delta and gamma numbers, it’s clear that, contra Mishel, marriages in the United States grew more “sorted” as inequality rose from 1980 to now. Jeremy Greenwood, the Penn professor who was one of the paper’s four authors, explained to me today that Mishel’s actually grabbing the least useful statistic (but the one that was apparently featured in the *VoxEU* blog post that publicized their paper). Both the delta (the green dotted line) and the gamma (the line on the left-hand graph above) are better assessments of whether American marriages are growing more or less sorted. The tau statistic is a more simplistic measure that, for one, doesn’t account for the fact that more people were attaining higher levels of education over time.

The gamma is the actual statistical analysis they did (notice the tight 95 percent confidence interval around it) and it rose as inequality rose — their analysis suggests the sortedness of married America, if you will, certainly wasn’t dropping and discouraging the growth of inequality, as Mishel suggests. It’s true that the degree to which American marriages are sorted doesn’t increase as quickly over the past few decades as it did between 1960 and 1980, but this is hard to assess on its own: The authors only had data measurements at ten-year intervals, so one shouldn’t read too much into the correlation at each of those measurement points.

Further, that wasn’t even entirely their point. When income inequality for *individual earners *rose from 1980 to 2000 (as it did), the fact that American marriages were sorted to a greater degree in 1980 than in 1960 (and in 2000 than in 1960) exacerbated the inequality increase that would have occurred otherwise, so it’s part of the inequality story. As Greenwood explained to me, “If you look at wages across educational groups, they got much more skewed recently. The fact that you get much more skewed mating, too, means that you will have even more inequality.”

Filip Spagnoli points out that we can’t compare this to a world in which marriages are totally random and spouses’ education levels are uncorrelated to each other, since that’s not in any way realistic — but if marriages are much more sorted than they were in 1960 (they are), any increase in top individual incomes (via technological change, returns to education, etc.) will create more inequality among households than it would have in 1960. Marriage does matter, and Greenwood’s data back that up.

Of course, these patterns aren’t the whole story of inequality, or anywhere near it — in fact, the paper’s point rests on the fact that wages have increased more quickly for well-educated individuals than less-educated ones. The fact that there are more marriages with positive assortative mating, Greenwood argues, has exacerbated these effects when you look at household income (which most measures of inequality do), so it explains some of the huge inequality increases that are always grabbing headlines.

This is certainly not necessarily something we should reflexively condemn or attempt to discourage (to a lesser degree, that disclaimer applies to other acknowledged drivers of inequality, such as technology and globalization). But it’s important to recognize the role it seems to be playing — not least because marriage also determines the fate of the next generation (both nature- and nurture-wise), so it could drive inequality in the next generation in an entirely different way.