Swipe to match.

Why do similarly-educated people pair up so much more often than chance would predict? Do they prefer one another, for shared outlook, lifestyle, or earning power? Or do universities and workplaces simply place them in the same rooms, year after year, until partnering becomes the likely outcome?

And does the same answer hold across the whole education distribution, or only part of it?

Three forces shape who we end up with: who we prefer, who we get the chance to meet, and who our families and communities steer us toward (Kalmijn 1998). This study isolates the middle one. Following Blau's argument that social structure sets the odds of contact, and Feld's idea that everyday life is organised around shared foci, a workplace, a campus, a neighbourhood, it treats the partner pool as something built by where people are routed before any choice is made.

The catch is that opportunity and choice produce the same thing: homogamous unions. You cannot tell them apart from the outcome alone. The strategy is to model the opportunity channels we can measure, then read whatever educational sorting remains as an upper bound on what choice could explain.

Swedish population registers from Statistics Sweden, covering the full resident population. The analysis follows every first union from 1991 to 2019, dating a union to the year two people move from separate addresses into a shared one.

For each person we reconstruct a full history: where they lived, at 100-metre resolution; every employer; and every university enrolment. Around 957,000 first unions among women anchor the main analysis, with men estimated separately.

The tool is a conditional logit, or discrete choice, model. For every person who formed a union, we build a choice set: their actual partner plus around 100 realistic alternatives, drawn from the singles they could plausibly have met in the same local market that year. The model then asks which features predict who was actually chosen, shared education, short residential distance, a shared employer, a shared campus.

Adding the opportunity channels one at a time, and watching how much of the education effect each one absorbs, lets us attribute a share of educational homogamy to where people live, work, and study, and see what is left over.

Opportunity does a lot of the work, but mainly near the top, as the interactive above shows. For couples where both hold a long tertiary degree, residence, workplaces, and universities together account for around half of their educational homogamy; for the middle of the distribution it is closer to 30 percent. The channels show a clean dose-response: each extra year sharing an employer or a campus raises the odds of a union.

The surprise is at the bottom. For the least educated, these channels explain essentially none of their homogamy. Accounting for where they live even nudges their estimated sorting slightly upward, not down. Whatever brings similarly low-educated people together, it is not the residence, workplace, or university overlap that registers can see.

So the mechanism of educational homogamy is not one thing. It runs through visible institutions at the top of the distribution, and through something registers cannot see at the bottom.