The U.S. marriage rate has fallen 26% since 2000, and that number gets treated as a single fact about American life: people are retreating from marriage. I wanted to check something narrower first. Is the retreat uniform, or is it a story about one kind of marriage in particular?
Married couples have made up a rising share of same-sex households every year since the Supreme Court legalized nationwide same-sex marriage in 2015, even as the overall marriage rate keeps sliding: same country, same decade, same dating apps, opposite direction.
An estimated 21% of same-sex couples were married in 2013, before Obergefell v. Hodges. By 2024, Census data puts that share at 60.3%, still climbing, not leveling off. The overall marriage rate spent the same stretch doing the opposite: two real trends, running in opposite directions.
My first guess was politics. The 2024 election produced a real, measurable gender gap in the vote, and I wondered whether it lined up with where marriage rates were falling hardest. I built the state-level gender gap myself from the AP VoteCast microdata and regressed it against each state's marriage-rate decline: no relationship (r² = 0.03, p = 0.22). I checked whether the states with the widest gender gaps were simply the more conservative ones, since conservative states are assumed to marry more: they're not. The widest gaps are concentrated in blue states like Rhode Island and Connecticut, not red ones, and state-level conservatism has essentially zero relationship with either the level or the decline of the marriage rate (r² < 0.001 both ways). Nor are the widest gaps in the swing states that decided the election, despite the coverage those states got for exactly this reason: the seven core battlegrounds average 28th out of 50 nationally, a median gap statistically indistinguishable from the country as a whole. The politics angle doesn't hold up from any direction.
The more serious candidate is a recent NBER paper by Caitlin Myers and Ezekiel Hooper, which argues the iPhone itself did much of this. Their identification is clean: from 2007 to 2011 the iPhone was sold exclusively on AT&T, so a county's exposure to the device depended on whether AT&T's network reached it. Comparing high- and low-coverage counties, they find iPhone access reduced births 4.5 to 8.0% among 15-to-19-year-olds and explains an estimated 33 to 52% of the fertility decline over that window. They propose the effect runs through two behavioral channels: the phone displaces the in-person time in which relationships form, and it substitutes pornography for partnered sex.
That didn't square with my own experience. I went through college from 2010 to 2016, the exact years smartphones went from novelty to universal, and the phone in my pocket coordinated meetups more often than it replaced them. Group chats, event pages, and location sharing got people into the same room; they didn't obviously keep anyone out of it.
Start with the simplest version of the test, before getting into either mechanism directly. Smartphone ownership went from 35% of U.S. adults in 2011 to 91% by 2024, and nearly all of that growth happened before 2019, when it was already at 81%; the curve has been close to flat ever since. The marriage rate didn't flatten alongside it. It kept falling at the same pace or faster the whole time, and that's not just a quirk of the national number: the middle half of U.S. states show the identical shape, flat through roughly 2017 and falling hard from 2018 on, regardless of how saturated phone ownership already was in that state. If access to a phone were the operating ingredient, the decline should have leveled off once everyone had one. It didn't.
That's suggestive, not a real test of either mechanism on its own; it's a background check before getting into one. So I tried to test the two mechanisms directly rather than take them on priors.
Using the American Time Use Survey's own respondent-level data, minutes per day spent with friends present, ages 20 to 34, fell by more than half between 2003 and 2023, matching the shape of the decline the NBER authors report for teenagers. That's the displacement channel, measured directly rather than assumed. Line the two series up and the naive correlation is strong, r = 0.78, because both have been falling for two decades. That's the wrong test: any two declining trends will correlate that way regardless of whether one has anything to do with the other. The real question is whether a sharper drop in friend-time in a given year comes with a sharper drop in the marriage rate that same year, and there the relationship disappears (r = 0.26, p = 0.28), with the same null at a one- and two-year lag. A shared multi-decade slope isn't evidence of a live mechanism; the year-to-year wiggles have to move together too, and they don't. The substitution channel is weaker still: Google search interest in pornography actually peaked in 2013 and has fallen since, likely because people stopped needing to search for it once bookmarking and apps took over, which makes it an unreliable proxy in either direction, not just a null one.
None of this contradicts the NBER paper's own estimate. Their identification is a clean four-year natural experiment, from 2007 to 2011, when phone adoption itself was the thing varying across counties, and nothing here touches that. What it does mean is that displacement and substitution aren't doing visible, ongoing work on the marriage rate a decade past that window, on top of the saturation point above. The clearest single data point: 2021 and 2022, the two years when time with friends hit its lowest point on record, were also the two years the marriage rate rose. If isolation were the operative lever in any given year, those are the years it should have kept falling instead.
The same-sex comparison is the cleanest version of this argument. Whatever is displacing in-person time or substituting for partnered sex, it's hitting the entire population at once: gay and straight Americans carry the same phones, use the same apps, and live in the same media environment. If that environment were suppressing relationship formation broadly, married-share among same-sex couples should have been pulled down too. Instead it rose from 21% to 60% over the identical span that opposite-sex marriage fell. A shared mechanism can't produce opposite outcomes in two populations exposed to it simultaneously.
Two things temper that comparison, and both are worth naming. Same-sex couples captured in Census data skew more urban and more educated than the general population, and either trait independently predicts higher marriage propensity, so some of the rise is likely compositional rather than purely a catch-up effect from legalization. And a population gaining a legal right for the first time is a poor comparison group for a story about behavioral displacement in a population that has always had it. The two groups aren't just experiencing different trends; they started the period in structurally different positions, which limits how much weight the contrast alone can carry.
What the data does support is narrower than a rebuttal: the leading explanation for the broader marriage retreat doesn't leave the kind of footprint you'd expect it to leave, either in the years after phones saturated the population or in the one large group whose marriage rate is heading the other way.
Paper: Myers, C.K. and Hooper, E. "Is the iPhone Birth Control? Causal Evidence from AT&T's 2007–2011 Carrier Monopoly." NBER Working Paper No. 35310, June 2026.
Marriage rate: CDC/NCHS, National Vital Statistics System, provisional marriage rates 2000-2023. Same-sex married share: Williams Institute (UCLA) for 2013 and 2016; U.S. Census Bureau, American Community Survey 1-year estimates, Table B11009, for 2019-2024. Gender gap and swing-state ranking: author's calculation from AP VoteCast 2024 general election public use file, all 50 states. Smartphone ownership: Pew Research Center, Mobile Fact Sheet. Time-with-friends: American Time Use Survey (BLS) respondent file, 2003-2023, variable TRTFRIEND, ages 20-34, weighted. Pornography search interest: Google Trends. Author calculations throughout.