attraction.lab

Best Age to Get Married: The Famous 28-to-32 Window Came From a Blog Post

The 11%-a-year figure comes from a 2015 blog post, and its own model turns at 28, not 32. In the same survey, in a journal, the curve flattens instead.

Research-backed writing on attraction, dating and relationships. Drafted by AI against a named bibliography, published only when every claim links to its source.
34 min read
Editorial illustration for: Best Age to Get Married: The Famous 28-to-32 Window Came From a Blog Post

Before the article, one note, because this topic is one people are pressured about. If someone is pushing you toward a wedding date — a partner, a family, a community — and the pressure includes being told who you may see, having money or documents controlled, or being afraid of what happens if you say no, that is not a question about timing. Find A Helpline, in the sources at the bottom, lists free and confidential services by country. In the United States the National Domestic Violence Hotline is 1-800-799-7233, or text START to 88788, around the clock.

With that said. If you have searched for the best age to get married you have met the same window everywhere: 28 to 32, with the odds of divorce falling 11% a year until then and rising 5% a year afterwards. It is quoted by lifestyle magazines, therapy practices, divorce firms and at least one page that calls it math.

It comes from a blog post published on 17 July 2015, and this page is what happens when you read that post, the author's own methods page behind it, his replication four days later, the critique another demographer published the same day, and the peer-reviewed literature on the same survey that none of the pages competing for this question mentions.

Four things turned up that change what the number means. The model everyone is quoting turns at 28, not 32 — the 11% and the 5% were estimated at the lower age, on the author's own methods page. That page also says the curve applies only to marriages contracted since 1998, which no one repeats. The survey behind it stopped interviewing people at 44, so nobody who married in their late thirties could be followed long enough to know how it turned out. And three peer-reviewed analyses of the identical dataset — one of them published two years earlier, all three free to read — report the curve flattening after the late twenties rather than turning upward, with the clear U-shape appearing in one racial and ethnic group out of three.

There is also a question nobody asking about a best age is being told exists. The studies that count divorces and the studies that measure whether a marriage is any good give opposite answers, because people who marry late are more reluctant to leave.

What follows is sixteen sections: where the figure came from, what its percentages mean, the limits the author himself published, the journal articles on the same data, the 1977 hypothesis this is a retest of, the mechanism that explains why the answer differs by group, the quality-versus-survival trade-off, the critique and the reply, the cross-national test, the two attempts at a causal estimate, what the top of the search results is actually built on, and the version of the question that has an answer. Our editorial policy explains why each figure carries a linked source, and why the places where the evidence contradicts itself are named rather than smoothed over.

#1

The number everyone repeats, and the page it was published on

Ask the internet for the best age to get married and you get a window: 28 to 32, or sometimes just 32, with a pair of percentages attached. Marry a year later and your odds of divorce fall by 11%, until a point, after which each extra year adds 5%.

That sentence has one source. Nicholas Wolfinger, a professor of family and consumer studies at the University of Utah, published it on the Institute for Family Studies blog on 17 July 2015, in a post called Want to Avoid Divorce? Wait to Get Married, But Not Too Long. He analysed the 2006-2010 National Survey of Family Growth and wrote: “prior to age 32 or so, each additional year of age at marriage reduces the odds of divorce by 11 percent. However, after that the odds of divorce increase by 5 percent per year.”

The day before, Slate's Jordan Weissmann had covered the same analysis and given it the name that stuck: the Goldilocks theory of marriage. “Getting married too early is risky, but so is getting married too late. Your late 20s and early 30s are just right.” Eleven years later the label is still travelling. The qualifications are not.

And this is a blog post. We can find no peer-reviewed journal article by Wolfinger presenting this analysis, then or since. That is not a scandal on its own — a researcher publishing his own models on a blog, with a methods appendix, is a legitimate way to put a finding into the world, and he did provide one. But a finding that never goes through review arrives without the things review attaches: nobody is obliged to answer the objections, and in this case the objections were filed the same week by another demographer, in public, with his code.

What follows is the record. The primary post, the author's own methods page, his replication, the three peer-reviewed analyses of the same survey that nobody citing this number mentions, the critique, and the two studies that tried to make the question causal. Our editorial policy explains why every figure here carries a linked source.

#2

The model's turning point is 28. The famous number is 32

Wolfinger's post links to a methods page on his own website, and the methods page does not say 32.

It says this: “The best fitting model had a two-piece spline with a knot at age 28. This model showed that each year of age prior to 28 reduces the odds of divorce by 11 percent. However, past 28 the odds of divorce increase by 5 percent per year. The change in slopes at age 28 is statistically significant.”

Same two percentages. Different age. The IFS post attaches the 11 and the 5 to “age 32 or so”; the model they were estimated in turns at 28.

Both descriptions can be defended, and it is worth saying how. He fitted two kinds of model: a piecewise linear spline, which forces a single breakpoint and found it at 28, and generalized additive models, which let the curve bend freely. A freely-bent curve has a flat bottom rather than a corner, and where you call the minimum of a flat stretch is a judgement call. So “32 or so” is a reading of the GAM, and 28 is the spline's knot. They are not the same claim, and the slopes everyone quotes belong to the one that turns at 28.

Others reading the same post landed on the lower number. Jaap Dronkers, writing on the same site three weeks later, summarised it as “the lowest divorce odds are for those who marry around the age of 28.” Wolfinger's own replication post split the difference and called it a range: “the 28 to 32 age range remains the period of lowest divorce risk.”

Four years is not a rounding error when the output is advice. A 30-year-old reading “after 32 it gets worse” hears that they have two years left; a 30-year-old reading “after 28 it gets worse” hears that they are already past it. The same model produced both sentences. This is how a figure drifts — the same drift that turned a film joke into the three-day rule and produced the numbered rules examined in what the 3-6-9 rule actually is.

#3

What 11% and 5% actually mean, and what they don't

The two percentages are odds ratios per year of age at marriage, which is a narrower thing than most retellings imply.

An 11% reduction is not an 11% chance of anything. It means that in this model, a person who married at 26 had about 11% lower odds of divorcing than an otherwise similar person who married at 25 — a relative change between two neighbouring ages, not an absolute risk. Stacked across several years it compounds into something substantial, which is why the teen-to-late-twenties stretch of the curve is steep.

The absolute rates are harder to get at than they should be, and the reason is in Wolfinger's own update to the post. The figures were generated with generalized additive models, “so the Y-axes are in a standardized metric that wouldn't resonate with most readers.” In other words you cannot read a divorce rate off the famous chart. He added a second figure estimating five-year divorce percentages by age group so readers would have something concrete.

Those are the percentages that now circulate, and they circulate second-hand because the figure is an image. Psychology Today's piece on the topic reports them as: roughly 38% of teen marriages ending within five years, 27% for the early twenties, 14% at ages 25-29, 10% at 30-34, and 17% at 35 and over. We cannot read the chart ourselves, so those numbers appear here as that page reports them rather than as something verified against the source.

Read the shape rather than the headline. The fall from 38 to 10 is almost the whole story. The rise from 10 to 17 is the part that got the coverage, and 17 is still well under half of 38. A page that tells a 36-year-old they are now in the danger zone is describing a group whose five-year divorce rate is less than half that of the group everyone agrees is at risk.

The uncontested part of the post is the part nobody argues with: “someone who marries at 25 is over 50 percent less likely to get divorced than is someone who weds at age 20.” That is the same arithmetic trap as the divorce figures attached to age gaps, where a large relative difference on a small base rate reads as a cliff.

#4

The curve only applies to marriages since 1998, and he says so

This is the single most limiting sentence in the whole record, it is on the author's own methods page, and no page competing for this question repeats it.

“The median respondent in this sample married in 1998. I divided the sample into two groups based on this median. The curvilinear relationship between age at marriage and divorce risk only applies for people married since 1998. For earlier cohorts, the pattern I describe in Wolfinger (2003) persists: marrying later produces incremental declines in the odds of divorce.”

The NSFG respondents in that sample married between 1976 and 2009. So the U-shape is a property of the more recent half: marriages formed in roughly the eleven years before the data were collected. In the older half, the old finding holds and later is simply better.

He presents this as evidence that something changed, and the post is explicit that it is a new development rather than a law of marriage. That framing is defensible. What it also means is narrower than the way the number is used. The finding is not there is a best age to marry. It is in American marriages formed in the decade before 2010, divorce risk stopped falling after the late twenties. Those are different sentences, and only the first one survived into the search results.

It has a second consequence that is rarely noticed. Restricting the curve to post-1998 marriages restricts it to short ones. A couple who married in 1998 and were surveyed in 2008 have been married ten years at most; a couple who married in 2007 have been married two. The claim about late marriage therefore rests almost entirely on what happens in the early years of a marriage, which is both the highest-risk period and the period least informative about whether a marriage lasts.

That is the same trap as the window problem in does living together before marriage lead to divorce, where studies of recent marriages and studies of long ones reported opposite findings for years because they were measuring different stretches of the same curve.

#5

The survey stopped at 44, and that decides what late marriages can show

Every number above comes from a survey with an age ceiling, and the ceiling sits right where the interesting part of the claim is.

The National Survey of Family Growth interviewed people aged 15 to 44 for both of the waves Wolfinger used. The range was only widened to 15-49 in 2015, years later. So a respondent who married at 40 could be observed for at most four years of marriage, one who married at 43 for at most one, and nobody who married after their mid-thirties could be followed long enough to know how their marriage turned out.

Wolfinger flags this himself, and flags it as a bias downward: “the NSFG is a sample of adults ages 15-44, so people in the 35+ group are especially likely to have married within a year or two of the data collection. Thus their divorce rates appear lower than they'd actually be if these people had been more fully observed.” He also reports that the 35-and-over group in 2006-2010 holds N = 379, which he calls more than adequate.

Philip Cohen flags the same ceiling as a bias upward: “He did not control for duration of marriage, which is messed up in his data anyway because of the age limits in the NSFG.” The logic is mechanical. Divorce risk is highest in the first years of a marriage. If the late-marrying group contributes almost nothing but early years, and marital duration is not flexibly controlled, then the coefficient on age at marriage can absorb what is really a duration effect.

Two careful people looked at one age ceiling and drew opposite biases from it, and both are plausible. Which one dominates is an empirical question about the model specification, and it cannot be settled from outside, because the models were never published. It is exactly the question a referee would have asked first.

It is also why a dataset's design limits deserve as much attention as its results — the same lesson as the research on friends-first relationships, where the famous claim turned out to rest on work that could not have measured what it was credited with measuring.

#6

Four days later he replicated it, and the second post says something the first one denied

On 21 July 2015, four days after the first post, Wolfinger published Replicating the Goldilocks Theory of Marriage and Divorce on the same site, using the 2011-2013 NSFG.

The headline result held. “The primary result… was almost identical to what I obtained from the 2006-2010 survey: the 28 to 32 age range remains the period of lowest divorce risk.” And: “I have now shown the Goldilocks effect using two different data sets… and more than 10,000 respondents. Its existence is beyond question. Explaining the Goldilocks effect, however, will require additional scholarship.”

Replication on an independent wave is a real strength and should count as one. Most of the folklore this site examines has never been tested twice.

But put the two posts side by side on the question of what happens when you control for demographics. Post one, listing sex, race, family structure of origin, age at survey, education, religious tradition, religious attendance, sexual history and metro size: “None of these variables seem to have much of an effect on the relationship between age at marriage and divorce risk,” and “the relation seems to function more or less the same for everyone: male or female, less or more educated, religious or irreligious.”

Post two, four days later, with the same list plus whether the respondent had a child before marriage: “Demographic differences between respondents can partially explain the Goldilocks relationship between marriage age and divorce,” and “including these controls yields a gentler increase in divorce risk for people marrying after their early thirties.”

That is not a contradiction in the data — different waves, one extra variable — but it is a different description of the same operation, and the second description is the weaker claim. The honest version of the finding, taking both posts together, is that part of the late-marriage upturn is demographic composition and part of it is not, with no stated split between them.

Every page that cites this topic quotes the first post. The concession is in the one that also carries the appendix with the variable coding.

#7

The same survey, in a journal, and the curve flattens instead

Two years before the blog post, the same wave of the same survey was analysed in a peer-reviewed, open-access journal, and the answer was different.

Evelyn Lehrer and Yu Chen published Delayed entry into first marriage and marital stability in Demographic Research 29(20), 521-542, in 2013. Data: the 2006-2010 NSFG — the identical wave. Sample: non-Hispanic white women, N = 3,184 after exclusions.

Their description of the shape is unambiguous. The association between age at first marriage and marital instability “is strongly negative until the late twenties, with the curve leveling off thereafter, i.e., there is no U-shape.”

Levelling off and turning upward are different findings, and only one of them licenses the advice. A flat curve after the late twenties says that waiting stops helping. A rising curve says that waiting starts hurting. Everything in the popular coverage depends on the second.

The difference between the two analyses deserves stating plainly, because it is not a case of one being wrong. They are not the same model. Lehrer and Chen restrict the sample to non-Hispanic white women on purpose, for a reason that matters later. They use age brackets rather than a continuous curve, and a bracket cannot find a bend inside itself. They model the probability of dissolution by the fifth year. Wolfinger pools everyone, adjusts for race with a dummy variable, and fits a flexible curve that can bend wherever the data suggest. Flexible curves find bends that brackets miss; they also find bends that are not there.

So a reader cannot adjudicate it, and neither can this page. What a reader can note is that one of these analyses went through peer review and is free to read, and that not one page on the first page of results for this question cites it. The disagreement is not hidden. It is in a journal that costs nothing to open, about the same survey, published first.

This is the same situation as the three papers arguing about one survey in the cohabitation literature, and the same reason our corrections log exists.

#8

Split that data by race and the U appears in one group only

The sharpest result in this literature is also the least cited, and it is a follow-up to the paper above.

Evelyn Lehrer and Yeon Jeong Son published Women's age at first marriage and marital instability in the United States: Differences by race and ethnicity in Demographic Research 37(9), 229-250, in 2017. Same 2006-2010 NSFG female sample, n = 12,279 in total, analysed group by group rather than pooled: white n = 3,184, Hispanic n = 1,269, black n = 724.

Three different curves came out.

  • Non-Hispanic white women. The fifth-year dissolution probability “falls steadily and substantially with age at marriage up to ages 30-32, and thereafter the curve flattens out.” They pre-empt the obvious objection: the 30-32 and 33-and-over cells hold 223 and 159 cases, “suggesting that the lack of significance is not a sample size issue.”
  • Hispanic women. Dissolution falls from age 20 to ages 21-23, flattens there, “beyond ages 30-32 it turns upward, producing a clearly U-shaped relationship.”
  • Black women. Falls until ages 24-26, flat afterwards. The point estimates hint at an upturn at 33 and over, but those cells hold n = 65 and n = 72 and the authors draw no conclusion, saying so explicitly.

Their stated contribution reads as a direct comment on the pooled approach: “The common practice in studies with US data of conducting pooled analyses with simple controls for black, Hispanic, and other can lead to misleading conclusions.” The pooled analysis of the same data had reported that the relation “functions more or less the same for everyone.”

And their explanation is not cultural, it is arithmetic about resources. What age at marriage buys differs sharply by group: 31% of the white women in this sample had completed sixteen or more years of schooling by the time they married, against 17% of black women and 10% of Hispanic women. If waiting brings a large gain in education and economic footing, it offsets whatever else waiting costs. If it brings a small one, it does not.

It is the second time in two weeks that the most-quoted figure in a divorce literature turned out to be hiding a racial interaction: the same thing happened to the 20% figure about in-laws. And the most-quoted figure of all in this field turns out not to be a figure at all: the claim that money arguments are the top predictor of divorce is a sentence from a press release, whose own authors published two papers that qualify it.

Further illustration for: Best Age to Get Married: The Famous 28-to-32 Window Came From a Blog Post
#9

The idea is from 1977, and it already has a name

The 2015 post is usually described as a discovery. It is a hypothesis from 1977 being tested again, and the literature has argued about it under a name for five decades.

Gary Becker, Elisabeth Landes and Robert Michael published An Economic Analysis of Marital Instability in the Journal of Political Economy 85(6), 1141-1187, in 1977. Using 1960s data they found a U-shaped relationship between age at first marriage and divorce, and explained the upturn after 30 as a poor match effect: someone searching longer in a thinning market ends up with a match further from the one they would have chosen, and that mismatch shows up later as instability.

That is the mechanism the Goldilocks coverage gropes toward without naming. Wolfinger's own guess is a version of it — “my money is on a selection effect: the kinds of people who wait till their thirties to get married may be the kinds of people who aren't predisposed toward doing well in their marriages” — followed, to his credit, by “this is all conjecture.” In a blog comment the same week, asked about causality, he answered: “As to causality, I have no idea.”

The peer-reviewed line of work testing the 1977 hypothesis is easy to find, because one of the papers is named after it. Lehrer's Age at marriage and marital instability: revisiting the Becker-Landes-Michael hypothesis appeared in the Journal of Population Economics 21(2), 463-484, in 2008, using NSFG cycles 5 and 6. Its published finding: the relationship “is strongly negative up to the late 20s, with a flattening of the curve thereafter.” Flattening again, on the two waves before the one the blog post used.

None of the pages ranking for this question names Becker, Landes and Michael, the poor match effect, or Lehrer. A reader is told that a sociologist found a startling new reality in 2015, when what happened is that a fifty-year-old question got another answer.

This matters beyond credit. A question with a research tradition comes with a vocabulary and a standard set of objections attached, and a finding presented outside that tradition arrives without them — the same thing that happened to the five love languages, where the popular version outran the measurement for thirty years.

#10

The poor-match effect is real, and in the data it loses

Lehrer and Chen's useful contribution is not the null result on the U-shape. It is what they found in its place, which is a better answer to the question than any single age.

They tested the 1977 mechanism directly and confirmed it. Women who marry after their late twenties do disproportionately make what the literature calls unconventional matches — heterogamy in age, in religion, in education — and those characteristics are, on their own, associated with higher marital instability. The poor match effect exists and is measurable.

Then it gets outweighed. In their words: “the destabilizing effects associated with indicators of unconventional matches are also present in marriages contracted late, but are dwarfed by the stabilizing influences associated with higher levels of education and older ages.”

Which produces the most counterintuitive sentence in this literature, and it is in their conclusions: “couples with traits normally associated with high risk of divorce actually tend to have very stable marriages if their unions are contracted at relatively late ages.”

So the picture the numbers support is two forces pulling against each other. Waiting makes the match less conventional, which costs stability. Waiting also brings education, income that has settled, and age itself, which buys stability. In this sample the second is much larger, which is why the curve flattens rather than rising.

And that is what makes Lehrer and Son's split by race the hinge of the whole topic rather than a footnote. In a group for whom waiting brings a large gain in schooling and resources, the stabilising force wins and the curve goes flat. In a group for whom it brings a small one, the poor-match force can win and the curve turns up. The answer to is there a best age is then: it depends on what waiting actually buys you, and that is not the same for everyone.

Stated that way the question stops being about a birthday and starts being about what is true of a particular life at a particular time — which is closer to what the markers of a relationship with real potential and the questions that reveal real compatibility are about.

#11

Survival and quality point in opposite directions

Everything so far measures one thing: whether the marriage ended. There is a peer-reviewed study that measured both whether it ended and whether it was any good, and the two answers run opposite ways.

Norval Glenn, Jeremy Uecker and Robert Love published Later first marriage and marital success in Social Science Research 39(5), 787-800, in 2010 — five years before the blog post. They pooled five American datasets: the General Social Survey plus four marriage-initiative surveys, mostly random-digit-dial telephone samples. Their published summary: later marriages “fare very well in survival but rather poorly in quality.” The greatest indicated likelihood of being in an intact marriage of the highest quality was among those who married at ages 22-25.

Their proposed mechanism is uncomfortable and worth saying out loud: “the constraints against divorce apparently are considerably stronger among those who marry relatively late.” People who marry late are more reluctant to leave. And a stable unhappy marriage counts as a success in every model that measures divorce — the same blind spot that makes the research on why people stay in unhappy relationships a necessary companion to any divorce statistic.

Then the brake the authors applied themselves, which is quoted by exactly one page in the current results and ignored by the rest: the negative relationship beyond the early-to-mid twenties “is likely to be at least partially spurious, and thus it would be premature to conclude that the optimal time for first marriage for most persons is ages 22-25.”

They tried to test the spuriousness. A maximising-personality item — “no matter how satisfied I am, I'm always on the lookout for something better” — moved the 22-25 coefficient from .498 to .449, about ten per cent, and not significantly. They report it as a crude test and say why: a single item from a scale with a reliability of .409, measured years after some of the marriages ended.

The consequence for the question is large. The two most-cited answers — 22-25 and 28-32 — come from studies measuring different outcomes. If both are right there is no best age at all. There is a trade-off between a marriage lasting and a marriage being good, and nobody searching for a number is being told that is the choice.

#12

The critique was filed the same day, with the code attached

On 17 July 2015, the day the IFS post appeared, Philip Cohen — a sociologist at the University of Maryland — published a critique on his blog Family Inequality, under the title No, you should get married in your late 40s (just kidding).

His four objections, in his words and in order of weight:

Nothing was shared. “Wolfinger has not shared his code, data, models, or tables, so it's hard to know what he really did.” Cohen published the codebook for his own IPUMS extraction and his Stata code in the same post.

The bump is small. Using the American Community Survey, women in first marriages, controlling for marital duration, education, race, Hispanic ethnicity and nativity: “Yes, there is a little bump up for the late 30s compared with the early 30s, but it's very small.”

And it is a duration artefact. In an update added two days later, from a model with age-at-marriage-by-duration interactions, “the late-30s bump is concentrated in the first five years of marriage” — which is the same objection as the age ceiling, now measured.

The variable may not be separable at all. Age at marriage and education cause each other, and “college graduates are only the majority of women getting married for the first time in the age range 27-33.” His closing line is the deepest thing anyone has said about this question: “It really may not be reasonable to try to isolate the age-at-marriage effect after all.”

Wolfinger's reply is a fair technical objection: the ACS lacks the marital history needed for event-history analysis, Cohen's model counts only divorces in the twelve months before interview, and so “Cohen's analysis is biased and therefore not dependable.”

There is one wrinkle in the record worth noting, because both halves are the same author's. Wolfinger's methods page reproduces Cohen's ACS figure with the caption: “and here are data courtesy of Phil Cohen from the American Community Survey which show the same finding.” The replication post, four days later, calls that same analysis undependable. The ACS appears as corroboration on one page and as unreliable on the other.

When two competent people get different answers from different surveys and neither can see the other's models, the honest summary is not one of the two answers — the same conclusion the research on couples therapy reaches by another route.

#13

Across countries, the shape depends on how much divorce there is

Three weeks after the original post, the same site published a cross-national test, and its result reframes the whole question.

Jaap Dronkers, then chair of international comparative research at Maastricht University, used the Generations and Gender Programme: a survey of 18-to-79-year-olds in 19 European countries from the first decade of the 2000s, restricted to people born after 1940 who started a first union between 14 and 50, men and women, with education and the partner's existing children as the only background variables.

What he found was not one curve but a condition on it. For marriage, the quadratic term was highly significant in four places: Estonia, Russia, the UK and the USA. For first unions including cohabitation — which is how most European relationships start — it was significant in more than half the countries.

Where it was absent is the informative part. He reports no significant quadratic term for Georgia and Romania, the two countries with the lowest dissolution rates in the data, and concludes that the late-marriage upturn “exists only in societies with higher rates of union dissolution across the board.”

He tested two explanations and neither survived. Late starters do more often pair with someone who has already been through a dissolution, because the pool of never-partnered people thins with age — but adding an indicator for that “does not change the importance of the curvilinear relation.” Nor did controlling for the partners' education. His own closing sentence is the appropriate one: “we do not know why that's the case.”

Two honest limits. This is another blog post rather than a reviewed paper. And its evidence sits in tables published as images, so no number from inside them is reported here — only the claims the text makes about them.

The logic is the same one that dismantles most of this genre's correlations, and it is set out at length in does living together before marriage lead to divorce. When a behaviour is rare, the people doing it are unusual, and what makes them unusual predicts the outcome. When the behaviour is ordinary, the signal goes. If the age effect were a fact about human development it would not care what country it was measured in. It does.

#14

Two studies tried to make it causal, and got different answers

Nobody can assign people at random to marry at 24 or at 34, which is why every figure above is an association. Two designs have got closer, and they disagree in an instructive way.

Dana Rotz, Why Have Divorce Rates Fallen? The Role of Women's Age at Marriage, Journal of Human Resources 51(4), 961-1002, 2016. She combined the Survey of Income and Program Participation, the 1979 NLSY and the NSFG, used state minimum-age-at-marriage laws as instrumental variables, and compared sisters' marital histories to test whether family background was driving the estimates. Her conclusion is that the rise in women's age at marriage is the main proximate cause of the fall in American divorce since 1980, with factors like women's labour-force participation and contraception working largely through it. In the 2012 working-paper version: “all estimates suggest that age at marriage can explain 60 percent or more of the decline,” and holding the bride's age constant, marriages formed between 1980 and 2004 are at roughly equal risk. (Those state laws exist because marriage below 18 is associated with coercion, which is a separate literature and a real one.)

Note the direction. In Rotz's analysis, marrying later is the reason divorce fell.

Jorge García-Hombrados and Berkay Özcan, Age at marriage and marital stability: evidence from China, Review of Economics of the Household 22(1), 297-328, published online in 2023. They used a 1981 Chinese reform that made marrying before 25 much easier for urban women, in a fuzzy regression kink design on census data. They first reproduce the association — the U-shape is present in urban China too — and then estimate the causal effect of delaying marriage by one year. It is “consistently small and statistically indistinguishable from zero at conventional confidence levels,” and their summary is that “the causal effect of advancing marriage on the probability of divorce is close to a precise 0.” Their reading: “unobserved factors drive this association.”

Their own caveat is the kind worth copying. The upper bounds of those confidence intervals “reach 3 and 4 percentage points,” so a modest real effect is not excluded. They also note the same null in a study of American men who married young to dodge the Vietnam draft.

The two fit together once you see which ages each one moves. Rotz's instruments move teenage marriage, where every study agrees the effect is large. The Chinese reform moves marriage around the early twenties, where the causal estimate is zero. Nothing instruments the choice between marrying at 29 and marrying at 34 — which is the only choice the popular version of this question is addressed to.

#15

What the pages at the top of the results are actually citing

Measured today, the first page for best age to get married runs: Focus on the Family, Brides, two Pew Research Center pages, a Bonobology piece about age gaps, Psychology Today, YourTango, a content site of the kind that has appeared in the last two years, and a Reddit thread. No Healthline, no Cleveland Clinic, no Verywell Mind — this question sits in the half of the niche the medical authorities have left alone, like the numbered rules and the tests.

What those pages are built on is worth itemising, because it is not what a reader would assume.

Pew's number is an opinion poll. “On average, Americans say the best age to get married is 26.5” comes from asking 3,600 US adults what they think is best. About half of them answered that there is no best age at all. It is a survey of beliefs, and it ranks for a question about divorce risk.

YourTango's 28-and-32 is a therapist's clinical impression. The page gives 28 for women and 32 for men, sourced to a relationship therapist describing what she sees in her practice, on a page that calls the whole thing “backed by science and math.” The 32 has travelled from a spline knot in a statistical model to an opinion about men, with the number intact and the meaning gone.

On the sibling query, the top result is a TikTok. For best age to get married to avoid divorce, the first result and one of the others are built on a divorce lawyer's video, with IFS's own news aggregator carrying it.

The one page that cites a journal has a position. Focus on the Family cites Glenn, Uecker and Love properly, quotes the authors' own caution, and adds a 2023 master's project — and never mentions Wolfinger, because 22-25 supports its argument for earlier marriage and 28-32 does not.

The most careful page is citing the blog. Psychology Today's 2016 piece reports the finding accurately and lists two references: the two IFS posts.

That is the evidence base for a question millions of people type: one unreviewed analysis, one opinion poll, one therapist, one lawyer, and three peer-reviewed papers nobody is reading. It is the same pattern as the statistics that circulate about modern dating and the numbers attached to dating apps.

#16

The version of the question that has an answer

What is the best age to get married? cannot be answered by evidence, and the reason has little to do with the quality of the studies.

Age at marriage is not a treatment anybody received. It is a summary of everything that happened to be true about a person in the year they married: how much education they had finished, whether their income had steadied, whether there was a child from an earlier relationship, how thin the pool they were choosing from had become. The literature has measured all of those separately, and in Lehrer and Chen's models they do the work that age appears to be doing. Cohen's version is blunter: it may not be reasonable to try to isolate the effect at all.

What survives, stated as narrowly as the data allow:

  • Marrying as a teenager is strongly and consistently associated with divorce. It is the one part of the curve nobody disputes, it is where the quasi-experimental evidence is strongest, and it is a much bigger effect than anything else on this page.
  • The decline continues into the late twenties in every analysis.
  • What happens after that depends on which survey, which subgroup, which outcome and which country. Three peer-reviewed analyses of the very dataset behind the famous number report the curve levelling off rather than rising, and when that data is split by race and ethnicity the clear U-shape appears in one group of three.
  • The one quasi-experiment that has estimated the effect of marrying a year later found something close to zero.

None of that supports a deadline, and it especially does not support marrying sooner than you otherwise would in order to land inside a window. Nobody has studied people who did that. The one thing that was measured about deliberate timing runs the other way: Glenn and colleagues found that people who marry late are more reluctant to leave a marriage that is not working, and treating a date as the goal is a way of manufacturing exactly that constraint.

If you want a question with an answer, pick one about the relationship instead of the calendar. How the two of you handle a disagreement that does not resolve. Whether you have settled what you are to each other before changing anything structural — the finding that keeps surviving in the cohabitation literature and in what predicts a relationship with real potential. Whether you would both describe the relationship the same way, which is the variable that did the work in the research on in-laws.

The conflict style test scores the pattern that tends to decide how a long disagreement goes, the relationship satisfaction test scores the dimensions the literature actually uses, and the attachment style test covers what you bring to the reading of any of it. More of this kind of thing is in the Mindset archive.

A 29-year-old wondering whether they have missed a window has been handed a number generated by a model that turns at 28, fitted to the recent half of one survey, in one country, in marriages mostly shorter than a decade, by an analysis whose code was never published. That is not nothing. It is also not about them.

Pulling it together

The honest one-line answer to this question is that the evidence does not contain one, and the most confident pages are the least sourced. Marrying as a teenager carries a large, repeatedly measured risk. Risk keeps falling into the late twenties. After that the picture splits by dataset, by subgroup, by outcome measure and by country, and the single analysis that says it turns upward is a blog post whose own appendix puts the turn four years earlier than the headline and restricts it to marriages formed after 1998.

What is more useful than any of the numbers is the reason they disagree. Waiting makes a match less conventional, which costs stability, and it brings education, settled income and age, which buy stability. Whichever is larger for a given person decides the shape of their curve — which is why the clear U-shape shows up in the group for whom waiting bought the least.

So the quantity that matters was never the birthday. It is what the extra years are actually doing, and whether the relationship in front of you works. The second of those is measurable, it is measurable now, and no survey needs to be consulted about the year you were born.

Frequently asked questions

What is the best age to get married?

No study can answer that, and the widely quoted answer is weaker than it looks. The 28-to-32 window comes from a 2015 Institute for Family Studies blog post by Nicholas Wolfinger analysing the 2006-2010 National Survey of Family Growth, which reported that the odds of divorce fall about 11% per year of age at marriage and then rise about 5% per year. His own methods page puts the turn at age 28, not 32, and says the curve applies only to marriages contracted since 1998. Three peer-reviewed analyses of the same survey — Lehrer 2008, Lehrer and Chen 2013 and Lehrer and Son 2017 — report the curve flattening after the late twenties rather than rising. What is agreed is narrower: marrying as a teenager carries a large and consistently measured risk, and risk falls steeply into the late twenties.

Is it true that divorce risk goes up after age 32?

It is contested, and the number itself is unstable. The claim comes from a blog post, not a journal article, and the model behind it has its breakpoint at 28; “32 or so” is a reading of a different, freely-bending version of the same curve. Philip Cohen, analysing American Community Survey data the same week, found “a little bump up for the late 30s compared with the early 30s, but it's very small,” and in a later update found that bump “concentrated in the first five years of marriage,” which points at marital duration rather than age. Lehrer and Chen, using the identical survey wave in Demographic Research in 2013, concluded there is “no U-shape” for non-Hispanic white women. The upturn is real in some analyses and absent in others, and nobody has settled which specification is right because the original models were never published.

What is the Goldilocks theory of marriage?

It is a label a journalist gave to a blog finding. Jordan Weissmann coined it in Slate on 16 July 2015, describing Wolfinger's analysis: “Getting married too early is risky, but so is getting married too late. Your late 20s and early 30s are just right.” The underlying idea is older than the label by nearly forty years. Becker, Landes and Michael reported a U-shaped relationship between age at marriage and divorce in the Journal of Political Economy in 1977 using 1960s data, and explained the upturn after 30 as a “poor match effect” — longer search in a thinner market producing matches further from the one someone would have chosen. The 2015 finding is a retest of that hypothesis, and the peer-reviewed retests of it mostly find a flattening rather than a U.

Does marrying in your early twenties mean a worse marriage?

It means a measurably higher chance of divorce, and possibly a higher chance of a happy marriage, and those are not the same question. On divorce, the association is strong and uncontested: Wolfinger reports that someone marrying at 25 is over 50% less likely to divorce than someone marrying at 20, and that teen marriages have by far the highest five-year divorce rates. On quality, Glenn, Uecker and Love found in Social Science Research in 2010 that the greatest likelihood of being in an intact marriage of the highest quality was among people who married at ages 22-25 — while explicitly warning that this relationship “is likely to be at least partially spurious” and that it would be “premature” to call 22-25 optimal. Nothing in either literature supports marrying earlier than you otherwise would.

Does marrying later mean a happier marriage?

The evidence suggests the opposite trade-off, and it is the finding most often left out. Glenn, Uecker and Love pooled five American datasets and reported that later marriages “fare very well in survival but rather poorly in quality.” Their proposed reason is uncomfortable: “the constraints against divorce apparently are considerably stronger among those who marry relatively late.” People who marry late appear less willing to leave a marriage that is not working, which produces more stable-but-mediocre marriages — and a stable unhappy marriage counts as a success in every study that measures divorce rather than satisfaction. So a survival-based answer and a quality-based answer to “what is the best age” can point in opposite directions, and almost every page offering a number is quoting only the survival one.

Why do studies disagree about the best age to marry?

Four reasons, all of them documented. The outcome measured: divorce and marital quality give different answers. The sample: the National Survey of Family Growth interviewed only people aged 15 to 44 until 2015, so late marriages could only be watched for a few years. The specification: with one plausible set of choices the curve bends upward, with another equally plausible set it flattens, and Philip Cohen's point is that age at marriage and education cause each other so thoroughly that “it really may not be reasonable to try to isolate the age-at-marriage effect after all.” And the subgroup: Lehrer and Son found the fifth-year dissolution curve flattening after 30-32 for white women, turning clearly upward for Hispanic women, and flat after 24-26 for black women, in the same data, and wrote that pooled analyses with a simple race control “can lead to misleading conclusions.”

Does waiting too long mean you end up settling for the wrong person?

That is the oldest explanation on offer and the data treat it as real but outweighed. The “poor match effect” was proposed by Becker, Landes and Michael in 1977. Lehrer and Chen tested it on the 2006-2010 NSFG and confirmed the first half: women who marry after their late twenties do disproportionately make matches that are unconventional in age, religion and education, and those traits do predict instability. But the destabilising effects “are dwarfed by the stabilizing influences associated with higher levels of education and older ages,” and the paper's striking conclusion is that “couples with traits normally associated with high risk of divorce actually tend to have very stable marriages if their unions are contracted at relatively late ages.” Waiting changes who is available; it also changes what you bring, and in that data the second effect was larger.

Has anyone shown that marrying at a particular age causes divorce?

For teenage marriage, close to it. For the ages people actually ask about, no. Dana Rotz, in the Journal of Human Resources in 2016, used state minimum-age-at-marriage laws as instrumental variables and sisters' marital histories as a check, and concluded that rising age at marriage is the main proximate cause of the fall in American divorce rates since 1980 — her working paper puts it at 60% or more of the decline. García-Hombrados and Özcan, in the Review of Economics of the Household in 2023, used a 1981 Chinese reform that made marrying before 25 much easier, in a regression kink design on census data: they reproduce the U-shaped association in urban China and then find the causal effect of delaying marriage by one year “close to a precise 0,” concluding that “unobserved factors drive this association.” Honestly reported, the confidence intervals still allow 3 to 4 percentage points. And no study instruments the choice between marrying at 29 and marrying at 34.

Sources

What this article draws on. For the broader bibliography see our sources page.

Keep reading