Does Living Together Before Marriage Lead to Divorce? Four Things Change the Answer
The famous 1.31 comes from women under 45 in first marriages. Control for the age they moved in together, not the age they married, and it drops to 1.09.
Before the article. The literature below includes a mechanism — inertia — whose whole claim is that moving in together makes a relationship harder to leave before it makes it better. If that describes a relationship you are in and leaving feels unsafe rather than merely complicated, the last section of this page covers it and the sources at the bottom link to Find A Helpline, which 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, 24 hours a day.
With that said. This question has one famous answer and it is narrower than it sounds. The figure almost every page repeats — a 1.31 times higher odds of divorce — comes from a single harmonised dataset of 24,888 American women in first marriages, aged 44 and under, published by Michael Rosenfeld and Katharina Roesler in 2019. It is a solid finding about that population. It is not a fact about marriages.
Four things move it, and each has a named study behind it. Which age you control for: switch from age at marriage to the age a couple moved in together and the hazard ratio falls from 1.37 to 1.09 and stops being significant. How long the marriage has run: the famous paper's own data show cohabiters doing marginally better in the first year and worse later. Whether a commitment came before the move: the split that keeps replicating. And which country you are in: across sixteen European countries, and again across twenty-one, the direction of the effect tracks how common cohabitation is.
There is also a part of the record that none of the ten pages competing for this question mentions. The same two authors published a follow-up in the same journal in 2024 reporting that premarital cohabitation predicted dissolution for non-Hispanic White wives and not for non-Hispanic Black wives, with the difference statistically significant.
Fifteen sections follow, and the evidence is contested throughout. Our editorial policy explains why every empirical claim here carries a linked source, and why this page says where the researchers disagree rather than picking a winner.
The number everyone repeats, and the sample it came from
Almost every page that answers this question is answering it with one study, so it is worth knowing exactly what that study measured.
Michael Rosenfeld and Katharina Roesler harmonised six waves of the National Survey of Family Growth into a single event-history dataset and published the result in the Journal of Marriage and Family in 2019. The dataset holds 216,455 couple-years drawn from 24,888 women in first marriages. Using discrete-time event history models, they found the odds of a marriage ending were 1.37 times higher for couples who had cohabited before marrying with no controls applied, and 1.31 times higher once controls were added. The confidence interval on that adjusted figure runs from 1.25 to 1.38, and the likelihood ratio test came out at 117.54 on one degree of freedom, which is a p value somewhere around two in a thousand trillion trillion. Statistically, it is not a fragile result.
Now the part that almost never travels with the number. The sample is women only. The authors say so plainly: the NSFG did not add male respondents until its 2002 wave, so the usable marital histories run through women aged 15 to 44. It is also first marriages only, excluded deliberately because second and third marriages behave differently. And 18,674 of the 24,888 women were white.
So the headline finding describes first marriages of American women under 45, mostly white, surveyed about their own histories. That is a real and useful population. It is not marriages, and anyone over 44, anyone on a second marriage, and every husband in the country is outside it.
The underlying rates are worth holding onto too, because they are small. Across the whole dataset the annual breakup rate was 4.70% for couples who had cohabited and 3.47% for couples who had not. The question is about the gap between those two numbers, not about a cliff edge.
The first-year advantage is two-tenths of a percentage point
The same paper is also the source of the more hopeful claim now circulating: that cohabiting couples do better at the start of a marriage. That is what the authors found, and the size of it is rarely reported.
In the first year of marriage, couples who had not cohabited had a breakup rate of 4.1%. Couples who had cohabited had a breakup rate of 3.9%. The authors state in the text that the raw difference between the two was not statistically significant at twelve months, though it was significant up to six months in the Kaplan-Meier analysis. In the multivariable models the early advantage holds up; in the raw curves, at the one-year mark, it does not.
Their explanation is practical rather than psychological. A couple who already shares a home has done the adjusting — the dishes, the sleep schedules, the money — that other couples start doing the week after the wedding. They call it an experiential advantage, and they argue it lasts about a year, after which the couples who married directly have caught up and the longer-term pattern takes over. The paper's own summary is that premarital cohabitation has short-term benefits and longer-term costs for marital stability.
This is the pivot on which the whole public argument turns, so it matters that the pivot is two-tenths of a percentage point. When a page tells you cohabiting protects a marriage at the start, that is the quantity being described. It is a real finding in the models and it is not a reason to do anything.
It is also why the literature looked for years as though it were contradicting itself. Studies that measured recent marriages were catching mostly short durations, where the two groups look alike or the cohabiters look slightly better. Studies that followed marriages for longer were catching the later divergence. What the research on couples who stay together reports runs into the same problem from the other direction: the window you measure decides the answer you get.
Which clock you control for decides the answer
Here is the finding that should be the headline of this topic and is on none of the pages that compete for it.
Arielle Kuperberg asked a question about measurement rather than about cohabitation. Every study of this question controls for age, because marrying young predicts divorce and cohabiters marry at different ages. But which age? The convention is age at marriage. Kuperberg pointed out that for a couple who lived together first, the marital roles — shared money, shared household, shared daily life — began when they moved in, not when they married. So she ran the same models both ways.
Using NSFG waves from 1995, 2002 and 2006-2010, N = 7,037 women, with Cox proportional hazards models predicting dissolution:
- With no age control at all, premarital cohabitation carried a hazard ratio of 1.40.
- Controlling for age at coresidence, it fell to 1.29.
- Controlling for age at marriage, it rose to 1.63.
Then she added demographic controls — previous cohabitation with someone else, race, education, mother's education, family structure at 14, growing up with no religion, a pre-coresidential birth or moving in pregnant, husband's relative age and race:
- With no age control, 1.21.
- Controlling for age at marriage, 1.37, significant at p < .001.
- Controlling for age at coresidence, 1.09 — not statistically significant.
Same survey. Same women. Same covariates. One switch in which age you treat as the start of the clock, and the effect that the whole genre is built on stops being there. Kuperberg reports that the difference between the two hazard ratios was itself significant at p < .05 on a one-tailed test, and that the model fit statistics favoured age at coresidence as the better predictor.
This does not prove the cohabitation effect is an artefact. It proves that a reader cannot be told what the effect is without being told which clock was running.
What age at coresidence means if you are the person deciding
Strip the statistics out and Kuperberg's argument is about what counts as the same event.
Two couples marry at 26. One met at 19, moved in at 20, and married six years later. The other met at 25 and married within the year without living together. Standardising those two by age at marriage treats them as the same starting point, and they are not: the first couple has been running a joint household since one of them was twenty. If moving in at twenty predicts the same difficulties that marrying at twenty predicts — immaturity, unfinished education, income that has not stabilised, a self that is still changing — then the comparison is loaded before the analysis begins.
Kuperberg's data say the loading is substantial. Age at coresidence accounted for more than a quarter of the gross association between cohabitation and divorce before any demographic controls, and once those were added, measuring age the conventional way left a hazard ratio four times larger than measuring it her way.
What that supports saying is narrow and worth saying anyway. Of the risk that gets attributed to living together before marriage, a large and measurable share is attributable to living together young, and the research design that produced the famous number could not tell the two apart.
What it does not support is a rule. Nothing here says that moving in at 28 is safe or that moving in at 20 is a mistake; the data are observational, the people who move in at twenty differ from those who move in at twenty-eight in dozens of ways nobody measured, and age at coresidence is itself a proxy for all of them. The same caution applies to the age numbers in what age gaps actually predict, where the gap turns out to carry much less than the ages themselves. What the ages themselves carry, and where the famous window of 28 to 32 came from, is in the best age to get married.
If you want the version of this question that a measure can actually speak to, it is not should we move in but how is this relationship working, and the relationship satisfaction test scores the dimensions the literature uses for that.
Three papers in one journal, arguing about the same survey
The reason this question has no settled public answer is that it has no settled academic answer, and the disagreement is a matter of public record in a way that is unusually easy to check.
In 2021 the Journal of Marriage and Family ran a formal comment by Wendy Manning, Pamela Smock and Arielle Kuperberg on the 2019 paper, and Rosenfeld and Roesler's reply, back to back in the same issue. Both are short. Both are about modelling choices rather than about couples.
The comment argues that the 2019 models include too many overlapping indicators of time — calendar year, marriage cohort and marital duration at once — which makes the interaction terms hard to interpret. Reanalysing with marriage cohort as the time axis, as earlier studies had, they find the association weaker among recently married couples and, for cohabitation with the eventual spouse in the most recent cohorts, not present at all. Their conclusion is that nobody has been misled.
The reply is blunt. It says the comment discarded more than 70% of the valid couple-years in the NSFG and then used the resulting loss of statistical power to call a key interaction non-significant; that it treated the presence of children as a fixed trait when children arrive partway through a marriage; and that it dropped one survey wave without explanation.
A reader cannot adjudicate that, and neither can this page. What a reader can take from it is the shape of the thing: the question of whether the cohabitation effect still exists in recent American marriages turns on whether you use calendar year or marriage cohort as your time axis, and on how much of the data you keep. Those are defensible choices with different answers attached, made by careful people who have read each other's work.
When two expert teams get opposite results from the same survey, the honest summary is not one of the two results. It is that the data do not settle it — which is the same conclusion that the research on couples therapy reaches by a different route, and the reason our editorial policy commits to saying so when it happens.
The authors' own follow-up says it is not a general finding
This is the part of the record that nothing on the first page of results mentions, and it comes from the same two authors in the same journal.
In 2024 Rosenfeld and Roesler published a second paper in the Journal of Marriage and Family, covering seven decades of American marital histories from the NSFG, 1950 to 2017, and asking which predictors of divorce have changed over time. Premarital cohabitation was one of the predictors they examined. They also did something the 2019 paper did not: they tested whether the non-racial predictors of divorce operate the same way across racial groups.
They do not. The paper reports that premarital cohabitation was associated with marital dissolution at a hazard ratio of 1.30 for non-Hispanic White wives, and that there was no such association for non-Hispanic Black wives — with the difference between the two groups statistically significant. They report the same pattern for low education and for teen marriage: the predictors that carry risk for white wives carry less of it, or none, for wives from racial minority groups. They call it the Racial Minority Moderation hypothesis, and they find support for it.
Read that against the headline. The 1.31 is not a fact about marriages, and after this paper it is not even a fact about all the women in the dataset it came from. It is a fact about a subgroup, and the subgroup it fails for is the one the original paper noted was a quarter of its sample.
Two honest limits. The 2024 paper is still the same survey, the same women-only design and the same observational logic, so it inherits every caveat above. And a null result in a subgroup can mean the effect is absent or that the subgroup is smaller and the estimate noisier — though the authors report the difference as significant, which is the test that matters and the one usually skipped.
What it rules out is the version of this claim that gets repeated. Whatever premarital cohabitation is doing, or standing in for, it is not doing it to everyone.
Whether you had decided anything before you moved in
Across studies that split cohabiters rather than treating them as one group, the split that keeps mattering is not how long or how young, but whether a commitment existed before the boxes were carried in.
Scott Stanley, Galena Rhoades, Paul Amato, Howard Markman and Christine Johnson surveyed marriages formed in the 1990s across four states and followed first marriages to 2001. By then 28% of people who had cohabited before being engaged had divorced, against 21% of those who cohabited only after engagement and 20% of those who did not live together until marriage. In the event history model with the full set of covariates, the annual probability of divorce at covariate means was .027, .019 and .015 for the three groups. The odds of divorcing in a given year were 45% lower for the after-engagement group than for the before-engagement group, and the difference between the after-engagement group and the never-cohabited group was not significant.
The sample is the limit here and the paper says so: only 124 divorces occurred in the window, which is not many events for an event-history model, and the four states were Arkansas, Kansas, Oklahoma and Texas rather than the country.
Nationally representative data point the same way with an extra wrinkle. Wendy Manning and Jessica Cohen used the 2006-2008 NSFG to look at marriages formed since 1996 — 2,003 women and 1,483 men — and found that a simple yes-or-no measure of premarital cohabitation was not significantly associated with marital instability for either sex, at the zero order or in the full model. When they split cohabiters by whether they were engaged or had definite plans to marry when they moved in, women who were committed first had significantly lower hazards than women who were not. For men, the commitment split did not predict anything.
That gender asymmetry is not a throwaway. It is the pattern Rhoades, Stanley and Markman had already found when they asked about commitment directly, and it shows up again in the markers of a relationship with real potential: the two people in a couple are often not describing the same relationship.
The only meta-analysis, and nobody on the first page cites it
One pooled analysis of this literature exists. Anita Jose, K. Daniel O'Leary and Anne Moyer published it in the Journal of Marriage and Family in 2010, combining 16 studies of marital stability and 12 of marital quality.
Its findings, as the published abstract states them: cohabitation had a significant negative association with both marital stability and marital quality. Moderator analyses found the effects had remained consistent over time, even though cohabitation had become normative — which is itself an argument against the comfortable idea that the association must be fading as the behaviour spreads.
And then the result that matters most to a reader who is already living with someone: the negative association with marital stability did not remain when the analysis was restricted to people who had cohabited only with the person they went on to marry. The authors' reading is that those cohabiters attach more long-term meaning to living together.
That is the single most decision-relevant finding in this literature, because it is the only one addressed to the actual situation most people are in when they type the question. Living with a series of partners and living with your fiancé are different behaviours, they have been pooled together for decades under one label, and when you stop pooling them the effect on stability is where the pooling was.
Two things have to be said about it honestly. First, this paper is behind a paywall and the open-access indexes list no readable copy, so everything above comes from the published abstract and the effect sizes are not reported here — a meta-analysis described without its effect sizes is a weaker thing than one read in full, and that is the state of it. Second, 2010 is a long time ago for a literature that has been fighting since, and the k values are small.
Even so: of the ten results on the first page for this question, none cites this paper. The comparison that changes the answer for most readers has been sitting in the same journal as the famous number for fifteen years.
Across countries, the effect changes sign
Everything above is about the United States. Every page competing for this question is about the United States. The literature is not.
Aart Liefbroer and Edith Dourleijn published a test in Demography in 2006 using data from 16 European countries, asking whether the cohabitation effect depends on how far cohabitation has spread in a given society. Their published conclusion, in their own words: former cohabitors run a higher risk of union dissolution than people who married without prior cohabitation only in societies in which cohabitation is a small minority or a large majority phenomenon. In the middle, where roughly half of couples have cohabited, the gap is smallest. They describe the relationship as U-shaped, with its low point around 50%.
That paper is also paywalled with no readable open copy, so the description above comes from its abstract and from how a later team describes its method; no numbers from inside it appear here.
The later team is the useful part. In December 2025, Maike van Damme and Fernando Ruiz-Vallejo posted a replication using the Harmonized Histories from the Generations and Gender Programme — 21 countries and 14 five-year birth cohorts, from those born in the early 1930s to those born in the late 1990s — with multilevel discrete-time event history models. They find the U-shape again, with the low point at around 40% rather than 50%, and they attribute the difference to more recent data and a multilevel rather than fixed-effects design.
They also report something that points the opposite way from the American headline: in their pooled models, married couples who had cohabited first had a lower separation risk than couples who married directly.
This is a preprint and has not been peer reviewed, which is the whole caveat and a real one. Its descriptive table is also coarse — raw five-year breakup percentages to two decimal places, no controls, no intervals — and across the 21 countries the direction of the former-cohabitor column against the directly-married column is genuinely mixed, lower in some, higher in others, equal in several. The modelled result is the claim; the table is not.
Why the same behaviour predicts opposite things in two countries
The diffusion idea is worth sitting with, because once it clicks, most of the apparent contradiction in this literature stops being contradictory.
An association between a behaviour and an outcome tells you about the people who do the behaviour. When almost nobody cohabits, the people who do are unusual, and what makes them unusual — weaker attachment to the institution of marriage, more tolerance of separation, a different relationship to religion or to family expectation — predicts divorce on its own. The cohabitation is a marker. When almost everybody cohabits, the people who don't are the unusual ones, and what makes them unusual — often strong religious commitment, often a low tolerance for divorce — predicts staying married on its own. The cohabitation is still a marker; it has just changed which direction it points.
The American numbers show the ground moving underneath the question. In Rosenfeld and Roesler's data, the share of women who had lived with the man who became their first husband was 11% in 1970, then 34% in 1980, 46% in 1990, 60% in 2000, peaking at 70% in 2011. The behaviour went from a marker of nonconformity to the default route into marriage inside one working lifetime. Whatever cohabitation selected for in 1975 it cannot plausibly select for now, and the comparison group has changed just as much.
This was the reasoning that made American sociologists expect the effect to disappear, and several studies of recent marriages reported exactly that. The 2019 paper's contribution was to argue it had not, and the 2021 exchange is about whether that argument survives. The cross-national work adds the piece neither side of that argument can supply from one country: if the effect really is driven by who is unusual, then its size should track the base rate — and in sixteen European countries, and again in twenty-one, it does.
It is the same logic that undoes most relationship folklore built on a correlation, including the numbers examined in what the data on dating apps actually shows and the figures in modern dating statistics that hold up. The clearest parallel case is the division of housework, where the association with staying together runs one way in Norway, another in Sweden, and in opposite directions in West Germany and the United States: what splitting the housework actually predicts.
Sliding versus deciding, and what it actually claims
There is a mechanism story attached to this topic that is better than the correlations, and it is worth separating from them.
Scott Stanley, Galena Rhoades and Howard Markman proposed in 2006 that what matters is not the living together but how a couple arrives at it. Their distinction is between commitment as dedication — wanting the relationship, planning as though it will continue — and commitment as constraint, the accumulating costs of leaving: the shared lease, the joint furniture, the dog, the fact that your friends are now a single social unit. Moving in builds constraint immediately and dedication not at all. If a couple slides into cohabitation rather than deciding on it, constraints pile up before either person has clarity about whether this is the relationship, and a couple can become hard to leave without having become more committed. They called it inertia. It is the same frame that explains the markers of a situationship, where the undefined arrangement is itself the finding.
The measured evidence for the mechanism is thinner than the frame and points in one specific direction. Rhoades, Stanley and Markman surveyed 120 cohabiting heterosexual couples about their reasons for moving in. The most strongly endorsed reasons were spending more time together and convenience. But the degree to which someone reported cohabiting in order to test the relationship was associated with more negative couple communication, more physical aggression, lower relationship adjustment, confidence and dedication, higher attachment insecurity, and more symptoms of depression and anxiety. Men endorsed testing more often than women; women endorsed convenience more often than men.
Read that carefully, because it is easy to over-read. These are 120 couples measured at one point in time, so nothing in it shows that testing causes the trouble — a relationship already in difficulty is exactly the kind someone would describe themselves as testing. What it does establish is that moving in to find out whether this works is a flag, not a method, and the thing it flags is the state of the relationship right now. The research on what builds attraction over time says the same about using proximity as a substitute for a decision.
The belief survey, and who paid for it
One recent dataset speaks to this question directly and belongs here with its provenance attached.
In April 2023 the Institute for Family Studies published a report by Stanley and Rhoades called What's the Plan? Cohabitation, Engagement, and Divorce. YouGov collected the data for them between 28 July and 29 August 2022; the analytic sample was 1,621 people, all aged 50 or under, all married for the first time between 2010 and 2019, all still married to or divorced from that first spouse.
Their primary finding is the engagement split again, in recent marriages: 34% of marriages had ended among those who started living together before being engaged, against 23% among those who moved in only after being engaged or married — a relative difference of 48%. The report also cites the belief side of it, that 50 to 65% of Americans think living together before marriage improves their odds, with younger Americans especially likely to believe it.
That gap between what people believe and what the studies report is the actual story of this topic, and it is worth stating that the gap is about improving the odds. The belief is not merely that cohabitation is neutral; it is that it works as a test. No study reviewed on this page supports that, and several are built on the finding that it does not.
The provenance matters too, and is not a reason to dismiss the report. The Institute for Family Studies is an advocacy organisation with a stated position on marriage, the survey was commissioned by it, and the report is not peer reviewed. The authors are also two of the people whose academic work is under argument elsewhere in this literature. None of that makes the 34 and the 23 wrong — the design is reasonable, the sample and the dates are published, and the result is consistent with Stanley's 2010 paper and with Manning and Cohen's national data. It means the number should be cited as what it is: an advocacy organisation's own commissioned survey, pointing the same way as the peer-reviewed work on the same split.
What none of this can establish, and why not
There is no experiment here, there never will be, and that limit is doing more work than any of the numbers.
Nobody can assign couples at random to move in together or not. Every figure on this page comes from asking people what they did and watching what happened next, which means every one of them is open to the possibility that the people who cohabit differ from the people who don't in ways that cause the divorces all by themselves. This is the selection problem, and it is not a technicality that careful controls dispose of. Kuperberg's models are the clean demonstration: with one plausible set of controls the association is 1.37 and highly significant, with another equally plausible set it is 1.09 and absent, and nothing in the data tells you which set is right.
Three further limits apply across the whole literature. The American work rests overwhelmingly on one survey, the NSFG, which for most of the relevant period interviewed women only and capped them at 44, so long marriages can only be observed for people who married young — a bias the comment and the reply both acknowledge and handle differently. The outcome measured is almost always divorce or separation, which is a legal and practical event rather than a measure of how a marriage is going; a stable unhappy marriage counts as a success in every model here, which is why the research on why people stay in unhappy relationships is a necessary companion to this one. And the effects are small in absolute terms: an annual breakup rate of 4.70% against 3.47%.
What survives all of that is narrow but real. Living together before marriage has been associated with higher odds of divorce in American data for forty years; the association shrinks or vanishes when you account for the age people moved in, restrict it to people who lived only with their eventual spouse, or look at marriages where a commitment preceded the move; it does not appear at all for some groups within the same dataset; and outside the United States its direction tracks how common cohabitation is. No number in that sentence can tell one couple what will happen to them.
The risk this literature actually describes
If you came to this question because you are weighing a decision, the finding most worth your attention is not the divorce odds. It is the constraint.
The inertia argument is, stripped of its statistics, a claim that moving in makes a relationship harder to leave before it makes it better. That is presented in the literature as a risk to marital quality. It is also, plainly, a risk to someone's safety, and the two are not usually discussed together. A shared lease, a pet, pooled money, a car in one name, a partner who manages the bills — each of these is a constraint in the technical sense and an obstacle in the practical one, and they are the obstacles that make leaving a relationship that has become frightening slow and expensive instead of simply difficult.
Some of what gets described as closeness after moving in is not closeness. Having to account for where you were and who you saw, being discouraged from seeing friends or family, money controlled or monitored by one partner, your name kept off or forced onto accounts, threats about what will happen if you leave, pressure after you have said no: these are patterns of control, and they are on the recognised warning lists for abuse. They are not a severe version of an ordinary conflict about housework. Rhoades, Stanley and Markman's 120 couples are a reminder that the question should we move in to find out is sometimes being asked inside a relationship where physical aggression is already present.
If any of that is familiar, the relevant information is not in this literature. Find A Helpline lists free and confidential services by country, and in the United States the National Domestic Violence Hotline is 1-800-799-7233, or text START to 88788, around the clock. Both are linked in the sources at the bottom of this page.
And if the question is less acute than that but still unresolved, what the research on thinking about breaking up reports covers what is actually known about how that thought behaves over time, which is less conclusive and more reassuring than the genre suggests.
The version of the question that has an answer
Does living together before marriage lead to divorce? has four right answers depending on when, where, who and with whom, which is another way of saying it is not a question evidence can close.
What the evidence can do is replace it with questions that have been measured. Whether a commitment existed before the move is measured, repeatedly, by different teams on different data, and it keeps coming out the same way — and it is a question about a conversation, not about a lease. How old you were when you started sharing a household is measured, and it carries a large share of what gets blamed on cohabiting. Whether this is the only person you have lived with is measured, and in the one pooled analysis of the literature it is where the association on stability went. How common cohabitation is where you live is measured, and it changes the direction of the finding.
None of those is advice about whether to move in. This page is demography, and demography describes populations; it has nothing to say about one couple, and the people most confident about what you should do with your lease are generally selling something — a service, a worldview, or a number without its denominator.
The one practical thing the literature does support is unglamorous and it is not about housing. Across the engagement studies, the group that did well was not the group that avoided cohabiting; it was the group that had settled what they were to each other before they changed their address. The never-cohabited and the cohabited-after-engagement groups were statistically indistinguishable in Stanley's data. What separated both from the third group was a decision, made explicitly, in advance.
That is a conversation, and it is a conversation most couples find easier to postpone than to have. The questions that actually reveal compatibility covers what is known about having it, and the conflict style test scores the pattern that tends to decide how it goes.
The same shape turns up in the other demographic variable couples get warned about. In the sixteen-year panel that produced the famous figures on in-laws and divorce, neither spouse's own account of how close they were to the other's family predicted whether the marriage lasted; what predicted it was the two of them not describing the same family the same way. The prior agreement is the variable in both cases, and the living arrangement is not.
And the clock Kuperberg argued about becomes a variable in its own right the moment you stop treating it as a control: what the research says about the best age to get married follows the same measurement question into a literature where three peer-reviewed analyses of one survey disagree with the blog post everyone quotes. The other variable couples are warned about is money, and the research on money arguments and divorce lands in the same place this page does: what predicted the outcome was a standing disagreement nobody had settled, not the arrangement itself.
Pulling it together
The short version is that the cohabitation effect is real in American data, smaller than it is usually made to sound, and conditional on four things that are rarely named alongside it: the age at which a couple started sharing a household, how long the marriage has run, whether anything had been decided before the move, and the society the couple lives in.
The second thing worth keeping is that this literature is a live argument, not a settled result. A comment and a reply ran side by side in the same journal in 2021, and they disagree about whether the effect still exists in recent marriages because they disagree about which measure of time to use and how much of the survey to keep. Nothing a reader does can settle that, and a page that reports one side as the answer is not reporting the evidence.
The third is the one with a decision attached. Of everything measured here, the variable that keeps separating groups is whether a commitment existed before the address changed — and in the clearest dataset, couples who moved in after getting engaged were statistically indistinguishable from couples who did not live together at all. That points at a conversation rather than at a lease.
And the last is the limit. There is no experiment here and there cannot be one. Every figure on this page describes people who chose, and choosing is exactly what cannot be controlled for.
Frequently asked questions
Does living together before marriage lead to divorce?
In American data it has been associated with higher odds of divorce for about forty years, but "lead to" overstates what the research can show. The most cited study, Rosenfeld and Roesler's 2019 analysis of 216,455 couple-years from 24,888 women in first marriages, reports odds of dissolution 1.31 times higher for couples who cohabited first, once controls are applied. That association shrinks or disappears under several reasonable changes: Kuperberg's 2014 analysis of 7,037 women found a hazard ratio of 1.09, not statistically significant, when the models controlled for the age at which couples moved in together rather than the age they married. Jose, O'Leary and Moyer's 2010 meta-analysis reported that the association with marital stability did not remain when restricted to people who had cohabited only with their eventual spouse. And no experiment is possible here, so selection — the possibility that the people who cohabit differ in ways that cause the divorces independently — cannot be ruled out. The same limit runs through most relationship statistics, including the ones examined in modern dating numbers that hold up.
Is the cohabitation effect still real today?
That is the exact point of an unresolved academic argument, and it is worth knowing it is unresolved. In 2021 the Journal of Marriage and Family published a comment by Manning, Smock and Kuperberg arguing that Rosenfeld and Roesler's models used too many overlapping measures of time, and that with marriage cohort as the time axis the association is weaker in recent cohorts and absent for cohabitation with the eventual spouse. Rosenfeld and Roesler's reply, in the same issue, argues that the comment discarded more than 70% of the valid couple-years and then used the resulting loss of power to declare a key interaction non-significant. Both teams are working from the same survey. The honest answer is that whether the effect persists in recent American marriages depends on modelling choices that competent researchers make differently, which is not the same as the effect being settled in either direction. Our editorial policy explains why this page reports the disagreement instead of picking the side with the better headline.
Does it matter whether you were engaged before moving in together?
This is the split that keeps replicating, and it is the most consistent finding in the area. Stanley, Rhoades, Amato, Markman and Johnson followed first marriages formed in the 1990s across four states: by 2001, 28% of people who had cohabited before being engaged had divorced, against 21% who cohabited only after engagement and 20% who did not live together until marriage. In the model with full covariates the annual probability of divorce was .027, .019 and .015 for the three groups, and the difference between the after-engagement group and the never-cohabited group was not statistically significant. Manning and Cohen found the same pattern in nationally representative data on marriages formed since 1996: among women, those who were engaged or had definite plans when they moved in had significantly lower hazards of dissolution than those who were not. For men, the commitment split did not predict anything, which is a real asymmetry rather than a rounding error. The samples are modest — Stanley's event history rests on 124 divorces — so treat the direction as better supported than the exact percentages. Because the finding is about having settled something rather than about housing, the questions that actually reveal compatibility is the more useful companion page.
Does the age you moved in together matter more than whether you cohabited?
Arielle Kuperberg's 2014 paper is the one that asked, and its answer is that a large share of what gets attributed to cohabiting is attributable to cohabiting young. Using NSFG data on 7,037 women, premarital cohabitation carried a hazard ratio of 1.40 with no age control. Controlling for age at marriage raised it to 1.63; controlling for age at coresidence lowered it to 1.29. With demographic controls added, the contrast is starker: 1.37 and significant at p < .001 when standardising by age at marriage, 1.09 and not significant when standardising by the age the couple moved in. Her argument is that for a couple who lived together first, the marital roles began at coresidence rather than at the wedding, so age at marriage is the wrong clock. The model fit statistics in her paper favour her version. This does not make the cohabitation effect an artefact, but it does mean no one can state the size of the effect without stating which age they controlled for. Ages carry the same hidden weight in what age gaps actually predict, where the gap turns out to matter far less than the ages themselves.
Is the cohabitation effect found outside the United States?
Not uniformly, and the pattern of where it is found is itself the interesting result. Liefbroer and Dourleijn tested this in Demography in 2006 with data from 16 European countries and concluded that former cohabitors run a higher risk of dissolution than people who married without cohabiting only in societies where cohabitation is either a small minority or a large majority phenomenon — a U-shape with its low point around 50% prevalence. A 2025 replication by van Damme and Ruiz-Vallejo, using the Harmonized Histories for 21 countries and 14 birth cohorts, found the U-shape again with its low point nearer 40%, and reported that married couples who had cohabited first had a lower separation risk than couples who married directly. That replication is a preprint and has not been peer reviewed, which is a genuine limit. The mechanism both papers propose is selection: when few people cohabit, the ones who do are unusual in ways that predict separation, and when almost everyone cohabits, the ones who don't are the unusual ones. It is the same selection logic that undoes the folklore examined in the 3-6-9 rule in dating.
Does living together first work as a test of the relationship?
Nothing in this literature supports that, and the one study that measured the motive directly found it associated with worse outcomes rather than better ones. Rhoades, Stanley and Markman surveyed 120 cohabiting heterosexual couples about why they had moved in. The most endorsed reasons were spending more time together and convenience. The degree to which someone reported cohabiting in order to test the relationship was associated with more negative couple communication, more physical aggression, lower relationship adjustment, confidence and dedication, higher attachment insecurity, and more symptoms of depression and anxiety. Men endorsed testing more often than women. Because that study measured everything at one point in time, it cannot show that testing causes the trouble — a relationship already in difficulty is exactly the one someone would describe as being tested. What it does suggest is that the motive is a signal about the present state of the relationship, not a method for finding out about its future. Meanwhile, 50 to 65% of Americans believe cohabiting improves the odds. If what you want is a read on how the relationship is going rather than a test of it, the relationship satisfaction test scores the dimensions this literature uses.
How many couples live together before marriage now?
In the United States it is the normal route into marriage. Rosenfeld and Roesler's NSFG figures track the share of women who had lived with the man who became their first husband: 11% of those marrying in 1970, 34% in 1980, 46% in 1990, 60% in 2000, peaking at 70% in 2011. That change is large enough to matter for interpreting the research rather than being mere background. A behaviour performed by 11% of people is a marker of something unusual about those people; a behaviour performed by 70% of them is not. It is the reason many sociologists expected the association with divorce to fade as cohabitation became normative, and the reason the argument about whether it has faded is as sharp as it is. It is also worth noting that these percentages, like most of the American evidence here, describe women's reports of first marriages, because the survey they come from did not interview men until 2002.
Why do different articles give completely different answers to this question?
Mostly because they are reporting different slices of a genuinely divided literature, and sometimes because they are reporting one study without its conditions. Four things change the answer, and a page that fixes any one of them can honestly report a different result: how long the marriage has run, since Rosenfeld and Roesler's own data show cohabiters breaking up marginally less in the first year of marriage and more thereafter; which age the models control for, where Kuperberg's switch moves the hazard ratio from 1.37 to a non-significant 1.09; whether cohabiters are split by prior commitment, which removes the association in several datasets; and which country the data come from, where the direction tracks how common cohabitation is. On top of that sits the 2021 comment-and-reply exchange about modelling the same survey. A page that gives you one number without one of those four conditions attached is not wrong so much as incomplete. The rest of this section is under Mindset.
Does the cohabitation effect apply to everyone?
No, and the clearest evidence for that comes from the authors of the most cited study. In a 2024 follow-up in the Journal of Marriage and Family, covering NSFG marital histories from 1950 to 2017, Rosenfeld and Roesler report that premarital cohabitation was associated with marital dissolution at a hazard ratio of 1.30 for non-Hispanic White wives, that there was no such association for non-Hispanic Black wives, and that the difference between the groups was statistically significant. They report the same moderation for low education and for teen marriage, and call it the Racial Minority Moderation hypothesis. The follow-up shares every limitation of the original — same survey, women only, observational — but it rules out the most common way the headline figure gets stated. Whatever premarital cohabitation predicts, it does not predict it for everyone, and none of the pages currently ranking for this question mentions the paper.
Sources
What this article draws on. For the broader bibliography see our sources page.
- Rosenfeld, M. J. & Roesler, K. (2019) — Cohabitation experience and cohabitation's association with marital dissolution. Journal of Marriage and Family, 81(1), 42-58 — harmonised NSFG, 216,455 couple-years from 24,888 women in first marriages aged 15-44; adjusted odds ratio 1.31; first-year breakup rates 3.9% vs 4.1%
- Rosenfeld, M. J. & Roesler, K. (2019) — author's copy — Accepted manuscript with appendices, posted by the author at Stanford. The source for the first-year breakup rates and the 1970-2011 prevalence series quoted on this page
- Manning, W. D., Smock, P. J. & Kuperberg, A. (2021) — Cohabitation and marital dissolution: A comment on Rosenfeld and Roesler (2019). Journal of Marriage and Family, 83(1), 260-267 — argues the 2019 models include too many indicators of time
- Rosenfeld, M. J. & Roesler, K. (2021) — Premarital cohabitation and marital dissolution: A reply to Manning, Smock, and Kuperberg. Journal of Marriage and Family, 83(1), 268-279 — the reply that reports more than 70% of valid couple-years discarded
- Rosenfeld, M. J. & Roesler, K. (2024) — Stability and change in predictors of marital dissolution in the US 1950-2017. Journal of Marriage and Family, 86(1), 154-175, first published online 14 August 2023 — hazard ratio 1.30 for non-Hispanic White wives, no association for non-Hispanic Black wives, difference significant
- Kuperberg, A. (2014) — Age at coresidence, premarital cohabitation, and marriage dissolution: 1985-2009. Journal of Marriage and Family, 76(2), 352-369 — N = 7,037; hazard ratio 1.37 controlling for age at marriage against 1.09 and non-significant controlling for age at coresidence
- Kuperberg, A. (2014) — open postprint — Author's accepted version in the Maryland Shared Open Access Repository, the source for the hazard ratios and model-fit statistics quoted here
- Jose, A., O'Leary, K. D. & Moyer, A. (2010) — Does premarital cohabitation predict subsequent marital stability and marital quality? A meta-analysis. Journal of Marriage and Family, 72(1), 105-116 — k = 16 for stability, k = 12 for quality; the association with stability did not remain for cohabitation with the eventual spouse. Paywalled with no open copy located; described here from the published abstract and no effect sizes are quoted
- Liefbroer, A. C. & Dourleijn, E. (2006) — Unmarried cohabitation and union stability: Testing the role of diffusion using data from 16 European countries. Demography, 43(2), 203-221 — higher dissolution risk for former cohabitors only where cohabitation is a small minority or a large majority. Paywalled; described here from the published abstract
- van Damme, M. & Ruiz-Vallejo, F. (2025) — Unmarried cohabitation and union stability: The cohabitation-diffusion-hypothesis revisited — 21 countries, 14 birth cohorts from the Harmonized Histories; U-shape minimum near 40%; prior cohabitors with lower separation risk than the directly married. PREPRINT, not peer reviewed
- Stanley, S. M., Rhoades, G. K., Amato, P. R., Markman, H. J. & Johnson, C. A. (2010) — The timing of cohabitation and engagement: Impact on first and second marriages. Journal of Marriage and Family, 72(4), 906-918 — four-state sample of 1990s marriages; 28%, 21% and 20% divorced by 2001; annual probabilities .027, .019 and .015; 124 divorces in the window
- Manning, W. D. & Cohen, J. A. (2012) — Premarital cohabitation and marital dissolution: An examination of recent marriages. Journal of Marriage and Family, 74(2), 377-387 — 2,003 women and 1,483 men married since 1996; the dichotomous cohabitation measure not significantly associated with instability for either sex; engagement before cohabiting predictive for women only
- Rhoades, G. K., Stanley, S. M. & Markman, H. J. (2009) — Couples' reasons for cohabitation: Associations with individual well-being and relationship quality. Journal of Family Issues, 30(2), 233-258 — 120 cohabiting couples; cohabiting to test the relationship associated with more negative communication, more physical aggression and lower dedication; cross-sectional
- Stanley, S. M., Rhoades, G. K. & Markman, H. J. (2006) — Sliding versus deciding: Inertia and the premarital cohabitation effect. Family Relations, 55(4), 499-509 — the constraint and dedication distinction this page refers to as inertia
- Stanley, S. M. & Rhoades, G. K. (2023) — What's the Plan? Cohabitation, Engagement, and Divorce. Institute for Family Studies — YouGov survey fielded 28 July to 29 August 2022, analytic N = 1,621, first marriages 2010-2019; 34% of marriages ended among those cohabiting before engagement against 23% after. An advocacy organisation's own commissioned report, not peer reviewed
- Find A Helpline — Free, confidential support lines by country, searchable and operated by throughline
- The Hotline (National Domestic Violence Hotline, US) — Warning signs of abuse, including monitoring, financial control and isolation from friends and family. 1-800-799-7233, or text START to 88788
Keep reading
- MindsetHow to Reconnect After a Fight: What the Repair Research ShowsResearch says little about how to fight well and a great deal about the hours afterwards. Most advice for those hours is untested or points the wrong way.
- MindsetThe Psychology of Jealousy: The Feeling, the Behavior, and the Line Between ThemThe feeling is ordinary and almost everyone has it. What a minority do with it is a different subject, and the research is clear about where the line falls.
- MindsetLove or Anxious Attachment? How to Tell the DifferenceFrom the inside they feel the same. They come apart in your body, in how long reassurance lasts, and in whether the intensity needs uncertainty to survive.
- SignsSigns She Likes You: What the Evidence Actually SupportsThe behaviours are real. What no study has ever found is one that tells you which kind of liking you are looking at.
- DatingFirst Date Mistakes: What the Research Says Actually Costs YouThe standard advice is to talk less. In the one experiment that assigned strangers how much of a conversation to fill, those who spoke more were liked more.
- TextingHow to Get Someone to Open Up: What the Research Actually ShowsThere is no phrase. There is a two-person process, one measured thing that matters more than the rest, and a fear that turns out to be wrong.