"We found no evidence for a cohabitation effect once we controlled for age at coresidence."

That sentence appears in Kuperberg's 2014 paper in the Journal of Marriage and Family. She reprocessed National Survey of Family Growth data spanning three decades of American relationship formation — the same dataset earlier researchers had used to establish that living together before marriage elevated divorce risk. Same sample frame. Different control variable. The effect disappeared.

This matters because it is not a refutation. It is a reattribution. The statistical signal that two decades of research had assigned to cohabitation was tracking something else: the age at which people began sharing a household. Younger coresidence predicted instability. Cohabitation status, once that variable was isolated, did not. The finding that launched pastoral warnings, a shelf of advice books, and over half a billion dollars in federal marriage-promotion funding accounted for less explanatory variance than when a couple moved in together.

What the Numbers Actually Say

The original cohabitation-effect literature rests on three foundational data streams.

Teachman's 2003 analysis in the Journal of Marriage and Family used NSFG Cycle 6 data and reported that women who cohabited with a partner other than their eventual husband showed elevated dissolution risk. The finding was specific. It applied to serial cohabitation — living with multiple partners sequentially — not to cohabiting exclusively with one's future spouse. Popular summaries flattened that distinction within weeks of publication. "Living together before marriage increases divorce risk" became the headline. The qualifier about partner identity dropped away, and what remained was a clean, portable, wrong claim.

Jose, O'Leary, and Moyer published a meta-analysis in 2010 aggregating results from 26 studies conducted between 1970 and 2008. Their pooled effect size for the cohabitation-dissolution association was modest but statistically significant. The abstract reported an odds ratio indicating cohabitors faced higher dissolution risk. Media coverage stopped there. The meta-analysis itself did not.

Jose et al. noted substantial heterogeneity across the 26 studies. Effect sizes varied by cohort, by decade of data collection, and by the control sets each study included. Studies drawing on 1970s and 1980s data showed stronger effects. Those using 1990s data showed weaker ones. The trend line was visible in 2010: the association was attenuating across successive cohorts. Attenuation across time, however, is not what gets cited. Static pooled effect sizes are what gets cited.

The older studies were not wrong on their own terms. Teachman's serial-cohabitation distinction was genuine. It still holds partial explanatory weight in certain subsamples. The Jose et al. meta-analysis followed standard aggregation protocols. Concede all of that. The foundational scholarship was competent.

Now observe what it did not control for. Age at coresidence. Relationship duration before coresidence. The socioeconomic composition of the cohabiting population in the 1970s and 1980s — which was structurally different from the cohabiting population of 2005 or 2015. The foundational analyses treated cohabitation as the independent variable and dissolution as the dependent variable, without adequately isolating the confounds that were moving both. The conclusions drawn from that framework shaped public policy for a decade. The framework itself was incomplete.

What Nobody Mentions

Two papers published within eight years of each other arrive at claims that, read without care, appear contradictory.

Stanley, Rhoades, and Markman developed the "sliding versus deciding" framework across several publications beginning in 2006 in Family Process. Their argument: couples who drift into cohabitation without explicit commitment — who slide — face higher dissolution risk because inertia, shared leases, and accumulated sunk costs generate "constraint commitment" that substitutes for genuine dedication. The mechanism is psychological. Cohabitation itself alters the decision architecture around staying or leaving.

Kuperberg's 2014 reanalysis says the mechanism is demographic, not psychological. Once age at coresidence enters the model, cohabitation status carries no independent effect. People who move in younger make less stable partner selections because they have narrower comparison sets, fewer financial resources, and less relational experience. The cohabitation variable was absorbing age effects the original models left uncontrolled.

Both papers are methodologically defensible. Neither has been retracted or substantially challenged in peer review. They are both operative. The resolution — and this is what almost no popular summary ever explains — is that they are measuring different things. Stanley et al. measured a psychological process: how commitment forms under conditions of residential entanglement. Kuperberg measured a demographic confound: what happens in population-level data when you properly separate age from cohabitation status. The sliding hypothesis can be true for individual couples while the cohabitation effect disappears at the population level once age is held constant.

This distinction determines what advice is actually useful. If the sliding hypothesis drives risk, the intervention is communication: discuss commitment before signing a lease. If the age-at-coresidence hypothesis drives risk, the intervention is timing: do not form a coresidential union before your mid-twenties, regardless of whether you call it cohabitation or marriage. These are different prescriptions. Most advice columns conflate them.

Manning and Cohen's work through the 2010s using data from the National Survey of Families and Households added another layer. Cohort analysis showed the cohabitation-dissolution association declining across successive birth cohorts. For couples entering coresidence after 2000, the gap between those who cohabited before marriage and those who did not had narrowed to a range several researchers characterized as not meaningfully different from zero. Selection effects — the kind of person who cohabited in 1975 was systematically different from the kind of person who cohabited in 2010 — explained much of the original signal. As cohabitation became normative, the people doing it stopped being a distinctive demographic subgroup, and the statistical artifact dissolved.

The Real Cost

The measurable cost here is not in divorce rates. It is in misallocated attention.

The cohabitation effect, as popularized, directed relationship decision-making toward a binary question: did you live together before the wedding? The accumulated research suggests that variable is either a weak proxy for age and socioeconomic status, or a marker of a specific psychological process — undiscussed commitment — that can be addressed without avoiding cohabitation. Either way, the binary framing pulls the wrong lever.

Federal marriage-promotion programs in the United States allocated over $600 million between 2002 and 2014 under the Healthy Marriage Initiative. Messaging included explicit warnings about premarital cohabitation. The Building Strong Families randomized controlled trial — the largest evaluation of these programs — showed minimal or no effect on relationship stability across most sites. Couples assigned to the intervention group did not stay together at higher rates than controls. The programs targeted a variable that, by the time the evaluations were complete, the longitudinal data had already shown was attenuating toward irrelevance.

For individuals, the cost is subtler but real. A couple in their late twenties with stable employment, explicit mutual commitment, and a clear long-term trajectory faces no additional dissolution risk from cohabitation based on post-2010 data. But if that couple has internalized the popularized cohabitation effect, two things may happen. They rush into marriage to "do it right" — introducing timing pressure that carries its own risk. Or they experience sustained anxiety about an arrangement that the research no longer flags as independently dangerous. Neither outcome is adaptive. Both flow from acting on a finding that the data has since qualified nearly to zero.

The predictors that do hold up across the longitudinal literature are well-documented and rarely the ones that make headlines. Age at union formation. Educational attainment. Income stability. Gottman's research on negative interaction patterns — criticism, contempt, defensiveness, stonewalling — remains the most replicated behavioral predictor of dissolution. Attachment security, measured through validated instruments, predicts long-term relationship satisfaction more consistently than relationship structure. Cohabitation status does not rank among these. It did, briefly, in a specific era with a specific population. It does not now.

If You Only Remember One Thing

The cohabitation effect was real in the data that produced it. It is not fabricated. But it was a product of who cohabited in the 1970s and 1980s — a self-selecting, socioeconomically distinct population — and of analyses that did not separate cohabitation from the age at which it began. Once that separation happened, the effect collapsed.

Whether the sliding-versus-deciding mechanism operates independently of age and selection — whether there is a genuine psychological cost to undiscussed cohabitation that no demographic control can capture — remains unsettled. Kuperberg's age variable absorbed the population-level effect. It has not addressed whether the mechanism persists at the couple level for specific subgroups who enter cohabitation without deliberation. That question is open. The longitudinal data to answer it with adequate controls and adequate follow-up do not yet exist in published form.

FAQ

Does living together before marriage still raise divorce risk in current data?

In cohorts forming unions after 2000, the association between premarital cohabitation and subsequent dissolution has narrowed substantially. Kuperberg's 2014 reanalysis found no independent cohabitation effect after controlling for age at coresidence. Manning and Cohen's cohort analyses confirmed the attenuation across successive birth cohorts. For couples who begin coresidence in their mid-to-late twenties with stable economic circumstances and explicit commitment, current data does not support an elevated dissolution risk attributable to cohabitation itself.

What is the "sliding versus deciding" framework and does it still apply?

Stanley, Rhoades, and Markman proposed that couples who drift into shared living without deliberate commitment discussion develop constraint commitment — obligations based on shared leases and logistics — rather than dedication commitment. This psychological mechanism may elevate dissolution risk independent of demographic variables. The framework remains active in clinical research and couples therapy, though its population-level effect is difficult to isolate from the age-at-coresidence confound that Kuperberg identified. It likely applies to specific couples, not to cohabitation as a category.

Why did earlier studies find such a strong cohabitation effect?

The cohabiting population of the 1970s and 1980s was demographically distinct from the married-without-cohabiting population in ways that independently predicted instability: younger on average, lower income, less formal education, more likely to hold non-traditional attitudes toward relationship permanence. The early studies captured a selection effect without fully controlling for it. As cohabitation became normative across socioeconomic strata by the 2000s, the selecting-in population changed, and the statistical association weakened accordingly.

What predicts relationship dissolution more reliably than cohabitation status?

Age at union formation, educational attainment, income stability, and behavioral interaction patterns during conflict are consistently stronger predictors in multivariate models. Gottman's work on four negative communication patterns — criticism, contempt, defensiveness, and stonewalling — remains the most replicated behavioral predictor of marital dissolution. Attachment security measured via validated instruments predicts long-term satisfaction with greater consistency than whether a couple shared a household prior to legal marriage.

Did the U.S. marriage promotion programs reduce divorce or separation rates?

The Building Strong Families randomized controlled trial and broader evaluations of the Healthy Marriage Initiative found minimal or no sustained effect on relationship stability. Federal funding exceeding $600 million between 2002 and 2014 supported programs that included cohabitation warnings and marriage-encouragement curricula. Post-program evaluations across most sites showed no significant differences in dissolution or relationship quality between treatment and control groups at follow-up.

Is serial cohabitation — living with multiple partners — different from cohabiting once?

Teachman's 2003 analysis found elevated dissolution risk specifically among women who had cohabited with someone other than their eventual spouse. Cohabiting exclusively with one's future husband showed no statistically significant effect on marital stability. This distinction between serial and exclusive premarital cohabitation is frequently omitted from popular summaries. Serial cohabitation may signal different relationship formation patterns or partner-selection tendencies that carry independent risk.

Kuperberg's 2014 findings point strongly in that direction. When age at coresidence replaced cohabitation status as the predictor variable in regression models using NSFG data, the cohabitation variable lost significance while age remained robust. Couples who begin coresidence — regardless of whether that coresidence is labeled cohabitation or marriage — before their early twenties face measurably higher dissolution risk than those who begin in their mid-twenties or later.

Can couples reduce any risk associated with cohabitation through specific actions?

If the sliding-versus-deciding mechanism is operative for a given couple, the research suggests that explicit discussion of commitment, relationship trajectory, and shared expectations before or soon after beginning coresidence addresses the primary psychological risk pathway. This does not require avoiding cohabitation. It requires making cohabitation a deliberate, discussed decision rather than an unexamined default. The age-at-coresidence pathway, by contrast, is not addressable through conversation — it is structural and demographic.