Number 3 in 2021: Is Cohabitation Still Linked to Greater Odds of Divorce?

Editor's Note: With 20,000 pageviews, the following article on cohabitation and divorce by IFS senior fellow Scott Stanley is our third most popular post of 2021 (this article was first published on January 12, 2021).

You might think the question about the link between premarital cohabitation and divorce would have been settled long ago, but researchers have puzzled about it for decades and the puzzling lives on. Part of why the issue draws so much interest is that the vast majority of people believe that living together before marriage should improve the odds of doing well even though research has not supported that belief. This is an update on the latest in this long-running saga of research on the cohabitation effect. 

In 2018, Michael Rosenfeld and Katharina Roesler published a study that contradicted the growing consensus in sociology that premarital cohabitation was no longer associated with greater odds of divorce, even though it had been associated with poorer marital outcomes for decades. The explanation various scholars had given for the cohabitation effect going away are based on the diffusion perspective, which suggests that cohabitation has become so common it no longer selects for those already at higher risk, and also that it has lost the stigma it once had. But Rosenfeld and Roesler showed that the association between premarital cohabitation and divorce has not declined over the years in any substantial manner. They argued that prior studies showing no negative associations were based on samples that did not include marriages that had lasted long enough to fully capture the increased risk for divorce. 

Rosenfeld and Roesler also showed something new in their 2018 study: cohabitation before marriage was associated with a lower risk of divorce in the first year of marriage but a higher risk thereafter. They interpreted this finding in light of experience theories, noting that living together before marriage could give couples a leg up at the very start of marriage because there is less of an adjustment to being married and specifically to living together. But they found this advantage to be short-lived. Other factors related to experience may take over from there, such as how cohabitation can increase acceptance of divorce. 

Rosenfeld and Roesler’s study caused a stir in the field, and this past December, the Journal of Marriage and Family published two pieces related to their 2018 findings. The first is a comment on the study by Wendy Manning, Pamela Smock, and Arielle Kuperberg and the second is a response by Rosenfeld and Roesler. The articles illuminate the complexities of cohabitation and the challenges of studying the effects in social science. 

In a prior IFS article on Rosenfeld and Roesler’s 2018 publication, Galena Rhoades and I described the study and competing theories for why living together before marriage can be associated with lower odds of success in marriage (i.e., selection, experience, and inertia). I refer you to that article for more background information.

The Critique by Manning, Smock, and Kuperberg

Manning, Smock, and Kuperberg make two primary criticisms of Rosenfeld and Roesler’s study. First, they argue that their statistical models include multiple and confounding measures of time. Second, they emphasize the important decisions one has to make about truncation based on age when using the National Survey of Family Growth (NSFG), upon which all of the studies suggesting the association has disappeared are based. Here is a sample of that complexity:

Another age truncation issue is that relatively long marriages cannot be observed with these data without bias toward those that occurred at young ages. For example, a 15-year marriage can only be observed for women who married at age 29 or younger. (p. 3)¹

This is the basis for their assertion that it is best to limit the analytic sample for this research to marriages of 10 or fewer years duration. In essence, Manning, Smock, and Kuperberg argue that Rosenfeld and Roesler made a number of decisions about the sample and statistical modeling that are inconsistent with the prior literature and therefore not sound. They present further analyses in their response and stand by their claim that the cohabitation effect has disappeared.  

Rosenfeld and Roesler’s Reply

Rosenfeld and Roesler respond that Manning, Smock, and Kuperberg misinterpreted how time-related variables had been handled in their original study, noting that the authors of the critique could have asked for clarification instead of building arguments around false assumptions. More importantly, they further explain their belief that prior works (along with new analyses by Manning, Smock, and Kuperberg) are based on decisions that leave out 70% of the relevant, available sample. This is primarily the result of that decision to limit the analytic sample to marriages of 10 years or less duration. Manning, Smock, and Kuperberg contend that this is standard, best practice when using the NSFG, while Rosenfeld and Roesler argue the decision unnecessarily limits sample and statistical power, causing a data-based bias in favor of finding that there is no longer a divorce risk associated with premarital cohabitation. 

Their reply also makes clear just how methodologically important their prior finding is showing that premarital cohabitation is associated with lower odds of divorce in the first year of marriage but greater odds thereafter. Manning, Smock, and Kuperberg attempted to replicate that finding and did not obtain it (but using options they prefer, not the same set up as Rosenfeld and Roesler). 

Rosenfeld and Roesler point out that their critique actually does display evidence of this finding, but that the effect was not statistically significant because of the smaller sample.² Thus, Manning, Smock, and Kuperberg do not account for that effect in other models they run. In practice, that is not an unusual decision, but Rosenfeld and Roesler believe that this decision, along with the decision to restrict the sample based on duration of marriages, leads to analyses less likely to find the increased risk for divorce. 

Filtering out the couples who have been married longer (as MSK do) enhances the Recent Cohort Fallacy because in the very early stages of marriages, premarital cohabitation reduces the risk of marital breakups. (p. 6)

Rosenfeld and Roesler also assert that Manning, Smock, and Kuperberg do not adequately account for the timing of children. They explain that cohabiters are much more likely than non-cohabiters to already have children at the time of marriage, and this difference has nearly doubled over the decades. Thus, cohabiting couples who married in later cohorts were quite a bit more likely than those marrying earlier to already have a child when they married, and the extra stability from having children that is changing by cohort is another factor that lowers the apparent cohort-based association between cohabitation and divorce.³

Rosenfeld and Roesler stand by their conclusion that the average increased risk for divorce associated with premarital cohabitation is mostly unchanged over the last 40 years. 

Comment and Implications

As I stated at the outset, most people believe cohabitation should improve one’s odds of marital success. Rosenfeld and Roesler’s work suggests this may only be true very early in marriage. Otherwise, not so much. As ever on this subject, questions abound. Are marital outcomes truly worse for those who live together before marriage, and, if so, for whom? For example, it is less clear that things work the same way, on average, for African Americans who cohabit, and economic disadvantage is deeply embedded in how cohabitation relates to risk in marriage.⁴

One of the most intriguing questions remains: why is there any association with risk? As Manning, Smock, and Kuperberg note, the long-accepted conclusion in sociology is that differences in marital outcomes based on premarital cohabitation are due to selection—that the added risk is really about who cohabits and who does not. Selection is surely a large part of the story. Of course, on top of that, they argue the risk is no longer evident. Rosenfeld and Roesler disagree. 

Although there are strong arguments on each side, I believe Rosenfeld and Roesler get the better of the debate. They make a compelling case for their analytic decisions and findings. Further, they clearly describe how the choices affect the findings (theirs, and that of others). 

The argument that the overall cohabitation effect will disappear has not been compelling to me, although I have no trouble accepting the possibility. There are two explanations for how the experience of cohabitation might increase risks for some couples, net of selection: changes in attitudes⁵ and inertia. My colleague Galena Rhoades and I are leading proponents of the latter theory, which contains no obvious reason to anticipate a negative effect going away for a large subgroup of those who cohabit prior to marriage. 

Inertia emphasizes that when two people move in together, all other things being equal, they are making it harder to break up. If so, the state of the relationship—and especially the understanding between partners at the time—should matter. Some couples are, in essence, increasing the constraints to remain together (including, for some, on into having children and marrying) prior to dedication being clear, mutual, and high.⁶ We believe that is part of why waiting until marriage, or at least engagement, is associated with lower risk in seven studies. In fact, one of those studies is among those suggesting that the overall cohabitation effect is gone. A differential effect can easily live within an overall average effect—or average non-effect. 

Also, it is worth noting that all of the studies related to the controversy about whether or not the cohabitation effect still exists focus only on the odds of divorce and not on marital quality. In one of our studies, we show that marital quality is lower among those who started living together before engagement or marriage (as inertia theory predicts), and in marriages occurring during the period of time when others have argued that the overall cohabitation effect no longer exists.⁷

One of the other stories in this controversy is endemic to social science. Researcher degrees of freedom is a concept referring to the fact that the reported findings in social science come at the tail end of a great many consequential decisions by the researchers on matters of data sets, included or excluded variables, and statistical models. Rosenfeld and Roesler make a strong plea for transparency in how researchers make their decisions. They are also circumspect in stating that the extraordinary complexity of changes in marriage and cohabitation in the last five decades make it impossible to account for all that may matter when analyzing and interpreting data on this subject. 

There is no simple answer for questions about premarital cohabitation. There is no experiment one can conduct to prove X leads to Y. As Rosenfeld and Roesler put it, “. . . all models of complex reality are flawed.” Count on that, and count on the interesting saga of research on premarital cohabitation to continue. 

Scott M. Stanley is a research professor at the University of Denver and a senior fellow of the Institute for Family Studies (@DecideOrSlide). 

1. These page numbers are those in the advance, online publications of these paper. Once the articles appear in the printed journals, they will have different page numbers. 

2.This is possible because an estimate of an effect can be noisy, having a lot of variability in a sample around whatever average size of effect is obtained.

3. Although it is true that cohabiting parents are more likely to break up than married parents, including those having children prior to marrying, it is also true that having children makes it more likely a couple will stay together or stay together longer—which makes the matter a big deal in analyzing outcomes related to divorce. Rosenfeld and Roesler argue that the specific way Manning, Smock, and Kuperberg control for children at marriage makes the control variable a proxy for cohabiting before marriage, and since having children before marriage is differentially changing across cohorts, they argue that the net effect favors the overall finding that the cohabitation effect has gone away. Related to this issue of children before marriage, Tach and Halpern-Meekin showed that some portion of the premarital cohabitation effect is driven by premarital cohabiters being more likely to have non-marital births before marriage. One can easily argue that cohabitation and child effects are hopelessly intertwined. Still, either factor can easily be seen to have the same implications for a causal risk of the sort Galena Rhoades and I have focused on, where relationship transitions fit a pattern of constraints to stay together increasing substantially prior to maturing of dedication to be together. Such factors can prematurely create inertia for a relationship to continue when a different path may have seen the relationship end or helped a couple form clearer decisions supporting commitment.  

4. As one example, an important matter running through all these themes is how two people can signal commitment to each other and those around them. Cultural context is important, as I wrote long ago: “I do, by the way, believe that cohabitation can signal higher levels of commitment (compared to not cohabiting) among some who are very poor. I think it likely that the potency of a signal is partially related to what other signals are available. For many complex reasons, marriage is so far off the radar screen in terms of experience for many in poverty that another signal like cohabitation can take on signal value.”

5. This paper by Axinn and Barber in 1997 is one of the most brilliant conceptual pieces in the literature on cohabitation. To me, the arguments are as fresh now as when they were written. 

6. Norval Glenn had made a similar suggestion around the same time we were developing our theory, focusing on the idea that “premature entanglement” foreshortened a solid search for a good match between mates (Glenn, 2002). 

7. It is a fair point to note that this study of ours, in particular, is based on vastly simpler sample and design (using a random phone sample) than studies using the NSFG. On the other hand, analyses of relationship quality based on cohabitation history in existing marriage have a built-in bias against finding lower marital quality for those who cohabited prior to marriage or engagement. Such samples have already selected out those who divorced and are no longer married (thus, not in the sample), likely biasing tests for differences in marital quality toward non-significance. Still, if you think about either the experience theory of cohabitation or the inertia theory of cohabitation, we see no reason to believe the risk should abate for those who move in together prior to having figured out their intended future.