Forty-one-year-old Oksana Chusovitina placed seventh in the vault finals at last month’s Olympic Games in Rio competing against gymnasts younger than her own son. Although I was twelfth in the nation in Masters Ladies figure skating back in 2006, I’m sure I could rank at least seventh this year—provided that the judges made some statistical adjustments, such as counting three of my double jumps as triples (one for each child I have born in the ensuing decade!), and inflating their jump quality ratings to compensate for the fact that it is harder to get good amplitude when carrying about a dozen more pounds. Statistical adjustments are common in sociological studies, as shown in Éva Beaujouan’s recent work, which essentially awards handicaps to second unions through statistical adjustments that allow her to conclude that second unions in France are more stable than first unions.

Here are the comparisons from her article in European Journal of Population without any adjustments: In France between the 1970s and the 2000s, the “share of unions dissolved within 10 years after the start of the relationship rose from 11% to 32% for first unions, and from 23% to 34% for second unions.” This does, indeed, represent profound change as a big stability gap has become relatively minimal. But note that the convergence resulted from first unions breaking up almost three times as often, while second union instability increased by “only” about 50 percent.

Why have first unions become so much less stable? A large part of the answer is that French women today are much less likely to marry their first cohabiting partner than in the past. They are also much less likely to bear a child in their first union. Both marriage and joint childbearing are associated with union stability, but these features have become relatively scarce in first unions.

Today, second unions commonly occur after first childless cohabitations; in the past, they more often occurred after first childbearing marriages. This means that the women entering second unions today haven’t put on as much “weight” as their predecessors: they are younger (typically in their late 20s or early 30s, ages that may be optimal for forming stable unions), their second partners typically do not have the difficulty of forging relationships with stepchildren, and they rarely need to stay in touch with their exes to coordinate co-parenting.

Beaujouan explains all of this quite nicely. She shows that first unions used to be more stable, but that more recently they have lost their advantage. In fact, recent second unions are longer lasting than first ones—but only after controlling for personal characteristics that aren’t measured in survey data (like discipline, impulsiveness, loyalty). The practice of controlling for “unobserved heterogeneity” has become quite common; the idea is that it isn’t really fair to directly compare the longevity of first and second unions because on average, people entering second unions have more traits that work against stability—those with fewer rarely enter a second union because they are more likely to stay in their first. Second unions always look more stable after adjusting for unobserved heterogeneity. But Beaujouan’s discussion helps us to not over-interpret this finding as she cautions that the person-specific control does not vary over time. People themselves change over time. One example of how this could matter is that getting married could make you a more committed person.

Simply staying in the game can also improve prospects for success. A statistical model controlling for unobserved heterogeneity would find me much more similar to skaters who have competed every year for the past decade than I really am because it would adjust for the fact that people who drop out for a decade have less drive than those who persist. It would predict that Michael Phelps would have been just as likely to win the 200M butterfly if he had taken the last two Olympics off than if he had competed continuously through five. Not only does staying in the game matter, but setbacks (like Olympic silver) can also enhance drive. Making it through relational difficulties doesn’t just prove that you have traits that promote stability; it can also produce traits that promote stability.

I applaud Beaujouan for presenting her results with and without controlling for unobserved heterogeneity for these reasons. Absent that control, second unions didn’t look more stable than first ones until she adjusted for whether the union was marital or cohabiting. Beaujouan had expected to find that marriage and cohabitation become more similar in their stability over time, but in fact, she found that the stability advantage to marriage grew. When statistically adjusting for the advantage first unions have by being more often marriages, second unions looked more successful. This adjustment effectively judges cohabitations by a different standard than marriages: five years can be a long cohabitation, but a short marriage, just like fifth place can be a disqualifying result in a semifinal, but an outstanding result for a final. She rates them the same.

Beaujouan further showed that second unions had the largest stability advantage over first unions when controlling for whether the partners had children from previous relationships (adjusting for some of the added weight that second unions more often bear). Again, I do not question the validity of her findings, only their importance. On the one hand, it is meaningful to say that second unions without stepchildren last longer than first unions without stepchildren—especially when today’s second unions are less likely than in the past to include the challenge of blending families. On the other hand, part of what it means to be in a second union is that there was a first one, and controlling out some of the results of first unions is a bit like pretending that I get better amplitude on my skating jumps. Oksana Chusovitina really does get excellent amplitude off the vault. Some 41-year-olds really are fitter than teenagers, and some second unions really are longer lasting than first unions. But on average, second unions are only more stable than first unions in France after much statistical adjustment.

Beaujouan suggests that second unions could possibly be replacing first unions as the more established and stable type of partnership. While first unions have clearly lost stability, second unions only seem more stable based on predictions that don’t seem to correspond well to people’s lived experience of repartnering. I doubt anyone walks away from a second marriage that was shorter than the first consoled by the fact that after adjusting for stepchildren, the second union was more stable. I also don’t believe that French women who enter second unions after dissolving a childless cohabitation should be encouraged by second unions being more similar to first ones when neither relationship lasts more than 10 years about one-third of the time.