Philip Cohen shares this amusing example of selection bias:
The Washington Post magazine has a feature out today called “The secret to a long-lasting marriage.” . . . the third paragraph is funny:
They have beaten the odds of death and divorce: Of all current U.S. marriages, only 7 percent have reached the 50-year mark, according to the National Center for Family and Marriage Research at Bowling Green State University.
As Cohen points out, this statistic is misleading:
It is certainly true that making it to the 50-year mark of marriage means you have beaten the odds of death and divorce. But that 7% figure has nothing to do with it, because it includes people who got married yesterday!
Here is the breakdown of when people got married, among people married right now (in the 2014 American Community Survey, which has to be the source for that statistic):
So the statistic is correct: only 7% of currently married people have been married for 50 years or more. Good for them! To bad for all those other people they were born so recently.
It’s all in the denominator. Sure, 50-year marrieds are rare, but compared to what?
With the ACS we can answer a more relevant question, which is this: among living people whose most recent marriage was 50 years ago or more, what is their current marital status? This is a little more encouraging: half are still married.
So let’s restate the original congratulatory message like this:
They have beaten the odds of death and divorce: Of all people who most recently tied the knot 50 or more years ago, and who haven’t yet died, only 50% percent have made it this far without divorcing or becoming widowed, according to the American Community Survey.
Yup.
P.S. I added “most recently” to “tied the knot 50 years ago” to clarify in response to comments.
And among the people who haven’t died or become widowed, ~84% are still married.
That is amazing. Just last week, I attended the 50th anniversary celebration of a college friend. At the time I was remarking that they were the only couple in my circle of friends from that era who are still married. So even the misleading 7% statistic struck me as an _over_estimate; the better 50% statistic truly shocks me. Evidently there is some severe adverse selection associated with having been my friend in that era!
Not necessarily. I find myself a bit confused by the wording here, but if it’s about their “most recent” marriage being 50+ years ago, it wouldn’t count people who were originally married 50 years ago, but then got divorced and remarried. At least if I’m understanding it correctly.
My parents celebrated their 50th wedding anniversary in May (in Texas), and there was a wedding reception going on next door at the venue, and one of the groomsmen was a Ph.D. student in my class at Columbia in 2013. What is the conditional probability of that given that I haven’t died?
It seems like the “most recent marriage” filter is going to miss at least some of the people married 50+ years ago who either divorced / were widowed and then remarried? Shouldn’t we want the stats for everyone who had a marriage within the right time span and then see the status of those marriages?
As some folks have been pointing out on Bluesky, the new data don’t seem to answer the question/confirm the final statement either. By looking at people whose most recent marriage was at least 50 years ago, it excludes living people who were married at least 50 years ago, had that marriage end, and were remarried.
For instance Joe Biden married 59 years ago, was widowed, and remarried 48 years ago. So he would be excluded from the sample, even though he is a person who tied the knot 50 or more years ago and hasn’t yet died (to the best of my knowledge, right now).
The statement we can confirm is “Of all people who most recently tied the knot 50 or more years ago, and who haven’t yet died, only 50% percent have made it this far without divorcing or becoming widowed, according to the American Community Survey.”
Yes, I was assuming that this was their most recent marriage. It’s funny how confusing these sorts of conditional statements can be!
Yes, it is unfortunate that they only ask the date of the most recent marriage. (They also ask how many times you’ve been married, but no dates on the earlier ones.)
There is another problem in the ACS data, which is recall, and specifically the tendency of people to report having been married in years ending with 0. This is equivalent to age heaping, which is less common in good surveys these days, but in the case of marriage in the ACS it’s pretty strong. I defined the Decadally-Biased Marriage Recall index, which is 10-times the number of people married in years ending in 0, divided by the number of people married in all years (starting with a 6-year and ending with a 5-year). Times 100 (to make it comparable to the Whipple index for age heaping). In the ACS, last I checked, the DBMR is 110.8 https://familyinequality.wordpress.com/2018/11/16/decadally-biased-marriage-recall-in-the-american-community-survey/
I looked it up and it doesn’t seem like there were actually more marriages in years ending in zero (see here: https://www.cdc.gov/nchs/data/dvs/marriage-divorce/national-marriage-divorce-rates-00-23.pdf) so I guess it really is reporting error.
In theory it could be something weird like that people who marry in years ending in zero are less likely to divorce, but that seems unlikely.
The bit about the age heaping reminds me of the trick that they’ll use at bars: instead of asking your age (“I’m 22”), they ask what year you were born (“Ummm, 2002?”). But maybe this trick doesn’t work so well now that the years are in the twenties: even a confused teenager knows how to subtract 22 from 2025.
There’s still the “beating the odds of death” part. Counting only living people excludes that portion of the population who are not living. So the truly analogous question is: among all people, dead or alive, who were most recently married 50 or more years ago in the U.S., what is their current marital status? Interestingly, nobody born in the 1700s or 1800s seems to have beaten the odds.
Okay, a better, well-posed question for both death and divorce would be: of all marriages that began in 1975, what percentage are still in effect?
This WAPO story dates from 2016 and thus, predates Trump’s rise to power, so is sort of ancient history. Lots of things/norms have changed since then.
Divorce rates have been falling steadily since at least 2008.
https://bsky.app/profile/philipncohen.com/post/3lq6cneyon22n
+1. After we get an acceptable estimate for the cohort of people born long ago whose marriage we are able to observe as having lasted at least 50 years, we have to now deal with the lack of generalizability of that knowledge to more recent cohorts.
Thanks, Andrew. I should update that! The post still gets 10,000 clicks a year.
I have no idea how many clicks my posts get!
The problem with all of these is, yes, definitions are complicated. Presumably mainly this is legally married people in 2016, so there would not have been any same sex marriages with the possibility of surviving 50 years. And wouldn’t it make sense to do e.g. a time to failure model censoring for death? But it’s not the exact number that really matters when the difference is so huge, edge cases like the Bidens wouldn’t change the overall number. What matters I think is that the Post reporter displayed such innumeracy and that no editor questioned it. It tells a story about marriage that is completely wrong and feeds into some kind of culture collapse narrative..
I attended parties for my great grand parents’, grand parents’ and parents’ 50th wedding anniversaries! (Well one set each of the grands.) Also my in-laws’ 50th.
Elin:
I looked into the issue of the same-sex divorce rate a decade ago, when statistics on the topic were just beginning to appear:
– Same-sex couples less likely to get divorced than straight couples . . . so far
– Same-sex divorce rate not as low as it seemed
At the time, I concluded:
Yes, the exposure is key, not only is it wrong to assume they were all at risk for 5 years, you also should not assume that the risk accounting for that is the same over the 5 years.
Another interesting version of this question is, “what is the proportion of people who could be in a marriage that’s lasted 50 years who are”? We’d need to carefully explicate what we mean by “could be” to get the right denominator, but a first stab might be to include all straight people at least 68 years old.