I think a big chunk of the discrepancy here comes from comparing against US expectations rather than a weighted average over countries.
Case in point, the ‘ever married 25-34’ stat (which is the one with biggest sample size) is
EA: 18%
US: 45%
UK: 22%
I can’t find directly comparable stats for Australia, Canada or Germany—which make up the other major countries of the 2018 survey. What is available is age-at-first-marriage:
US: 29, UK: 33, DE: 33, AU: 32, CA: 31
Doing linear interpolation on the UK/US values, estimated ever-married 25-34 stats are
US: 45%, UK: 22%, DE: 22%, AU: 28%, CA: 34%
which, weighted by the survey fractions of .36/.16/.07/.06/.04 gives an expected ever-married 25-34 rate of 35%.
Some other things worth noting:
EA is overwhelmingly male, and men marry a year or three later.
The EA pop distribution inside the 25-34 bucket is going to lean towards the 25 end, while the national pop distributions are going to be even across it.
All in all I’d expect a properly-calibrated expected rate to be 25-30%-ish.
I’m also curious if the higher-education-higher-marriage-rate thing holds in the UK/Europe, but damned if I can find solid stats. Anecdotally it doesn’t, but anecdotes are awful for this kind of thing. Sample bias’ll kill you dead.
I had trouble finding the ever-married stat, so I went to the 2017 marriages dataset. Table 12 is ‘Proportions of men and women who had ever married by certain ages’, and for the 1984 men’s cohort - the last one with complete data—the rates are 6% for 25yos through 39% for 34yos. The average over that age range is 22%.
So upfront, I didn’t notice it was the mens’ cohort. My bad, and I’ll fix my post once I’ve figured out the other problem: these figures are decidedly inconsistent with the figures you’ve found. It’s not obvious to me what the difference is.
First, momentary thought was I’d been totally daft and averaged married-this-year data. Second thought was that I’d averaged over birth-cohort rather than reporting year. But that gives 20%, an even lower figure. Third thought was that ‘married by 34’ is different from ‘married at 34’, but shifting the age range up a notch only gets the figure up to 26%.
The data I gave is ultimately survey data, the table you post is based on marriage certificates issued. This has advantages but has one large disadvantage, namely ignoring marriages that take place overseas, while possibly counting marriages between two overseas residents that take place locally. It’s mentioned on the ‘Table 12 interpretation’ tab:
These statistics are based on marriages registered in England and Wales. Because no adjustment has been made for marriages taking place abroad, the true proportion of men and women ever married could be higher.
I followed that link to get any context on how big a deal this might be.
In 2017, an estimated 104,000 UK residents went abroad to get married and an estimated 8,000 overseas residents married in the UK.
To put that number in context, there are roughly 240k marriages per year in the UK, presumably involving around 480k people, so that’s a large chunk of the total.
I think survey data is just better for our current use case since we don’t much care about sample noise; apart from the ‘destination wedding’ issue, I definitely want to count two immigrants who arrived in the UK already married, and I think they’ll also appear in the survey but not the certificate-counting.
Thanks very much for figuring that out! I’ve retracted my original comment; my estimate of the background rate is now 30-40%ish—I think the various perturbations’ll near-enough cancel - and with that, the diff against the American rate is no longer the majority of the anomaly.
I think a big chunk of the discrepancy here comes from comparing against US expectations rather than a weighted average over countries.
Case in point, the ‘ever married 25-34’ stat (which is the one with biggest sample size) is
EA: 18%
US: 45%
UK: 22%
I can’t find directly comparable stats for Australia, Canada or Germany—which make up the other major countries of the 2018 survey. What is available is age-at-first-marriage:
US: 29, UK: 33, DE: 33, AU: 32, CA: 31
Doing linear interpolation on the UK/US values, estimated ever-married 25-34 stats are
US: 45%, UK: 22%, DE: 22%, AU: 28%, CA: 34%
which, weighted by the survey fractions of .36/.16/.07/.06/.04 gives an expected ever-married 25-34 rate of 35%.
Some other things worth noting:
EA is overwhelmingly male, and men marry a year or three later.
Those fractions add up to .7; another .2 come from a scattering of countries that are largely European. Much of Europe has ages-of-first-marriage of 33 or more.
The EA pop distribution inside the 25-34 bucket is going to lean towards the 25 end, while the national pop distributions are going to be even across it.
All in all I’d expect a properly-calibrated expected rate to be 25-30%-ish.
I’m also curious if the higher-education-higher-marriage-rate thing holds in the UK/Europe, but damned if I can find solid stats. Anecdotally it doesn’t, but anecdotes are awful for this kind of thing. Sample bias’ll kill you dead.
Source for the UK:22% figure? The ONS figures for 2019 (for married, not ever married) are:
Men 25-29: 15.7%
Women 25-29: 25.4%
Men 30-34: 42.4%
Women 30-34: 52.3%
These groups are all roughly the same size, so a combined 25-34 group would be around 34%. ‘Ever married’ should be 1-4 percentage points higher.
https://www.ons.gov.uk/peoplepopulationandcommunity/populationandmigration/populationestimates/bulletins/populationestimatesbymaritalstatusandlivingarrangements/latest
https://www.ons.gov.uk/visualisations/dvc4
I had trouble finding the ever-married stat, so I went to the 2017 marriages dataset. Table 12 is ‘Proportions of men and women who had ever married by certain ages’, and for the 1984 men’s cohort - the last one with complete data—the rates are 6% for 25yos through 39% for 34yos. The average over that age range is 22%.
So upfront, I didn’t notice it was the mens’ cohort. My bad, and I’ll fix my post once I’ve figured out the other problem: these figures are decidedly inconsistent with the figures you’ve found. It’s not obvious to me what the difference is.
First, momentary thought was I’d been totally daft and averaged married-this-year data. Second thought was that I’d averaged over birth-cohort rather than reporting year. But that gives 20%, an even lower figure. Third thought was that ‘married by 34’ is different from ‘married at 34’, but shifting the age range up a notch only gets the figure up to 26%.
Any ideas?
The data I gave is ultimately survey data, the table you post is based on marriage certificates issued. This has advantages but has one large disadvantage, namely ignoring marriages that take place overseas, while possibly counting marriages between two overseas residents that take place locally. It’s mentioned on the ‘Table 12 interpretation’ tab:
I followed that link to get any context on how big a deal this might be.
To put that number in context, there are roughly 240k marriages per year in the UK, presumably involving around 480k people, so that’s a large chunk of the total.
I think survey data is just better for our current use case since we don’t much care about sample noise; apart from the ‘destination wedding’ issue, I definitely want to count two immigrants who arrived in the UK already married, and I think they’ll also appear in the survey but not the certificate-counting.
Thanks very much for figuring that out! I’ve retracted my original comment; my estimate of the background rate is now 30-40%ish—I think the various perturbations’ll near-enough cancel - and with that, the diff against the American rate is no longer the majority of the anomaly.