Three issues on the decline of marriage: how universal is it, is delay the same thing as decline, and how can you predict it?
1. The worldwide decline in marriage
I basically argued that marriage decline in the U.S. is universal and inevitable. The headline for the post at the Atlantic was “How to Live in a World Where Marriage Is in Decline,” and at Sociological Images it was, “Marriage Is Over: Live with It.”
For context, I hinted at the globalness of this trend, but here I’ll give a little more detail. I consulted three sources: This Eurostat database on demographics from Europe, this United Nations compilation of marriage rates and marital status from 2008, the Demographic and Health Surveys (DHS) conducted by USAID, and this amazing IPUMS compilation of microdata from censuses around the world.
From Eurostat I got the crude marriage rate, which is marriages per year for every thousand people in the population. They have data for every year for most countries, so I took decade averages for each decade since the 1970s. Here it is:
The big countries on the left account for 78% of the population, and they all show a decline in marriage rates for every decade. Most of the little countries show the same pattern. Overall, 89% of the population lives in a country with that pattern.
Because it is affected by changes in the age distribution, the crude marriage rate is not ideal. So with the UN data, like I did with the US data in original post, I use the percentage of women married. Here it is for those ages 30-34 for almost the whole world, 174 countries, using two data points each, the first around 1985 and the second around 2002 (I’ve scaled the dots to represent population size). Countries below the diagonal line have falling proportions of women married:
In this selection of countries 87% of the population is living in a country with falling marriage prevalence, but that includes the biggest – China and India – which still have virtually universal marriage (98+%). Still, almost all the other big ones show declines (I’ve labeled Pakistan, Russia, Mexico, Brazil, the US, Japan, Italy, Germany and France). The only countries with more than 50 million people that show increases are the Philippines, Vietnam, Egypt, and the Dem. Rep. of Congo.
That is an impressive array of countries with declining marriage prevalence. However, the big countries with big drops in marriage are mostly rich: France, Italy, Germany, Japan and the US. Is this a rich-country phenomenon? One thing that sets apart some of those countries — as Gøsta Esping-Andersen pointed out in a comment here — is their high rates of non-marital cohabitation, either replacing marriage or as part of a pattern of delayed marriage. Delayed or reduced marriage is often part of package of demographic changes, including high income and low fertility. (The exception is Japan, where cohabitation rates, although rising, are still low, with just 21% of Japanese women born in the 1970s having ever cohabited, compared with two-thirds of US women.)
One of the sources for that UN compilation is the DHS, which has been collecting fertility-related data in poor countries since the 1980s. To look more closely at some poor countries, I used the DHS data compiler to get the proportion of women ever married in the ages 25-49, for the 17 countries that had a data for the 1980s, 1990s (except missing a few), and 2000s (most from late in the decade):
This shows the decline of marriage is common in poor countries as well, albeit smaller declines from high levels in recent decades (note I’ve scaled it from 70% to 100%).
Finally, after registering as an IPUMS International user, I downloaded data on 40 million women ages 25-49 from 26 countries, using the oldest census from the 1980s, one from the 1990s, and the most recent one from the 2000s for each (so two to four censuses per country).* This source is great because it’s got age and education variables, which I’ll use below. Here is the before-and-after graph using these data, again showing the percentage of 25-49 year-old women married:
Here, only Vietnam and Indonesia show increases in marriage prevalence. The big countries with declines are France, the US, and Brazil; medium-size countries include Argentina, Spain, Colombia, Venezuela, Chile and Turkey. And all the little countries show declines: Morocco, Ecuador, Greece, Malawi, Canada, Austria, Senegal, Thailand, Portugal, Hungary, Ireland, Switzerland, Costa Rica, Panama, Jamaica (I don’t know what’s up in Jamaica).
In almost all of these countries, women with more education are less likely to be married (the U.S. and Jamaica are exceptions to this; in Hungary, Argentina, Canada, and Ireland only those with university completed are less likely to be married). And in all countries older women are more likely to be married. So if world populations are aging (which should increase marriage prevalence) and women are gaining access to education (which should decrease it), we might get misleading results from simple rates. With the IPUMS data there is an indicator of whether people have finished secondary schooling and university education in all these countries. So I can adjust for those (and run a linear probability regression with 40 million cases across 26 countries in the time in takes to walk the dog). Here is the result:
With the adjustments, Thailand, Indonesia and Vietnam have increased marriage prevalence. In most cases the adjustments reduce the time effect, reflecting the spread of education.
Conclusion: The decline of marriage is worldwide although not universal, and it’s probably not accounted for by age and education changes.**
2. Is delay the same as decline?
But what if some or even all of this worldwide trend – or the trend in some countries, like the US – is really from people delaying marriage rather than forgoing it? Some people argue, reasonably, that delaying marriage can be a sign of its increased social significance. This is the “capstone” concept popularized by Andrew Cherlin, and it’s been raised in blog comments many times.
There is good evidence for this. In the US at least, despite the delay in marriage, and the drop in the marriage rate, the great majority of Americans have been married at least once by the time they reach their fifties. Here are the percentages ever married (that is, including separated, divorced and widowed) by age for cohorts born from 1930 to 1990 (using IPUMS data).
The steps downward from back to front show that, at each age, successive cohorts are less likely to have been married. But the high levels on the far right show that much of this is just delay. By 2010, 87% of about-50-year-olds – those born around 1960 – were ever married. That said, the decline is still pronounced – that 87% is down from 95%. Thus, marriage is being both delayed and foregone, albeit to different extents.
The measure used is very important. Consider these scenarios:
- If everyone suddenly put off their marriage till next year, the marriage rate this year would be zero.
- If everyone got married at age 50, the percentage of people married in the ages 25-49 would be zero.
- Even if no one got married ever again, we would still have a majority of adults married for a while. (Idea for demography exam question: how long?)
- If everyone decided marriage was just so important as a capstone that they would only get married on their death beds, the marriage rate would be low, the percent of the population married would be low, but the percentage who ever got married would be 100%.
So what’s the right way to look at this? Questions of meaning in demography are amorphous. I can’t say from these numbers how people think about marriage, especially across all these countries. That doesn’t mean the question isn’t important, but in one sense these numbers mean it’s undeniable that marriage is less important – fewer people are doing it, and doing it for a smaller portion of their lives. (More on this here.) That goes especially for the question of family structure for children, who are increasingly likely to be born to parents who aren’t married.
3. Says who?
So marriage decline is both worldwide and real. Can we predict what will happen, that the decline will continue? The scenarios above all are possible. But I think they’re unlikely. In the post I started with, I included an illustration of the marriage rate from 1940 to the present, and argued we should plan for a future in 2040 where the rate would be somewhere between 0 (complete collapse) and the 2000 level (strong rebound but still low by historical standards). Some tweeter gave that the award for “today’s most hilariously amateur data analysis.”
I’ve made short-term predictions by extrapolation (such as the name Mary), which is fine unless the trend changes. And I’ve made some based on evidence from leading indicators, such as Google searches for divorce- or fertility-related terms. But what about the long-term trend on marriage? Here’s a cautionary tale.
In 1955, Talcott Parsons wrote the introduction to his co-authored book, Family, Socialization and Interaction Process. The book advanced a theory about how families worked that relied on a static concept of complementarity between male and female roles. The idea was that this structure was pretty ideal, the breadwinner-homemaker model was highly evolved, and change was unlikely. So Parsons’ introduction sought to tamp down speculation that a sea change was underway. Unfortunately, he did that using data through about 1950, focusing his attention on whether changes observed in the 1930s were continuing, especially increasing divorce and falling fertility. Not to worry, he wrote, mistakenly seeing the beginning of the baby boom as a new plateau. With regard to women’s employment, he went on:
…there can be no question of symmetry between the sexes in this respect, and we argue, there is no serious tendency in this direction. … The number [of women] in the labor force who have small children is still quite small and has not shown a marked tendency to increase.
And the rest is history:
Parsons and I are only two generations apart (I took theory from a guy who co-authored a book with him). And all this change was pretty fast. So, why should I be confident in predicting the future of marriage – continued decline, with chance of rebound not exceeding 2000 rates by 2040? Were Parsons and I both blinded by our assumptions?
Sure. But one difference between us is our access to demographic data. It’s surprising to me that the great thinkers of social science mostly wrote in the olden days, when they had so little information. In just a few hours I summoned these global statistics covering decades and assembled them into a worldwide trend. Doesn’t it increase our confidence in the power of the trend to know that it’s so universal? And of course that’s just a drop in the bucket. Shouldn’t all this information yield more, better knowledge and better predictions?
Even if I’m not as smart a theorist as Parsons, I can eyeball the US and global trends and conclude that a major reversal in marriage is not likely: there is no recent precedent for that, and, with the exception of some countries that still have universal marriage, no major exceptions in the direction of change. That’s how I get from “marriage has declined globally” to “marriage is declining globally.”
(This post was a lot of work so gimme a couple days off now, okay?)
*This IPUMS data includes people who are in “consensual unions” that aren’t legally married, which is less than 5% of the total. IPUMS does have some other censuses, but I excluded ones that weren’t from the right period or had something unusual about the coding, sample or measures. In the regression version I excluded individuals with missing education levels, a very small percentage. The IPUMS data are weighted to reflect population totals.
**In addition to IPUMS International, which compiled the data, these statistical offices made it available for use.