Fertility and social liberalism in Europe

A bunch of graphs from the European Values Survey

Yes, I’ve been on hiatus for months. Sorry. I don’t really know what to say, just that I wanted to burrow into myself for a bit, rather than just churning out crap for an audience. Now I’m re-emerging with something, maybe — no guarantees, sorry again.

Especially when overall fertility is low, who is fertile can transform society drastically. This is an important enough topic to maybe become a book, so I plan to put some early sketches and drafts out here.

I wrote earlier about value change in Europe:

The cultural surrender of the European middle class: part 1

I argued that these people’s low fertility was not great news for them, and claimed that this was linked to their values:

Cultural surrender: a comment from a reader

But I didn’t really prove that. So let’s look again at the European Values Survey.¹ These are the countries I’ll use:

All countries in waves 2-5 of the EVS

We’ll start with a graph, showing waves 2-5 of the EVS. The x axis is “social liberalism” (more on that later). The y axis is how many children people had.

This is already pretty interesting:

1. 45-49 year olds are done having children (well, mostly ☺️). So this is a measure of cohort fertility, starting with the cohort born in 1941-48 and ending with those born in 1968-76.

2. There are massive differences in fertility by social liberalism. The earliest cohort has 2.4 children versus 1.6 children at the two ends of the graph. Even between, say, the 25th and 75th percentile, there is a difference of half a child. I don’t know any modern pro-natalist policy that achieves anything close to this.
   They’re also bigger differences than usually seen in the literature on left-right orientation and fertility. (E.g. this paper finds differences of about 0.2-0.3 children in Europe.)²

3. But the differences gradually shrink, and by the latest cohort they are completely gone and overall fertility has declined to modern values much below the replacement level of 2.1, marked by the dotted line.

4. There’s another effect, which you can see in the graphs on the bottom row. Those are density plots, showing how many 45-49 year olds were socially liberal in each wave. The humps move to the right: there’s a big shift towards social liberalism over time. So it’s not just that fertility is declining at each given value of social liberalism; also the population is becoming more socially liberal, and that will also correlate with lower fertility, until the most recent cohort.

But what, you must be asking, is this “social liberalism” measure?

Creating the measure of social liberalism

The EVS contains numerous questions about people’s personal, religious, social, political and environmental values. But individual survey questions are quite noisy: people may answer them at random, or be uncertain where to put themselves on a given scale. A better way to get at their underlying views might be to aggregate their answers to many questions. I used all questions that were asked in waves 2-5, and which seemed like values (so, not things like whether someone trusted others, or membership in organizations). Then I ran a principal components analysis. That is a piece of maths which finds one or more dimensions that explain the variation in the data. I was hoping to find multiple dimensions, but really only the first looked promising: it explained about 12% of the variation in the data in every wave. (All the other dimensions were at around 6% or less.)

Does the first principal component measure anything real? It seems so, because I computed it separately for each of the four EVS survey waves, and its relationship with each individual question was almost the same in every wave. I applied the analysis for wave 2 to respondents in all waves, so it’s always exactly the same summary measure of the questions.

Here’s what it measures. This is the set of questions that “loaded” at more than 0.10 on the first principal component in every wave:

Important in life: Religion
Important child qualities: religious faith
Belong to religious denomination
How often do you attend religious services
Religious person
Believe in: God
Believe in: life after death
Believe in: hell
Believe in: heaven
Personal God vs. Spirit or Life Force
How important is God in your life
Justifiable: Homosexuality
Justifiable: Abortion
Justifiable: Divorce
Justifiable: Euthanasia
Justifiable: Suicide

In other words, this is a measure of traditional religious and social values, and their opposite. Higher scores on the principal component are associated with less religious and traditional answers to these questions, so I called it a measure of “social liberalism”. That is what’s shown on the x axis of the graph above.

Sidebar: here is how I think about liberalism. Political liberalism is the version of liberalism that emerged in Europe in response to the religious wars of the sixteenth and seventeenth centuries. It is a purely political set of beliefs, designed to let people of different belief systems get along. You can be an observant Christian or Jew and a political liberal. Social liberalism is another offshoot of liberalism’s tree, one which insists that the liberal vision can’t be fulfilled by purely political measures. For that reason it tends to become a total vision of society and values (which is what leads people in the extreme to talk about “woke religion”). I would say that both ends of the social liberalism measure capture strong, coherent world views or ideologies: various traditional religions at one end; social liberalism at the other. In between, I imagine people who are more confused, uncertain or indifferent — Christmas churchgoers, say, who might also be pro-choice.

It was surprising to me that even in the most recent survey wave, around 2020, the big dimension separating Europeans is religion, belief in hell, church service attendance and conservatism on social issues! Perhaps I spend too much time in a world of effective altruism, AI doomerism, and other highbrow intellectual debates. Anyway, it is useful that a single unchanging dimension captures variation in values for more than 30 years of data.

About causality

Is this causal? Do socially liberal values cause people to have fewer children? That is a difficult question which I shall basically duck, for two reasons.

First, it is just very hard to answer. Taken literally, the measure summarizes respondents’ answers to some survey questions. Obviously, we don’t care about the answers on their own — we are only interested in them because they correlate with some other things, nebulous mental objects which we think of as beliefs, attitudes or values. But then what would even count as a “clean” policy measure or natural experiment, which would alter socially liberal beliefs and nothing else, and thus let us talk about causality? How would we tell? It’s not by chance that economists, who focus on policies which can be deliberately introduced, obsess about causality. But much of demography is about grand-scale social change, which is bigger than any single policy.

Second, when fertility changes the distribution of values, that is an example of cultural selection — a kind of natural selection, which operates on the level of culturally transmitted beliefs rather than biologically inherited genes.³ And natural selection is about correlation, not causality.

(Example: suppose there’s a nuclear war between Oxford and Cambridge, and Cambridge is destroyed. That would be an event selecting against intelligence.⁴ It doesn’t matter if people’s intelligence caused them to die in the bomb blast, or simply correlates with it, via living in Cambridge; the change to the population is the same.)

So, I won’t try very hard to pin down causality. Correspondingly, you shouldn’t think of the graph as “these values lead to fewer children”. Instead, just think of it as about two groups of people, tribes if you like. Social liberals in Europe have been having fewer children than traditional conservatives; this has shrunk their tribe compared to the other.

Some obvious confounds

Still, there are some causal-ish questions you might be interested in like “is this really just explained by education?” Maybe traditionalists are less educated, and controlling for education would remove the association with social liberalism.

To investigate this, the next graph shows children by liberalism, but splitting our respondents by when they completed full-time education:

Because we’re splitting up respondents into smaller groups, I’ve used a simpler smoothing line, which captures the overall association without allowing too many sample-dependent wiggles.⁵ But the overall story, shown by the dashed line, is the same: very big at first, then shrinking over the waves. Although more educated people seem to have slightly fewer children, this doesn’t really affect the association between liberalism and fertility, which is still there within each group. There’s some evidence that in the last wave, liberals who left education early have more children than conservatives, but this is quite a small group of respondents so I wouldn’t read much into it.

We could ask the same question about religion: maybe the association is due to highly conservative and fertile Muslims? In fact, that’s impossible because there are too few Muslims in the survey — only about 2% of 45-49 year olds even by the most recent wave. And splitting up respondents by some religion or none, or by Roman Catholic versus others, also doesn’t get rid of the association with social liberalism. I also checked if religious practice (e.g. how often the respondent attended church services) explained the association: it didn’t.⁶

What about income? Here’s the same graph, split up by income groups.

This is a more interesting picture. At first, differences between rich and poor are basically unimportant. Over time, the liberalism-fertility relationship becomes stronger for the poorest group. It’s interesting to speculate why. Still, the basic association holds across all income groups, until the final panel when the overall association is flat anyway.

We might wonder if the liberalism-fertility relationship holds in all countries. I needed a bit more data for this, so I included everyone aged 45-54, and estimated just a simple linear relationship with the raw liberalism score rather than its quantile. I did this mainly for the joy of including a splendidly tall graph:

Clearly, though there isn’t always enough data to learn about every wave, the overall relationship holds in every single country.

Lastly, could this simply be reverse causality, i.e. people get more traditionalist when they have children? I can’t tell from these surveys, but others have tested this idea and found little evidence for it in Europe.

Mechanisms

In short, whatever makes social liberals have fewer kids, it’s not that liberalism is standing in for some obvious other demographic. So, what are the possible mechanisms?

Maybe marriage? Here’s a similar graph to before, but this time showing what proportion of 45-49 year old respondents were married:

This shows the same basic pattern: social liberals are less likely to be married, and also the likelihood decreases over time across the board. Unlike for number of children, the relationship with social liberalism holds up in later waves.

Another mechanism might be how many children people want. One survey question asked “What do you think is the ideal size of a family - how many children, if any?” I don’t love this question: does it mean ideal for you, ideal for society, or what? It’s also only asked in one of my waves. But it’s what we’ve got:

The age differences combine the effect of aging, and of birth cohort: older respondents in 1990 were also born earlier. Anyway, there are certainly big differences in ideal family size by social liberalism. As with marriage, those differences hold quite steady across ages/cohorts. On the other hand, basically no group wants less than 2.1 children, so while ideal family size probably helps explain the differences in actual family size, it can’t be the only explanation.

Let’s see how much these variables contribute to the fertility-liberalism relationship. The next graph plots number of children, split by respondents’ marital status.

The relationship between values, marriage, and fertility changes over time. In the 1990-1993 wave, marital status does not make much difference: although married people always have more kids, the slope of the relationship with liberalism is still strong for either group. By 1999-2001, that’s no longer true for unmarried people, but it still holds for married people. In the next wave, the relationship is clearly weaker among both groups than it is overall. In other words, by then, marital status is explaining some of the relationship between values and fertility. In the most recent wave, the fertility-liberalism relationship actually seems to reverse among the unmarried. The flatter relationship in this wave combines three facts: liberals are less likely to be married; married liberals have fewer children than conservatives; unmarried liberals have more.

Here’s a similar graph splitting respondents by their ideal family size. (Almost 90% of people say either 2 or 3 kids for this.)

For wave 2, family ideals make a small-ish difference to the liberalism-fertility relationship, which is still pretty strong among either subgroup. We don’t know if this changed in later years, though.

Here’s another illuminating way to split this up: we can calculate the gap between respondents’ ideal and achieved family size. A positive gap means having more children than the respondent’s ideal; a negative gap means having less.

Among the (small) group whose ideal family was no or one child, conservatives tend to have more than their ideal. Those whose ideal is 2 children mostly tend to achieve that. But those whose ideal is 3 children or more, have fewer than their ideal on average; among them, conservatives get closer to their ideal.

Millennials

One last thing. Our data so far stops with generation X, people who were at least 45 in 2021. What has happened more recently? We can’t tell for sure because respondents under 45 might still have more children in future. But let’s just look anyway. We’ll plot children by age for under 45s in the most recent wave. Again, this combines cohort effects and age effects, since older people in this wave were born in earlier years.

Although the first three quartiles are very close to each other (and probably statistically indistinguishable), the most liberal group has consistently fewer children at all ages. So, the association between social liberalism and fertility may not have vanished.

The big picture

Among European baby boomers, born after the war and 45-49 in the EVS 1990-1993 wave, religious social conservatives had higher fertility than social liberals, and were well above replacement level while liberals were below it. Over the next thirty years, that changed: conservative fertility fell to liberal levels. At the same time, society changed and liberals grew from a minority to a majority.

Fertility has two aspects: how many children you want, and whether you achieve that number. The data on ideal family size shows that conservatives consistently say they want more children than liberals. But while earlier generations achieved that goal, the 16-24 year old conservatives of 1990-1993, who wanted more children than their liberal peers, ended up having the same number.

Why is that? My best guess is that when liberals predominate, they set up liberal laws and institutions and propagate liberal social norms, and this affects conservatives too. There is some supporting evidence for this from a paper by Brandon Schnabel which finds that a country-level measure of secularism predicts the fertility of religious conservatives in that country. The kinds of laws and norms I mean might include easy divorce and large state pensions (both known to link to lower fertility); less public discourse supporting the family and marriage; or public spaces that are not designed with families in mind. There is obviously plenty to investigate here.

A famous argument about fertility and values is Eric Kaufmann’s excellent Shall The Religious Inherit The Earth? It focuses on extreme religious sects which mandate high childbearing. Here I’ve looked at more mainstream values. But if social conservatives indeed have lost whatever magic kept their fertility high, then maybe the future does belong to the extreme sects which avoid this.

The most important conclusion is the simplest. (It is not the sort of thing I could put in an academic paper!) The basic function of social values is to help society survive. No matter how attractive they seem, values whose holders die out have something wrong with them.⁷ It remains to be seen whether European social liberalism is compatible with the long-run existence of the societies it has moulded.

I started off by talking about how fertility affects society’s values. But to understand this, we need to know not just which values are associated with having children, but also whether those children’s values are the same as their parents. I’ll look at that in future.

Code to reproduce the plots.

1. The EVS has a trade-off between country coverage and time coverage: there are five waves, at nine year intervals going back to 1981, but early waves had fewer countries. I’ve used just waves 2-5, and countries that were present in all of those waves, so results aren’t driven by new countries entering the sample.

2. One reason is that I’m relying on a composite measure instead of a single survey question; if single questions are noisy measures of ideology, then that will shrink estimates of ideology’s effect towards zero.

3. Genes could matter too, but I doubt they play a large role, and anyway I won’t try to separate out genetic from cultural inheritance.

4. If Oxford is destroyed, the direction of selection is less clear.

5. A Poisson regression.

6. I added “Attends church services” and “Belongs to a religious denomination” as controls to a Poisson regression of number of children on my social liberalism measure. They were negatively signed (switching from positive if I didn’t include social liberalism) and including them made social liberalism have a bigger negative coefficient. I also fitted a factor analysis, separating out “religious” and “social components” of the most important variables. Both factors had large and significant coefficients on number of children, though religion’s coefficient was bigger.

7. Certainly if they are the cause of the die-out. Arguably, even if they are not, because they failed to protect against it.