The Great Baby Panic of 2017

Twitter is all aflutter at the idea that the US is going through an unprecedented baby bust. It started with this Medium post by blogger and economist Lyman Stone, which was tweeted by Ross Douthat. A day later, Matt Yglesias tweeted about the economics of parenting, and in the replies I was dismayed to see that folks were already (still?) referencing the “plummeting fertility” narrative. Now that the baby bust has been taken up as a motif in a Douthat column, with link to the original post, I feel less bad about airing my frustrations on the mathematical problems with the idea.

It is true that fertility, as measured by the total fertility rate (TFR), is falling in the U.S. We enjoy a relatively high fertility rate compared to most high-income countries, but it slipped below two children per woman around the time of the Great Recession, and it hasn’t yet recovered. The Human Fertility Database puts us at 1.844 children per woman for 2015.

However, it is not true that TFR data alone indicates a massive crash in fertility. Stone writes in appealing expert-ese (“remember, total fertility is demographically-controlled”), tinged with a showman’s flair for the dramatic (“guys, this is NOT good”), but his use of TFR is misguided. I’m here to talk about why.

The fundamental problem with fertility measures is that humans are likely to bear children anytime between the ages of, say, 15 and 45, but we need data about fertility every single year. Do we know what the current cohort of 15-year-olds’ lifetime fertility behavior will look like? Not too clearly, no; that’s a long time horizon.

How to solve this? The TFR is a composite measure that allows us to imagine what a woman’s lifetime childbearing would look like if she were like the women of today. We take a snapshot of age-specific birth rates — that is, what every age group is doing right now — and imagine that that represents what some real group of women (like today’s 15-year-olds) will do over their lifetime. It’s a quick and handy way of summarizing fertility, but here I want to present three illustrations of the pitfalls of using it as a predictor of long-term trends.

Example 1: A Simple Model

Imagine all childbearing happens within three age groups — 20-somethings, 30-somethings, and 40-somethings. At time T, the 20-somethings are having a lot of babies. The 30-somethings are having a few babies, but they’re mostly done. The 40-somethings are basically done having babies.

At time T+1, yesterday’s 20-somethings are now 30-somethings, and they’re having a few babies, but they’re mostly done. Yesterday’s 30-somethings are now 40-somethings, and they too are done. What’s up with today’s 20-somethings?

Well, suppose we find that they aren’t having babies! Maybe they graduated from college during a recession. Maybe they’re not interested in kids. Who knows. Bottom line, if we take a snapshot of fertility right now, at time T+1, we’ll end up with a very low TFR — neither 20-somethings, nor 30-somethings, nor 40-somethings are having many kids. It looks like fertility has totally and utterly crashed.

But maybe at time T+2 we’ll find that those 20-somethings, now 30-somethings, have gotten around to kids, and the fertility rate will be back to normal — one age group having a lot of kids, other age groups not. Here’s a simple chart illustrating what that would look like:

In the transition from time T, when the youngest age group is producing the bulk of the children, to time T+2, when the middle age group is producing the bulk of the children, there’s a temporary trough in childbearing.

This is what demographers refer to as a tempo effect. TFR is subject to tempo effects any time the mean age at childbearing is changing, and they can be quite large. (Here’s a classic article on it if you would like to read more.)

Mean age at childbearing is indeed rising in the U.S., as you may suspect if you are between the ages of 20 and 35+ and have recently reflected on what your parents were up to when they were your age. This depresses the yearly TFR measure, regardless of how many kids our cohorts will end up having. For now, it’s much more useful to think of the current slouch in TFR as a baby gap than a baby bust.

Example 2: Real Data

This graph, from my own research, shows the total fertility rate and the cohort fertility rate for Russian women during some of the worst of Russia’s demographic crisis, when fertility fell and mortality rose. The blue line shows the “snapshot” view — what we might imagine fertility would look like if a real group of women did, over their lifetimes, what women in that year were doing. The dotted orange line shows the actual fertility behavior of some of the groups that were having kids during those lean years — the birth cohorts of 1963 through 1980. (Note that it only includes fertility up to age 40, so these cohorts actually had *slightly* higher fertility than shown, and is centered on the year each cohort turned 30, purely for graph readability. Note also that for the last few cohorts, there’s a bit of projection involved.)

Data source: Russian Fertility and Mortality Database. Center for Demographic Research, Moscow (Russia). Retrieved from http://demogr.nes.ru/index.php/ru/demogr_indicat/data

We can see that Russia’s fertility decline, and subsequent recovery, were real — the birth cohort of 1973, who hit peak childbearing age during the tumultuous 1990s, really did have fewer kids than those born 10 years earlier. For later cohorts, the line trends back upward. But both the decline and the rebound look way bigger in the period “snapshot” view than in actual realized behavior. The overall effect of those lean years on Russia’s population size has not been as big as was initially feared. TFR is not a good indicator of CFR.

Example 3: Of Emmas and Emilys

If graphs aren’t your thing, let’s bring it back to the U.S. and illustrate with Stone’s own tragic example ladies, who failed to reproduce in their 20s.

In two years, Emma meets someone on Tinder — I know, right? — and gets married. They go on to use donor sperm to have two kids in their late 30s.

Olivia continues to excel at her beloved job, gets transferred to San Francisco, gets egg-freezing covered by the company(!). But that city’s high-rent, child-unfriendly climate, combined with their own ambivalence about children, leads her and Bob to decide against having kids.

Harper and co. re-accrue their vacation time, but are still undecided about the third kid. When Harper accidentally gets pregnant (the US has a high rate of unintended pregnancy for a wealthy country!) they decide to roll with it. Yes, they name him Lyman.

The “big changes” in these women’s late-20s fertility behavior partly ironed themselves out over time. Looking at their sad, sad lives in 2017 did not turn out to be a good predictor of their cohort’s fertility.

In conclusion: demographic data literacy is hard. Unless you’re in a certain corner of the social sciences or public health research, you won’t ever learn this stuff. But a good first rule of thumb is to be very skeptical when people use short-term trends to make a case about something that involves human behavior over the long term.

A good second rule of thumb is to be skeptical when someone tells you something is really bad, but is vague about why. But that’s a topic for another post.

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.