New Developmental Milestones from the CDC

Developmental milestones hold a certain combination of joy and fear for parents. On one hand, we look forward eagerly to the first smile, the first step, the first word. On the other hand, milestones are an opportunity to fear that our children are behind, that they aren’t measuring up. When the book says “the average age for walking is 12 months” and our 15-month-old is still crawling around, it can be a source of stress (I speak to this one from experience).

With this context, it wasn’t surprising when people sat up and took notice of the recent CDC update to developmental milestones. It was the first time in many years that the agency has updated these. One change was to include additional sets of milestones at 15 and 30 months. A second was to, generally, make the milestones “easier” — to lower the expectations around things like language development.

This newsletter will do three things: First, explain why the CDC made these changes. Second, talk about the new guidelines around language in particular. And third, take a look at some broader data on language development, which gives more of a sense of the distribution.

Why the change?

I have seen some suggestions online that the change in guidelines is reflecting the fact that children are developing more slowly, and that the CDC is trying to hide this. In a few cases, the narrative surrounds the role of masks (lately, I feel like it always comes back to masks). These claims are untrue.

The guidelines seem to have been changed with the goal of making it easier to identify children who are eligible for early intervention. Early intervention refers to a suite of services available to children with developmental delays or disabilities. Children are identified as potentially eligible for these services by a pediatrician, a social worker, or child-care personnel.

Early-intervention programs can help children catch up to their peers. Given that, identifying eligible children is a high priority. Which means we want to make it easy to do. The goal of having developmental milestones is to do just this — to have a simple way to flag a child as possibly needing some help. But to do that, the milestones need to be organized to identify children in this group rather than to (say) identify the average child.

The last round of developmental milestones from the CDC focused more on what we would expect from the average child. This update is intended to capture something like: at least 75% of children at this age would meet this milestone. This means, then, that if a child is not meeting it, they are in the lower 25% in terms of that area. And this is the group we want to flag for possible early intervention. Prior to the update, the guidelines were closer to identifying the average — the 50th percentile. But, of course, moving from a milestone that identifies the 50th percentile of the distribution to one that identifies the 25th percentile will change the expectation.

All that is to say that much of the reason for the change (as I understand it) was to enhance the usefulness of the guidelines for pediatricians. It is unrelated to masking, and doesn’t reflect any systematic change in realized child development.

The guidelines on language

That’s the background. With that, I wanted to dive into the guidelines on language in particular, because they got a big update. (Another big update was to drop crawling as a milestone. This is a source of debate, as it is not uncommon for kids to skip crawling altogether, but there are those who think it’s an important milestone to identifying possible issues. I’m going to leave this for another day.)

You can see the full CDC guidelines here, but I’ve pulled out the “productive” language and gesture guidelines at each age group in the table below.

Again, these milestones are intended to represent levels of concern, not averages. The median child (more on this below) does say several words by one year of age. But: at 12 months, there isn’t a flag for concern unless we do not see the use of a caregiver word.

One thing that this table makes clear is the relatively slow increase in language up to 18 months, and then the explosion after. In moving from 24 months to 36 months — that’s two to three years — the expectation for what most children can do moves from putting two words together to “talks well enough for others to understand, most of the time.”

This large change is a big part of the reason that the new guidelines included the 30-month check-in. This has, however, been a source of significant debate. For one thing, counting words is perhaps helpful when we are talking about a small number, but it’s hard to know if a child has 40 or 50 or 60 words. Additionally, commenters have suggested that this milestone is too conservative — that it will miss many children who actually would benefit from additional intervention.

On reflection, I thought it might be interesting to return to the data I explored in Cribsheet. So let’s look at that, and then circle back to the 30-month question.

Word distribution

(Note: A longer discussion of this topic was in the newsletter in August 2020.)

Part of the issue with “50 words” is that it’s hard to count words when you get above a small number. If we want to get a sense of the full distribution of language across the population at various ages, it’s necessary to approach it more systematically.

A standardized approach is to identify a target word set — a compilation of common words — and collect data on the number of these words that children know. It’s possible to then look at the distribution across age in how many words children know. The most popular word list is from the MacArthur-Bates Communicative Development Inventories (MB-CDIs). It has a list of simple words — blanket, kitty, uh-oh — and asks parents if their child understands the word and if they can say it.

These forms are commonly used by many researchers in child development, and there’s a project at Stanford University called Wordbank that compiles a significant body of the data on this. You can visit the project here and produce neat graphs like the one below, on the size of vocabulary by age. It’s possible to divide these by gender or parental education and to look across languages (versions of this inventory appear in virtually all languages).

Even cooler, from a pure parental navel-gazing perspective, the Wordbank also allows you to analyze the data by word. It produces graphs like the one below, from which we learn that a word like windy appears, on average, later than moo or baa baa.

The data on this site is interesting and fun. You can do the inventory yourself if you are so inclined.

Based on these data, we might imagine evaluating the “50 words at 30 months” milestone. One significant caveat, though, is that much of this data comes from research samples that are likely to be selected in various ways. They’re not what is called a “norming sample,” which is what the CDC ideally uses to generate its guidelines. On average, the kids in the Wordbank are differently selected.

However: looking at that first graph above, it does seem like the 50 words at 30 months could be a bit too conservative. In the Wordbank data, even the 10th percentile of children have considerably larger vocabularies. This is true even if we limit the Wordbank data to norming samples, which have a less serious selection problem. To be clear: I do not think there is anything nefarious here, or anything masking-related; just perhaps this particular milestone isn’t quite correctly calibrated.

Final thought

Asking ourselves how our kids measure up — to milestones, to other kids, etc. — is a natural parenting instinct. We tend to regard early achievement of milestones as a sign that our children may, in fact, be wizards.

These CDC milestones are not focused on identifying wizards. They are very sensibly focused on helping find children who might benefit from additional scaffolding and public services. There is a wide range of normal development, and, by and large, the exact speed of development in these areas has relatively little predictive power for long-term outcomes. So by all means, explore the Wordbank database to find out whether the word hat is generally learned before the word tired. But don’t overthink it.