I was comparing a measurement to a memory
2026-08-25
The pooled reference everyone cites — Galland's systematic review, dozens of studies — puts baby night wakings at about 0.8 a night across infancy. My own 670 logged nights of one child say 2.2, 2.4, 3.5, 2.3 on the review's own age bands. Three times as many, in every band after the newborn months.
My child is not the anomaly. That reference is built from parents remembering, and parents under-report; the review's own authors say so. I log in the moment, so I find more. Obviously I do.
Which makes what follows my problem and not the literature's. A parent who tracks honestly and then meets the published figure will conclude their child is broken — and my app is what handed them the comparison.
The likeliest explanation is not that my child sleeps badly
It is that the two numbers are measuring different things. Most of the studies pooled in that review ask parents, retrospectively, how many times their baby wakes at night. The review's own authors flag it: parents under-report night wakings, because the ones a child resolves alone are invisible, and because recall flattens a month of nights into an impression of one. Night waking is the least reliable variable in the whole review — the one its contributing studies agree on least.
What I hadn't appreciated was the size, and that the gap opens after the newborn months rather than during them. My guess, and I'm labelling it a guess: early wakings are memorable because each one is a feed and you're up anyway; later ones are brief, blurry and gone by morning. The under-counting gets worse exactly as the wakings get cheaper.
Why this is a problem an app creates
A parent installs a sleep tracker. They log honestly, night after night, in the moment. Then they read anywhere on the internet that a one-year-old wakes 0.7 times a night, look at their own log showing 2.3, and conclude that something is wrong with their child.
Nothing is wrong. They have compared a measurement to a memory. But the act of tracking is what set up the comparison — the app handed them a number precise enough to be alarming and no way to know it wasn't commensurable with the published one. That is a harm which exists only because the tool exists, which makes it my problem specifically.
The fix is small and I've taken it: say both numbers, and say plainly that they are not the same measurement. Compare logs to logs. The feature I will not build is the obvious one — the reference figure quietly shown beside the parent's own. It would look like a helpful benchmark right up until the moment it wasn't.
I am wary of that one because I can feel the pull of it. And because the industry's version is worse than useless: the market leader's article on early waking name-drops millions of logged sleep moments and derives not one number from them. Data as a credential rather than as an argument. If you are going to hold a hundred thousand times my dataset, publish a base rate.
One child, hand-logged, and a waking we slept through is a waking we didn't record — which, note, biases my numbers the same way as everyone else's, just less. The gap is if anything the floor. The other thing those 670 nights turned out to be hiding is in the entry alongside this one.