Instagram parenting wins, the mom whose kid actually does homework, my kid's "I'm fine" — all of it is survivor data. And the day after a good day is usually just regression to the mean, not proof that yelling works. Knowing this didn't fix how it looks to me, so I put my kid's days on a moving-average chart instead.
World War II. Bombers come back from missions with bullet holes: wings, tail, the edges of the fuselage. The military looks at the pattern and says, reasonably: let's add armor where the holes are.
A mathematician named Abraham Wald stopped them1.
Backwards, he said. Put the armor where the holes aren't.
Because these planes made it home with those holes. A hole in the wing is survivable. The spots with no holes on returning planes are the spots where, if you get hit, you don't come back. The planes that got hit there are at the bottom of the Channel. They never made it into the dataset.
When your sample only contains survivors, you draw exactly the wrong conclusion. That's survivorship bias.
At some point I realized I run this exact error in parenting, daily.
Open Instagram, or r/Parenting. A feed full of things that worked. "We tried this chart and it changed everything!" Same. Same exactly. So I try it too.
But hold on. The parents it failed for didn't post.
Nobody writes "tried this, made everything worse." Failure posts don't get engagement, so they barely exist. What I'm scrolling through is a bullet-hole map of the places that don't need armor. I'm staring at the planes that made it back and deciding where to armor mine.
There's a mom at pickup whose kid does homework without a fight and hasn't quit piano. I asked how. I copied the method.
Didn't work. Actually made dinner tense for a week.
And it doesn't stop there, that's the part that gets you. Next comes "it works at their house, so what's wrong with ours," and you sink a little.
Pure survivorship bias. Maybe she tried a hundred things and this was the one that happened to stick. Maybe it just happened to fit her kid. Different plane, different holes. The armor goes somewhere else at our house, probably.
I do it inside the house too.
"How was school?" "Fine."
Great, he's doing fine. I relax.
Except I'm only counting the days he could say "fine."
The day he said it and his eyes slid sideways. The day he went straight to his room, earlier than usual. Those are the bullet holes in the spot that needs armor. He didn't report them, so I never saw them.
There's no fix. A bias doesn't dissolve just because you named it. What I try to do:
The third one lives in TroubleNote at our house. Six months of entries and the bad-day patterns start to show.
Since we're on planes: the Israeli Air Force.
Daniel Kahneman (Nobel laureate, later) was lecturing flight instructors early in his career. His point: praising improvement trains people better than punishing mistakes2.
(Yes, I know. Modern parenting says praise. I know. The praise doesn't make him do it, though.)
A veteran instructor pushed back:
When I praise a cadet for a clean aerobatic maneuver, the next one is worse. When I scream at him for a sloppy one, the next one is better. So praise makes them worse and screaming works.
Field data. Sounds airtight. Except praise and screaming have nothing to do with it.
Fly better than your average, and the next run tends to land back near your average, worse than the great one. Fly worse than your average, and the next run drifts back up, better than the disaster. That's the whole trick.
The praise didn't break anything, the yelling didn't fix anything. Either way, the next flight was going to be average.
It has a name: regression to the mean. Francis Galton spotted it first, in parent-child height data3.
Our house version:
Homework goes smoothly on Tuesday. Wednesday falls apart. "You did this fine yesterday — why not today?!" I'm about to say it. Fine, I do say it.
But Tuesday was just above his average. Wednesday came back to it. And since he's growing, that average itself is probably creeping upward the whole time.
A worse-than-yesterday is not a reason to scold. ...Which I know, and it still doesn't look that way when the meltdown is in front of me. That's what makes a cognitive bias a cognitive bias — knowing about it doesn't turn it off.
Side note on how to praise: Carol Dweck's research found that praising ability makes kids avoid challenges, while praising effort makes them pick the harder problems4. Different mechanism than regression to the mean, so that's a post for another day. (I know, it's written everywhere. He still doesn't do it when I praise him.)
Reading ADHD research for app design, there's also this: kids who get scolded a lot tend to end up with low self-esteem, to the point that praise doesn't even register as praise. Stacking up small wins matters. I know. I know...
(Dropping paper links is an engineer habit, not a medical claim.)
Roll it all up and the advice is:
...and none of it sticks, because the bias doesn't care what I know. Like an optical illusion: the two lines stay different lengths no matter how many times you measure them.
So instead of knowing better, I made it visible.
Log the day's small wins, one tap each. Each day becomes a dot; the dots scatter up and down. That's the wave. Run a 7-day moving average over them, and even with the waves, the line climbs. Gently, but it climbs. Should, anyway.
And when the line points up, I don't need the lecture. "Eh, today's a wash" becomes a thing I can actually say, because there's one line on a screen backing me up.
That app became GrowthTrend. It's not really for praising more. It's for yelling less. Watch the trend instead of today, cancel out my own cognitive bias, buy the parent (me) some peace.