Are 20% of Children Really Not Their Father's? What Amazon Reviews Teach Us About Conditional Probability
The other day, I came across a post like this on an online forum.
“Out of 2,000 DNA paternity tests, about 400 showed that the man was not the child’s father.”
That’s about 20%.
Some people who saw this probably thought, “So one in five children in the world isn’t biologically related to their father?”
The moment I saw it, I knew it was clickbait. These 2,000 cases are not 2,000 families chosen at random from society.
The Same Trap as Amazon Reviews
Something similar happens with Amazon reviews.
When a product arrives normally, works normally, and you have no particular complaints, few people bother writing a review. I don’t feel like writing, “Arrived today. Works fine. Exactly as expected,” either.
The people who write reviews are the ones who were impressed, “This is fantastic,” and the ones who are furious, “It broke on day one. Unbelievable.”
In other words, the review section doesn’t represent all buyers. It’s just people who have something to say, gathering there on their own. In statistics, this is called self-selection bias.
The DNA test story is basically the same thing.
Ordinary Fathers Don’t Get DNA Tests
If you think about it calmly, it’s obvious.
A father who has no doubts about his wife or his child doesn’t usually wake up one day and think, “Maybe I’ll get a DNA test, just in case.” It costs money, and if it comes out, the atmosphere at home turns ice-cold in an instant.
People who go as far as a DNA test have their reasons. They suspect an affair. The timing of the pregnancy doesn’t add up. The child looks nothing like them. There’s a divorce or an inheritance involved.
In other words, by the time they walk into a testing company, their situation is already different from that of an ordinary family.
“20% of people who took a DNA test found they weren’t the father” and “20% of children in the world aren’t biologically related to their father” are completely different statements.
In fact, read it from the other side, and you get this:
80% of fathers who were suspicious enough to get a test turned out to be the real father.
It’s the same 20%, but the impression changes completely depending on which side you look from.
What Matters Is the Denominator
Online articles love to emphasize numbers. 20%. One in three. Ten times more.
But what really matters in statistics is what the number is a fraction of. In other words, the denominator.
In the DNA test example, what we want to know is “P(not the father),” but what we’re actually looking at is “P(not the father | took a DNA test).” There’s one condition to the right of the vertical bar. Ignore it, and the story changes completely.
Back in school, when I learned conditional probability, it was all problems like “Given that you drew a red ball, find the probability that it came from Box A.” I had no idea what it would ever be useful for.
It turns out the lesson was this: before you judge, look at which numbers were collected.
Online, strong numbers win.
Write “20% are someone else’s child,” and people click. “20% of people who had doubts about paternity and applied for a DNA test on their own” is too long and not exciting at all.
But the second one is what’s statistically correct.
When you see a number, the first thing to think about isn’t the number itself. Who was studied? Why did those people end up in the data? Are they the same as the group you actually want to know about?
Do you read it as “20% of these fathers were raising another man’s child,” or as “80% of the suspicious fathers were suspicious for nothing”?
Statistics is a scary thing.
And marriage is probably even scarier.