Survivorship bias
Survivorship bias is drawing conclusions only from what made it through a filter, like the businesses still open, without counting what fell away.
Select any words here to ask your Maestro about them.
Of 200 funds launched in 2016, 121 are still open, and they averaged 5.8% a year. What did all 200 average, counting what the closed ones earned while they were open?
Where it shows up: in mutual funds that are still open, in successful companies studied so others can copy them, in surveys of current customers and in any table someone already cleared of what didn't work.
How to spot it: ask what had to happen for a row to reach your data. If the answer depends on the same outcome you're measuring, the sample is biased, even if it's random within what's left.
How to fix it: count from the start, everyone who began, not only those still around. It's the same question as separating a population from its sample: who does this data describe, and who does it leave out?
How it's used
The planes that came back
Where to put the armor
- Planes that came back had bullet holes mostly on the wings and fuselage
- The obvious idea: armor where the holes are
- marked: Abraham Wald argued the opposite: armor where there are no holesPlanes hit in the engines more often didn't come back, so they were missing from the count.
When to look closer
The customer survey
Satisfaction survey
- marked: Average rating: 4.6 out of 5
- Sent to: every customer with an active account
Your current customers rate you 4.6 out of 5. Who is that average missing?
Ask your Maestro about Survivorship bias
It answers from the free course page closest to this word, the way it answers while you study.
You: What does "Survivorship bias" mean, in plain words?
Your Maestro: I'll explain it with a case of yours: it sticks better than a definition. Which rows had to survive to reach your table?
Asking about: Answers from the page "How do I open a CSV in Excel without it changing anything?"
Only this browser keeps what you ask here. How it works
Learn more
Related terms
- Sampling
Sampling is how you pick your sample. The four types: simple random, systematic (one in every so many), stratified (random within each group) and cluster.
- Population and sample
The population is the whole group you want to know about; the sample is the part of it you actually measure to estimate the rest.
- Correlation vs. causation
Correlation means two numbers move together; causation means changing one changes the other. Moving together doesn't prove that one causes the other.
- Mean (average)
To find the mean (the average), add up all the values and divide by how many there are: 15, 7, 8, 13 and 7 add up to 50, and 50 ÷ 5 = 10.