What Does a 90% Confidence Interval Actually Mean?
What Does a 90% Confidence Interval Actually Mean?
Most people who quote a 90% confidence interval are wrong about what it says. 🤦
It does not mean there is a 90% chance the true value sits inside the interval. If you think that, you are confusing two completely different worlds.
The Truth Is Less Comforting
In frequentist statistics, the true parameter is a fixed number. It is not wandering around. Your interval is the thing that moves.
A 90% confidence interval means: if you repeated your sampling process infinitely, 90% of the intervals you build would trap the true parameter.
The other 10% would miss completely. And here is the kicker — you never know if the interval you just calculated is one of the misses. 🎯
Why the "90% Probability" Idea Feels Right
Because Bayesian credible intervals actually give you a probability about the parameter. But unless you explicitly fed in a prior and ran a Bayesian model, you did not compute one of those.
You computed a frequentist confidence interval. Stop calling it a probability. It is a coverage rate over infinite repetitions.
What 90% Costs You
A 90% interval is narrower than 95%. That looks nice on a chart. 📊
But narrowness comes from accepting a 1 in 10 chance of being wrong. A 95% interval accepts 1 in 20. That difference matters when you are making decisions under uncertainty.
| Confidence Level | Interval Width | Chance of Missing Truth | Best Used When |
|---|---|---|---|
| 80% | Narrowest | 20% | Early exploration, massive samples |
| 90% | Medium | 10% | Industry standard for some engineering tolerances |
| 95% | Wider | 5% | Academic publishing, medical studies |
| 99% | Widest | 1% | High-stakes safety decisions |
How To Calculate a 90% Confidence Interval
Here is the blunt formula for a mean when you know the standard deviation:
Sample mean ± 1.645 × (standard error)
That 1.645 comes from the Z-distribution. For 95% you use 1.96. See the pattern? Lower confidence means a smaller multiplier means a tighter interval means more risk. ⚠️
Step-by-step
- Collect your sample and calculate the mean.
- Compute the standard error (standard deviation divided by the square root of n).
- Multiply the standard error by 1.645.
- Add and subtract that value from your mean. Done.
If you are using software, stop clicking buttons blindly. Know that when you select 90%, you are telling the computer to let 10% of imaginary future intervals fail. 💻
The Real-World Translation
Your boss sees a 90% interval and thinks "pretty likely." You should think "I am comfortable being wrong one out of ten times."
That might be fine for ad click-through rates. It is probably not fine for bridge load limits. 🌉
Stop treating confidence intervals as magic shields. They are gambling odds dressed in Greek letters. Know your tolerance for losing before you pick 90%.