21st of December ... here's my #stats advents calendar:
In case you learned that a 95% confidence interval (e.g., [0.3,0.6] for some true parameter) is interpreted like: "With probability 0.95 the true parameter is between 0.3 and 0.6", then STOP IT 🛑! It's wrong! ❌
1/3
In case you learned that a 95% confidence interval (e.g., [0.3,0.6] for some true parameter) is interpreted like: "With probability 0.95 the true parameter is between 0.3 and 0.6", then STOP IT 🛑! It's wrong! ❌
1/3
Correct is:
If you could resample data from the true data generating process infinitely many times, then 95% of the resampled confidence intervals will contain the true (unknown) parameter (e.g. the mean).
Best visualization I know:
rpsychologist.com/d3/ci/
2/3
If you could resample data from the true data generating process infinitely many times, then 95% of the resampled confidence intervals will contain the true (unknown) parameter (e.g. the mean).
Best visualization I know:
rpsychologist.com/d3/ci/
2/3
The one single confidence interval [0.3,0.6] computed from your data will either contain the true parameter or not... you never know.
This makes it extremely difficult to properly interpret a given single (frequentist) confidence interval.
3/3
This makes it extremely difficult to properly interpret a given single (frequentist) confidence interval.
3/3
• • •
Missing some Tweet in this thread? You can try to
force a refresh
