AI summary by TwinMind
Confidence Intervals, Clearly Explained!!!
StatQuest with Josh Starmer
6m 41s
Bootstrap Refresher
- Bootstrapping method using 12 female mice weight measurements
- Sample mean calculation from 12 measurements (not global population mean)
- Bootstrap process:
- Randomly select 12 weights from original sample
- Duplicates allowed (sampling with replacement)
- Calculate mean of random sample
- Repeat steps 1-2 until thousands of means calculated (sometimes >10,000)
Confidence Intervals Definition
- 95% confidence interval = interval covering 95% of bootstrapped means
- 99% confidence interval = wider interval covering 99% of bootstrapped means
- Visual representation: black bar spanning specified percentage of means
Statistical Testing Applications
- Confidence intervals enable visual statistical tests
- 95% interval means anything outside occurs <5% of time
- P-value of anything outside confidence interval <0.05 (statistically significant)
Single Sample Testing Example
- Test: true mean of all female mice <20
- Method: draw confidence interval, check if area left of 20 falls outside 95% interval
- Result: if outside interval, P-value <0.05, statistically significant difference
Two Sample Comparison Example
- Compare female vs male mice weights
- Both samples bootstrapped separately
- Non-overlapping 95% confidence intervals = statistically significant difference (P<0.05)
- Overlapping confidence intervals = still possible significant difference, requires t-test
- Caveat: overlapping intervals don't guarantee non-significance
Key Takeaway
- Non-overlapping confidence intervals guarantee statistical significance
- Overlapping intervals require additional testing
