AI summary by TwinMind
The standard error, Clearly Explained!!!
StatQuest with Josh Starmer
11m 43s
Introduction to Error Bars and Standard Errors
- Error bars closely related to standard errors
- Example: measurements from 3 samples (A, B, C) plotted on scatter plot
- Calculate means for datasets (shown as green horizontal bars)
- Calculate standard deviations, add to graph as red error bars
- Manuscripts/presentations often show "dynamite plots" (mean + standard deviation only, no raw data)
Three Common Types of Error Bars
- Standard deviations: show how data distributed around mean
- Large standard deviations = data points far from mean
- Use in graphs to show information about collected data points
- Standard errors: show how mean is distributed (not just data, but means themselves)
- Confidence intervals: related to standard errors (covered in future StatQuest)
Normal Distribution Example - Mouse Weights
- Normal distribution example: weighed many mice, plotted distribution of differences from mean
- Y-axis: proportion of mice weighed
- X-axis: difference from mean
- Most mice had weights close to average
- Few mice weighed much less/more than average
- Usually can't afford to measure all mice, so take sample (example: 5 measurements)
- Most samples close to zero since most mice have weights close to average
Sample Calculations
- Calculate mean and standard deviation from sample
- Example sample: mean = -0.2, standard deviation = 1.923
- Plot mean ± standard deviation on graph
- Rule of thumb:
- 1 standard deviation each side of mean covers ~68% of data
- 2 standard deviations each side covers ~95% of data
Multiple Samples Analysis
- Take additional 5-measurement samples, overlay means and standard deviations
- Means relatively close to each other compared to raw data
- For mean to be far from middle, most/all raw data points would need to be in single cluster far from middle
- More likely: most points close to zero, only 1-2 far away
Standard Error of the Mean
- With multiple means, can calculate standard deviation of those means
- Standard deviation of means much smaller than standard deviations of individual samples
- Standard error of the mean = standard deviation of multiple means from same population
- Standard error = 0.86 (much less than individual standard deviations)
- Gives sense of variation expected in means from independent samples
Other Standard Errors
- Can calculate standard error for any statistic:
- Standard error of standard deviations
- Standard error of median, mode, percentiles
- Process: calculate standard deviation of multiple values of any statistic
Calculating Standard Errors Without Multiple Experiments
- Formula method (rare cases): Standard error of mean = standard deviation ÷ √(sample size)
- Bootstrapping method (for everything else): when no simple formula available
Bootstrapping Procedure
- Example: 5 measurements (note: usually need 10+ measurements for good bootstrapping)
- Step 1: Pick random measurement from existing sample (not new measurement)
- Step 2: Write down value (example: 1.43)
- Step 3: Repeat step 1, pick new random measurement
- Continue 5 times total (okay to pick same measurement multiple times)
- Example bootstrap sample: 1.43, -1.38, -3.11, 1.43, -0.10
- Step 4: Calculate desired statistic (mean, median, etc.) from bootstrap sample
- Step 5: Repeat steps 1-4 until have many calculated statistics
- Step 6: Calculate standard deviation of all generated statistics = standard error
Key Takeaways
- Standard error = measure of how means from different samples might vary
- If no formula available, use bootstrapping to calculate standard error yourself
- Bootstrapping conceptually simple, easy for computers to execute
