One Sample, A Whole Distribution

Added: 2026-05-27

[00:00:00]You measured 30 trees in a forest and the average came out to 14.2 m. But if you went back tomorrow and measured a different 30 trees, that number would wiggle. So how sure are you about it when all you have is one sample.

[00:00:16]Here is the magical idea. Take that one sample and treat it as if it were the entire population. Then draw a new sample of the same size but with replacement. So, some trees get picked twice, some get skipped, and some show up three times. That's a bootstrap resample built from nothing but the data you already have. Now, do it not once, but hundreds of times. Each reample gives you a new mean, and stacking all of them into a histogram gives you the bootstrap distribution. Chop off the bottom 2 and 12% and the top 2 and 12%.

[00:00:55]And what's left is your 95% confidence interval. No normal distribution assumptions, no zcores, just honest uncertainty straight from the data.

[00:01:07]You might be wondering why on earth this should work. The answer is the empirical CDF. With 10 points, it's a jagged staircase, but with 100, it tightens up.

[00:01:18]And with a thousand, it's almost indistinguishable from the truth. Your sample, however small, is the best window you have onto the population. And resampling from it is the closest thing to time travel statistics will ever offer you.