Understanding Uncertainty · Mini-Lesson · pairs with Class 04 & Class 10

One Sample, One Point — A Skewed Population

Same collapse, same layout, a population that isn't symmetric at all

Understanding Uncertainty · in-class group work

A companion to the die version of this mini-lesson, rebuilt on a population that looks nothing like a die: customer wait times, modeled as Exponential with true mean \(\theta = 2\) minutes — most customers wait a short time, a few wait a very long time, and the histogram has a long right tail instead of six even bars.

A companion exploration built to accompany the sample-mean material in Class 04 and the sampling-distribution material in Class 10 — no external dataset, the widget generates its own wait times in the browser.

New vocabulary is underlined like this — hover or tap any term for a one-line explanation that appears right below the line you are reading.

Think · Pair · Share · does skew break the collapse?

Think · 2 minIn the die mini-lesson, the top histogram was symmetric (roughly flat across six faces) and the bottom strip of averages still ended up looking clustered and roughly symmetric around θ. If the top histogram is instead heavily lopsided — mostly small values with a long tail of rare large ones — do you expect the bottom strip of averages to also look lopsided? Why or why not?

I've done the Think step — reveal Pair & Share

Pair · 3 minCompare predictions, then watch what the widget's bottom strip actually looks like once you've drawn 30+ samples.
Share · 2 minAgree on what's surprising (or not) about the answer.

Same mechanic as the die version, different population. The top histogram is one sample of N customer wait times — most bars piled up near zero, a thinning tail stretching right. Averaging those N numbers collapses the whole histogram into \(\bar X_N\), which joins the sampling distribution in the strip below, against the true \(\theta = 2\). Click any dot to bring its histogram back up top.

Top: histogram of one sample's N raw wait times — visibly lopsided, long right tail. Bottom: every sample's mean collapsed to one dot, plus the solid θ̂ line averaging every dot so far.

The top stays lopsided; the bottom does not. No matter how many wait times you draw into one sample, that sample's own histogram keeps the same long-tailed shape — skew is a property of the population, and it doesn't go away. But the bottom strip of averages behaves completely differently: at small N it's a little lopsided too, but by N = 30 or 40 it already looks close to symmetric, clustered tightly around θ = 2. Averaging is what launders the skew away — a preview of exactly what the Central Limit Theorem formalizes.

Check your intuition

Set N = 2 and draw 30 samples. Look at the bottom strip's shape — is it symmetric? Now set N = 40 and draw 30 more. What changed about the strip's shape, not just its spread?

Show answer

At N = 2, the strip of averages is still visibly skewed right — averaging just two numbers barely dents the population's own lopsidedness. At N = 40, the strip looks much more symmetric and bell-shaped, clustered near θ = 2, even though the original wait-time population never stopped being skewed. The averaging step, not the population, is what produces the symmetry.

FAQ

Why does the mean equal the standard deviation for this population?

That's a specific fact about the Exponential distribution: with rate \(\lambda\) and mean \(\theta = 1/\lambda\), the variance is also \(\theta^2\), so the standard deviation is \(\theta\) too. It isn't true of skewed distributions in general — it's a coincidence of this particular one, worth not over-generalizing from.

Is this what the bootstrap actually looks like for skewed data?

Close, with the same substitution as the die version: instead of drawing fresh wait times from the real (unknown) population, the bootstrap resamples with replacement from the one real sample you have, and treats each resample's mean the same way — one more dot in the same kind of bottom strip.

More estimator-properties mini-lessons: One sample, one point (die) One sample, one point (variance) Unbiased, but never consistent What the sample mean minimizes Why the sample variance undershoots How bad is my estimator?