Module 8 · Lesson 8.3
Expectation, Variance & the Normal Distribution
Learn how a distribution's balancing point and spread summarize its whole shape in two numbers.
Intuition
Picture a distribution's histogram cut out of cardboard and balanced on a finger. The point where it balances perfectly, without tipping either way, is the expectation — the long-run average value. How far the weight is spread out from that balancing point is the variance: a tight clump balances easily near the center, while a widely scattered shape needs the finger to support material that's flung far in both directions.
Core concept
Expectation, written $E[X]$, is a weighted average: multiply each possible value by how likely it is, then add those products up. Variance measures spread by averaging the squared distance of each value from that balancing point — squaring keeps distances on either side from cancelling out and punishes values that sit far from the center. Together, expectation and variance are often enough to describe a distribution's shape without ever drawing it.
Interactive Lab
Drag the mean to shift where the curve balances, and drag the spread to squeeze or stretch it. Notice how a wider spread must also flatten the peak.
Key Idea
This is the normal distribution — the bell curve you just shaped in the lab above. It's built entirely from the balancing point $\mu$ and the spread $\sigma$: no other numbers are needed. In fact $\mu$ is exactly $E[X]$, and $\sigma^2$ is exactly the variance, $Var(X)$. Two A/B test groups can share the same mean conversion rate yet have very different variances — the group with the tighter spread gives you a far more confident read on whether a measured difference is real or just noise.
Think Further
- Why must the peak of the curve get shorter whenever you increase the spread?
- Two A/B test groups have identical mean conversion rates but very different variances. Which one gives you more confidence in the result, and why?
- If a distribution has $E[X] = 0$ but a very large variance, what does that tell you about the individual values, even though they average out to zero?
Show suggested answers
- The total area under the curve always represents total probability, which must stay fixed at 1. Spreading that area over a wider range means less of it can pile up at any single point, so the peak has to drop.
- The group with the smaller variance, because its outcomes cluster tightly around the mean — a measured difference is less likely to be an accident of noisy, widely scattered data.
- Large variance with a mean of zero means individual values swing far above and below zero, even though the positive and negative swings cancel out on average — the mean alone hides how volatile any single outcome can be.