Skewed Distributions Explained Simply

Full transcript

Word for word what is said in the video — the same text as the subtitles. Times are minutes:seconds into the video.

0:00Three worlds, one lean ▶

Incomes, waiting times, house prices: three different worlds, and one and the same lean. That is no coincidence — last time, one billionaire dragged our average into exactly this kind of tail. By the end of this video you will spot a lean in one glance, without doing a single calculation. And you will know why the world keeps building these shapes, with nothing going wrong at all.

0:26The anatomy: a hill and a tail ▶

Every skewed distribution has the same two parts. A hill, where most of the data lives — and a tail, where a few stretched-out values sit. Now the naming trap, and it catches everyone: the skew is named after the tail, not the hill. In this picture the hill sits on the left, but the tail points right — so this is right-skewed. Say it with the tail, and you will never get it backwards. Now, can you see a lean without reading a single number?

1:01The detector: the boxplot gives it away ▶

Bring back the boxplot from video four — it gives skew away instantly. Draw the box for these incomes, and the median slides off-centre inside it. The right whisker runs long — the picture of a right tail. Half of this country earns below 40,000, and the box shows that immediately. And the dots past the fence are not errors here — in a leaning country, one in twenty simply earns that much. One warning: read the whole picture, not just the box — the whisker and the flags are where the story is. And the mirror image exists too. Scores on an easy exam pile up near the ceiling, and age at death in a rich country leans the other way. Median to the right, long left whisker — left-skewed. The picture shows the lean; three numbers can reveal it without drawing anything at all.

1:59The compass: mode, median, mean drift apart ▶

Now watch the three centres as the bell leans over. The mode stays on the hill — it is the most common value, and the hill is where most people are. The median slides to the right, to 40,000 here. And the mean gets dragged furthest into the tail — the balance point chases the extremes, all the way to 49,000. Remember the seesaw from last time — this is that same drag, now built into the shape itself. For a single hill like this one, that order is your compass. Mean above the median means the data leans right; mean below means it leans left. But where does a shape like this come from? Before you watch: if only luck moves everyone's salary, does the middle person end up richer, poorer, or unchanged?

2:52The why: a floor, and luck in percentages ▶

So why does the world keep producing these shapes? Two ingredients are enough: first, a floor — incomes, waiting times and prices cannot go below zero. Second, change in percentages: this year's income times ten to fifteen percent up, or down. And the percentage differs from person to person, year after year. A five percent step is small for a small salary and large for a large one, so the steps grow with the income. Start everyone at the same 40,000 and run that luck for twenty-four years. The middle person still earns 40,000, the average has drifted to 49,000, and a long right tail has built itself. House prices do the same dance: a floor near zero, change in percentages, and a tail full of mansions. So a lean is not a flaw in your data — nothing went wrong. Floors and percentages simply do this, everywhere. Which is exactly the problem: one rule from an earlier video was written for the bell, and it will not survive this.

4:04What breaks: the bell's promise ▶

But the lean does break something. Video one gave us the 68–95–99.7 rule — and that promise belongs to the bell. Lay those bands over our incomes, with this country's own sigma of about 35,000, and they fail twice. The wide bands drop below the floor: 49,000 minus two sigma lands at minus 20,000 — an income nobody can have. And beyond the top band, the bell promises about one earner in seven hundred and forty; this country has thirteen times as many. Predictions built on those bands will quietly keep being wrong. The bell made those promises — the world never did. So which of your tools still stands?

4:54What still works ▶

And yet nothing here is broken, because you already own the right tools. Three of them came from earlier videos — which one showed you the lean before you read a number? The boxplot — and it works because none of these three ever assumed a shape, they only sort and count. The median stays honest when the mean gets dragged, and percentiles keep working whatever the shape. There is one more way out — and it is the strangest one.

5:28In one picture ▶

Five things, in one picture: skew is named after the tail, not the hill. The boxplot gives it away — median off-centre, one long whisker. For one hill: mean above median, it leans right. Floors plus luck build these everywhere, so a lean is never a flaw, and the median toolkit works — which is where we go next.

5:56What's next: change the ruler ▶

Next time we take this same leaning data and do something strange: we leave every number in place, and change the ruler instead. One axis trick, and the lean disappears. Tell me below — where does skewed data show up in your world? And if you want to keep learning, subscribe — the next video will find you. Until then: happy learning.