Hacker News Reader

Comment by google234123 | Hacker News Reader

Even if there is a large skew, the mean and median will be often still be pretty close together. Especially with a large sample size.

View on HN

That's not how it works... The larger the skew, the farther away they are. If your median goes up significantly because of whales, that means they're not really "whales" as usually understood.

If you have 90 creators earning 5$ / month and 10 creators earning 50k$ / month, they earn 5,4k$ / month on average while the median is 5$ / month. On my books this is not close together.

If it’s somewhat Gaussian or normal sure. But if it’s a different distribution then this isn’t true at all.

What use is median in a Pareto distribution (which this phenomenon falls under)? From what I can tell median is useless, I can't think of a use case, and the internet fails to give me one. Average gives you an expected value to work with. Median though, can be worthless in a pareto distribution where the 90% of the value is in the top 10% of the distribution.

I feel like people have this default reaction to assert that median is as or more important than mean. When, in reality, median is useful for in only specific distributions.

I think people say "median", but what they are really interested are the moments of the distribution. Median just is a key moment in Guassian distributions, other distributions have different important moments, and median isn't necessarily one of them.

Mean and median give you strictly greater information than just mean (or just median).

Given assumptions about the distribution, you may even be able to fit (or at least narrow down) the exact shape based on these two values.

If the median and mean are significantly far from each other than that's an immediate indication that the distributon is skewed, and we also know to what side. If you know the sample size you can also get a degree of skew from this. I challenge you to tell me any other two numbers that can give you that much info by themselves with no knowledge of the underlying distribution.

Reply on news.ycombinator.com