Hacker News Reader

Comment by shubb | Hacker News Reader

The choice of graph here is really bad.

Basically the graph is trying to show two things at once - the difference between the categories and the difference between each category and normal. For multiple variables.

By showing a lot of information on one graph, it becomes hard to see any particular pieceof information.

If you are trying to show a comparison between two categories, it's better to normalise those two categories. If you want to show that being far from the centre of the bell curve is more significant (which doesn't come accross on this graph at all by the way), then factor that in when you normalise.

Two clear graphs are better than one bad one.

View on HN

Also, the y-axis is totally off. I zoomed in and drew black lines from the top of each of these bars (60%, 61%, 62%, 63%):

https://imgur.com/a/wzbI7mC

I honestly don't understand what they are plotting:

> So for example, the 74th-%tile in Competition among Hardcore gamers means that the average Hardcore gamer scores higher on Competition than 74% of the gamers in the full data set.

But together casual and core make up 78 % of the sample, so that the some hardcore gamers must have scores lower than the "hardcore average". Those hardcore gamers only make up 4% of the total sample, which means that 0.04/0.21 = 0.19 -> 19 % of hardcore gamers score lower than the hardcore average!

That's so skewed an arithmetic average is clearly of little use, but at the same time they talk about percentiles, so I don't know what's happening here, did they take the sample average and pretended it's the median?

Aren't you assuming (probably incorrectly) that all hardcore gamers rated Competition more highly than all core/casual gamers?

I'm not sure how much less weird allowing for that makes things, but your reasoning for concluding that hardcore gamers with lower Competition scores than the hardcore average make up only 4% of the sample isn't valid.

For example, if half of the hardcore gamers (around 10.5% of the sample) are above the hardcore average and half below, that would put about 81% of the nonhardcore gamers (about 63.5% of the full sample) below the hardcore average and the rest above (about 19% of the full sample).

Agreed. It's relatively useless information without normalizing the differences by the variances (Cohen's d) or at least providing confidence intervals.

Reply on news.ycombinator.com