Hacker News Reader

Comment by bobbylarrybobby | Hacker News Reader

And if the circles’ areas, not their radii, were the failed assets. As it is, a circle’s apparent size (we judge 2D shapes by their area, not any one linear dimension — although this is difficult to do in practice, and linear marks are generally superior) is the failure size squared, which distorts the data a boatload.

Here is a, er, more faithful representation of the data. The recent failures don't look quite so crazy anymore!

https://observablehq.com/d/d85ce0c639bc4df7

View on HN

The circles' areas are showing the failed assets. Observable Plot defaults to using a square root scale when encoding a quantity with the radius of a circle.

https://observablehq.com/plot/marks/dot#dot-options

TIL, thanks. Guess they are crazy after all.

You'd think that radius meant radius, but that'd be too simple.

So weird, should it not be the area of the circle being the value? The radius should be √(Area/π) not the square root of the value or am I misunderstanding?

Circles don't have volumes?

The volume of a circle would presumably be zero, but yes I've used the wrong word. I'll change it above. I sometimes wish there was a forum that allowed editing of spelling in other peoples posts and they could just accept the changes (same with Twitter). You could spend all your days correcting obvious mistakes...

> The volume of a circle would presumably be zero, but yes I've used the wrong word.

i guess?

> I sometimes wish there was a forum that allowed editing of spelling in other peoples posts and they could just accept the changes (same with Twitter). You could spend all your days correcting obvious mistakes...

nice and interesting idea except some people are awful and would abuse that feature :(

maybe you should make a forum with that feature!

In this case, the x-axis is a different unit from the unit of data measured by the radius, so the fact that it's off by a constant multiple doesn't matter; it's just an arbitrary scale.

If both the x- and y-axes were encoding data in the same unit as the area encoded, then it might make sense to correct for the area, but I don't think I've ever seen a plot like that.

I have also built an alternative as a stacked bar chart https://observablehq.com/@mjbo/bank-failures-as-a-stacked-ba...

The circle version is terrifying. The stacked bar chart is even more terrifying.

Right, it makes me wonder whether we are really good at looking at areas or not.

You ploted the size on a log scale and claim "the recent failures dont look quite so crazy anymore?" Is this a joke?

Log 10, and the scale doesn't start st zero as if to compensate.

What i find with graphs is that they scales and ranges are often chosen to look "alright" and always to communicate an intended message.

Aside those datasets anyway what's truely going now is most likely under the surface still. RemindMe! In a year.

When I posted this the only change I'd made was to sqrt the assets (which I now see was wrong). The lollipop chart came later.

> we judge 2D shapes by their area

I've seen this claim a few times in discussions about such charts, but it's far from obvious to me that this is really true. My best guess based on my own perception is that we estimate quantities represented by circles as proportional to something between the diameter and the area.

Wouldn't that be even worse for interpretability?

Well is the unit is the diameter, then why add extra dimensionality at all then?

That’s literally what bar charts are for.

I can feel a Tufte monologue coming on, even.

Human beings are spectacularly shit at comparing areas. If you put two circles next to each other, one with double area of the other, most humans will say their relation is 3:2.

FWIW, I think the size of the insolvency is more telling of the failure than the AUM. E.g. First Republic was about ~$13B in the hole when it was taken over. Otherwise a large bank with a large hole looks way worse than a small bank with a huge hole.

They still look pretty crazy to me! https://www.fdic.gov/bank/historical/bank/index.html

Total assets of banks failed in 2008 + 2009 => 373 + 171 = 544 Billion

Total assets of banks failed in 2023 (so far, still 2/3 to go) 548 Billion

The total bank assets doubled in the meantime (from 11-12TN to 23TN)[1] so proportionally it's half than 2008 + 2009, but that's still gigantic.

[1] https://fred.stlouisfed.org/series/TLAACBW027SBOG via https://nitter.lacontrevoie.fr/2Steady4U/status/165337475754...

That's a log scale...

I would also have a 2nd plot that has all banks / banking / bank assets with the failures in a different color, so we could identify the relative size of the failed banks, and also see the actual size of the big players in banking.

This uses a log y axis. The circles are actually better for visualizing.

Thanks for that!

Reply on news.ycombinator.com