Comments on: The Visual Display of Quantitative Information

By: mq

mq — Tue, 11 Mar 2008 23:20:44 +0000

It’s a telling graph. Could you just quickly cite the sources (possibly adding them right into the post?).

I’ll bet it’s from here:

http://www.cbo.gov/ftpdocs/88xx/doc8885/Appendix_tables_toc.xls

By: The Ambrosini Critique » Blog Archive » Beautiful evidence

The Ambrosini Critique » Blog Archive » Beautiful evidence — Tue, 11 Mar 2008 10:57:42 +0000

[...] UPDATE 2: I swear to all that is holy that I titled this post before reading the post over at CT. [...]

By: Sortition

Sortition — Tue, 11 Mar 2008 03:30:18 +0000

I find this graph more informative. Everybody knows that some people are extremely rich, but it is surprising to find that 1% of the families control about a fifth of the total income and 10% control about half of it. Thus not only are some people very rich, they are so rich as to make the rest significantly poorer.

By: Barry

Barry — Tue, 11 Mar 2008 02:10:50 +0000

19 comments, and note one jsutifying the situation, or denying it. A new record!

By: John Emerson

John Emerson — Mon, 10 Mar 2008 22:34:56 +0000

I disagree, this graph is virtually unreadable. I see one line going up but cannot tell at all what is happening to the other two lines.

Hint: they’re not going up much, if at all. The top line more than triples and almost quadruples. If the bottom two lines did that, you’d easily be able to tell. One goes from slightly under $50,000 to slightly over $50,000. One stays significantly below $25,000. Significant change in either number (more than about +/- 20%, which would be $45,000 to $54,000) would have been clearly visible. There was no such change. That’s what the graph was designed to show.

If you can’t read the graph you probably you should major in art history or aromatherapy.

By: smaug

smaug — Mon, 10 Mar 2008 19:20:46 +0000

Re my 17, the “it” is the log-scale.

By: smaug

smaug — Mon, 10 Mar 2008 19:19:06 +0000

I think an index is better, largely on legibility grounds. The overall effect is the same, however, in terms of the gap. The magnitude difference between incomes and between initial positions is just not high enough to make it more informative rather than misleading.

By: notsneaky

notsneaky — Mon, 10 Mar 2008 18:57:04 +0000

And he’s right that the Gini measure of inequality is pretty much useless when talking to non-inequality-studying folks. For one, there’s some problems with it that you always need to be aware of. Second, how much of a Gini in exactly “a lot”? If it changes from, say, .45 to .49 is inequality skyrocketing or could that just be due to measurement error or something?
On the other hand, using quintile shares and the like has its own problems.

By: notsneaky

notsneaky — Mon, 10 Mar 2008 18:53:24 +0000

Of course it depends on who your target audience is (just like writing say a Lit paper). If it’s “professionals” who are down with log scale, use log scale. If it’s the general public which has trouble with that sort of thing then use the absolute values and maybe state in words what’s going at the bottom. Or you can go completely nutzoid and have not one, but TWO, yes TWO, graphs.

By: mpowell

mpowell — Mon, 10 Mar 2008 18:18:41 +0000

If one believes in log-utility a logarithmic scale should be used for the y-axis imho.

That would give you an idea of the utility growth. Then we could sum up the different parts by weight, and I’m sure we’d see a decrease in net utility over the time period. But the point is to show the vast difference in linear allocation, b/c if you could swap these amounts efficiently, they would swap linearly. The only problem is that we can’t see much regarding the relative changes in those other groups.

I looked into this issue quite a bit recently. It seemed to me that for each percentile in the US economy, you could talk about their growth (or loss) in their share of US income. This growth seems to have been a mostly monotonically increasing function of the percentile. It seems, depending on your time period of measure, the zero-crossing is somewhere in the 90-95 percentile range. This is pretty astonishing and also convenient. Since I fall approx in that range, I can be very bothered about this issue w/o having to feel guilty that I benefited from an unfair growth in my income share. Or in other words, there is perfectly good reason for me to advocate a tax increase on those making more than me. (though I am willing to pay more taxes as well, if it comes to that)

By: David Estlund

David Estlund — Mon, 10 Mar 2008 17:01:03 +0000

It’s a telling graph. Could you just quickly cite the sources (possibly adding them right into the post?). Not that I doubt them. They’d just help when using the graph to argue with Republicans.

By: Andy

Andy — Mon, 10 Mar 2008 14:47:54 +0000

I disagree, this graph is virtually unreadable. I see one line going up but cannot tell at all what is happening to the other two lines. It looks like the 40-60 group is going up, but who knows by how much. And who can tell about the bottom series. Just terrible presentation.

You don’t need a log-scale to fix this. Just graph it in %-changes year-to-year or scale all series to 100 in 1980 so you can see differential trends clearly. Another possibility is the ratios of these three series over time (which is a standard presentation in the inequality literature).

By: wolfgang

wolfgang — Mon, 10 Mar 2008 13:42:55 +0000

If one believes in log-utility a logarithmic scale should be used for the y-axis imho.

By: Barry

Barry — Mon, 10 Mar 2008 13:26:47 +0000

Somebody posted a very tall, skinny graph, so that they could use a linear scale, and show all quartiles at the same time (with the top quartile broken down into pieces).

I wish that I had saved that graph.

By: Matthew

Matthew — Mon, 10 Mar 2008 13:23:17 +0000

Only if you believe its the proportional difference that counts, not the absolute difference. Normally we talk about the former, but I can see a case for the latter.