The Visual Display of Quantitative Information
Nancy Pelosi’s office has produced the following chart to illustrate the immediate need for stimulus (h/t Andrew Sullivan):
Of course there are a couple of odd things about this. First, it shows absolute job numbers, rather than unemployment rate (that is, job losses per capita). This matters, because the U.S. labor force is a lot bigger now than in prior recessions. Second, it ignores the recession of 1981 – 1982, which was by far the most serious of recent recessions.
Here is what the chart looks like if you look at the unemployment rate and include the ’81 – 82 recession:
The unemployment rate is increasing at about the same pace as it has in every recession for the past 25 years (from very different starting points).
Why does this matter?
Well, I think that the second way of looking at this data tends to lead naturally to two important observations:
1. What seems to matter in getting to really bad job losses is the duration of the recession. So, speed in passing a stimulus bill is probably a lot less important than getting our countermeasures right. This is, of course, diametrically opposed to the natural conclusion one would reach in looking at the first chart.
2. The structural work on the economy is at least as important as how we deal with the recession. The stagflation of the 1970s meant that we started the ’81-82 recession at about the unemployment level that we have taken more than a year of recession to reach today. Doing something, anything, to stop the pain of the current recession, no matter what its structural effects on the economy, might seem practical, but it is not.
Jim,
I’m a big fan of your writing and thought, and being an analyst myself I very much appreciate your quantitative approach to things, but you’ve got to use something better than excel to make your charts—and I’d like to think that Tufte would back me up in this particular case ;-).
There are a number of packages out there—including one made by the company I work for. You can get an eval copy of it here: http://registration.spotfire.com/eval/default.asp, but another good option is Tableau Software (http://www.tableausoftware.com). Please have a look at these, just for the sake of beating back the nonsense that’s being spouted as gospel about this recession. In the same way that a well-articulated but incorrect argument often beats a poorly-worded but correct one, a pretty graph will often leave a more profound impression than a ugly one regardless of accuracy.
— Tim Wormus · Feb 9, 12:54 PM · #
Well, look, Mr. Manzi, not to dispute the expert up there (though I thought of amking a snide comment about the Excel, too), but, what you want to be able to use is R.
— Sanjay · Feb 9, 01:18 PM · #
Yeah, tell me about it. I blame Bill Gates for all my troubles.
BTW, I know spotfire from my professional life in software, and it’s a great product. Also, we integrate certain R routines into our product.
In short, it’s really pathetic that I’ve not gotten around to improving this.
— Jim Manzi · Feb 9, 01:48 PM · #
R is the bomb. (We have copies of the book that is the title of this post all around the office. I hadn’t encountered it until I got to the Bay Area.)
This graph too though. This could split either way – but I’m not sure if beginning unemployment rate is the right Y-axis. Unemployment can be defined exogenously with labor unionization rates and job search means available (monster.com) and monetary policy on the table among other things, so the relative change from the start point is what we should be looking at when comparing different time regimes.
— Rortybomb · Feb 9, 02:10 PM · #
Rortybomb:
Thanks for the link. I thought about displaying it that way (which is, of course, equally mathematically valid). I think this tends to obscure two things:
1. A given reduction in per capita jobs seems like the more intutive measurement of the job destruction of a recession, rather than what % increase in the rate this represents; and,
2. These different recession started at very different levels of joblessness, which I think is important when considering the relative importance, over time, of recssion-fighting vs. structural reform, and therefore I wouldn’t want to throw this information away if it can be conveniently and intuitively represented in the chart.
That said, as always what Y-Axis to use depends on what exct question we’re trying to answer.
— Jim Manzi · Feb 9, 02:35 PM · #
Decent-looking graphs (no gray background, no horizontal lines, more aesthetically-pleasing colors, etc.) can be made with Excel. Try right-clicking on the gray area and using “Format Plot Area…” and “Chart Options…” or right-clicking on a datapoint or curve and using “Format Data Series…” Also, drag the legend inside the graph so you don’t have that odd right margin of whitespace.
— Christopher Monnier · Feb 9, 03:28 PM · #
Jim: doesn’t your second point cut the opposite way of the way you think? If unemployment is structurally low and cyclically high, that would mean a focus on short-term is more important, while if it’s structurally high and we hit a cyclical high on top of that, you have to be careful that counter-cyclical measures don’t worsen the structural situation. Right? So Keynsian solutions weren’t appropriate to the recession of the early ’80s because that recession was itself the consequence of the first serious effort to tackle structural inflation. Or did I miss your meaning?
— Noah Millman · Feb 9, 03:33 PM · #
Christopher, I’ve no doubt that you can tweak those Excel graphs more and more and make ‘em look better. But acquiring the level of skill to use Excel to do and to display stats that much better — something better done with any of many programs — will just get you that much more of a sneer from real quantitative types. It’s like people who’ve mastered Word’s ugly Equation Editor when they should just get over it and use LaTeX already.
— Sanjay · Feb 9, 04:06 PM · #
see this graph too
http://www.calculatedriskblog.com/2009/02/job-losses-during-recessons.html
and i vote for R!
— razib · Feb 9, 05:55 PM · #
Wow! People will really jump on any old chance to feel smarter than Jim Manzi. Obviously, we need to
spamadvertise The American Scene on the r-help list.That said, I too used R all afternoon.
— Matt Frost · Feb 9, 07:43 PM · #
Noah:
I take your point. What I was trying to get at is that I think poor macroeconomic decisions in the 15 years or so leading up to the 1981 recession made the economy less productive, with higher structural unemployment, than it would have been if it had been managed differently. By extension, we should be careful (in my view) not to start down a road that helps ameliorate the short-term pain, but at the cost of moving to structurally higher unemployment for decades.
— Jim Manzi · Feb 9, 10:20 PM · #
It looks like the current 15 month old recession consists of first 5 months of no unemployment increase, then five months of average unemployment increase, then 5 months of rapid unemployment rate growth. That’s not good …
— Steve Sailer · Feb 10, 03:31 AM · #
Along these lines, here’s a thread on Crooked Timber in which Keynesianism and R collide, complete with guest appearance by the indespensable Hadley Wickham. Awesome.
— Matt Frost · Feb 10, 09:47 AM · #
Let’s try that again. How will Obama and his team calculate the “created or saved” number? I suspect it will be by ascribing any slowing in the unemployment rate to the stimulus and then drawing two trendlines (w/stimulus and w/o) and subtracting the different. Poof! Jobs saved!
— Klug · Feb 10, 01:04 PM · #
They will say that the CBO told them 4 million, and that whatever number turns up is therefore 4 million better than what the GOP would have delivered. It will be goofy, sure, but that’s what they will say.
— J Mann · Feb 10, 03:53 PM · #