Karl Denninger: Bad Math, Bad Statistics: Trying to get a blogger to admit a mistake

2009-07-15

Sometimes someone on the internet is just so wrong that I become engulfed with rage and have to write about it.

In this case, an article was linked to on reddit about "Economic Fundamentals." The point of the article was that we are spending way too much on debt and credit. Which may be a perfectly valid fact, but it couldn't have been argued more poorly. His first argument that we are spending way too much on debt is the following:

In 1981, US GDP was $3.1 trillion dollars. In 1992 it was $6.3 trillion, a double. In 2005 it was $12.4 trillion dollars, another double. Doubling in roughly 12-13 years. Not bad, right?

Let's look at it a different way, this time in "current" (not inflation-adjusted, since GDP isn't) dollars. In 1981 the per-capita income in the US was $8,476. In 1992 it was $14,847, a 75% gain. In 2005 it was $25,036, a 69% gain.

Notice anything? Its not really that subtle, is it? GDP slightly more than doubled in each of those above periods, but per-capita income lagged, and the lag rate is increasing.

So what's wrong with this picture? If you guessed, "he's comparing a gross value with per capita value" you win a cookie. US population is increasing all the time, and therefore, even if per capita incomes have dropped, that doesn't mean total income hasn't. So if you multiply those figures by the population and then compare them, you get this (source):

1981: 229465714 * 8476.0 = 1.944 trillion
1992: 255029699 * 14847.0 = 3.786 trillion (94% gain)
2005: 292892127 * 25036.0 = 7.332 trillion (93.6% gain)

Er, doesn't look like a lag to me. In fact, it looks like it's doubling every 12-13 years just as much as GDP is. I also looked up total income statistics for the US, and found the following figures (source). (Note these figures are different. More on that later.)

1981: $2,580,600,000 (2.58 / 3.1 = 83% of GDP)
1992: $5,349,384,000 (more than double!) (5.34 / 6.2 = 86% of GDP)
2005: $10,252,973,000 (another double!) (10.25 / 12.4 = 82% of GDP)

Er, looks like the same thing. So if his thesis is correct, it certainly doesn't rest on this. I emailed him twice about this fact. You think he'd be happy to correct this unfortunate error of math. But no, instead he told me to "READ IT AGAIN":

GO BACK AND READ IT AGAIN. The issue isn't GDP-per-capita - IT IS PERSONAL EXPENSES IN EXCESS OF PERSONAL INCOME, BOTH PER CAPITA. Both of those series are from the BLS - SAME DATA SOURCE. We went from spending $400 more than we make (less than 5%) to 16% more than we make ($4,000), both per-capita per year. How? DEBT. As I said: GO BACK AND READ IT AGAIN.

First problem here is that the listed sources on his blog weren't the BLS: it was the Census Bureau. Anyway, that seemed besides the point. So I went back and read the rest of the article. In it, his next claim is the following:

In 1981 we spent $1.941 trillion, or $8,600 per capita, about $200 more than we made. That is, we barely spent more than we made - this was the start of the "gluttony of credit."

But in 2005 we spent $8.694 trillion, or $29,000 per capita, or about $4,000 per person more than we made, or a spending deficit of some sixteen percent!

Looks damning. But I checked his source for the spending amounts, and he uses the BEA. Here's where those different results for personal income come into play: I realized that statistics groups have different ways of measuring personal income. So I checked the BEA's personal income figures so they would be consistent with the personal expenditure figures. Here's the result (source):

(Percentages are spending divided by income)
1981: 11,266 (spent "about $8,600", actual spending $8,439, 74%)
1992: 20,870 (you didn't calculate this one, actual spending $16,485, 78%)
2005: 34,691 (spent "$29,000 per capita", actual spending $29,368, 84%)

So in every case, the expenditures are less than the income. The percentage is increasing, which is definitely cause for concern. But it's nowhere near the debt spending the article claims. So I emailed him about this, and he responded that the BEA make some kind of adjustment to their income numbers:

Because the BEA has ADJUSTMENTS they make to income. Care to try to back 'em out? I don't know if you CAN - this, by the way, is a big part of where the problem with their "savings rate" comes from - its false, because it counts paying down debt as "savings".

My response:

Source, please. I looked at the BEA's methodologies and see nothing about "adjustments" made to income. I don't see it in the table. And the numbers I got weren't even adjusted for inflation. Perhaps I'm ignorant of some government trickery to falsify the income numbers, but getting two different numbers from two different sources who may very well use VASTLY DIFFERENT methodologies is just bad science, bad math, and even bad statistics.

He responds:

Now you've gone from a discussion to ad-hominen attacks. I have twice asked you to re-read the original article. You keep coming back with canards; the entire issue is the efficiency of the economy - that is, what is the "spread" between spending and income, and what is the "spread" between GDP and income. I've presented my case and have no more time for someone who wishes to argue that the RATIOS aren't the important issue when that's exactly how the original argument was framed.

Now, I did say that his math was bad. I did say that his article was flawed. But people really need to learn what "ad hominem" actually means instead of just throwing it out there every time they feel threatened. Also, I'm pretty sure I never sent him any canards and my complaints were legitimate. This is the part that gets me frustrated enough to type this blog out.

In any case, what's this "spread" he's referring to? Well, he wrote a follow-up article in which he talked about the GDP to Personal Income "spread":

In numbers, in 1981 per-capita GDP was $13,600 while per-capita income was $8,476, a "spread" of 60%. In 2005 per-capita GDP was $42,200 while per-capita income was $25,036, a "spread" of 68%.

Lower "spreads" denote a greater return of GDP into the hands of people - that is, a more efficient economy. The wider the spread the more "parasitic drains" there are on GDP - that is, the greater the amount of GDP that winds up somewhere other than in the people's hands.

But here's the problem: he is still using Census values. The Census web site itself says: "Income data obtained in household interviews are subject to various types of reporting errors which tend to produce an understatement of income."

So here's the numbers with the BEA's figures:
1981: PCI $11,266 / GDP $13,601 = 82.8%
2005: PCI $34,691 / GDP $41,961 = 82.6%
So the "spread" really only dropped by 0.2%. But let's look at 2007:
PCI $38,654 / GDP $45,670 = 84.6%

It's clear that this "spread" value is meaningless. Confronted with this fact, Karl replies:

You haven't proved anything Stephen. You have steadfastly refused to go to the point of the article, and this has consumed over an hour of my time today.

You know, I was at least hoping he'd make some kind of correction or at the very least acknowledge that his article has data in it that, at the very least, conflict with other sources. But this is the internet, where you never have to say you're wrong.

Karl eventually wrote back, saying:

No. [BEA includes] "records of disbursements to individuals by governmental agencies" If you don't understand why that can't be included for the series I was presenting then there is nothing more to discuss. If you DO understand why that can't be included then go ahead and tell me how you back it out - because you have to. You did the basic research ONLY AFTER I BASICALLY FORCED YOU TO but you didn't understand what you were looking at. Now cogitate on the "why" for a while and the light should come on as to WHY the BEA numbers can't be used for that particular piece of information. (See, progress, even if I had to drag you kicking and screaming to it!)

Of course, I responded that 1) the eventual "answer" had nothing to do with what he was talking about and 2) if if you're an Ayn Rander that things fiat currency is fake money, if you get a disbursement from the government and spend it, you're not spending outside your means, and therefore is perfectly acceptable to be included in a per capita income statistic, especially in an article like he wrote..

Am I wrong? I repeatedly asked him why he refused to accept the BEA figures, or what kind of "adjustments" they made, and he kept on saying that "If you read the report like I do every month, you'd know that you'd have to go through every year and adjust!!" He never articulated exactly what kind of thing might be unacceptable in the income figures. He said that they "made ADJUSTMENTS" but after I explicitly asked him, out of ignorance, not spite, what kind of adjustments they were since I couldn't find any mention of them on the BEA's site. If someone out there reads this and knows the answer, I would really like to know. Karl, on the other hand, refused to answer, saying he doesn't want to do my work for me.

Anyway, the point of the whole story is that I was silly and expected a blogger on the internet to recant their mistakes when I pointed them out. The best bloggers do, but I find that crazy bloggers and unwillingness to admit wrongdoing are strongly correlated. I strongly suspect that Karl won't write me back. Sometimes you can't win. EDIT: He did write me back; updates you've probably already read.

I also think that this is a symptom of our poor national discourse, especially in the world of blogging and the internet. Blogging is just throwing your opinions out there. As pro-science as the internet demographic is, it seems none of them really enjoy partaking in reasoned debate about politics. I suspect that maybe it's the case that some people just aren't good at reading or expressing their thoughts clearly enough to efficiently communicate with people on the internet without shouting in the most vague terms or repeating talking points ad infinitum. I am also worried about yet another Latin phrase being misused -- ad hominem doesn't mean someone insulted you.

(email conversation)