Stupid Beyond Belief

i-c1c9665d7dd68924c47beade3c7d1e09-pinata.jpg

Latest Tim Blair attempt to refute the Lancet study:

Lancet's number of documented deaths in Iraq, upon which the respected medical journal based its Iraqi mortality study, is but a mere 0.0835% of Lancet's estimated post-invasion death total.

The "estimate" part of Lancet's equation is 99.9%.

Well, I guess that's it for the entire field of statistics.

More like this

You think that's dumb? Try reading the comments -- only one of the first 17 commenters even noticed.
Here's an exercise for the masochists among us:

Spend a few minutes reading the comments on the same subject over at Crooked Timber, until you're adrift in a warm sea of confidence that your fellow humans are a thoughtful lot --then, real quick, go over to Blair's site and read the comments there.

YEEEEAGGGH!

It's like plunging your head into a bucket of ice water.

Perhaps someone should make a list of important and widely accepted statistics that rely on cluster sampling?

Bad Jim, bad!

Bad Jim - making me read Blair's comments like that! I need a chemical bath now.

Best,

D

The complaint isn't about cluster sampling, but about sampling per se. It's like how those Gallup guys pretend to predict election outcomes, when they've only interviewed 0.01 per cent of the US population.

By John Quiggin (not verified) on 12 Oct 2006 #permalink

Yeah. If they'd interview the 0.01 percent who own and control the voting machinery, they'd have much more reliable results.

Oh, wait ...

And a random sample that included the Tims Lambert and Blair would conclude that 10,000,000 Australians are innumerate. Fortunately, statistics does not work like that.

(I do like that Blair's site requires registration, so it includes an inherent sampling bias with regard to the quality of its commenters.)

Yeah imagine how bad the comments would be without the small intelligence test of registration!

"I do like that Blair's site requires registration, so it includes an inherent sampling bias with regard to the quality of its commenters."

Since when does registration ensure any kind of quality?

Millions of people registered to vote in the last US Presidential election but nonetheless voted for George Bush.

Why the Lancet study, when little effort would be needed to compile total REPORTED deaths from all available press and other media reports? The current average daily rate in Baghdad, the highest ever, of around 100, yields aound 150,000 over 4 years; the slaughter rate is less in most other cities, especially those in the north and south. However if it was the Lancet's 600,000++, unless diminishing returns to the killers (almost all Iraqis) set in, since most victims are men, within 10 years the problem will have been solved by removal of almost the total adult male population to all those virgins in the sky. What is the Lancet's projection of the adult male population in 2026 based on its cluster sample? I think we should be told the Lancet's experts' projection - but here's a rough guide: the adult male population of Iraq in 1999 was c.5 million, so in 8 years there will be hardly any male bombers and executioners left, and the sacred duty of eliminating all Shia/Sunni will fall to their respective women. How about "secticide" to describe what is happening in Iraq?

By tam o'shanter (not verified) on 12 Oct 2006 #permalink

Beyond the obvious problems of the insurgency and civil strife, there might be another factor in the disproportionate military aged male death rate.

By djotefsoup (not verified) on 12 Oct 2006 #permalink

Tam, that's one of the most innumerate "criticisms" yet. Congratulations.

The surveyed population had roughly 730 male births in the study period, of 13.5 male births per thousand per annum. There were approximately 330 male fatalities under the age of 75 in the same group, or 6.1 male deaths per thousand per annum.

Do the math. Tell me how long at the going rates before Iraq runs out of fighting-age men.

llewelly,

Almost any survey where interviewers make personal visits uses cluster sampling. It's just too expensive to go traipsing all over the country. For example, the U.S. unemployment and poverty rates are based on a survey that samples clusters of 4 neighboring housing units, and in more rural areas only goes to a random sample of counties.

genocide

By ThinkTank (not verified) on 12 Oct 2006 #permalink

Just a note that Tam is a sock for Tim Curtin.

By John Quiggin (not verified) on 12 Oct 2006 #permalink

I do like that Blair's site requires registration, so it includes an inherent sampling bias with regard to the quality of its commenters.

Since when does registration ensure any kind of quality?

It ensures two qualities:

(0) It ensures the registrant is sufficiently interested in commenting to register.

(1) It ensures the registrant's previous comments have not caused Tim Blair to revoke said registration.

Both qualities may bias comments towards Tim Blair's opinions. Perhaps that is what ahem meant.

Bruce R: on your figures, Iraq despite everything has the highest population growth rate in the world. My projection was based on the growth rate in the Lancet's death estimates over 2 years.

By Tam o'Shanter (not verified) on 12 Oct 2006 #permalink

llewelly,
Almost any survey where interviewers make personal visits uses cluster sampling. It's just too expensive to go traipsing all over the country

Thanks, exactly what I was expecting - a list of reputable surveys depending on cluster sampling would be large, and people with many different views would find some surveys they agreed with in the list.

BruceR "Do the math. Tell me how long at the going rates before Iraq runs out of fighting-age men.'

Really, Bruce, don't you know that maths are just another left-wing conspiracy?

By Ian Gould (not verified) on 12 Oct 2006 #permalink

I once tried to register on Tim Blair's site to counter some particularly nasty and inaccurate stuff on France's record in various wars over the past couple of centuries
- you can imagine what people were saying - but the system didn't work - perhaps its because I actually live in France.

"Well, I guess that's it for the entire field of statistics."

Well, it would be if all statistics were extrapolations of very small and very suspect datasets...but, of course, that isn't the case.

"Stupid beyond belief"

The appearance of being so can be avoided by thinking before one types.

By Dave Surls (not verified) on 13 Oct 2006 #permalink

Just another example of cluster sampling that many people trust - try the TV ratings. A few thousand households are used to work out exactly how many people watch television programs. And people believe those cluster samples enough to hand over millions of dollars.

Here is a post where Tim Blair quotes a Newspoll opinion poll:

http://timblair.net/ee/index.php/weblog/comments/howard_up_labor_down/

Newspoll opinion polls are typically obtained by door-knocking or telephone calls, and political polls are generally conducted by asking 500-1000 people their opinion. By survey weighting methods quite similar to those in the Lancet study (and drawn from exactly the same realm of statistical theory) they make an estimate of the national vote. If one assumes 8 million voters (probably 4 million too few) and 1000 respondents, then according to Tim Blair's logic, this very newspoll he quotes would be 0.0125% data, and 99.9875% "estimate".

Here's another post:

http://timblair.net/ee/index.php/weblog/comments/vote_leaves_home/

oh look, he's quoting statistics based entirely on "extrapolation" as if they were fact. Shame on you Tim, how can you use such left-wing rubbish to support your view?

And isn't it funny how often those polls turn out to be right on the money on election day ... it's as if the statistics worked or something.

According to the Electoral commission there are about 13 million registered voters in Australia. Roy Morgan polls on Voting intentions usually include about 2,000 people, but NewsPoll are apparently usually 1,000 to 1200. So 1200 out of 13 million is about 0.009%

So Tim has been fighting the good fight for Statistics! Basing political opinion on the results of surveys where almost none of the numbers came from actual people! And he claims the Lancet is biassed!

Tim (or Tam):

"Why the Lancet study, when little effort would be needed to compile total REPORTED deaths from all available press and other media reports?"

IIRC that the Iraq Body Count Project is compiling a subset of that. Considering that they need to collect media reports to meet their criteria for an 'official' death, they should be able to produce a list of all reported deaths.

As for your question, consider the word that you capitalized - 'REPORTED'. Consider the difficulties of covering Iraq. Consider the high background level of violence in Iraq - single deaths are rarely newsworthy; doubled-digit death batches are newsworthy but not headline news.

Tam, I see your point, but by the Lancet's figures the male adult death rate even in this latest worst year is still only about 10 per 1000 (+/-3), still under the 13 per 1000 male birth replacement rate.

Obviously the killing cannot go on at this rate forever, and there's also little doubt there's now effectively a "Lost Generation" effect in Iraq that's going to have demographic effects for decades. But, to take a counter-example, several American black ghettos have proven able to absorb male adult death rates above 5 per thousand for years without running out of young black men.

You know, if one adds 1 to 199,999 to get 200,000, half of the input is coming from only .0005% of the total. It's very suspicious. I await an explanation from Mr. Blair.

So this Tim Blair's version of science? Just imagine the impact if his kind of reasoning actually held sway.

Ooops - sorry, forgot for a moment.

llewelly was on the mark: to comment chez Blair, one must care enough to register, feel confident enough to supply details to the site, knowing the track record of right-wing sites, and survive the knee-jerk censorship of Blair and Andrea Harris.