First of all, the number 100,000 is really the number 98,000. Minor point, really, but since we’re talking numbers let’s be accurate. Don’t exaggerate.

AND (and this is the important one)

The number 98,000 is really a number 95% likely to be between 8,000 and 194,000.

In the arcane world of mathematical statistics, this is what’s called a B.F.S., or “Big Spread”.

Although not a statician in any sense, I do believe that results with a 95% accuracy rating are considered not all that terribly accurate in the statistics world. Someone straighten me out if I’m off base here.

Nukevet on Random Nuclear Strikes commented about this number on a post that was originally about the Marines around Ramadi:

The Lancet study has been widely slammed as pseudoscience by a large group of epidemiologists – the study is flawed, the conclusions bogus, and completely unsupported by any semblance of facts.

I respectfully point out that calling this study pseudoscience is a direct insult to real pseudoscientists everywhere.

But some either don’t understand or don’t care.

95% accuracy is two standard deviations. This is a pretty typical measure of error in statistics, espcially for inherently hard to measure things like Iraqi casualties. For easier to measure quantities (like failure rates of industrial processes) companies target much lower errors: famously, TQM (‘Total Quality Management’) targets an error rate less than six standard deviations (literally, one in a million). Also, I disagree with your point that you have to use 98,000 instead of 100,000. Even if the standard deviation was more reasonable, you really should be rounding a number like that. Saying 98,000 instead of 100,000 actually increases the perception that its an exact number, when it CLEARLY isn’t. The only reason to use 98,000 is the same reason retailers charge you $19.99 instead of $20.00 — it looks like like a much bigger difference than it is. Frankly, considering the error range, its not even accurate to say ‘around 100,000 casualties,’ which would be the preferred wording if the error rate was something reasonable like plus/minus 20%. I know plus/minus 20% might sound big, but it would be hard to reduce the error rate below that, given the difficulty of collecting accurate data, and 20% would be a narrow enough error rate that any number within that range ‘feels’ the same politically. What I mean by that is that the headlines and rhetoric associated with 80,000 Iraqi casualties are not that much different than associated with 100,000 Iraqi casualties. As it stands, though, there is a huge difference between 8,000 Iraqi casualties, which would be a great outcome (all things considered), and almost 200,000 casualties, which would be a terrible outcome. Bottom line: The error range proves the study is useless, but not because they targeted a 95% range, but because the range is too wide.

Don’t use 98k instead of 100k? 2000 dead people (or not) seems significant to me. This isn’t a $19.99 shirt we’re talking about. It’s lives (or lack thereof), so even though I know I’m splitting hairs I’ll ask those trying to use statistics to prove a point to use the actual results. At least to the closest thousand. Yes, I realize that the 95% is less of an inaccuracy than the 8k-194k range. I thought I had addressed point that by invoking the B.F.S. theorem, but maybe not. My point was that basically nothing in the report seems to hold any amount of water, at least not in a sense that it’s valuable to discussing justification (or lack of such) for the war.

Its not about whether 2,000 people are important. Its about whether its any more accurate to say 98,000 or 100,000. Using the 98,000 implies a level of confidence in the accuracy of the number that we all clearly believe should not be there. (Basically you are insisting we round to the nearest 1,000 and I’m saying why not the nearest 10,000.) I was further trying to make the point (maybe not that well) that in this case the error range is so wide that its probably even misleading to quote any single number (whether 98,000 or 100,000). I only addressed the 95% thing because you asked if 95% was too low of an accuracy threshold. It’s in fact the standard accuracy threshold for determining the range of error. The problem is that its not standard in statistics to come up with an error range of plus/minus 90%. Such a range is completely useless. Let’s not lose sight of the fact we are agreeing that the study is completely bogus. As far as justification for or against the war, I don’t think Lancet’s numbers, even with better accuracy, provide any insight into that question. If you were against the war, any causalties were too many. If you were for the war, casualties are the cost of freedom and security (at least within the ranges we’re talking about — its potentially a different discussion if we’d reached the levels of indigenous casualties that we had in Vietnam, but we’re no where near that).

Chuck: Right on. The study is totally bogus, and the 98,000 remark was just directed at people who might be trying to use these numbers to prove the invasion was wrong. You hit the nail on the head when you say that no single number represents the study’s findings. There isn’t enough information in this study, what I’ve read anyway, to mean anything. I must admit that I’m more than a little pleased to see that US mainstream media didn’t pick up on this and splash the numbers all over. I was expecting it to, and I’m glad I was wrong.

I’ve never been able to figure out why the media picks up on some stuff and not others. You’d like to think they are dismissing bad science in this case, but I’m not sure history shows the media is that smart.