Sunday, December 14, 2003
Meatgrinder Metrics: A Statistical Analysis
There’s a very useful site on coalition forces casualty figures here
Unfortunately, I don't see very many journalists using it very well.
The media focuses overwhelmingly on the number of lives lost. It’s easy to see why—it tugs more heartstrings in Peoria. But if you’re going to accept coalition casualties as a metric for measuring the progress of the struggle for Iraq, then I believe it’s much more instructive to focus on the numbers of wounded.
Why? Because the sheer numbers of wounded dwarf the number of dead, any statistical analysis of wounded is going to provide a much smoother graph with a smaller margin for error. Further, when you focus exclusively on the number of soldiers killed, then your analysis excludes the vast majority of attacks on coalition forces.
Let’s look at the numbers.
The United States took 270 wounded in September, 433 wounded in October, 344 in November, and 119 through 11 days of December. I’ll count three more wounded I have personal knowledge of from the 12th of December (plus one KIA, and another dead in a separate possible suicide, from the site’s press releases), for a total of 122 wounded through 13 days of December.
You can see the table on the Website. I would adjust the December daily average figures to 8.7, based on the information available to me here.
To reduce increase the number of data points and reduce the standard deviation of daily figures, I would add the number of those killed and wounded from enemy action together, while excluding non-battle casualties, which don’t tell me much about the enemy.
That yields:
September Total casualties: 277 9.23/day
October: 466 15.03/day
November: 441 14.70/day
December: 128* 9.8/ day
So what can we learn from the numbers? The first thing you see is that from the point of view of total U.S. casualties, November was not the worst month since the President declared an end to major combat operations; that honor actually belongs to October.
Moreover, the November figures for those killed in action were skewed upwards by two statistical outliers: two downed helicopters—one in Fallujah killed 17, I believe, and another near Mosul killed, for a total of 23. Eliminate the outliers, and the average number of wounded/day drops below 14. The difference between 15.03 and 13.93 is not particularly significant, except that the trend continues into December, which thus far has only produced 9.8 casualties per day—the best day since September.
Again, though, December’s figures are skewed by two significant outlying data points: last week’s car bombing in Mosul (wounding 26), and Friday’s car bombing at the 82nd Airborne Division headquarters in Ar Ramadi, which wounded 14 and killed one. Subtract those two events, and the remaining guerrilla activity throughout the country caused 6.69 U.S. casualties per day so far in December—well under the September’s pace.
This is not to suggest that the casualties in these events ought to be dismissed as insignificant. They themselves reflect disturbing trends about the enemy’s ability to 1.) obtain and use surface-to-air missiles against U.S. aircraft, and 2.) Recruit people actually willing to blow themselves up.
Equally significantly, though, the figures suggest that a smaller and smaller number of insurgent cells may become responsible for a greater number of casualty-producing operations.
I’m also struck by the casualty figures’ seeming lack of correlation with Ramadan. If Ramadan were going to bring a grass-roots rash of religiously motivated attacks, then you would expect that November’s total casualties would be significantly greater than October’s. I’m not entirely surprised by it, though. Littlegreenfootballs.com links to a good statistical analysis of intifada casualties, and I could find no evidence of a Ramadan boost in terrorist activity there, either. Ramadan isn’t about killing, anyway.
Lastly, the overall drop off in casualties may well be weather-related. Night time temperatures are dropping into the 30s, which probably makes it tough for Joe Sheik’s-pack to motivate himself to get off the couch at night and pick up his remote control and go out and kill some infidels. Why not wait until spring, when it’s warmer?
Someone with more time than I have might be willing to go back through casualty figures and weather data and do a regression analysis of temperatures vs. guerrilla activity in Afghanistan and other areas. Anyone with an econometrics background ought to be able to do this.
I’m surprised I haven’t seen any respectable statistical analysis of casualty figures yet from the NY Times. Maybe they should get someone on their financial desk on the story.
I have no solid answers for why total U.S. casualties seem to have markedly increased in October. My own unstudied sense is that the number of total incidents seems to have remained constant, or even declined since July. One possibility: somewhere along the line, the insurgent changed his tactics. Direct fire engagements on U.S. troops are less common now. His emphasis has shifted to improvised explosive devices, and more recently, to more spectacular car bombings.
I haven’t seen anything like this 10th grade level of statistical analysis from the New York Times or other major media outlets yet, though. Maybe they should get their political and financial desks in on the story. Time, Inc., on the other hand, should probably contract it out to Morningstar.com.
*Note: The KIA count is updated through the 13th, so I’m assuming the Ar Ramadi fatality is included in this figure. Monthly wounded figures prior to September are apparently not available, and any aggregate would be highly skewed from the March and April casualty figures. I’m therefore excluding them for the purposes of this post.
Splash, out
Jason
Unfortunately, I don't see very many journalists using it very well.
The media focuses overwhelmingly on the number of lives lost. It’s easy to see why—it tugs more heartstrings in Peoria. But if you’re going to accept coalition casualties as a metric for measuring the progress of the struggle for Iraq, then I believe it’s much more instructive to focus on the numbers of wounded.
Why? Because the sheer numbers of wounded dwarf the number of dead, any statistical analysis of wounded is going to provide a much smoother graph with a smaller margin for error. Further, when you focus exclusively on the number of soldiers killed, then your analysis excludes the vast majority of attacks on coalition forces.
Let’s look at the numbers.
The United States took 270 wounded in September, 433 wounded in October, 344 in November, and 119 through 11 days of December. I’ll count three more wounded I have personal knowledge of from the 12th of December (plus one KIA, and another dead in a separate possible suicide, from the site’s press releases), for a total of 122 wounded through 13 days of December.
You can see the table on the Website. I would adjust the December daily average figures to 8.7, based on the information available to me here.
To reduce increase the number of data points and reduce the standard deviation of daily figures, I would add the number of those killed and wounded from enemy action together, while excluding non-battle casualties, which don’t tell me much about the enemy.
That yields:
September Total casualties: 277 9.23/day
October: 466 15.03/day
November: 441 14.70/day
December: 128* 9.8/ day
So what can we learn from the numbers? The first thing you see is that from the point of view of total U.S. casualties, November was not the worst month since the President declared an end to major combat operations; that honor actually belongs to October.
Moreover, the November figures for those killed in action were skewed upwards by two statistical outliers: two downed helicopters—one in Fallujah killed 17, I believe, and another near Mosul killed, for a total of 23. Eliminate the outliers, and the average number of wounded/day drops below 14. The difference between 15.03 and 13.93 is not particularly significant, except that the trend continues into December, which thus far has only produced 9.8 casualties per day—the best day since September.
Again, though, December’s figures are skewed by two significant outlying data points: last week’s car bombing in Mosul (wounding 26), and Friday’s car bombing at the 82nd Airborne Division headquarters in Ar Ramadi, which wounded 14 and killed one. Subtract those two events, and the remaining guerrilla activity throughout the country caused 6.69 U.S. casualties per day so far in December—well under the September’s pace.
This is not to suggest that the casualties in these events ought to be dismissed as insignificant. They themselves reflect disturbing trends about the enemy’s ability to 1.) obtain and use surface-to-air missiles against U.S. aircraft, and 2.) Recruit people actually willing to blow themselves up.
Equally significantly, though, the figures suggest that a smaller and smaller number of insurgent cells may become responsible for a greater number of casualty-producing operations.
I’m also struck by the casualty figures’ seeming lack of correlation with Ramadan. If Ramadan were going to bring a grass-roots rash of religiously motivated attacks, then you would expect that November’s total casualties would be significantly greater than October’s. I’m not entirely surprised by it, though. Littlegreenfootballs.com links to a good statistical analysis of intifada casualties, and I could find no evidence of a Ramadan boost in terrorist activity there, either. Ramadan isn’t about killing, anyway.
Lastly, the overall drop off in casualties may well be weather-related. Night time temperatures are dropping into the 30s, which probably makes it tough for Joe Sheik’s-pack to motivate himself to get off the couch at night and pick up his remote control and go out and kill some infidels. Why not wait until spring, when it’s warmer?
Someone with more time than I have might be willing to go back through casualty figures and weather data and do a regression analysis of temperatures vs. guerrilla activity in Afghanistan and other areas. Anyone with an econometrics background ought to be able to do this.
I’m surprised I haven’t seen any respectable statistical analysis of casualty figures yet from the NY Times. Maybe they should get someone on their financial desk on the story.
I have no solid answers for why total U.S. casualties seem to have markedly increased in October. My own unstudied sense is that the number of total incidents seems to have remained constant, or even declined since July. One possibility: somewhere along the line, the insurgent changed his tactics. Direct fire engagements on U.S. troops are less common now. His emphasis has shifted to improvised explosive devices, and more recently, to more spectacular car bombings.
I haven’t seen anything like this 10th grade level of statistical analysis from the New York Times or other major media outlets yet, though. Maybe they should get their political and financial desks in on the story. Time, Inc., on the other hand, should probably contract it out to Morningstar.com.
*Note: The KIA count is updated through the 13th, so I’m assuming the Ar Ramadi fatality is included in this figure. Monthly wounded figures prior to September are apparently not available, and any aggregate would be highly skewed from the March and April casualty figures. I’m therefore excluding them for the purposes of this post.
Splash, out
Jason
Comments:
Post a Comment
