Five years ago, President Bush called out Iraqi insurgents, telling them to "Bring 'em on!" and they did.
The only thing that has changed the rates of US fatalities seems to be a cease-fire in recent months by the forces of Moqtada al-Sadr. Evidence suggests that he is using this time to rebuild his organization, and to build ties to Iran.
Since Bush invited anti-occupation forces to "Bring 'em on!," nearly 4,000 American soldiers have died in Iraq, with roughly 29,000 wounded. The number of Iraqis killed directly by the fighting, or indirectly through stress, disease, and weakened infrastructure is hard to say, but is probably over three quarters of a million by now.
Uh, is there any meaningful reason to use a linear regression here?
Not really, other than making it clear that the trend is nonlinear. I show it for convenience, but spend more effort on getting the nonlinear loess regressions right.
Now if you had the standard deviations as well as the means of the monthly figures, we would be able to do the usual SPC (statistical process control) charts to see if the process is stable or has changed in any way.
Sadly, a job only half done, Josh :-(
These are exact counts, not samples. No standard deviations.
Really random question I know, but regarding U.S. military casualties, has anyone looked at any hypothetical increase in mortality rate for soldiers who have returned home, and attempted to factor that in to counts?
these are averages per day over the month. That's what the scales say. That's why the Y-numbers are not integers, OK?
If you have the exact integer counts each day, then we can calculate the standard deviations over the month, or even per week.
Given those data, we can then do statistical process control to see if the process has changed. Clear now?
I see what you're saying, OPS. In my mind, averaging is done not so much to create a statistic of daily mortality, but to correct monthly figures (exact counts, measured without error) to a constant time frame. Months vary in length, so using raw monthly data introduces error.
I suppose you're looking at it as a Poisson process on a daily level that adds up to the monthly figures. And then we could use SPC to see whether lambda of the process changes. I'll dig around on that.
Martin: I'm not aware of a dataset of that sort, but it would be interesting to look at. It'd surely be a longer fuse for problems to show up once soldiers get home, but suicide is a growing problem in the field, so it would be worth looking at suicide rates among soldiers after they come home.
Yup. Just using the averaged monthly figures in your graph SPC tells us that the Sadr cease-fire changed the process,
whereas the surge didn't. With daily data we could do a Fourier transform too (spectral analysis) to see
a) how many processes there are
b) if there are periodic effects to look for (e.g. more or less killings on a friday, which might have religious grounds)
Of course the best way for the US to reduce their death toll would be to withdraw their troops (:wry grin:)
Your anti-Bush friend,
Ole Phat Stu