[stormer]

Given a normal distribution (bell curve), 67% of the sample will fall within 1 standard deviation of the mean, and 97% of the sample will fall within 2 standard deviations of the mean (if I recall). By comparing 2 sets of samples (e.g. 2 different locations, or two different date ranges), one can determine to what degree changes in the second can be explained by changes in the first, and the level of confidence that can be placed in that determination (you’re taking me back here). The Pearson Correlation (R-value) reflects that level. With 10 degrees of freedom (we have 11), the critical value of R needs to be < .576, assuming a P-value of .05 (very strong correlation); in comparing Aug 05 - Jul 06 with Aug 06 - Jul 07, the R-value is .218, indicating that variations between the two years in the number of deaths is statistically insignificant. In other words, there is no down trend other than the seasonally expected one. Whew - does that make sense?

[Enchante]

There are two aspects that are not accounted for in any comparision of year-to-year troop fatalities:

1) we have 20% more troops in theater

2) by all accounts many more of the units are doing more aggressive missions within the ‘surge’ which could easily yield more US casualties unless/until the terrorists are pretty well quashed.

Thus, what cannot be quantified (it seems to me) is that with substantially more troops performing a much larger number of dangerous missions now, US casualties could easily have gone up a lot. Even to stay in the same ballpark, statistically, may prove to have been a remarkable positive accomplishment given the missions the troops are currently performing.

The real test will be whether the new counter-insurgency tactics prove to work, as many reports indicate they will. IF in 6 months’ time the ‘insurgency’ turns out to be much diminished, then this summer’s efforts will have accomplished a great deal at less than the casualties that might have been expected.