I have compiled, graphed, and posted the statistics about the mutinies provided by Denis Rolland in La grève des tranchées (Paris: Imago, 2004). You can see them here.

First is a simple graph showing Rolland's count of the mutinies, sorted into three day bins. Second is a graph showing both Rolland's count and that from SHD 16 N 1525, which allows for a comparison of their respective counts. Third is a graph showing the difference between the two counts, with positive y-values indicating that Rolland's count was higher and negative y-values indicating that SHD's count was higher.

While some differences exist between the two, the average difference for the three day periods provided comes to 1.66, which is not very large. The difference is larger--3.06--for the three day periods covering the most active period (15 May to 1 July). Nevertheless, these two averages suggest that, while Rolland and SHD differ in their counting, this difference is relatively small. The two sets of figures largely agree on the general trend over time. Their similarities are far larger than their differences, and, within reason, either one probably provides a reliable overview of the timeline.

I have also uploaded a digitization of Rolland's statistics, as well as several of my own analyses thereof, here. I plan to go into more detail about this data set later, but for now I will make two points:

The 95% confidence interval for the total number of active mutineers between 29 April 1917 and 1 January 1918 is [60096, 80141]. I found this by taking the average size of a mutiny as reported by Rolland as my x-bar, the standard deviation of these sizes, and a total sample size of n=159. Additionally, the 85% interval is [63803, 76434]; 90% is [62130, 77928]; 99% is [55944,84294]. I calculated these other intervals in order to see if they would differ substantially from the 95% interval. They do not--the difference between the low-end 85% and 99% interval is 12.3% and the high-end 10.2%. The estimates provided by the 95% interval are surprisingly high, much higher than what Mariot called the "usual" estimate of 30,000 to 40,000 men (Nicolas Mariot, "Pour compter des mutins, faut-il soustraire des moutons," in Obéir/désobeir, 346).
Indeed, the interval is about double what Rolland himself concluded about the total number of mutineers. Granted, he did not use a confidence interval in his estimation. This discrepancy can be explained in one of four ways:
1. Rolland's estimate is off by about 100% (this is unlikely, but is what the statistics themselves suggest)
2. The mutinies whose sizes are known were particularly large mutinies, which may push the confidence interval to the right (this is possible, but size estimates were provided for 30% of the 159 mutinies, which should be enough for a representative sample)
3. The size estimates provided by officers' reports are uniformly exaggerated (this is likely, but I am not sure if it can explain such a large discrepancy)
4. Some combination of the above

What I plan to do now is complete whatever gaps I can in Rolland's numbers and run similar operations on those provided by other scholars. I will also double check all my math to make sure that I myself have not made a silly error or series thereof. The goal is to provide the most accurate estimate possible as to the total number of mutineers in 1917.

Of course, the outcome depends on the source material, which has many of its own problems. GIGO: if the officers' estimates contained in the archives are themselves garbage--and there are many compelling arguments that, at least to some degree, they are--then the result will be garbage as well. A later post will address some of the shortcomings of the source material.