I'd be interested in seeing that paper. My guess is that they are talking about looking at individual cause errors as compared to the total number of causes for fires. In other words it is OK if one investigator mistakenly called a fire as being ignited by a cigarette, when in fact it was an intentional act because another investigator called a cigarette ignition an arson. This statistical averaging accounts for arguments that the numbers of PARTICULAR causes are either under or over reported. The individual cause error rates may "cancel out" because (the argument goes) the 20 fires that were called cigarette ignitions that were actually arsons are offset by the 20 fires that were wrongly coded as arsons that should have been cigarette ignitions. Clearly the problem with this argument is that it is intuitive or anecdotal and not based on real studies. It may be impossible to develop real statistics on this because investigators would have to go back, re-evaluate their calls and change them in order to correct errors. In real fires, it has to be the original investigator who admits an error. If another investigator comes in and says that I was wrong and the call should have been accidental, isn't it possible that they are wrong and it was, in fact, an arson? (This example might look familiar to some readers of this forum... or those who've been to court!) Even if I change my call... maybe I was wrong about being wrong. (I think you see the problem.)

An additional problem with this type of statistical analysis is that it really doesn't account for the impact "wrong" individual scene calls have in the real world. Statistically, it might not matter that a few accidental fires are mislabeled as arsons because they will be offset by "wrong" accidental calls. However, in the real world, real accidents that are called arsons might result in real people being charged with someone's mistake or a real arsonist may go unpunished and remain free to place more innocent people at risk. While, intuitively, I agree that the overall causes probably average out and there's not much sense in arguing the error rates of individual calls, it doesn't mean we should be cavalier about the impact the errors may have in either financial or human terms.

Steve