Those are pretty much all valid points, however at first glance a drop from 57 incidents to 17 is at least statistically significant. The standard deviation can

Those are pretty much all valid points, however at first glance a drop from 57 incidents to 17 is at least statistically significant. The standard deviation can be assumed to be at most around sqrt(57) = ~7.5, so a difference of 40, which would be over 5 standard deviations, shouldn't normally happen. You're right that relative percentages are sometimes a bit dubious (and even in this case you'd fluctuations on the order of 1/sqrt(57) = ~13 percentage points) but 57 is high enough that it actually has some meaning.