The obvious solution is to use probabilities rather than absolute judgements of true/false. Although we still have the issue that in general the average of two products is different from the product of two averages. This inconsistency is much smaller though, and can be dealt with a more nuanced calculation (accounting for the possibly correlated distributions behind the point estimates) if absolutely necessary.
The obvious solution is to use probabilities rather than absolute judgements of true/false. Although we still have the issue that in general the average of two products is different from the product of two averages. This inconsistency is much smaller though, and can be dealt with a more nuanced calculation (accounting for the possibly correlated distributions behind the point estimates) if absolutely necessary.