I generally agree, and CEARCH uses geomeans for our geographic prioritzation WFMs, but I would also express caution—multiplicative WFM are also more sensitive to errors in individual parameters, so if your data is poor you might prefer the additive model.
Also general comment on geomeans vs normal means—I think of geomeans as useful when you have different estimates of some true value, and the differences reflect methodological differences (vs cases where you are looking to average different estimates that reflect real actual differences, like strength of preference or whatever)
I don’t see any strong theoretical reason to do so, but I might be wrong. In a way it doesn’t matter, because you can always rejig your weights to penalize/boost one estimate over another.
Good point on the error sensitivity. The geometric mean penalizes low scores more so it increases the probability of a false negative/type II error: an alternative that should be prioritised is not prioritised.
I generally agree, and CEARCH uses geomeans for our geographic prioritzation WFMs, but I would also express caution—multiplicative WFM are also more sensitive to errors in individual parameters, so if your data is poor you might prefer the additive model.
Also general comment on geomeans vs normal means—I think of geomeans as useful when you have different estimates of some true value, and the differences reflect methodological differences (vs cases where you are looking to average different estimates that reflect real actual differences, like strength of preference or whatever)
Naively, is there a case for using the average of the two?
I don’t see any strong theoretical reason to do so, but I might be wrong. In a way it doesn’t matter, because you can always rejig your weights to penalize/boost one estimate over another.
Good point on the error sensitivity. The geometric mean penalizes low scores more so it increases the probability of a false negative/type II error: an alternative that should be prioritised is not prioritised.