To what extent do the aggregate intervals for a given parameter reflect the number of experts and the variability in the responses of individual experts?

[ChrisBeeley]

My intuition, from designing and running the aggregation process is that the aggregate intervals will be narrower if:

1. The variation in the central estimates of the experts, var[P90+P10)/2)], is smaller (i.e. more agreement between experts)
2. The average expert interval width, mean[P90 – P10], is smaller (i.e. more confident experts)
3. The number of experts, n, is larger (i.e. more experts)

I suspect these are independent effects although there may be some limit cases exceptions. (i.e. If all experts offer the same view, then the number of experts will not affect the aggregate interval width.)

In general I’d expect effect 3 to be modest relative to effects 1 and 2.

If we need to be more sure then I guess we could run some simulations.