I'm benchmarking procedure groupings done in a hospital and all the items associated with the cost of the procedure. As you can imagine the range can be from .01 to 50k dollars per item. With the majority of the procedures being in the 100−2000 range the distribution spikes and drops very quickly with the tail trailing to the right.
I've experimented with using the geometric mean removing outliers with the standard deviation 1.96 above and below the geometric mean.
As well as using the median and removing the 5th and 95th percentiles.
When I state benchmark I'm referring to a cost a physician should try to meet. The benchmark is more of a target and not a range. Currently i'm using the geometric mean for that "target".
What other methods are there to benchmark a right skewed distribution cost on a large amount of data over 29 million rows of data with a range between $1 and 3m?