What inspired this wish list request?
When using the "Fit best" or "Fit All" options in the distribution platform or Capability Platform there are often times that the Normal Distribution(ND) is just as good as the selected non-normal distribution(NND) but the NND has a slightly lower AICc. There is more functionality with the ND so I have to go and review all the NND that were selected and see if I could use the ND just as well.
What is the improvement you would like to see?
I would like to see the option to select a fit criteria (AICc or BIC) and the range in which I would consider the ND equivalent to the best selected distribution and if it is within that range the ND will be used. So in this case from the semiconductor example data set I could have set a run that if the AICc of the ND was within 10 of the lowest AICc, then the ND would be selected.
Why is this idea important?
It is always better to use the ND when possible but there is no easy way right now to identify and then change the best identified distributions from a NND to a ND when there is a large number of columns in your analysis.