Some more thoughts on quantiles...

In a previous post some weeks ago I wrote about calculating a quantile in MDX. Of course you could also think of using a SSAS stored procedure for the calculation as stored procedures offer the full range of .Net programmability and libraries for doing the calculations. In my last post about quantiles I simplified the calculation a lot by assuming that our computation results in a single element. This made the calculation easy to be written in MDX. But if you need the quantile at a higher precision you could approximate the values between two discrete elements.

First let's take a brief look at the simplification we did when calculating the quantile in MDX: The calculation of the quantile is simple in the case that (n-1) * p% gives an integer (n=number of cases, p%=percent of the quantile). In our example we had 10001 scenarios (n=10001) and computed a 5% quantile (p%=5%) giving (n-1)*p% = 500. So we simply have to take the 501th element from the sorted set of cases.