Median Calculation Wrong

I used the Median() function in a calculated measure in a cube. All seemed well until one of my users, a statistician, pointed out that the displayed values were incorrect. I investigated this and finally built the simplest cube with three records and the displayed median does not make sense, i.e. equal what a median value should be.

The 3 records have loan amounts as a measure with these values: $102,500, $168,400, and $172,181 and loan number keys of 1, 2, and 3. That's it.

The median should be $168,400 since the number of items is odd and that is the middle value.

But the cube calculated measure displays $170,290.50. It appears to be taking the average of the middle row and the next value.

When I increase the number of records I get the same odd behavior but it not only alternates (of course) with odd and even (since they use different formulats for odd and even) but the odd wrong results alternate within themselves. The apparent calculations for 5 or 7 records are different compared to 3, 9, 11, 13, 15, 17 records. The first set seems to be calculating odd number of items median by averaging the middle value + the value BEFORE the middle value while the second set of odd rows (3, 9, etc) seem to be calculating median by averaging the middle value + the value AFTER the middel value. The even numbers result in the larger of the two middle values being selected instead of the average (financial median) or the lower number (as one poster claimed for statistical median).

My calculated measure MDX is very simple:


MEDIAN( [Test Fact].[Loan Number].Members, [Measures].[Loan Amount])

Ideas anyone? Am I missing something? Is this a bug?

Thanks!

- Grant