Version history

v7
Implemented 3 x,y correlation graphs to show how the wa compares with wf. Seems to be an almost linear fit.

v6
fixed some glaring errors in the weighted averaging.

now the means seem to line up to expectations... .5

Interestingly, the geometric mean also seems to line up to my pre prescribed .5 mean. Ultimately, I think the geometric mean can replace the weighted average formula without the need for normalization. I'm hoping to do a correlating graph... but that's a diff story.

v5
fixed some errors with incorrect places brackets that were screwing up the weighted product results.

v4
Updated Weighted Product so the sum of the ^'weight's total to 1.

Apparently I'm implementing a geometric mean without needing to 1/nth the value? I don't get how that works, maybe it's because the weights are reduced to 1.

The method was independently correlated by checking results against these two methods:

http://www.gyplan.com/weighted_gm_en.html

and

http://www.excelforum.com/excel-formulas-and-functions/914895-how-to-calculate-weighted-geometric-mean-in-excel.html

v3
Weighted Factoring is done as such
(a^weight)*(b^weight)*(c^weight)

v2
Added how As Is currently redraws
Also shows how averaging As Is with Rank-ECDF compared to the Alternative Output compares.

many

This is basically a MCDA system.

Opens in libre office

ctrl-shift-f9 to gen new simulation results

Setup a spreadsheet environment to analyze various aspects/statistics of the game for role calculations.

Also shows more or less the approach to deriving "%"'s for each aspect/category (i.e. skills, attributes, traits; and how to Monte Carlo new values to bootstrap more quantifying analysis).

Explanation:

bootstrapping: taking a small amount of data and reusing it.

monte carlo: taking a pre defined distribution, generate randoms results based on the old data.

In short, derive midpoints (a+b)/2 by ordering columns of data. The standard deviation becomes (b-a)/4, I throw in a little margin error curve probability magic to make values in the center of all possible aspect values have a little wider standard deviation to give a little more randomness to the process.

I begin the Monte Carlo by generating a flat % (randbetween 1, n-1) which corresponds to each midpoint, then I gen a norminv(rand(),midpoint,stdev) to generate a number based on that midpoint. I do this for each aspect (traits, attributes, skills) and then I run the data # analysis that each transformation uses. Then it outputs either WA = Weighted Average (As Is method), and a WF (Weighted Factor) method. On the very first sheet I show the analysis comparing the two methods and suggest what it would like like all the way to the right what an average between WA and WF would look like, and an alternative WA/WS average which would draw the values nicely from 0 to 100%

The Correct As Is method is. The 2d cube at q11 is the closest as is method, but in drawing it's rescaled from 0 to 100% around median, which I'm about to implement and reupload.

I did all this to test out data transform/normalization methods for dt. It was a fun little exercise to see if I could model a Monte Carlo Data Analysis program in Excel. It's based on one static input, however I set a lot of it up to be dynamic and extendable based on the input. I think it can accept up to 300 dwarfs as input for those 3 categories. So if one had a current population, they could export their data in. However... it's not 100% DT, there's two algorithms not implemented for how skills have a skill rate adjustment, as well as for attributes having an attribute potential gain adjustment formula. To implement those I'd most likely have to use vba and that doesn't work with libreoffice.