Printable View
Hi All,
Just wondering if people might be interested in a new stat that can track a person's overall rank/place among many different DC projects? Here's an example of one such that I threw together using 27 people and 7 different projects (thanks to NinjaMicros website). It's a relative stat so there's a maximum score that one could get.
Below is a comparative chart:
Chart of 27 people, 7 projects. Blue = Mega Stat. Green = Potential Advancement. Chartruse line = Maximum Score Possible for 7 projects.
Raw Data that the Chart above was created from. Zeroman is artificially added as is Maxman. These are needed to calculate distance and the maximum possible stat respectively.
I'm not ignoring you, I promise. :D
Life's been busy the last week or so and I haven't had time to really look at the stuff you sent. I'm hoping things will calm down some by the weekend, at which time I'll dig in and see what I can do.
I think this a really good idea, and I'm going to try to support it. Of course, the more the merrier. I'm certainly not the only programmer around here! :thumbs:
Using this system, neat comparison graphs can be easily (?) created. Here's a radar graph comparing the second place person to the first place person.
Any point in the graph that is positive is where the second place person is ahead of the first place person.
Any point in the graph that is negative is where they are behind the first place person.
The Scale in this graph is proportional to the maximum score (of the 27 people) for a project. Thus if a point is at 0.8 then that represents being ahead of a person by 80% of the maximum value. Next post shows the table illustrating this a little better.
Cool, no probs. :cool: Just thought I'd throw out the idea for feedback and to see if anyone is interested (if not, then there's no need to bother).Quote:
Originally posted by Dyyryath
I'm not ignoring you, I promise. :D
Life's been busy the last week or so and I haven't had time to really look at the stuff you sent. I'm hoping things will calm down some by the weekend, at which time I'll dig in and see what I can do.
I think this a really good idea, and I'm going to try to support it. Of course, the more the merrier. I'm certainly not the only programmer around here! :thumbs:
Here's a table of the values used to create the radar graph: The second column shows the actual project score difference between the two people from the second person's perspective. So person 2 is ahead of person 1 in DF by 20915477 points and behind person 1 by 96767 in OGR-25 for example:
Finally, here's a link to NinjaMicros SuperStarStat system based upon rank instead of actual points (very similar, though you can't tell how far ahead (point wise) you are using rank data). Example is Here
At the bottom of that table, the left most column entry for the last row should read "Points per Project Per Rank". Also note that in that system, each project does not contribute equally to the stat. It is weighted to favor the less populated projects. The weight factors are in the second from the last row.
Weighting factors can easily be incorporated into the Mega Stat as well.
Also, comparisons can easily be done between groups and subteams (makes for a possible scoring system to be used in a multi-project gauntlet perhaps?).
Anyhoo, what do people think or would like to see?
prok
:notworthy /me likes!!! :D
You know about this site right?
http://www.msu.edu/~bercikda/rank_overall.html
I remember it from aways back. A nice analogy for this thread, substitute individual with team and it's very similar with some differences.
At the time that I saw it, the stat/ranking was based on some weird assumed distribution. That doesn't seem to be the case anymore (thank goodness :D).
Now it looks to be a relative stat like the one discussed in this thread, but instead of dividing by the number of WU's for the first place team in a project (this would give the first place team in a project a rating or value of 1.00 for that project) it's now dividing by the sum of all teams wu's. In that sites methodology the sum of the unweighted ratings for all teams in a project would equal 1.00. However, that system also has a weighting factor that multiplies this normed or commensurate value (if you will) by a measure of team popularity (the more teams a project has, the higher the weighting factor for that project). Again very similar (identical really in principle) to what's being thrown about in this thread.
There is however a difference that I can immediately see. In that system, the overall rating is akin to a Hamming distance measure (a team's overall rating is the simple sum of its rating over all of the projects within a particular class). Whereas in this thread's system, each project's rating is a co-ordinate in a project space where each axis of this space is represents a project. Thus comparisons made between teams or individuals follow a more intuitive measure of distance (Euclidean distance, what we're familiar with in our 3D existance).
I can think of one example where this discrepancy is illustrated. Consider a case with 2 projects. Say Team "a" has a rating of 0 for one project (it doesn't crunch for that project) and a rating of say 0.5 for the other project. Now say Team "b" has a rating of 0.25 for both projects. According to that sites method, both teams have an equivalent rating (0 + 0.5 = 0.5 and 0.25 + 0.25 = 0.50). However if you were to look at a plot of these teams on a standard Cartesian graph Team "a" is farther from the origin than Team "b" is and we would say that it is ahead of team "b" not tied. Euclidean distance holds to our intuitive feel of distance and shows that team "a" is 0.5 units from the origin and that team "b" is ~ 0.35 units away from the origin. Thus, the end effect is that team "a"s effort in the one project is discounted relative to the other teams effort (additionally to the weighting factor). Is this a bad thing? Don't know, depends upon your perspective and what is trying to be accomplished really. As a side note, if anyone cares, a Cartesian graph would have to be modified to look like bull's eye with a cross superimposed on top of it (like an old radar screen) in order to make the distances work. Net effect is that distances are progressively elongated as one approaches the 45 degree line. Not very intuitive to me :)
I've been thinking about this "mega-stat" thing myself for a while. I've come to the conclusion that there probably should be some element of time involved also.
What I tinkered with was trying to create something close to an average user production value per project. I was using something like the middle 80% of all active producers and averaging their output. I kept a running sum of this production number by project.
Comparing a user's total production against this sum of averages gave me a simple percentage. What was more interesting was that it seemed to dampen out short term high production and mildly reward long term sustained production. The down side was that it took a while to level out some and really couldn't reward past production. It's completely future weighted.
My biggest road block was relative weighting of projects to a master index. The best I arrived at was percentage weighting projects by the number of daily active users. It's a weighting by popularity, but it seemed to reward average production in large projects and lessen the value of huge production in small projects.
Another reason I went with the averages, user counts and percentages is that people seem to be able to understand them easily. Most people have a concept of the big fish in a little pond being a little fish in a big pond.