Something in skrum's personal belief system clearly prevents him from understanding the arguments people are presenting. His system must be so important to him that he can't actually see it's shortcomings.

It's actually really interesting to see the different way he tries to ignore the facts & examples presented in so many different ways.

I'm not sure we can give him any examples that are more trivial than we have. He's basically ignored them, because he has to.

He can't actually say that the player with the obviously better record is better because it would contradict his dogmatic TAZD position.

But he clearly can't say the worse record is better because it's so obviously not. It's such a trivial example that he realizes it on some level and has to try to explain it away.

He also can't choose the wine in front of me. ;)

skrum, we all understand that variance goes by the number of games, and stdev goes by square root of number of games. but why does that make it statistically relevant to multiply by stdev?

the rest of us are just looking at the effects of that multiplier, namely that it means people who play an inordinate number of games per month get rewarded for their average play moreso than players who play a statistically relevant number of games with superior play.

you yourself say that in order to qualify for the TAZD each month, you must have 35 games played. or for the yearly, you must have 60 games per month on average. i could only assume that these minimum game qualifiers are to minimize variance from the calculation. so, since youve already minimized variance enough in your own opinion, why do you go the extra mile and multiply by square of games?

it honestly confuses me. and im not sure what to think of you throwing out terms like "variance", "standard deviation", etc, and then using them in nonsensical methods (multiplying by the stdev? really?), and claiming that just because you used the terms, that makes them statistical truths. i will reiterate: it makes no sense to multiply a statistic by a stdev just to say its now void of variance.

P1. 35%. 30%. ANYTHING more than expected. I arbitrarily pick example numbers. Again: You are trying to measure positive skill. Why not use POSITIVE deviation?

P2. Sure, LOOKS like it. The question here isn't why you use the square root; it's why the multiplier is there are all. In a Pearson chi-square test there IS no multiplier for the number of instances. Your justification for the games multiplier to BEGIN with is the encourage players to play more. My justification for the cap stands- I don't want to encourage players to spend 30 hours a week on the game, so I don't reward them for it.

P3. I've said in the past that the TAZD* is stronger than the ASR or the ASRm. I'm not even arguing about those here. We're discussing the TAZD and the TAZD*.

P4. Exactly 0 of the decisions made about the TAZD* are arbitrary. They are based in logic. Too often do statisticians find themselves blinded by the numbers. They are not based on my personal preference; they're based on input from the people who care about it. You CANNOT game the TAZD*. It's not possible. The best statistics are the ones that can't be fooled. The TAZD can easily be fooled into saying a player is "very skillful" if that player has played a large number of games and has a large number of sevenths. The TAZD* can't. The world isn't perfect. You have to live outside the perfect world of statistics and adjust accordingly.

P5. Earlier you said you would wait to see me do it. Congratulations on finding the motivation to do it yourself.

P6. Your three-year-old wallpost doesn't do shit for the average person. I'm just gonna quote my previous posts here, because you clearly did not hear them:
"You must admit that the post is hardly comprehensible for the average joe. It isn't so hard to put it in terms the non-mathematically minded can understand- I've done it several times."
"You are unwilling to explain anything to the general public in terms they'll understand."
The fact remains that you care so much about this statistic that you are willing to blind yourself to the facts and defend it to the death. The TAZD works the same way now as it did then because you are unwilling to consider changing it. Your ego is so large that you believe it is perfect. It isn't. Myself and others have been telling you that for a long time now.

CHLOE

crazy new idea i have. ppl were talking about the relevance of the luck stat earlier. i totally agree (with dott i think?) who brought it up as having a significant effect on results.

off the top of my head, heres an idea to account for variations in luck. and i warn you it will take a shit ton of legwork. prepare for some calf burn.

for each player, look at all their monthly ASRs or ASRms or TAZD*s (but not TAZD since every thinking person in kdice has concluded the creator has not childproofed his brain). plot those numbers against the luck stat. hopefully there will be a linear correlation. this all assumes there is one. hell, if theres a correlation and its nonlinear, work with whatever makes the most sense.

hopefully this will provide a way to compensate for differing amounts of luck that everyone has each month.

ill actually take a bit of time now and try it on my own stats. ill keep yall updated.

monte, there are also all the past months that I've posted in my google docs. I'll plot those and see what I come up with. I'm gonna do TAZD* against luck. I am reallllly hoping for a low r^2.

how about we have a statistical showdown between the two calculations (TAZD vs TAZD*) on a common set of data, and plot ROC curves so we can visualize which test yields less false positives/negatives. that way we can end the debate on a note that everyone enjoys: a statistically relevant and testable result.

skrum will dogmatically ignore the results, no?

477 data points later and r^2 = .000241.
Here's the graph. x axis is luck, y axis is TAZD*. I'll rerun the analysis with influential points removed in a few minutes.
http://i.imgur.com/cOmvE.png

Nice job chloe.