so I was pulling the influential points and saw this:
1st 2nd 3rd 4th 5th 6th 7th ASRm TAZD* Name Rank
41% 16% 16% 10% 6% 5% 3% 2.43 16096 olkainry38 14th

GOOD GOODNESS. The next highest TAZD* is like 13000.

anyway the rsq value got HUGE after taking out those influential points! Giant! More than tripled! rsq = .000774

And here's TAZD vs. TAZD*! TAZD is on the X and TAZD* is on the Y for the same set of data (ie, all the qualifying players for the months that I had easy access to on my computer): http://i.imgur.com/pDNKA.png

And for a single player (namely montecarlo), for those months that qualify:
Luck vs. TAZD*: http://i.imgur.com/nvPcF.png rsq=.279
TAZD vs. TAZD*: http://i.imgur.com/xN4Td.png

well, nvm on that theory. at least the one outlier made sense. way low luck yields has way low tazd*. thanks for all the work chloe, youre the best <3

Chloe and monte:

There is a multiplier for the number of incidences. It is the expected number of incidences which goes as the number of games played. The chi square approximates a normal distribution whose mean goes as the number of games. To compare two normal distributions with different number of games, you have to scale them by **dividing** them by the standard deviations.

If you use the distribution of percentages rather than the distribution of incidences, there is no multiplier, but to compare distributions of different sample sizes (number of games) you have to *multiply* them by their standard deviations. This is all explained on my Wall.

The 60 or 35 game minimums have nothing to do with the variances but with having to avoid having to do Yate's Correction. For the TAPL, Yate's Correction would be needed for fewer than 5 games per cell, or 35 games total. For the TAZD, a bigger minimum would be needed because the expected values for 1st and 2nd would be less than 5 if you used 35. So I chose 60 because Ryan had once said that 60 was the minimum needed per month to make results significant.

The square root rule ensures that the players with more games have to play the square of the number of additional games of the players with a fewer number to overtake them. The purpose of the square root rule is both to reward players who play more games (but less and less for each additional game) but at the same time serve as a warning for the people who have played a few games that they need to play more games, with the same good percentages, to keep from being overtaken. Also, if players are pga'ing (unfortunately, pga'ing is a skill and is rewarded in the TAZD), it will be more difficult to arrange more pga's to stay ahead. On the other hand, I suspect that Dottir has not arranged 6,000 pga's to get to her high score.

I was off watching the Republican debates while you were doing the luck stuff. I don't think the luck stat is very useful; for one thing, it is rolls only, not inital stacks or restacking.

My "net dom" stat may have something to do with wimpiness in flagging, but I need a different data set to compare with. Perhaps I can use tourney data after the first two levels, since flagging is not an option for most players, who are all-in.

"If you use the distribution of percentages rather than the distribution of incidences, there is no multiplier, but to compare distributions of different sample sizes (number of games) you have to *multiply* them by their standard deviations. This is all explained on my Wall."

i respectfully disagree. the fact that you are using percentages inherently means that different sample sizes are just fine. all you need is a minimum number of games to remove an appropriate amount of uncertainty, so that the true mean dominates, and not variance's effect on the mean.

having a greater number of games doesnt mean youre percentages should be multiplied, it just means that they are closer to their true means.

btw how did the republican debate go? and who do you like so far skrum?

I went to my quick and dirty source for all things mathematical, Wikipedia, and pulled up Welch's t-test for two distribtions that have possiblly differet variances and different sample sizes. Note the weightings for the sample sizes:

http://en.wikipedia.org/wiki/Welch's_t_test

I like Newt and Perry the most. Romney just doesn't want to give any specific answers, like "what would you have told Obama was wrong about Romneycare?" Bachmann doesn't have enough command of constitutional law. She seems to think that if something is important enough, here has to be a federal solution for it. Santorum looks like he's just been to the dentist. I may want to look into the book Ron Paul mentioned, Bastiat's "The Law".

but i dont think your distributions have a different number of samples. they all have 100 samples (or slightly less) corresponding to the total percentage of all places.

if you were comparing the number of placements, yes, multiply by the stdev. but if youre comparing the percentages, then you shouldnt.

ill vote for ron paul, despite all the media and polls saying that he cant win. i like the guy for being true to himself over the past 30 years of politics, and not just trying to say what his audience wants to hear so that he can get elected/reelected. newt has a degree of that to him as well, so i respect him for that, but its only a fraction of how genuine paul is.

i dont agree with all his points, maybe 75% of them, but after watching american politicians for my short 14 years of voting eligibility, im sickened by all the chatbox manipulation that is poured into deceiving the american populace. and the utterly unbelievable amount of wasteful spending done by both parties. for some reason, i trust that paul will try to fix that, instead of maintain status quo. i dunno.

expect more fun from me tomorrow. I have a huge headache.