During the time of the poker-style scoring (on the August test server and in the first few days of September) it was said that “fifth is the new seventh” because it was often good strategy to flag quickly to avoid paying more points into the pot.

In October we had the first unbroken set of data since the Elo era and I thought to do a linear regression—using the free software at www.wessa.net—to see how point earned correlated with place of finish. I was expecting that first place would be the best point generator, second place second best, third place third best, and so on, but that turned out not to be the case.

I did a regression of the top 25 players’ final scores that I reported for October and got the following results (not all the coefficients are statistically significant):
Points = 22F1 + 21F2 -18F3 - 13F4 + 29F5 - 13F6 - 15F7

where F1 is number of first places, F2 is number of second places, etc. The standard errors for the seven coefficients are (11), (15), (27), (24), (20), (22), and (15).

The plus sign for fifth place is not a typo. Fifth place has the largest coefficient of all, though since its standard error is higher than that for first place, it ranks second in being the best predictor of success in scoring points.

What is suggested by these results is than in a typical kdice game there are two subgames: a battle among the top three (or four) and a battle among the bottom three. Two separate skills seem to be identified: the ability to play a winning game well, and the ability to play a losing game well.

In light of these findings, I thought I would test out a modification of the baseball-style standings that I have been reporting for the last two months. In the standard style standings, players will advance or fall back according to the following schema:

First: datum
Second falls back 1/6 game.
Third falls back 2/6 game.
Fourth falls back 3/6 game.
Fifth falls back 4/6 game.
Sixth falls back 5/6 game.
Seventh falls back 6/6 game.
Game not played falls back 1/2 game.

The final October standings under this system were:

12 kissygirl 9562â—† 492 19% 23% 16% 13% 12% 8% 6% ----
27 Grimbum 6156â—† 648 18% 19% 14% 16% 12% 8% 10% 8
9 Ssergio 10450â—† 557 18% 20% 16% 13% 11% 9% 9% 9
11 XxDiceyGirlxX 9894â—† 371 20% 25% 16% 11% 9% 9% 7% 12
10 uukrul 10235â—† 486 21% 17% 14% 14% 12% 12% 7% 17
18 fish28 8094â—† 811 14% 19% 17% 12% 12% 11% 11% 19
14 AnMoNiDa 9034â—† 326 19% 25% 18% 10% 8% 9% 8% 20
21 nuflis 7309â—† 485 17% 19% 16% 10% 12% 12% 10% 29
8 SodaPop 10627â—† 381 25% 12% 16% 11% 12% 9% 12% 30
7 vicsf 10754â—† 401 18% 21% 13% 12% 11% 9% 12% 31
16 Linch 8218â—† 361 21% 16% 14% 12% 14% 10% 9% 32
22 ndthorn 7251â—† 429 20% 17% 13% 10% 14% 12% 10% 33
19 yokobelgium 7959â—† 347 21% 15% 14% 12% 14% 10% 10% 36
6 Uyduruk 11116â—† 428 17% 19% 15% 10% 12% 10% 14% 40
3 dasfury 13732â—† 336 22% 18% 10% 11% 12% 10% 14% 40
1 Zosod 27030â—† 400 21% 15% 15% 12% 10% 12% 14% 40
4 Vohaul 13298â—† 607 19% 14% 11% 15% 11% 13% 13% 41
23 retareded rock 6745â—† 243 20% 21% 13% 8% 12% 9% 12% 41
15 Vermont 8312â—† 175 22% 16% 18% 14% 10% 12% 5% 42
24 tajmtoedaj 6479â—† 153 21% 22% 13% 10% 13% 6% 12% 46
17 hasiktirlan PUST 8100â—† 256 20% 12% 14% 13% 12% 12% 13% 51
25 Phoenix37 6352â—† 344 22% 15% 11% 9% 8% 14% 18% 52
20 Aiden 7735â—† 371 16% 15% 12% 15% 12% 12% 14% 54
13 El Destructor 9036â—† 431 18% 13% 12% 14% 9% 16% 14% 55
5 lesplaydices 11436â—† 370 17% 16% 11% 12% 11% 14% 16% 57
2 MadHat_Sam 18751â—† 346 19% 15% 10% 11% 10% 16% 17% 59

I now introduce the split-game baseball style standings. Players advance or fall back according to this schema:

First: datum
Second falls back 2/6 game
Third falls back 4/6 game
Fourth falls back 3/6 game
Fifth falls back 2/6 game
Sixth falls back 4/6 game
Seventh falls back 6/6 game
Game not played falls back ½ game.

Under this system the lower subgame has three players, the upper subgame has three players, and fourth is neutral. The lower subgame group falls 1/3 of a game behind the upper subgame group.

Under the split-game system the standings for the same 25 players would be:

12 kissygirl 9562â—† 492 19% 23% 16% 13% 12% 8% 6% ----
10 uukrul 10235â—† 486 21% 17% 14% 14% 12% 12% 7% 5
27 Grimbum 6156â—† 648 18% 19% 14% 16% 12% 8% 10% 6
9 Ssergio 10450â—† 557 18% 20% 16% 13% 11% 9% 9% 10
11 XxDiceyGirlxX 9894â—† 371 20% 25% 16% 11% 9% 9% 7% 12
22 ndthorn 7251â—† 429 20% 17% 13% 10% 14% 12% 10% 15
16 Linch 8218â—† 361 21% 16% 14% 12% 14% 10% 9% 16
8 SodaPop 10627â—† 381 25% 12% 16% 11% 12% 9% 12% 17
19 yokobelgium 7959â—† 347 21% 15% 14% 12% 14% 10% 10% 19
14 AnMoNiDa 9034â—† 326 19% 25% 18% 10% 8% 9% 8% 20
21 nuflis 7309â—† 485 17% 19% 16% 10% 12% 12% 10% 22
4 Vohaul 13298â—† 607 19% 14% 11% 15% 11% 13% 13% 22
3 dasfury 13732â—† 336 22% 18% 10% 11% 12% 10% 14% 23
7 vicsf 10754â—† 401 18% 21% 13% 12% 11% 9% 12% 23
18 fish28 8094â—† 811 14% 19% 17% 12% 12% 11% 11% 25
23 retareded rock 6745â—† 243 20% 21% 13% 8% 12% 9% 12% 27
15 Vermont 8312â—† 175 22% 16% 18% 14% 10% 12% 5% 28
1 Zosod 27030â—† 400 21% 15% 15% 12% 10% 12% 14% 29
24 tajmtoedaj 6479â—† 153 21% 22% 13% 10% 13% 6% 12% 31
6 Uyduruk 11116â—† 428 17% 19% 15% 10% 12% 10% 14% 31
17 hasiktirlan PUST 8100â—† 256 20% 12% 14% 13% 12% 12% 13% 34
25 Phoenix37 6352â—† 344 22% 15% 11% 9% 8% 14% 18% 36
20 Aiden 7735â—† 371 16% 15% 12% 15% 12% 12% 14% 36
13 El Destructor 9036â—† 431 18% 13% 12% 14% 9% 16% 14% 38
5 lesplaydices 11436â—† 370 17% 16% 11% 12% 11% 14% 16% 39
2 MadHat_Sam 18751â—† 346 19% 15% 10% 11% 10% 16% 17% 39

Look at the two sets of standings. I invite comments. Does one set have a better “feel” than the other?