Choosing the odd man in a score group

Running Chess Tournaments
1

OK, it’s time for a good old-fashioned pairing question!

After the first round of a Swiss tournament in a section with 10 players, we have the following five players in the 1.0 score group:

1514 B, 1121 W, 1096 B, 982 B, 978 W

(I’m using the rulebook notation here.)

Now, if I treat 978 W as the odd man, I’m left with three players due White and one player due Black. However, if I treat 982 B as the odd man, I have two plaers due White and two players due Black.

So, I look at these pairings (still using the notation from section 29 of the rulebook):

1514 B - 1096 B
1121 W - 978 W

There are two ways to fix the color equalization. I can make a transposition in the bottom half between 1096 and 978, which is a rating difference of 118 points. Or, I can make an interchange between 1121 in the top half and 1096 in the bottom half, a difference of 25 rating points.

According to rule 29E5e, the interchange is preferred over the transposition because the transposition exceeds the 80-point limit. Therefore, the pairings should be 1514 - 1121 and 1096 - 978. (See also example 2 on page 151 and example 4 of section 29E7 on pages 158-159.)

However, both SwissSys and WinTD choose to treat 1096 B as the odd man instead of 982 B. In that case, we have these pairings:

1514 B - 982 B
1121 W - 978 W

In this case, the rating difference between 982 and 978 player is only 4 points. So, the pairings are 1514 - 978 and 982 - 1121.

Is there a reason to prefer the pairings that treat the player rated 1096 as the odd man, perhaps because it avoids an interchange? Or is it preferable to choose the lowest possible odd man that allows the colors to “work out” (in this case, the player rated 982)?

2

Check to see if your WinTD pairing options settings have the Harkness odd-man option selected.

With Harkness, the middle player is dropped.

3

Ken, I think treating the 982 as the odd man is acceptable. This will give you two due for white and two due for black, then use a transposition to fix the colors.

Technically, the interchange is okay to use as the rtg difference is lower, however, an interchange is less preferable to a transposition because it forces the top half of the score group to play each other.

Therefore, while not exactly correct according to definition, I’d use the 200-pt option, and transpose the 978 with the 1096. This is 118-pts, but should be ok, this being early in the tournament.

I don’t know why WinTD would choose to drop the 1096, unless it’s set to use the Harkness, which drops the middle person. Check your settings for that.

4

Thanks. Actually, I use SwissSys. One of the assistant TDs at the event uses WinTD. He took the results of the first round and let WinTD try the pairings.

I’ll ask him whether WinTD has the Harkness odd-man option selected.

5

In the preferences for the pairing rules, Harkness (drop middle) is not selected and wasn’t selected when I tried the pairing.

Chris Bird

6

You didn’t post the entire crosstable, but there are at least three plausible explanations why the pairing programs might treat the 1096 as the odd player rather than the 982.

First, I note that you said you had 10 players, with 5 in the 1.0 group. This means you also had 5 in the 0.0 group, and no first-round draws.

So maybe:

  1. The 982, if treated as the odd player, would be paired against the same opponent he had just played in round 1. This would necessitate further transpositions (in the 0.0 group), which might distort the pairings even worse than treating the 1096 as the odd player.

Or:

  1. The 982, if treated as the odd player, would be paired against an opponent due the same color as he. This, too, could necessitate (or make desirable) further transpositions, which again could be worse than treating the 1096 as the odd player.

Or:

  1. The pairing programs fell into (or blindly followed) what I call the “Defective Example 5 Trap”.

Let’s say the first round pairings were:

1514 B vs 1040 W
1121 W vs 1020 B
1096 B vs 1000 W
1080 W vs 982 B
1060 B vs 978 W

– and there were upsets on the last 2 boards. The result would be the exact scenario described by WileCoyote, and furthermore, treating the 982 as the odd player would result in his being paired against the highest-rated player with 0.0, i.e. the opponent he had just played in round 1. This makes Explanation 1 plausible.

I’m sure there are also plausible ways Explanation 2 could come about, but this is left as an exercise for the reader.

Explanation 3 is more disturbing, and requires a digression, so please excuse (in advance) my long-windedness.

In the case of “normal” (not across score groups) transpositions, the rulebook goes out of its way to say that transpositions “should be evaluated based on the smaller of the two rating differences involved” (29E5c), and presents the following example:

2000 WB vs 1800 WB
1980 BW vs 1500 BW

If we switch the 1800 with the 1500 to improve colors, this is considered a 20-point (2000-1980) transposition, not a 300-point (1800-1500) one, even though the physical switch is done between the lower-rated players, so that the 2000 is still playing on Board 1 and the 1980 on Board 2.

But not so with transpositions across score groups. In the following situation:

1980 WB 2.0
1900 WB 2.0
1800 BW 2.0
1920 BW 1.5
1840 WB 1.5
1760 BW 1.5

– the “raw” pairings would pair the lowest 2.0 (1800) vs the highest 1.5 (1920). But colors can be improved by treating the 1900 as the odd player instead of the 1800. Such a transposition would be evaluated at 100 points (1900-1800), with the other difference (1980-1920) being ignored since those two players have different scores. In discussing transpositions across score groups, the rulebook says “In deciding whether to make a switch of either the odd player or the opponent, you should look only at the rating difference of the players being switched” (29D1b).

But then comes Defective Example 5 (page 160), where the rulebook ignores its own advice. The example presented is:

2100 BWB (3.0) vs 2080 BWB (3.0)
1990 WBW (3.0) vs 2050 WBW (2.5)
1980 BWB (2.5) vs 1800 BWB (2.5)

Only two of the three bad colors can be corrected here. Which is the better transposition, i.e. should the 2080 be treated as the odd player instead of the 1990, or should the original odd player (1990) be paired against the 1980 rather than against the 2050? The first appears to be a 90-point switch (2080-1990), while the second is apparently 70 points (2050-1980), and since we are supposed to look only at the players being transposed, the second seems better. Yet the rulebook contradicts itself by also considering the “other” differences. Looking at it this way, the 90-point switch becomes 50 points (2100-2050), while the 70-point switch remains 70 (because 1990-1800 is more than 70). So now the book recommends the 90-point switch because it is “really” only 50 points.

In my opinion, the rule (29D1b) is correct and the example is wrong (“Defective”), but your mileage may vary.

In the original example presented by WileCoyote, if the rating of the highest-rated 0.0 is 1514 (or within 3 points of it), then treating the 1096 as the odd player and using the Defective logic essentially results in a 0- to 3-point transposition which, supposedly, is superior to the more logical 4-point (982-978) transposition. So that’s Explanation 3.

Personally, I suspect Explanation 1 is the correct one in this case. Could we see the crosstable, please?

Bill Smythe

7

DOH! :blush:

The sound that you just heard is two senior TDs and an NTD all slapping their foreheads simultaneously.

1 1514 B 6 1.0 2 1263 W 7 0.0 3 1127 B 8 0.0 4 1121 W 9 1.0 5 1096 B10 1.0 6 999 W 1 0.0 7 982 B 2 1.0 8 978 W 3 1.0 9 952 B 4 0.0 10 945 W 5 0.0

Sure, it’s clear now. If you treat player 7 as the odd man, he’s already been paired against player 2. The colors don’t work out if you pair him against player 3. So he would end up paired against player 6.

Anyway, Bill, thank you for the interesting analysis.

8

Well, I’m glad it turned out to be a case of not looking ahead far enough, rather than a program falling into the Defective Example 5 trap.

Programs often do a better job of using “look-ahead pairings” than human TDs do. A human might be inclined to simply make the best pairings for the top score group, and if that causes problems in the next group, well, let’s cross that bridge when we come to it. A computer is more willing to go back and change things, to make the best pairings overall.

But I would like to start a raging debate (I love raging debates on pairings issues) over the validity of rule 29D1b, which states that, when transposing across score groups, only the rating difference of the players being switched (not the rating difference of their opponents) should be considered. Example 5 (page 160) suggests just the opposite.

In my opinion, the rule is correct and the example is wrong. But some of you may feel it’s the other way around.

To start the raging debate, consider the following example:

Score group A:
2000 WB
1900 WB
1800 BW
1700 BW
1600 BW
1500 BW
1400 BW
1300 BW
1200 BW
1100 BW
1000 WB

Score group B:
1801 WB

The raw (untransposed) pairings would be:
2000 WB (2.0) vs 1500 BW (2.0)
1900 WB (2.0) vs 1400 BW (2.0)
1800 BW (2.0) vs 1300 BW (2.0)
1700 BW (2.0) vs 1200 BW (2.0)
1600 BW (2.0) vs 1100 BW (2.0)
1000 WB (2.0) vs 1801 WB (1.5)

(I have listed the pairings with higher-ranked on the left, lower-ranked on the right, rather than white left, black right.)

The temptation would be to switch the 1100 and 1000, i.e. to treat the 1100 as the odd player, in order to improve colors on two boards. But this would violate the 80-point rule (in this example we are dealing only with color alternation, not with color equalization).

A proponent of Example 5 would agree that this transposition is too large, and would also say that the next player up (1200) should not be treated as the odd player either. But going up still further, treating the 1300 as the odd player, all of a sudden becomes acceptable because the opponents of the players being transposed are only 1 point apart?!?

I think this example pretty much refutes Example 5, but as I said before, your mileage may vary. Opinions, please?

Bill Smythe

9

Come on, raging debaters, where are you? Surely John Hillery and/or Terry Winchester and/or Jorge Garcia and/or other frequent participants in pairings debates will have some opinions here.

Bill Smythe