Executive Summary:
Some players worry that using spread to fine tune ratings will create a system that rewards counterintuitive behavior. I hope to dispel this notion.
---

It is a common perception that the player with more wins and less spread must be playing better endings than the player with less wins and big spread. I don't believe this is so.

(The following pertains to Stu's hypothetical example (way)below.) What actually occurred was that the distribution of results for the player who averaged 100 pts more per game than his opponents just happened to have more instances fall to the left of zero than his rival who averaged a mere ten points more than his opponents. Game-specific considerations may well necessitate times where spread is sacrificed to steer the distribution. This is why some people are opposed to factoring spread results into ratings. However, by and large, players do NOT have control over how this distribution is assigned. The player who garnered the 1000 point spread probably was not doing so at the expense of his winning chances, and he wasn't thowing away the close ones. He was probably playing better throughout all phases of the game than his putative rival.

The players who have the reputations as great ending players during my short time (Gibson, Morris and Tiekert) were not going through tournaments with large wins and small spread. Usually, it was large wins AND large spread. They were outplaying their rivals in the other dimensions of the game as well. As a famous example, one weekend in August of 95, Gibson finished 5th in a field that was less than Superstarrish. He still had a nice spread despite not winning many games. The following week he destroys the strongest tournament of all time. Question: Was Gibson "throwing away the close ones" the week before? Does anyone think he was playing to the level of his competition? Does anyone think he was a streak shooter? Does anyone think he was blowing endings? Well, the answer is no. It was just that that week the distribution of his differentials went askew a couple of sigmas. I'm sure the next week they were a little bit skewed the other way, but a strong spread in both cases does have significance in evaluating his playing strength.

Players should not oppose spread factoring because it might lead to perverse motivations contrary to winning the game. The (1% that I proposed) weighting of spread would not alter the hierarchy of motivations that govern decision making in a Scrabble game. Between two combatants whose K factor is 10 (the lowest factor, pertaining only to players > 2000), a situation would have to arise where someone would need an expected value of 1000 spread points to risk losing the game. I have not seen a situation like this since the famous game in Medleys-1 where the protagonist had to evaluate the reasonableness of fishing for the OVEREMPHASIZING tripletripletriple. From a ratings standpoint, it will still ALWAYS BE CORRECT to sacrifice spread for winning chances.

Comparatively, using spread as a tiebreaker DOES create situations where it may be correct to increase risk of losing to maximize net spread (short tournaments or large fields). Yet, no one rails against this. If you worry about perverse motivations, this should be your causa. In short, those who worry that rating spread will affect the game as we know it should realize that no situation will come up where rating matters would affect game play.

In a non-credit tournament, I still say give the trophy to the player who won the most games. But when crafting a number to rank the results of the players, by all means, don't ignore spread. It IS a measure of skill.

JG

Followup: This essay was so well written, it convinced someone to do some of their Statistics Ph.D. work on just this premise.

 From: Jim Miller

 Cc:  crossword-games-pro@MIT.EDU
 Subject: Re: Ratings, Spread, and Logic
 Date: Tuesday, February 03, 1998 7:40AM

 Stu D wrote:

   "You still have not answered my point, endorsed by Daveboys, that
   the 7-3 +1000 playier in a certain tournament is likelier to do better
   in the future than the 8-2 +100 player in the same tournament, and
   therefore deserves to be higher rated."

 Not if he keeps throwing away the close games.  Perhaps his endgame strategy
 isn't as hypertrophied as his bingo knowledge.  You're still equating "big
 spreads" with "better player", when that isn't appropriate.  Big spreads
 signify better accomplishment at one aspect and one strategy of the whole,
 complex game.

 Why are you concerned about the future, with the outcome of a tournament
 that's in the here and now?  It may well be true that the +1000 will do
 better in future tournaments; but he didn't in this one; and again, both
 players knew upfront the method by which their performance would be rated.
 So who deserves the win here? 

Last Modified 9/24/00

Back to Scrabble Correspondence Page
Back to Scrabble Main Page
Back to Jim's Main Page