From: Larry Kaufman <lkaufman <at> WIZARD <dot> NET>
Date: 14 mar 2001
Subject: Re: Evaluation of handicap games (please comment)

At 09:52 AM 3/14/01 +0100, you wrote:
>Hello all,
>
>can anyone give me a hint of how to evaluate handicap games? Right now I
>thought of simply adjusting the opponent's rating by a certain number of
>points according to the handicap, so that the game would simply be evaluated
>as if it were an even game against a stronger (or weaker) player. I'm
>considering the following magic numbers:
>
>L   200
>B   450
>R   575
>R+L 700
>2p  900
>4p  1250
>6p  1500
>
>So if a player with a rating of 2200 played against a player with a rating
>of 900 at 4 pieces, the game would be evaluated for him as if he had played
>an even game against a player with 900 + 1250 = 2150. If he plays at 6 pieces
>it would be evaluated as an even game against s.b. with a rating of 1500+900
>= 2400. For his opponent, the games would be equivalent to playing an even
>game against opponents with ratings of 2200 - 1250 = 850 and 2200 - 1500 =
>700, respectively.
>
>Please tell me your opinion about these numbers, how they should be adjusted,
>should other handicaps be taken into account (Right Lance, 3p, 5p), if there
>are different approaches, etc.
>
>Best
>
>Edi Werner, Shogi Ladder Coordinator
>
     As to your methodology, it is the same as the one used by the
Pan-Atlantic rating system.  We have some suspicion that using percentage
values rather than absolute numbers might work better (to allow for the
fact that stronger players are more adept at exploiting advantages and less
likely to blunder), but this has not yet been proven empirically.
     As to the values, they are much higher than the ones we use, but this
may in part be justified by the fact that there is no time pressure in
e-mail games (time pressure naturally favors the odds-giver, as it
randomizes the game somewhat thus making the advantage less meaningful).
     Our current values for the above seven handicaps are: 100, 200, 250,
350, 450, 550, 900 (we also use five-piece right handicap, value 750).
Your values are roughly double, which seems too high even with allowance
for the fact that yours are e-mail games.  The traditional values in
pre-war pro play were lance two ranks, bishop five, rook seven, rook &
lance nine, and two piece twelve ranks. I would suggest multiplying these
values by fifty, giving 100, 250, 350, 450, and 600, which total about 30%
more than the Pan-Atlantic values, which seems about right to me.  If so,
then I would suggest 750 for four piece, 1000 for five piece right, and
1250 for six piece.
     I think it is okay to rate five and six piece games, because although
theory shows how to achieve forced breakthrus and promotion of the major
pieces, it still takes some (minimal) skill to actually win the resultant
positions.  But, as my numbers suggest, the gap between four and six piece
handicap is quite large (despite your personal results with pros), which is
why we also use five piece.

     Larry Kaufman