Re: Hoshitori: New Banzuke Proposal

[Keith Moulton <km34@columbia.edu>]

	I agree that it would be nice to have a full numbered maegashira
ranking now that we have win-loss equivalents, but I have some
reservations.  Over the weekend I crunched some numbers to see how closely
our hoshitori compares to the real thing.  I compared the numbers of the
top 14 rikishi competing over the course of 6 consecutive bashos with the
corresponding top 14 cyber-rikishi (excluding rikishi with kosho, etc.).
I measured their average score and the standard deviation of the spread of
their scores over the 6 basho period and did the same thing with the
cyber-rikishi.
	I don't suppose it's surprising to see that their is about twice
as much luck involved in our hoshitori than the real bashos.  In Real
Life, the rikishis' consistency (at being good *or* bad) was shown by an
average standard deviation of 1.626, while our efforts varied with an
average standard deviation of 3.296 (a purely random sample would produce
an SD of 4.6).  It's been a long time since I took statistics but I think
my basic gist is at least right.
	I think this means that we can't expect our cyber-rikishi to
maintain the same level of consistency that we would expect from a real
rikishi.  This is particularly true for the Ozeki/Yokozuna candidates at
the end of the distribution.  For this reason I think we would be better
off using the *average* score over a couple of bashos for Ozeki promotion
as Dhuggy suggests here:

> One problem with the current system is that while it is still possible
> to be promoted with a three tournament record of 11-4, 11-4, 11-4, it is
> NOT possible to be promoted with a three tournament record of 15-0,
> 10-5, 15-0, for example.  Perhaps we could say that a Sekiwake needs to
> have 33 or 34 wins in their last 3 tournaments, with the most recent
> tournament being at least an 11-4.

	This sounds reasonable to me, but his idea as represented in the
promotion table, that it would usually require more than three 11-4
records to become Ozeki seems a bit strict.  As far as I can tell, nobody
in the history of the hoshitori has ever had four consecutive 11-4 (or
better) records.  Applying the proposed rules to the old results of the
hoshitori show that no one would have ever achieved Ozeki status, let
alone Yokozuna.  By my calculations there would presently be only 4
Sekiwake at the top: Ekigozan, Hibanayama?, Mimasumoumi, and Rikimaru.
Current Ozekis Hibijibi and Shimizuyama would only be Kw and M1.  The
promotions of the top cyber-rikishi notwithstanding, the rest of
the promotion scheme seems reasonable to me though.  A couple more
comments:

> I noticed that our current system does not reward 2 or 3 consecutive
> 14-1 records (or even three consecutive yusho!) any more than three
> consecutive 11-4 records.

	Actually, two 14-1 records are rewarded since that fulfills half
the requirements of becoming Yokozuna.

> One final issue to discuss here is that IF we decide to use a new system
> such as this, then how do we figure peoples new ranks?  Do we all start
> from scratch at M16 for the first tournament?  Or do we start from our
> current ranks, which were calculated under a different system?  Or do we
> use each player's entire results history (since they began competing in
> our contest)?

	I, too, prefer the last option but not as the system stands now 
since it would demote too many of the cyber-rikishis in the top ranks (I
would go from Ozeki to Komusubi-west! %-).  Maybe we shouldn't worry too
much that the hoshitori resemble the real thing.  Real rikishi put in 10
years or more, but hoshi-stats show that most contestants don't even last
a year.  So we shouldn't make it too time-consuming or too difficult to
reach the top ranks, since winning is what really makes the game fun!
Cheers,
Keith Moulton.
OperaStuff:  http://www.columbia.edu/~km34	email:  km34@columbia.edu