As a distraction from a particularly tedious task, I have now entered all
the data I needed for a program intended to rate chess players of the distant
past. My original intent was to rate players from about the death of
Labourdonnais to the 1851 tournament, but I found myself widening the
range, and now go from 1836 to 1863. This post will discuss general issues
of historical rating; I will give the results in a later post.

Why make another rating system? I have several reasons. One reason is the
inadequacy of certain well known rating charts. Most of these seemed to
be geared towards rating players at later dates, when information is more
reliable, but I must comment on certain inadequacies for rating these early
players so that readers don't think that I simply am unaware of these
projects which are taken seriously by some people.

Let us start with the most recent historical ratings systems; the Chessmetrics
system due to Sonas and the CAX system due to Schiller.

Sonas' rating system is impressive in many ways. However, when one looks
at the very early ratings, they are clearly ridiculous in a number of cases.
Several examples: Kipping, a Manchester player (who I am actually quite partial
to) is rated 5th among active players in 1857. He is rated above Anderssen, a
view that seems very hard to justify and definitely at odds with beliefs
at that time period. Sonas has only 9 active players for this year, by the
way, with odd exclusions; why does Paulsen (2d in NY 1857) not appear, for
example, and why not Boden and Pindar who played in Manchester 1857?

In 1862, what rational person (as opposed to rational program) would place
Steinitz behind Blackburne and Mongredien in their ratings? Mongredien won
only 3 games (one by forfeit) at London 1862, while Steinitz won 8 (2 by
forfeit), and other results from the year also clearly showing Steinitz
as the superior player. With such poor judgements, why should we give any
credence to more defensible but still controversial rankings on this site
(for instance, Staunton's very long period of being ranked above Anderssen,
from when this was debatable but controversial in 1851 to when it seems
absurd in 1859)?

I think some of Sonas' problem may be from relying on game databases. The
preservation of games was very biased in the old days, and a player such
as Staunton who ran a magazine and wrote books finds far more of his
wins in today's databases than such players as Buckle who never kept
a score, but had well documented matches. Schiller takes the other
extreme, relying only on particularly well documented matches and
tournaments. For these early years, he just does not have enogh information;
Kieseritzky was not a worse player than Mucklow, but by looking only
at the tournament results from London 1851 rather than the known other
results in less formal games played at that time, Schiller comes to this
conclusion. If you look at the list and know something about chess from
this time, it is hard to believe that the program does a better job than
you would get if you asked for the opinion of the players strengths to
any chess historian, and I do not give these ratings much weight at all.

Going back a bit, we get Divinsky and Keene's rating of players for
Warriors of the Mind, with details explained in Life Maps of the Masters.
However, their ratings start only in 1858, so there is not much overlap.
It is also hard to justify their technique (simply take the scoring
average agains other designated top players of the period) when the strength
of the top players was so different from each other; Staunton's avoidance
of Morphy certainly is a smart move if you want to do well on this scale!

In my opinion, the best historical ratings are those given by Elo. However,
these focus on rating peaks and lack that year to year information which
is fun to debate (who was really best in the world at a particular year?).
We also do not have the data itself available for checking, and giving the
huge uncertainty in what really happened in matches from this period, this is
quite important.

I also want my system to evaluate some issues which you do not get from
the existing attempts. I would like to see results which explain both why
people had certain beliefs about strength of players in the past, and why
we might disagree with them now. Therefore, I would like to show both
ratings if we have no future information (these should correspond roughly
tp what people believed at the time), and ratings with future information
taken into account; when these show large discrepancies, as they certainly
do, it makes for interesting discussion and evaluation. None of the systems
in place try to give both types of information.

Long enough post for now on these general issues; I will be back with
posts on the technique used, certain anomalies which must be considered,
as well as the results of my program.

Jerry Spinrad