From: "Stouten, Pieter [R&D/0467]" <pieter.stouten <at> PHARMACIA <dot> COM>
Date: 10 dec 2002
Subject: Tournament Performance Rating-based pairing

Hans Secelle <johannes.secelle <at> MINFIN <dot> FED <dot> BE> wrote:

>TPR stands for [...] Tournament Performance Rating
>This "theoretical" elo rating is calculated by...
>1. calculating the average of the official rating of all one's
>   opponents
>2. calculating the percentage of the points scored (e.g. 6/6 =
>   100 pct., 4/6 = 66 pct)
>3. calculating the TPR accordingly
>[...]
>It's as simple as that :-)
>
The definition of TPR is indeed that simple. Building a pairing program
around it that also leads to meaningful results is a different story. A TPR
tournament is actually a refined version of a McMahon tournament, but rather
than starting out with a crude measure of strength (zero or more McMahon
points), one starts with one fictitious game against oneself. This is how
everybody's starting Elo rating is set. Then, as the tournament progresses,
one gains and loses rating points as a result of one's games and the (also
changing!) ratings of one's opponents. Similar to a McMahon tournament, the
winner of the event is the person with the highest TPR at the end. Going
back to the table Albrecht posted in his report on the Italian Championship,
one will see that he and Terje Christoffersen both had a TPR of 1775.
However, the sum of the ratings of Albrecht's opponents was 39 points higher
than those of Terje's opponents, so Albrecht ended second. Asle's program
leads to meaningful results, but for example handling of byes and of players
with unknown or unreliable ratings needs some finetuning.

Ciao,

Pieter