The Trend-O-Matic

A few days ago I released the Respect-O-Meter, a measure of a team's public perception relative to its performance. I now present the Trend-O-Matic-the most complete measure of a team's performance I have ever come across.

To read the graph, first find the line that represents your team of interest. Then follow that line from left to right through the course of the season (the numbers at bottom represent the week). The numbers on the left hand side represent the team's "performance level" at that point in the season.

It is interesting to think how this graph reflects on the possibility of a tournament. I noted in an earlier post that tournaments struggle to identify the best team because of natural variance in a team's performance. Another issue is that it puts the emphasis to a team's performance at the end of the season and ignores earlier work. According to the Trend-O-Matic, Georgia and USC were the best teams in the nation at the end of the season, but LSU and Oklahoma were the best at the beginning. Why should we ignore LSU's and OU's early season accomplishments at the expense of Georgia's hot streak near the end? Georgia may have beat LSU in week 16, but I'm confident LSU would have beat Georgia in week 1.

The idea behind the Trend-O-Matic is simple. It is a graphical representation of a mathematical model with four inputs and five outputs. You put into the model the point margin of games, the week the game was played, the location of the game and the team's involved and out pops five parameters that track the performance of each team through the season.

The model is about as good of a description of the events of last season as you could hope to find--its predicted game outcomes and the real game outcomes have a correlation over .95. The nature of the model makes it useless during a season to predict game outcomes, but it is very useful for describing trends afterwards. The flaw that prevents it from being useful to predict games is that it is too sensitive near the extremes (precisely where the model would be generating predictions). You can see this in the graph above, for example, where Oregon's line spikes up right at the end after a very good performance from Senor Roper in the Sun Bowl. It is not inaccurate (it's, in fact, incredibly accurate), but that one bowl game was able to move the end of the line more than a good game in the middle of the season would have been able to move the line.

I hope to prepare a graph for each conference and I will also soon be releasing the Respect-O-Matic, which charts the public perception of each team over the course of the season.

Enjoy.