Trend-O-Matic 1.1

Before I went ahead and presented the rest of the data from the Trend-O-Matic and its companion, the Respect-O-Matic, I have made some minor modifications to make the model more robust. I now present the T-o-M 1.1. It is about as close to perfection as this blog will ever get.


Part of the challenge with this type of metric is that it attempts to account for four variables with only a single equation. The four variables are the quality of each team at that point in the season and the level of performance of each team on game day. The only solution is the score. Therefore, we don't know if one team performed extraordinarily well on that night, or if the other team played poorly, if the game outcome was the result of one team getting better over the course of the season and/or the other getting worse. All we now is that, on that night, Team A was x points better than Team B. I have found it to be relatively easy to rank a team's performance on any given night, the challenge is recognizing if the performance is part of a larger trend or just a fluke. This particular metric assumes that about 50% of the variation from week to week is due to larger trends and the other 50% due to what the kids ate for breakfast, if they got enough rest before the game, if the quarterback had recently been dumped by his girlfriend who left him for a male cheerleader, etc. It is impossible to ever know for sure where we should actually draw the line, so 50% seems like a decent compromise.