> Each SICL team plays at most 1 team from out of their own conference during the season. Is that enough of a comparison to make a valid rating?
I really like your question.
Here are the 4 SICL non-conference games.
Date Cl Rank Visitor Score Cl Rank Home Score Post Act 11/29/2010 1A 37 Morning Star 59 1A 75 English Valleys 42 -10.43 -17 11/29/2010 1A 8 Traer North Tama 68 1A 39 Belle Plaine 61 -13.36 -7 12/04/2010 2A 58 Eddyville-Blakesburg 43 1A 4 Lynnville-Sully 63 16.67 20 12/07/2010 1A 37 Morning Star 55 2A 126 Marengo Iowa Valley 22 -29.31 -33
——————
Bias:
Predictions are generally close to actual scores. In this case the SICL team is always the home team. Summing the games, SICL teams are predicted to lose by 36.43 points; they actually lost by 37 points.
I see no bias in evaluation of the SICL.
——————
Accuracy:
Residuals:
(-17 – (-10.43)) = -6.57
(-7 – (-13.36) = 6.36
(20 – 16.67) = 3.33
(-33 – (-29.31)) = -3.69
Standard deviation of residuals = 6.007
Overall standard deviation model = 6.8339
At this time, model is predicting SICL non-conference games more accurately than the average Iowa high school game.
——————
> But is one non-conference game enough?
As of now, and based on the bias and accuracy measurements above, one non-conference game per team is currently producting above-average predictions.
It is unclear to me that this analysis will works so well next week or next month. Prediction is hard, especially with only four games.
This should be monitored as more games are played.