There was some pretty big news today in Section 2 boy's basketball. From Laura Amato at The Troy Record, the Watervliet boys basketball team will need to vacate their first 5 wins this season. This is a difficult problem for me and the model because the model is based upon these games and needs the most information possible to be as accurate as possible.
Problem #1: The model is based on wins, both yours and your opponent's and margin of victory. How can I now have a team that has a large positive MOV and zero wins. The two variables aren't comparable if I change the team's record.
Problem #2: Watervliet has now played seven games with a team that doesn't represent the one that will play in sectionals. This is similar to when a player on a team gets hurt near the end of the season and misses the sectionals. The major difference being one player was ineligible to play in the ones he/she did and the other wasn't.
Problem #3: Every team that Watervliet has played now has tainted scores and records as well. This is a greater problem than just one team.
At this point I'm leaning toward excluding the games altogether. Since only three of them were against Section 2 teams, it's not a huge disruption and I'm certainly glad it happened now and with a team that has played so few in-section games as opposed to an end of the season proposition. This isn't like baseball where you have so many stats like Wins Above Replacement where you can fairly reasonably determine a player's value to a team. The point being there is no reasonable way I can adjust the standing games to what would have happened without the player. And leaving the games alone doesn't represent the true picture of where each team stands against each other.
If a player were injured, I would leave the games as is. Injuries are part of the game, though no one likes to see them especially at this level. A team is deserving based on what it has accomplished and that is the goal of the model anyway, to determine who has the best chance to win sectionals based on their accomplishments weighted against historical standards. This situation is different however, and essentially these games didn't happen at all.
I have four options on how to handle this. The first is just giving Watervliet an 0-7 record, which I don't believe works due to how the model is designed. The second is allowing the games as I would for an injured player, which also doesn't work as the player in this case was ineligible which basically renders the games meaningless. The third is removing the games completely. I'm going to proceed with this strategy for the moment and be thankful it's only 3 games in the database. The fourth is to run the model each of the three ways and use an average of the three. I haven't ruled it out yet, and with more time, or if there were more games played my level of consideration may change.
For now though, it's big news to swallow and I wouldn't be too worried about Watervliet's performance tonight in predicting how they finish the season. As the news settles so will the players. It did however, just make the next Forecast that much more interesting.