Bringing my prediction analysis of the 2015 season to a close, finally, though this is generally a lot more fun than the data I usually look at.
The projected playoff picture looks fairly similar to last year. In the AFC especially, we simply swap in the Houston Texans, and pull out the Bengals.
In the NFC, things are a bit more complicated: Lions, Cardinals, and Panthers are out, but the Saints, Eagles, and Falcons are in.
Overall, I hope to just get three out of six teams correct in each conference (which shows my models were at least somewhat predictive for good teams over the season).
MAKING PICKS, PROBABILITY AGGREGATION
You may have noticed that my above chart includes three columns for each team. The first is the output from my initial model, found here. The last column is how many games the team won last year. These columns are similar in many cases, though the projections are generally compressed towards the mean.
The middle column is a little more complex to explain. It has to do with the difference between picking which team will win each week, versus how many games a team will win in a season.
Let's choose a bland, but explanatory example. Let's say that my model predicts that a team has a 75% chance to win each of its games (it never does, each game has a different probability based on opponent and other factors, but this example works):
- Original Model: It seems redundant, but teams generally win 75% of the time when they have a projected 75% percent chance to win. This means that if all games are projected at 75%, they would win 12 of their 16 games on average that season.
- Picks Model: If I were a sportswriter who has to make picks each week about who will win a each game, it may seem that I would pick the team 12 times, and their opponent 4 times. But this isn't true, because the team is a 75% percent favorite in each game, so I should pick them each time to assure my best chance of being right. I know that they will lose 4 of their games in all likelihood, but I don't know which ones (without external knowledge) The issue here is that although they are a favorite in each game, they also have a 25% chance to lose each, which aggregates into 4 likely losses.
The picks model is valuable in a number of ways, especially in considering which teams have a potential to win a lot of games this year if they win in each game they are slightly favored (can they beat the odds?). It also helps to explain why teams that seem to be better than all of their opponents end up with 6 or 7 losses.
Finally, consider the case of my Kansas City Chiefs. This is a team that I project will win 9 games, but will be favored in 12. As a fan it's important to understand that the chiefs may look good to win a majority of their games, but it may not happen as a result of wins not being high-probability wins.