Week 6 Power Rankings and Season Predictions

As usual, here are the power rankings for the upcoming week 6.

Now that every team has played four games, I’m going to start something new, which is to predict the rest of the season.  Since I know the schedule and I have the performance of the teams so far, I can predict any game even if it isn’t in the upcoming week.  I’m doing this using Luigi’s predictions (you can look at the model descriptions in the link in the banner).  One thing to note about the projections is that they have updated what they think should have happened in already-played games.  For example, when Dallas played Chicago in week 2 Luigi thought that Dallas had a 43% chance of winning.  Having seen more performances from those teams, Luigi now thinks that Dallas had a 46% chance of winning that game; other games have changed by more.  So just like the power rankings, the season win projections are based on what we know about a team now.  Also, because I don’t predict week 1, I took what happened in the first week and simply added a win for the victorious teams to their counts.  Here are the inaugural season predictions:

Here’s what’s in the table: in each row you have a team followed by how many wins they have on the season.  Then you have a predicted number of wins called ‘expected value’; this is the sum of the win probabilities in the 15 games played from week 2 on, plus one if they won in week one.  They’ve been rounded to two decimal places.  You’ll note that since the two probabilities (one for home, one for away) add to one, there is one expected victory per game, and so these predicted wins add to the 256 regular season games (within rounding).  The final number is a predicted number of wins based on how many games the team is the favorite (from week 2 on, with one added if the team won in week one), defined as a win probability over .5.  This is also what the table is sorted by.

There are a few things you can see here.  First is that the EV wins and the ‘favored’ wins differ widely for some teams; like Atlanta only has 9 EV wins but 13 ‘favored’ wins.  That means that Atlanta is the favorite in many of its games, but not necessarily by a lot.  Arizona and Cleveland, on the other hand, have EVs of 7 but predicted wins of only 4, meaning that they are rarely the favorite but are probably close behind in many games.  Broadly speaking, I would expect the EV predictions to be closer to the actual final outcomes.

Another thing to note is that neither of the predictions account for what has actually happened.  So Oakland, for example, already has two wins yet is predicted (final column) to win only 1 game.  This can give us a sense of random luck; Oakland is only the favorite in one game according to the model (at home against Denver in week 15), but they have already managed two wins (what I think we would all call a surprising win over the Chargers this past weekend and against the Rams in week 2 when they were a small underdog).  This is another reason to prefer the EV predictions; they acknowledge that sometimes an underdog will win, and account for that by giving them credit proportionate to their chances of winning.  As more games are played I expect the EV predictions to close in on the actual victories number while the predicted wins will likely stay about the same.  At the end of the season, the disparity between the actual wins and EV wins (and to a lesser extent the predicted wins) will tell us which teams were lucky or unlucky, or if you have something against the model you might say that they were ‘special’ or did something great that the model doesn’t account for (special teams are a likely culprit).

Finally, using the predicted wins we can predict what the playoffs will be like.  The AFC division winners should be the Patriots, Steelers, Texans (EV prefers over the Colts, predicted wins says tie), and the Chiefs, with wild cards going to the Chargers and Ravens.  The NFC division winners should be the Falcons, Packers, Giants (over Philly like Houston/Indianapolis), and Seahawks, with wild cards going to Chicago and Tampa Bay.  I think perhaps the most interesting prediction here is that the Jets are going to slow down a lot, winning only 8 or 9 games total when they’ve already won 4.  That’s due to their remaining schedule, with tough games against the Packers, Texans, on the road against the Patriots, and Pittsburgh, and their easy games against Detroit and Cleveland being on the road which cuts into their win probability.  But, time will tell.