Useful analytics at the start of the NBA season are sometimes hard to come up with. Essentially you have a choice of noting how unprecedented Anthony Davis's start of the season is or how noisy any particular stat is.
There isn't anything wrong with either of those approaches, but I tend to want to estimate how much of this noisy data or unprecedented performance can we take forward. So, I have been experimenting doing in season projections with based on the expected regression using a Bayesian updating formula that weights the amount of noise in the sample against the stability expected in the entire season. The gist of the formula is that the noisier the sample is the less weight we give the new data.
I decided to apply that method to Dean Oliver's four factors, effective field goal percentage, rebounds, turn overs and free trow rate, on both offense and defense for each team as of Tuesday morning via Basketball Reference. I used a team based prior based on the last season regressed toward average based on the season to season correlation.
(See below for the individual team factors and the expected percent adjustment)
The adjusted expected four factors were then converted into net rating using a formula developed by Evan Zamir. In the table I have each teams current margin of victory, or net rating, and the expected rating after weighting with the prior.
The four factor method has the benefit of being relatively straight forward, but it leaves out few noisy areas that can especially distort early season success, three point shooting by the offense and even more by their opponents, both of which get hidden in eFG%, as well as my personal favorite, free throw defense. But the results above at least have air of reasonableness, at least for a sample based on three games for most teams, well the Warriors aside.
Individual four factors as of Tuesday, November 1 and the percent adjustment expected by the end of the season according to the model.