Estimating Wins From a +/- Model

[lukec]

Hi All,

I'm currently working on a project that integrates Hustle States into +/- models and I was wondering what the best way to estimate the total number of wins would be?
For example, if before the season I applied a minutes projection to my model to find a per 48 +/- number for each team, what would be the best way of going about calculating their expected number of wins?

Thanks in advance.

[Mike G]

The pythagorean expected win formula is:
pW = PPG^14/(PPG^14 + OppPPG^14)*G
Other exponents besides 14 have been used; likely depends on possessions per game.

Alternatively, you may use the straight line approximation:
exW = MOV*2.55 + 41
That says each PPG in margin of victory is worth 2.55 wins in 82 games.
This coefficient has also appeared as 2.7, I believe; but 2.55 currently makes a closer fit with this season's pythagorean model.

What are Hustle States? ... Nevada?

[GK5]

From looking at my data I've found Point differential to be closer to the 2.7 coefficient rather than the 2.55 depending on how far back you go.

You mentioned having a minute weighted +/- number for each team. It might be more accurate to see the correlation between team minute weighted +/- and wins going back x years and then translating directly to wins with that formula rather than converting your +/- numbers into point differential and then converting those into wins.

[Nate]

... Other exponents besides 14 have been used; likely depends on possessions per game. ...

You can also work out an exponent using logistic regression if you like.

...way to estimate the total number of wins would be?...

Another way to estimate the win probability is to make assumptions about the 'noise'.

A simple example would be to say that score difference varies along a normal distribution with a standard deviation around 12.5 points.

So if a team is expected to win by 5 according to the model, you can use something like "normdist(5,0,12.5,"TRUE")" in excel to generate a win probability estimate.

[lukec]

Thanks... I'm using the NBA's hustle stats (http://stats.nba.com/hustle/). If I were to use the pythagorean expected win formula that simply multiplies by the number of games, would there be a better way to calculate it than to do it individually for all 30 teams after calculating their PPG and OPPG(since every team plays a different schedule) from the +/-?

[Nate]

... would there be a better way to calculate it than ..

There's a trade off between how much work you do and how many things you want to account for. The (naive) pythagorean formula would just be win probability x number of games. If you want to do stuff with opponent strength or home field advantage, you'll want a fancier win probability formula that incorporates those factors.