I am debuting my Franken-Win Predictions for the upcoming NBA season. I am calling them Franken-Win Predictions because they come from an unnatural bastardization and stitching together of statistical models and predictions from other sources. Having never done this before I decided to start with a source I was familiar and semi-comfortable with, David Berri’s Wins Produced. If you are unfamiliar with Wins Produced, it is a box score based metric which converts a player’s individual statistical contributions into a measurement of how many wins they produce for their team.

Making predictions with Wins Produced requires some information about the position a player plays and the minutes they spend on the floor. To this end I borrowed the minute projections for next season from Basketball Prospectus, whose terrific 2010-2011 version is now available in PDF format. I had to make a few adjustments to the minute projections based on injuries which occurred after the publication of the Basketball Prospectus (Al Harrington, Carlos Boozer, etc.).

I then took the Wins Produced per 48 minutes numbers from last season. I made some slight adjustments to these numbers to account for player improvement or decline. These adjustments, in most cases, were made by dividing the Basketball Prospectus projected Win% for each player by their Win% from last season and then multiplying it by their WP48. I combined these new, adjusted WP48 projections with the minute projections and arrived at a projected Win Total for each roster.

I am sure there are several reasons why my method is mathematically ludicrous but at this point in the development of my statistical knowledge and analytical skills it was the best I could come up with. In addition, these projections don’t account for potential future trades and injuries which will almost certainly affect the fates of several different teams.

Here are my projected win-loss records for each team:

With these win-loss records the playoff seeds would break down like this:

The teams left out of the playoffs would stack up in lottery positions thusly:

A few points:

- My projections reflect a decline in league parity, especially in the Western Conference. This is partially due to the strength of several power teams; with Miami, Portland, Chicago, Orlando, San Antonio, and the Lakers all amassing 53+ wins. This will also lead to sub .500 teams, Cleveland and New York, to make the playoffs in the East.
- Based on my numbers, Washington will challenge the futility of last season’s Timberwolves and Nets. This projection is largely based on the huge minutes that should be played by John Wall, who has the potential to have a negative WP48 score.
- The surprise team in my projection is easily the Golden State Warriors, who if my analysis proves correct, would jump 23 wins from last year’s total and catapult themselves up to the 5th seed in the west. Wins Produced is a metric heavily influenced by rebounding. (David Berri would also point out that winning games is heavily influenced by rebounding). With Nellieball being abandoned, more minutes will be available for the above average rebounders David Lee, Andris Biedrins, Brandan Wright, and Louis Amundson. Simply put the Warriors will be beasts on the boards this year and should be an incredibly improved team.
- My projections have Denver missing the playoffs. If this scenario is in play as the trade deadline approaches it seems extremely likely that Carmelo Anthony would be traded, throwing a monkey wrench into all projections for how the season will play out.
- By my numbers Philadelphia is the team in the most tenuous position this season. I have them only projected to win 24 games, roughly 12 of which would be produced by Andre Iguodala. If Iguodala is injured or traded things could spiral out of control very quickly.

If anyone is interested, here is the link to the google spreadsheet which shows the detailed breakdowns of my projections. These include the individual WP48 and Wins Produced projections for each player.

I don’t get how you adjust for the rooks. Do you count those as having 0 wins produced during their minutes? You mention Wall as being on of the reasons for Wiz poor results and I guess the same could be said for Clippers. (Griffin)

I like the Franken-Win name btw.

JJJ, thanks for reading and commenting. I had to come up with a different system for rookies because they don’t have WP48 numbers from last year to adjust, and they don’t have 2 years of projected win% from the Basketball Prospectus to do adjustments with.

I started with a WP48 baseline number for rookies of 0.050. In the Wins Produced System a WP48 of 0.100 is considered an average player. Most rookies, although not all, produce at a below average level in their 1st season. The 0.050 number has stuck in my head from a Wages of Wins post, which I can’t locate, as the average WP48 of rookies. This may be incorrect, but for whatever reason, it’s what stuck in my head. To make the adjustments for rookies I looked at the Basketball Prospectus Wins Above Replacement Player (WARP) projections for the rookies. In cases of a positive WARP I simply multiplies the WARP times 0.050 to get a WP48 projection. For example, a rookie projected to have a WARP of 1.5, they would have a WP48 of 0.075. In cases of a negative WARP, I multiplied 0.050 by the absolute value of the WARP and then subtracted that from 0.050. So a player with a -1.0 WARP would have a WP48 of 0.00.

John Wall is projected to have a significantly negative WARP, which creates a negative WP48 which means he is actually costing his team wins instead of creating them. This is mostly due to his projected problems with turnovers and efficient scoring. But not all rookies fall into this category, some look to have a significant impact in my analysis. Here are the WP48, and Wins Produced projections of some other notable rookies:

Jordan Crawford/-0.070/-1.66 Wins Produced

Paul George/0.055/1.43 Wins Produced

Blake Griffin/0.150/6.53 Wins Produced

Evan Turner/0.125/6.67 Wins Produced

Wesley Johnson/0.075/3.56 Wins Produced

Derrick Favors/0.020/0.79 Wins Produced

Damion James/0.080/1.90 Wins Produced

DeMarcus Cousins/0.235/10.18 Wins Produced

James Anderson/0.050/1.58 Wins Produced

Tiago Splitter/0.205/9.09 Wins Produced

If these numbers hold up, the Rookie of the Year should be a two horse race between DeMarcus Cousins, and Tiago Splitter.

I hope this helps, Thanks again for reading and commenting!

Nice work with this list ilevy!

I found this quote from the Prof’s blog on rookie WP48. I think you were probably close enough with 0.050 though.

“an average rookie only posts a 0.042 WP48”

http://dberri.wordpress.com/2008/10/26/the-nba-rookie-polls/

Thanks Michael! I looked forever for that darn number. I remembered reading it, but it seemed like it magically disappeared after that. I don’t think that I’ll re-write the post with that adjusted rookie number, but will save it for next year. However that will make some of the rookie projections, especially for ones playing significant minutes a little less rosy. That would probably take a win or so away from San Antonio (Tiago Splitter) and Sacramento (DeMarcus Cousins) and maybe a half win or so from Philadelphia (Evan Turner) and the Clippers (Blake Griffin).

Thanks again for reading and commenting!

Thanks. I’m just getting into APBRmetrics (however I’m not new to university statistics or sabermetrics) so please bare with the noobish questions. I found this blog thru a some other page on the topic (don’t remember which) and it seems like good stuff so it’s bookmarked!

Keep up the good work.

I’m happy to answer questions or explain my stuff in more detail at any time. For the record, I’m pretty new to advanced stats myself. I don’t have a statistics or math background, and most of my knowledge base has come from reading other people’s work. Part of the reason I like this blogging thing so much is that I am constantly learning something new with each project. Thanks again for reading and commenting!

Comparing your predictions to one Vegas line these are the teams with the most variance (more than 5 or 6 wins):

Charlotte Bobcats

Washington Wizards

Cleveland Cavaliers

Toronto Raptors

New Jersey Nets

Portland Blazers

Utah Jazz

Dallas Mavericks

Houston Rockets

Denver Nuggets

Los Angeles Clippers

Golden State Warriors

Sacramento Kings

I could get into this spreadsheet. Thanks.

Sorry, about the comments going straight into moderation. I’m not sure why it happened with these comments but not others. Maybe the word Vegas triggers the spam blocker or something.

I have to be honest, that several of these predictions don’t exactly jive with my “gut” feeling. My gut tells me Cleveland is not going to be as good as the numbers predict. I think it’s just too hard to account for the inflation LeBron provided to everyone else’s production and efficiency. It’s also hard to imagine the Wizards really being that bad. You have to imagine that there will be some close games that Wall will be able to turn into wins simply with athleticism and force of will. Golden State has struggled so much recently it’s difficult to imagine them making such a huge jump. Portland’s win prediction falls apart completely if Pryzbilla and Oden don’t come back strong, or if Camby goes down with an injury.

If I had made the predictions based simply on instinct they would have looked VERY different. I tried to come up with a mathematical way of making my predictions, and rudimentary and flawed as it is I went with it. If anything this will be a good learning experience for me and I’ll have something more developed and comprehensive ready for next year.

I didn’t necessary mean to imply, if I did, that variance from the line was bad. I think it is worth thinking about. Applying instinct after viewing the results of a metric is what I have done in the past but unless you do a lot of tests I don’t think you can declare either pure metric or two-step objective/subjective an “approach” winner.

Pingback: Fear the Deer? and Forecasting the NBA | The Wages of Wins Journal

Pingback: Podcast Predictions « Nerd Numbers the Blog