What factors are most predictive of a game's outcome for the Rockets? A multiple/logistic regression

So the key to success is playing better defense, shooting better, and minimizing turnovers. Who would've thought?

Cool post, though.

 

For you stat heads take a look at this site:

http://www.sportsclubstats.com/NBA.html

Amazing stuff. I have no idea how accurate his models are. I am going to look into it further.

 

This may be a stupid question, but are you using in-game statistics as the independent variables, or the team and opponent season statistics? That wasn't clear to me. If the former, then its of course trivial to develop a 100% predictive model for whether the team will win or not (just points scored and points allowed).

 

Trade and injuries is definitely harder to normalize, but back to should be pretty easy to use as an independent variable on a game.

 

Yeah, which is why I think we should use opponent's TS% and our TS% at each game vs. Opponents avg offensive and defensive TS%.

 

Hmm, maybe then use TS% interacting with a percent of FGs 3pts variable?

 

The team that gets to the line and attempts the most FT's, usually wins the game. There are several reasons for this:

1. It puts the other team into a foul situation which breaks down their defensive intensity.

2. It increases the team's FG % by allowing shooters to get into a rhythm at the FT line.

3. It slows down the opponent's tempo, so that they are forced to play a half court set with defenders already set.

 

paging Daryl Morey

 

Maybe hollywoodMarine IS Daryl Morey. :grin:

 

I love stats myself and am only average with them, but I just want to say the usefulness of the stats has to be taken in the correct context. I don't know if the math has issues (I was too lazy to go through it myself) but a lot of people may read the results as unique to the Rockets.

I'm guessing the importance of the ranking isn't very useful as the results just show how FG%, oppFG%, and turnovers are inherently the most directly correlated stats to the generation or reduction of points per possession in a zero sum basketball game which should be expected (zero sum applies because comparison of points determine the winner of a game, generating more ppp and reducing your opponent's effective ppp have equal value, and there is a parity of limited resources (possessions in turns). This weighting of a player's total contribution to both teams' ppp is something many advanced stats take into account and attempt to do - for instance, PER.

So someone might ask, isn't the OP's math at showing that the Rocket's team defense has more value than their offense? Another guess of mine is that too might not be the case. A single team's offensive deviation is going to remain more stable because you have the same guys shooting in the same systems. My guess is the impact of the opponent's defense is less than impact of facing teams with different avg ppp every game, especially over the course of a regular season. Saying defense wins championships may be true but probably not something that can be extended to postseason lightly from regular season conclusions.

Let me try to say it another way with some mental math. There are roughly just under 100 possessions per side in an NBA game and 100 points per side, so 200 total possessions and 200 total points (which is roughly 1 ppp btw).

(references: http://www.teamrankings.com/nba/team-stats/)
Team Avg TS%: ~55% (composition mostly 2pters, but 3pters for high volume 3pt teams)
Team Avg TO per game: 15 (each TO is ~2% comparison of possessions, the larger the margin is)
Team Avg Stls per game" 7 (so about half of TOs)
Team Avg Reb differential between+/- 5 reb (also around the same impact as TO)
And then the idea that a team's offense is more regular and opposition offense is more non-regular.

Given the above, I could quickly ballpark and rank, the impact of stats vs. ppp for any given team:

Opp2FG%,2FG% ,Opp3FG%, 3FG% (all around +/- 0.55 and order depending on volume of 2pters vs. 3pters, but defensive metrics impact higher)
TOV and oppTOV around +/- 0.30
Steals around +/-0.15 (half of TOVs)
Reb around +/- 0.11 depending on differential
Other stuff that impacts ppp (PF * % of PF that result in FTs * FT% => FTs)

Which is pretty much the ranking and values of ratios the OP found.

My guess is all teams have the roughly same ranking of statistical impact on wins/losses and in roughly the same ratios. So the value of what the OP found is more exact as to how much more or less these things contribute to wins or losses for the Rocket's system. The ranking is expected. A good follow up would be to compare these to other teams or the league average. It is also the start of something like a team-specific PER stat, where you can match up and see if a certain player would fit the Rocket's system well.

 

Great post, very informative.

 

I thought from the last incident that it was commonly understood that amateur statisticians would need to submit their work to me prior to posting for peer review?

This "analysis" is the poster child for why that step is necessary. Where is the detail? Where are the graphs? A regression analysis without graphs? Is this even real life?

Tread lightly or be vanquished like my past rivals.

 

Those were the things that jumped out to me at first. Comparing to a league average or top performing teams (MIA, SAS, OKC, etc.) would be a natural direction. In my interpretation, your results more directly describe what kind of system a team runs (which then can be compared to win/loss), although I'll be honest again in that I only skimmed the work. It would be a great way to forecast team strategy vs. other teams head-to-head (ie. Detroit Pistons rely on 2pt% to generate wins, so usage of Dwight + Omer is key this game) and also see if a FA or trade prospect plays to the Rocket's systemic strengths/weaknesses (ie. Ryan Anderson scores a 85% compatibility (per dollar analysis could be easily tacked on here) as his strong points correlate well with Rocket system impact areas vs. average of 75% for next best option). Definitely a lot of cool extensions you could go from here!

 

But do your fancy pants equations take into account....

<iframe width="420" height="315" src="//www.youtube.com/embed/5-1jgNhopNo" frameborder="0" allowfullscreen></iframe>

....the heart of a champion?

 

I am new to this forum I have lived in the Houston area my whole life! I'm usually just creeping on this forum but after reading your analysis I decided I would comment! I have multiple bachelor degrees and really enjoy statistics they are even more enjoyable to do when you get paid to run statistical analysis for companies on the side (like me ). I just wanted to point out this is a very good start and I am impressed on the data you were able to collect and run. It is always good in my personal opinion to run statistical methods on topics you truly enjoy and understand. It makes the statistical process be come much easier and more fluid! The best information I was given while learning stats and doing my own beginning research analysis is no matter what the hypothesis you are trying to research always attempt to prove your hypothesis wrong along the way. When you do that then guys like me who go in to disprove your theories have a much harder and more difficult time. Meanwhile you will find that the statistical methods and analysis you are trying to use to disprove the hypothesis will end up proving the hypothesis!... However, you could completely disprove your hypothesis and end up throwing out all the research because the original hypothesis had flaw in the design... and you will have to start all over from scratch.... which sucks! lol However, if you enjoy stats just stick with it and you will found out some very interesting facts over MANY different topics.