Modelling the mvp voters

[browning]

Wasn't sure whether to put this in the history research or metrics forum.
But a couple weeks ago I decided to run an experiment to use a logistic regression model to model the way that voters choose the MVP.
It looks at individual stats + team record to pick the MVP each season. It was pretty accurate and it highlights some interesting years where the MVP was controversial.

Pretty fun to look back and think about how dominant some of those bigs were in the 70s

http://browning.github.com/2012/05/24/n ... egression/

[DSMok1]

Cool, nice job! I presume the percentages do not sum to 100% for each season--what seasons had the most "MVP odds" when summed? I.E. some season probably had 200% total among the players...

[browning]

Yea the numbers are 0-100 per player, they don't add to 100 for the season.

I think the best year was 1971-72. But its tough to say because I have some ABA players in the database during the 70s as well.
That year Kareem put up 35 and 17
Jerry west put up 26 and 10 assists
Both on great teams, plus some great ABA guys that year like Artis Gilmore.
Was an awesome year for basketball.

[kpascual]

Nice analysis! Love the simplicity of your analysis, but I may be biased because of your use of 1) R 2) Github and 3) Ubuntu. Open source FTW.

How's the machine learning course going? I've signed up and done the first couple of weeks, but haven't had much time to keep going at it.

[jimhk1975]

Your "Year" appears to be off by one? i.e. e.g. what you have as 2000 should be 2001, etc.

[bbstats]

Wow! This is great!

I'd love to see last year's MVP race. My bet is Lebron/Howard/Paul would win by the model, but of course D-Rose won.

[Mike G]

The only one of those guys to not actually win the MVP was Artis Gilmore. Its a shame because his numbers in 1971 were ridiculous! 24 points 18 rebounds 5 blocks on a team that won 80+% of their games.

As jimhk noted, the year is off by one. Artis did win MVP in 1971-72, in the ABA.
Julius Erving also won in '75 and '76, "as predicted", but also in the ABA.
You might fix the "years" and weed out the ABA guys -- better yet, run them separately.

I think the Cowens MVP (1973) might be the worst choice ever. Not only was Kareem great, that's the year Nate Archibald led the league in scoring and assists.
In 1977, Bill Walton missed a bunch of games. Did you find (or look for) any correlation with games played?

Maybe you could include a column for annual rank (in your prediction metric) of the actual MVP winner. Iverson and Nash may not have been top 10 in their winning years.

Nice job, for a first go-round!