Interview With Neil Paine Part I

"Statistically, he [Marcin Gortat] looks a lot like an Andris Biedrins or a Chris Andersen type, in terms of being a low-usage big man with great efficiency (he's got an eye-popping offensive rating of 123.7 right now, using 14.4% of Orlando's possessions when on the court). His defense -- specifically the block rate (4.8% of opponent 2FGA when on the floor) -- is closer to Andersen than Biedrins, though, and that's also good news for the Magic. I say he eventually becomes the Polish version of the Birdman, except without the drug problems, which would make for a pretty nice all-around player."

-- Neil Paine, Basketball-Reference

Readers of Third Quarter Collapse should recognize Paine's name, as he was the gentleman that helped create two what if scenarios with former Orlando Magic player, Grant Hill. The first scenario he ran (over at the BBR blog) involved Hill individually and the second scenario he ran (here at 3QC) involved Hill collectively with his former Orlando teammates. In case you haven't been able to read either of the articles, I highly suggest checking them out. They answer a lot of questions for Magic fans, with regards to 'what could have been' had Grant stayed healthy, in his prime, during his tenure with Orlando. 

I'm a big fan of Neil Paine and his work at Basketball-Reference. Paine does a ton of excellent stuff with advanced statistics, not only at the BBR blog, but on the APBRmetrics forum, and elsewhere. I recently was able to chat with Paine on a variety of subjects, some stat-related and some Magic-related. 

This interview will be a two-part series (Part II will be revealed on Wednesday).

In Part I of the Q/A, Paine shares his experience working for Basketball-Reference, provides some insight on advanced metrics like statistical plus/minus, shares his plans on his future as a statistician, and more.

Enjoy.

Click after the jump for the full transcript.

So, how long have you been contributing and helping run Basketball-Reference?

Only since last October, although I had worked with Justin on side projects and had a cordial relationship with him since 2005. Sports-Reference hadn't really tried to develop and market the BBR blog until right around the beginning of this season, so that's when they approached me about the job and I started officially working for them.

I understand that, blog-wise, it's you and Justin Kubatko that post on the BBR blog. Am I missing anyone? Likewise, how many people combine forces to make Basketball-Reference what it is today (one of THE most comprehensive statistical databases available on the internet)?

Yeah, it's basically just us (with the occasional post from Sports-Reference founder Sean Forman, of course). Justin's kind of the point guard of the blog operation: he does a ton of dirty work behind the scenes and "assists" me if I ever need a certain database query, and I "finish the play", so to speak, by researching, analyzing, and writing about the data. As far as the main site goes, Justin is 100% responsible for that content -- he has expended a lot of time, money, energy, and coding skills towards making BBR the absolute best basketball statistical site on the internet, and it has really paid off for the entire basketball community. Really, my only role there is that he'll sometimes pick my brain about new features and statistical categories.

I'm curious, as time progresses, will BBR continue to add more advanced statistics to every player page, as long as permission is allowed from the creator of said metric (ex: Kevin Pelton & WARP)?

I think we're always going to tinker with the advanced stats section and add new and better metrics as we find or invent them. That said, I don't think it's necessarily going to include proprietary metrics (like WARP) that people can readily find on great sites like Basketball Prospectus, especially when we already have a similar stat in Win Shares. But we will definitely keep adding categories and looking for better ways to express a player's particular skill set (passing, ballhandling, etc.) in the advanced stats section. Also, I'm pretty confident that similarity scores will make a triumphant return at some point in the future.

A month back, you indicated in a BBR blog post that you began fooling around with statistical plus/minus, which is a regression Dan Rosenbaum ran between "pure" plus/minus and box score stats. What specific things did you do in tinkering with the metric?

I was aware of Dan's +/- work from nearly the beginning, but I've gained a lot more appreciation for it in recent seasons, as I came to see the structural problems with a lot of box score-based player ratings. One thing he did that I liked the most was to connect "pure" adjusted plus/minus (which uses no box score input) to box score numbers, which I felt had a nice application when it came to historical players. Right now we don't know what Larry Bird or Magic Johnson's pure adjusted +/- was, but we can estimate it using some of the regressions he developed. So my tinkering was just to take the raw scores and make them fit team point differential, which helps add a defensive component and puts the results on roughly the same scale as the +/- you find at a site like BasketballValue.com.

What type of further refinement do you feel is necessary for statistical plus/minus to be even more accurate? What do you feel are the differences with statistical plus/minus, compared to adjusted plus/minus (which is an improvement of raw plus/minus)?

I think it can definitely be refined by adding additional variables like height, weight, age, and experience, all of which were absent from the initial regression. It's all about finding factors in the traditional stats that better reflect a player's impact on point differential, and I think things like height and experience in particular -- combined with the rest of the player's stat line, of course -- can help shine a light on some of the "hidden" value that isn't captured in the box score. But that's still the major difference between statistical +/- and pure APM: it has to rely on the box score, which, as you know, doesn't count as many actions as we wish it did.

 

On the other hand, the biggest advantage of SPM is that it has a lot more year-to-year consistency than pure APM does. For instance, we had a recent controversy because pure APM had Chris Paul rated as merely decent last year, and it sees him as being the league's 3rd-best player this year. By contrast, SPM hasn't really changed its stance on CP3 since last season, the stance being that he's probably the 2nd-best player in the NBA, behind LeBron James.

What are your current credentials at the moment? I know that NBA teams are like hawks, waiting to prey on available "stat geeks" with one felt swoop .. have you've been approached or collaborated with any NBA teams at this point in your career? If not, would you like to? If yes, why? If no, why not?

In terms of my credentials, I plan on graduating with a degree in STaC from Georgia Tech this summer, and I have a ton of experience in online sportswriting, which incidentally fits in really well with our blog right now. Like just about everyone in this business, I've been approached by an NBA team about work, and I would definitely be open to working for an NBA organization in the future... but I confess that I'm torn because I love the writing aspect too, and that's something which almost always falls by the wayside when you work for a team.

 

In a lot of ways, the APBRmetric revolution is markedly different from the Sabermetric one, because they did theirs from the outside, with great writers like Bill James and Rob Neyer pounding on the door of the establishment until somebody let them in. In our sport, it's almost the opposite -- by far the most innovative work is being done inside the establishment. And don't get me wrong, I think that's great. But if Bill James comes up in that environment, he gets snapped up by a team the second he develops Runs Created, and we lose probably 3/4 of what he wrote in real life. So, like most things in life, it's a give and take.