Today's mainpage link of BoxScoreGeek.com's Andres Alvarez features a statistical examination of LaMarcus Aldridge. Over the course of the article, Alvarez challenges the notion that LaMarcus Aldridge is a legitimate MVP candidate. He also goes on to say some very nice things about the rest of the team. However, Andres's core argument is that LaMarcus's True Shooting Percentage (TS%) disqualifies him from MVP consideration.
You can see my counter-argument here, but once I was done deconstructing his position, I realized that I wasn't offering any alternatives. Not that I had to, but now my wheels were turning. Was there a way we could objectively measure what it means to be an MVP? I won't say I have the answer. but I do think I have an answer, based on my personal understanding of what it means to be an MVP. Below I break down how I arrived at what I call Quantifiable Value (QV), then I apply these measures to both the Blazer starters and star players from some of the best teams in the league today.
Step 1: Building a Metric
What defines value? Is it how many points a player scores? Making the players around him "better"? You could spend all day spinning around this initial question, but I felt strongly about a couple of things coming into this process. The first is that I wanted to be able to capture both offensive and defensive contributions equally. The second is that I wanted to find a way to distinguish how an individual player's effort contributed to the overall success of the team.
Although I am no expert on advanced metrics, I settled pretty quickly on the notion of looking at a player's on/off splits for both their own team's ORtg and opponent ORtg over the course of the season. This should have the general effect of telling us how much better the team performs as a whole when that player is on the floor vs. when they arent. It should allow us to measure the contributions of great scorers and great defenders equally. It should be able to represent the benefits that are generated by assists and rebounds, and by avoiding turnovers and fouls. Fortunately for me, Basketball Reference will happily provide me with these exact on-off metrics on a single player basis. This might lead to problems later on when trying to rank players against each other, but its a good start.
Producing an on/off differential is not especially new or innovative, and I don't think that this value by itself can tell us who an MVP candidate is. What we've got so far is an ability to measure the on court impact of a specific player... but it doesn't automatically follow that impact = value. Lets think about Greg Oden for a moment. Here's a guy whose size and talent causes time and space to bend around him when he's on the floor. He's a scary dude. As of today, he's also not good for more than 6-8 minutes a night. Sure he has a real impact during that time, but how much does that benefit the team overall? For an impactful player to really benefit his team, he has to be on the floor providing real help in real time. So the next step is to take the player's differential and multiply it by the % of the game that the player is on the floor.
Now we're starting to get a real feel for how much contribution a specific player is providing for his team, but the whole thing is still a little bit abstract. A true MVP shouldn't be a statistical anomaly. Any decent player on a team full of practice fodder will be the "most valuable" relative to the other guys on his team, but MVP isn't about being the best guy on your team. It's about being the "most valuable" player in the entire league. Therefore I feel that we have to consider how much the player's value contributes to wins and losses. This helps us to filter out stat-hounds who fatten their personal numbers up on bad teams. It also helps to quantify the idea that it's much harder for a good team to get better than a bad one. So we'll go ahead and take the numbers we already have and multiply them again by the team's Win percentage. This provides us with a final formula:
[ (On-Off Ortg) + (Opp On-Off Ortg)] * MP% * WIN%
Step 2: Running the Numbers
Let's start with LaMarcus Aldridge. When LMA is on the court (in the 13-14 season), the team has an ORtg of 114. Off the court, the team's ORtg is 107.8, for a differential of +6.2. Opposing teams have a ORtg of 106.1 when he is on the court, and a 112.2 when he is off it, for a differential of 6.1. So LaMarcus's contributions offensively and defensively appear to be pretty similar, giving him a total differential of +12.3.
LaMarcus has played 76% of all minutes during Blazer games, and the team currently sports a winning percentage of .723. Putting it all together we find LaMarcus Aldridge has a QV of 6.76. Since this is an index of sorts the number by itself is pretty much meaningless. It's only relevant in comparison to other players. Let's look at the QV for the rest of the starters and see how LMA fares:
LaMarcus Aldrige: 6.76
Nicolas Batum: 4.60
Damian Lillard: 3.09
Robin Lopez: 2.27
Wesley Matthews: 1.95
As you can see, LaMarcus has the highest QV of all the starters and it's not even close. Given that win pct. is the same for all players, and the MP% is not all THAT different (LMA has the highest with 76%, Robin the lowest with 64%) what makes the difference? In looking at the individual numbers, it's all about defense. Nobody among the starters forces the opponent ORtg lower than LaMarcus when he's on the court, and he does so while also sporting one of the highest offensive differentials. Combined with his #1 MP% and you get an out-of-the-park QV.
Now let's compare LMA to his MVP competition. I arbitrarily selected one player off the top two teams in each conference:
LeBron James: 6.99
Paul George: 5.06
Kevin Durant: 2.99
Tony Parker: -1.07
Hmmm. Interesting. We've got some mixed results here. I think this is a good time to move onto our final step.
Step 3: Evaluate Results
As I was building my model initially, I knew I would run into two particular problems. The opponent ORtg numbers suggest that Parker is a defensive liability. Enough to completely offset his offensive bonus, resulting in a negative number overall. Here's where my formula breaks down. Because MP% and W% will never be higher than 1, the formula moves all QV ratings closer to zero. So we actually made his negative number look smaller aka better. I know there are some mathematical approaches that can be used to cancel this effect but that is totally over my head. Regardless of whether or not this method underrates Tony Parker in general, it skews the results.
The bigger problem is that in the real world, on/off splits are not solely a measure of one player. On/Off tells us how much better a player is than the guy that replaces him when he's resting on the bench. LaMarcus and LeBron benefit greatly from having reserve players behind them that cannot hold a candle to what they do. Kevin's differentials are surprisingly thin compared to his MVP competition.
Now this doesn't mean that QV is a broken stat, but it does raise some questions. Does the strength of the OKC bench make Durant less valuable? Closer to home, would having a healthy MLE-level reserve PF make LMA expendable? The answer depends on whether you believe that strong/weak reserves represent noise in the results, or whether the talent leap between a starter and a reserve player is exactly why you value that starter in the first place. At this point I don't have an opinion. I would be very curious to know what the more veteran stat monkeys around here think about "QV" and whether it points us in the right direction.