NBA League Size and Competitiveness, Part One: Introduction

Thanks to a generous friend, I've recently begun reading Bill Simmons's colossal The Book of Basketball.  I say colossal because, as you may not be aware, the book is about as long as Anna Karenina.  It's long, it's big, and so far, anyway, it's extremely entertaining.  A significant portion of the book seems to be Simmons - perhaps better known simply as "The Sports Guy" - taking digs at Vince Carter, Kareem Abdul-Jabar, and Wilt Chamberlain, whilst trumpeting (who else, as a kid who grew up in Boston?) anyone who played for the Celtics, and especially Bill Russel.  Which is all very fun.

Anyway, in the first few chapters, Simmons has already made a point - well, he's made many points - with which I disagree.  He is a firm believer, it seems, that expansion has diluted the NBA, and that the league's competitiveness was much higher when he was a kid.  "Back in my day," he never says, but might as well, "Basketball players had to try harder, because every night they played against teams filled with All-Stars."  Now, "my day," in this case, refers to the Russel era of the late 50s and early 60s, when the Boston Celtics - despite the huge competitive balance of the NBA at the time - won eight championships in a row (and nine out of ten, and ten out of thirteen).  If only we could have that again!

In all seriousness, though, Simmons does an excellent job describing what makes for success in the NBA, and, frankly, he knows way more about it than I do.  He points out - and rightly so - that basketball statistics are deeply flawed, because they don't capture the magical things that allow teams to win games.  I would say that Simmons is right: points scored, assists, rebounds, blocks, and steals do not accurately measure a player's contribution to his team.  Not even close.  That doesn't mean statistics have no place in basketball, it just means that basketball statistics have to get better and, what's more, that might be impossible to do because unlike in baseball, a team's success in basketball has more to do with how teammates work together than with how individuals perform.  (Inhales).  The linchpin of Simmons argument, here, is that Wilt Chamberlain - for all his statistical dominance - was a terrible teammate who's teams rarely won championships, while Bill Russel was actually a better and more valuable player, as evidenced by his bevy of MVP awards and Championship rings.  And you know what, I buy it.

I still don't buy, however, that the modern NBA is somehow watered down compared to the NBA of the 60s, and I do think that statistics can demonstrate why.  A while back I explored how NBA rosters are constructed, using Win Shares, and discovered that teams, as a whole, follow a highly predictable model.  That model, to rehash, is that the average "best player" on a team accumulates 9.3 Win Shares in a season, and each subsequent player accumulates less in a logarithmic way.  The stunning result was, at least for the season I looked at, a correlation coefficient of exactly one.  League wide, there's a very strong trend towards a regular distribution of success on the court.

What does this have to do with competitive balance?  Not much, but I want to point out that it jives well with the qualitative description of successful NBA teams that Simmons gives in his book.  He argues that teams need a great player, followed by a couple all-stars, followed by some key role players.  If you look at my old post and the graph with the Lakers and Celtics, it's easy to see that their model fits well with that description.

Now, I bring this up because Simmons points out how many All-Stars were on the Celtics and their rival Lakers and Warriors back in the 60s.  The teams were stacked, he tells you, replete with great talent.  Not like todays teams, where many teams are lucky to have even one All-Star.

Of course, the easiest hole to poke in this argument - that teams had more All-Stars back in the 60s - is a direct result of a smaller league.  Not because the talent level was necessarily higher, but because there were fewer players from which to draw an All-Star team.  Of course the Celtics had a bunch of All-Stars in the 60s, because the league only had eight teams.  That means that, even if every team was equal, an all-star roster of 12 would mean taking three players from each team in both of the four-team divisions!  Since the Celtics were also the best team in the league, it's only reasonable that they would have four or five All-Stars in any given season.

Compare that to today's league.  With 30 teams in the league, it's hard to have even two All-Stars from the same team, because an individual player has to out-shine so many others.  What's more, a second (or third) best player on a given team is going to have an even harder time, because he has to look better than the best player on many other teams, not easy to do given the limited and flawed statistics available in the modern NBA.  I realize that may be a bit opaque, so let me clarify using the simplest example: points.

Consider two teams that score 100 points per game.  On Team A, King Star scores 25 a game, whilst his brother Duke Star scores 20 a game.  Thing is, Duke takes way fewer shots, because he's more accurate, and is generally just a more efficient player than King Star, despite King's gaudy numbers.  Now, King Star is a perennial All-Star and fan favorite, and he's still plenty good, so he's going to the All-Star Game no matter what.  Duke is on the cusp, especially because Team B - which also scores 100 a game - features Selfish McGee (also known as Allen Iverson), a player who plays the same position as Duke, but scores 30 points a game in twice as many shots, thanks to a higher-paced offense and a team that has no other reliable scorers.  So Duke Star, in order to make it to the All-Star game, has to outplay either King or Selfish in the eyes of the people who make these decisions.

Of course, that's no different now than it was 40 years ago.  What's different now, instead, is that Duke is up against his equivalent on 14 other teams (whether they be like Duke, like Selfish, or like the heretofore unmentioned Crappy Sullivan), instead of 3.  Suddenly Duke, who's just as good - maybe even better - than the number two guy the Celtics had back in 1962, doesn't even make the All-Star game, while he would have been a shoe-in, at least as a backup, back in the 60s.

Phew.  OK, all that out of the way, let's actually get to the point of the post, which is how to actually assess whether the NBA is more or less competitive now.  How do we do this?  Is it best to look at players or teams?  What statistics should we use, in order to compare against eras?  In fact, there are many ways we could study the question, but the easiest and most intuitive, to me anyway, is simply to look at wins and losses.  I'm struck by a sentence in The Book of Basketball, which goes something like this: "I'm telling you, everyone had a good team back then."  Now I know what Simmons means is "All the good teams had good teams back then," because he knows that, even then, there were cellar-dwellers.  The reality is, every game played has a winner and a loser, and one of the constants in all sports is that, league wide, the average winning percentage is always exactly .500.  It goes without saying that, in order to get to .500, there will always be some teams that are much better and some teams that are much worse, and some teams that are right about in the middle.

What do we make of the claim, then, that everyone had a good team?  Well, what I think Simmons means is, there were more - a larger group of teams - well above average, and fewer really bad ones (and, likely, fewer really great ones).  He might phrase that as "more great teams, fewer average ones," but that's just a perceptual thing.  We live in a time where "average," in sports, has come to mean "bad," and "mediocre" has come to mean "absolutely terrible."  Ironically, "terrible" is something we don't actually dislike: the Timberwolves are terrible, but in a lovable kind of way.  It's mediocre teams we can't stand.

Anyway, how do we test whether or not everyone had a good team, given that we think it means that there was better competitive balance, that fewer teams were terrible, and fewer were so good that the games weren't even worth playing?  Well, there's a pretty easy - if tedious - way, that does not require digging into the deeply flawed player statistics of the 60s.  We can, in fact, compare across eras and leagues easily - as Simmons does when he says that the modern NBA is watered down compared to the old NBA - using wins and losses.  It's simple, really.  We just need to look at standard deviations of Win-Loss records throughout NBA history, and we'll see when the NBA has been at its most competitive.  In short, smaller standard deviations means the league is more competitive, while larger ones mean that the league is less competitive (more top and/or bottom heavy).

Now, there are some concerns here.  First off, those old leagues were so small that our sample size is going to be tiny.  Standard Deviations don't mean a lot when you're talking about 8 data points.  That is, they don't mean a lot if you're trying to be predictive based on only 8 data points.  But, in this case, I think we'll be fine, because we're just trying to deduce how "spread out" the quality of teams has been throughout NBA history.  Standard Deviation is exactly the statistic we want to use.  Since we'll be able to get a broad view of competitiveness, we'll be able to take the first steps towards assessing the competitiveness or watered-down-ness of the NBA across eras, irregardless of silly things like small league sizes making it easier to win championships (because, hey, fewer opponents) or make it to the All-Star game.

I honestly don't know what I'll find in doing this, even though my hypothesis is that the modern NBA is, if anything, more competitive than the NBA of the 60s.  I might be wrong.

As an extra outlet (for both me and Simmons), I'll also calculate the mean and standard deviation of the smaller set of "good" teams in the league.  I haven't yet decided how to draw this line, but I'm initially thinking that anyone above .500 makes the cut.  Basically, if we find a relatively constant standard deviation across time, we'll still want to test is maybe, in certain eras, the "good teams" are more evenly balanced with each other.  Now, this will be built into our bigger SD calculation, but we'll be cutting out noise like a team or two that finishes with a winning percentage of .130, and thereby makes the whole league's SD look way bigger than it is.  Indeed, I think Simmons would agree that the occasional really really bad team shouldn't count against any assessment of the competitiveness of the league as a whole, and so we'll do a parallel calculation that cuts out those really bad teams.

So to recap, here's the method: I'll be going through every season of professional basketball on basketball-reference.com (oh the wonders of being unemployed), and putting every team's W-L record into a spreadsheet.  From there, it's easy to calculate mean (which will always be half the games in the season) and standard deviation of wins per league per year.  The lower that SD, the more competitive the league.  I'll also pull out just the above .500 teams, and run the same calculations, to see if maybe there was more competitiveness amongst the good teams than in the league as a whole.  Finally, I'll do a smaller cut of outliers, removing just the really really bad teams (teams more than 2 SDs from the mean), and recalculate the league without their nefarious influence.

What will I find?  You'll have to come back to my next epically long blog post to find out, because I don't know yet.