The Brooklyn Nets obliterated the Golden State Warriors in the season opener and then, even more impressively, made minced meat of the Boston Celtics in TD Garden on Christmas Day. Kyrie Irving and Kevin Durant look happy to be back on the court. As a fan of the team, I’m not unpleased by these developments.
But I have the annoying “How real is this?” rattling around in the back of my head tempering my reaction to any small sample size. I know that weird things can happen over a short time period in the NBA. Outside shooting can ebb and flow as a prime example. The Basketball Gods fan be fickle.
Faced with this uncertainty, I like to try to quantify things. We know the Nets were projected to be a good, but not spectacular regular season team this year; sportsoddshistory.com had their win total over/under set at 45.5 games which translates into about 52 wins over a standard 82 game slate. Using this information (what I like to think of as a preseason prior) plus what we’ve seen on the court so far, what are our reasonable expectations going forward?
One simple way to combine the preseason prior with the on-court play is a linear regression. Let’s use preseason win total over/under and current season point differential as the independent variables. Here the dependent variable is the win pace for the team over the rest of the season. As an example of how to interpret this, if the model output is “45 win pace” that means that we expect our team to win the equivalent of 45 out of 82 games over their remaining schedule.
With this preamble out of the way, take a look at these projection lines:
Here I have plotted projections for our preseason 52 win team in four different time increments: after 10 games, 20 games, 30 games and 40 games. We see that the slope of the lines gets progressively steeper and this should make sense. A team’s point differential after 20 games has more predictive power than their point differential after 10 games, 30 games has more weight than 20, and so on.
Ok, now for the fun part! Let’s say we make it through 10 games of the season and the Nets are still dominating with a cool +10 point differential. We then consult the blue line in the graph and find that Brooklyn would be projected to play like a 55.5 out 82 win team going forward. That’s really impressive obviously, but not quite 60 wins level of impressive. We would want to have a bit of caution.
What about if they are still outscoring opponents by 10 points a game after 20 games? Then we can see from the blue line that we would update our view of the Nets higher still to a 57.5 win team going forward. If we make it through 40 games and the Nets are still +10 then this model sees them as a 61 win juggernaut.
I find it fun to look at a graph like this (which is probably a strange thing to find fun), but I should caveat this a bit. I projected what a hypothetical +10 point differential Nets team might look like, but this is on the high side of possible outcomes. Even though Brooklyn has looked like world beaters in the early going, we are much likelier to come back in a couple weeks and be talking about a +6.0 point differential team (which is still highly respectable).
Ok, two more important caveats. First, this type of analysis would be better if it took into account the strength of opponents played. In other words, something like basketball-reference’s Simple Rating System (SRS) would be better than point differential. The second caveat is that linear regression (and models in general) are better at making projections for more commonly seen input values. That is, the model knows how to handle a Nets team with a +5.0 point differential through 10 games better than if that team was at a gaudy +15.0, for example.
The reader may have noticed that I did projections for a team that has played at least 10 games already, not 2 games like the current Nets. I may be a coward, but I’m admittedly scared to use point differential to project after 2 games. Let’s sit back, enjoy the basketball at Barclays Center, and see where things stand in a few weeks.