How Predictable is NBA Pace?

Oct 7, 2015; Phoenix, AZ, USA; Sacramento Kings guard Rajon Rondo (9) dribbles the basketball up the court in the second half against the Phoenix Suns at Talking Stick Resort Arena. The Suns defeat the Kings 102-98. Mandatory Credit: Jennifer Stewart-USA TODAY Sports

Whether you are betting the NBA on a nightly basis or sporadically playing daily fantasy basketball, pace is a term you are undoubtedly familiar with.  Pace simply measures how many possessions a team is getting per 48 minutes.  With the recent addition of pace filters into the Bet Labs software, I decided to take a deeper look into how difficult it is to predict pace on a daily basis.

Using the game number filter, I only looked at teams that had played at least 20 games.  Numbers from very early in the season can be volatile so I avoided those in this analysis.

First I simply looked at three columns.  The home team’s pace, the visiting team’s pace, and the result pace for each game.  I first found the correlation coefficients for each input variable.  The home team’s pace coefficient was 38.8% while the visiting team’s coefficient was 42.5%.  Both inputs are positively correlated but the result has more dependence on the visiting team’s pace.  To put simply, the visiting team’s pace has more to do with the actual result than the home team’s pace.

Next, I looked at some common ways that people “calculate” pace for a game and compared to the actual result to measure the disparity.  First is simply taking the average of the two team’s pace values.  If one team has a pace of 98 and the other has a pace of 102, then this method would just set the projected pace to 100.  When taking the average pace and comparing to the actual result of each game, I calculated an average error of 3.5 possessions per 48 minutes.  While 3.5 possessions might not seem like a lot, that’s basically the difference between Golden State and Chicago in terms of average pace.

The next method I evaluated was again taking the average, but this time using ratios compared to the league average.  So if one team has an average pace (1.00) and another team is 5% above league average (1.05), then the projected pace would be 1.025 times league average.  Comparing that method to the actual result returned the same average error of 3.5 possessions per 48 minutes per game.

Next I took the same ratios but instead of averaging them, I multiplied them together.  So two teams above average would further increase the projected pace value.  For example a team 20% above league average (1.20) plays a team 10% above league average (1.10) would equate to a projected pace 32% above average.  Running this method against the actual numbers didn’t provide much improvement: an average error of 3.4 possessions per 48 minutes.

How about a regression?  Using the home pace and visitor pace as inputs to produce an equation should work, right?  After running a regression in Excel and comparing to the actual results, the average error was 3.2 possessions per 48 minutes.  This is an improvement over the other methods but still shows that there is a high variance in the calculation.  Again, this is the average error, meaning some games were exactly as predicted while others were over ten possessions off the projection.

What about overtime?  You might think that overtime games would really skew the data.  That’s why pace is measured per 48 minutes instead of per game as not all games are created equal.  So the important takeaway is to know the variance of your own pace projection and realize that your projected tempo of the game may have a larger range than you originally thought.

If you are interested in looking at all of these pace numbers and building your own systems using our new pace and efficiency filters, you can create a free account at

©2016 Sports Insights Inc.

Log in with your credentials

Forgot your details?