Picking a starting lineup, finding the best in-game rotations, and identifying when to switch things up are some of the hardest decisions a basketball coach has to make. Our lineup focused approach to basketball analytics makes this process a bit easier. As in the previous post, we encourage our readers to check out the data for themselves.
In this week's post, we're diving into the lineup data from Duke and selecting, based on the results, the best 5-player lineups for the team as well as identifying a handful of best case scenario substitution strategies. The key metrics in analyzing these lineups are offensive and defensive ratings and their resulting net rating. Additionally, we take into consideration the total # of possessions played to account for small sample size problems.
The first question is whether Javin DeLaurier or Marques Bolden is a better fit to join Cam Reddish, Zion Williamson, RJ Barrett, and Tre Jones in the starting lineup. We can start with a basic comparison of how those two five-player combinations performed throughout the season. Using the Lineup Analysis tool, we can filter using total playing time or plus minus, and easily drill down to visualize the two lineups of interest.
This snapshot gives us the ability to quickly compare the totals and effectiveness of these two lineups. After reviewing the data, the initial answer is clear (Plus-Minus Points Per Possession is the metric to review). The lineup with Marques Bolden is slightly more effective per possession (~0.3 net points per possessions > ~0.2) and each lineup has 300+ possessions played together, so there is no sample size issue.
Additionally, we can look at the two player combination data for each of these pairs. Examining the performance of each player combination at a more detailed level provides more clarity and data to base a decision on. Using the Player Combination Analysis tool, we can identify the joint rating for each of the combos as well as the "synergy" score. The Synergy scores gives us a gauge on how much better or worse that combination performs relative to the individual player's net ratings.
By averaging their respective synergy scores for the combinations with Cam, RJ, Zion, and Tre, we can identify the relative positive or negative impact for each Javin and Marques. In this case it is Javin DeLaurier who looks slightly better with an average positive synergy score of 0.67 (Marques Bolden's = 0.58).
This discrepancy from the initial lineup comparison data indicates that we may need to go a step further on the lineup review. To get even more specific and accurate, we can compare the performance of these combinations between halves - as this helps us determine who is the better starting lineup versus who is better to come in later and keep up the stamina.
Once again looking at the Plus-Minus Points Per Possession as the primary metric to compare, we can see the lineup with Marques is more effective in the 1st half, but becomes less effective in the 2nd. Breaking it down by half reduces the sample we have to work with, but due to the large minutes that these lineups logged, the sample is likely still adequate.
In just a few minutes using the Pivot tools, we were able to more easily understand how the lineups with Marques and Javin compared to each other, as well as identify when each 5 player combo may be more effective. Using a combination of the player combination data and lineup stats can illuminate starting lineup rotations, but also can be used to identify optimal substitution strategies. Using the same methods as above, we can identify optimal 5 player lineups accounting for rest for each of the starters, using an 8 player rotation.
Primary 1st Half Unit - Marques Bolden, Cam Reddish, Zion Williamson, RJ Barrett, Tre Jones
Primary 2nd Half Unit - Javin DeLaurier, Cam Reddish, Zion Williamson, RJ Barrett, Tre Jones
Alex O'Connell, Marques Bolden, Cam Reddish, Zion Williamson, Tre Jones
Jack White, Marques Bolden, Cam Reddish, Zion Williamson, RJ Barrett
Jack White, Javin DeLaurier, Cam Reddish, RJ Barrett, Tre Jones
Jack White, Javin DeLaurier, Zion Williamson, RJ Barrett, Tre Jones
Of note, when using a data-driven strategy to analyze and generate lineup decisions, it is extremely important to consider playing time. Generally speaking, the data becomes more reliable as the sample size increases. But ultimately, a data analysis platform should not be the only resource used. Stats and data analysis alone will never replace a coach's intuition and intimate knowledge of the team. Lineup data is important as a way to test theories or review results, and should be a part of a modern coaching and scouting strategy, but tools like these are much less effective when used on their own and disconnected from the context and knowledge of basketball professionals.
Overall, we hope this analysis demonstrated how easy it is too review and compare lineups from an overall season long standpoint, compare individual players and their affects on their teammates, and the ability to drill into some situational data and identify the best lineup by half.
Continue reading... Part 3: Anatomy of Loss - Using Pivot Analysis to Learn from a Loss