Note: Due to the extremely graph heavy nature of this article, as well as the aspect ratio of the graphs, this one is definitely best viewed on a device other than a phone. Also, it is recommended to click on the graphs to zoom in as necessary.
In my never-ending quest to rid the world of garbage sports cliches, it’s time for “rookie wall” to take its turn on the chopping block. The basic idea is that at a certain point in the season, rookies unaccustomed to a long, physical season filled with travel, accumulate an excess of chronic fatigue that causes their performance to drop. Additionally, veteran players allegedly do not experience this performance decrease, based on their prior experience with the physical and mental realities of an NBA season.
Game logs for all NBA players were acquired for the 2015-16 through 2017-18 seasons. Rookies were identified if their first season in the NBA was that specific season. Each game for each player was assigned 1-82, so each game was put in place across the 82-game season, regardless of how many games that player actually participated in. Several different metrics were used in the analysis; each will be discussed more in-depth in their specific section, but the following process holds true across the board.
Game statistics used were Game Score (GS), Defensive Rating (DRTG), Effective Field Goal % (EFG), True Shooting % (TS), and Usage % (USG). Keep in mind for DRTG, a lower number corresponds to a better defensive profile.
My personal definition for a rookie wall: a section of time that is consistently and meaningfully worse across multiple statistics, compared to either season long averages and/or the preceding weeks, that is also not present in veteran data.
Season Time vs. Individual Time
There are two possible time metrics for this analysis. The first, season time, simply numbers each game in the season 1:82 and assigns each game played to its respective place in the 1:82. For example, if Player X’s first game of the season comes on the 15th game of the season for his team, his statistics for that game are assigned to 15. However, if we use individual time, Player X’s first game is assigned to 1. Season time puts performances into essentially calendar time, whereas individual time simply places games in the order they occur. Unless specified otherwise, all graphs and data are in season time.
The players who contributed box scores to the data were divided into three categories. The first was all veterans (not their first NBA season), the second was rookies who finished Top-5 in Rookie of the Year voting (except Joel Embiid due to games played), and the third was all other rookies. Keep in mind that Karl-Anthony Towns, the ROY winner in the 2015-2016 season, contributed as a veteran in the following two years as well. The same holds true for other rookies in 2015-2016, and then 2016-2017 as well.
Three separate metrics were used to examine the data over the course of a theoretical season. The first was simply the raw numbers, the second was the difference from the mean, and the third was the first derivative of the raw numbers. The difference from the mean identifies if the performance was above or below the season average, and the first derivative indicates the direction of change of the statistic.
Game Score (GS)
Defensive Rating (DRTG)
Effective Field Goal % (EFG)
True Shooting % (TS)
Usage % (USG)
That’s certainly a huge wall of graphics. Now we are going to (by game) identify for each statistic if the mean difference is +/- and if the first derivative is +/-. Usage Rate is excluded for this section, as lower or higher usage does not necessarily correspond to better or worse.
Interpretation & Conclusion
It seems that if you isolate approximately games 45-55 (season time), the average ROY level rookie performs worse in comparison to his baseline than does a veteran player, whereas 55-65 appears to be the sweet spot for average non-ROY level rookies. When you flip the metric to individual time, games 45-55 are the nadir for both types of rookies (likely because ROY level rookies just play more games).
This is a generalized approach, but I think one that provides an indication that something is happening in the previously mentioned ranges. I think it’s important to note that I did not account for meaningful difference in this piece. If a residual was -0.00001 or -100, it was put into the “bad” bin. It’s certainly something that I could look into for a future piece. With the evidence gathered, I’m reasonably comfortable agreeing to the existence of a rookie medium incline, if not altogether a hard and fast boundary.