高阶数据能够衡量NBA球星的排名吗？

Are Advanced Stats Capable of Assessing the Rankings of NBA Superstars?

在2010年前后，NBA圈层经历了一次所谓的“高阶数据改革”，大量的诸如PER、BPM、RAPM、VORP、WS等五花八门的高阶数据开始主导整个NBA以及球迷群体对球星的评价体系。在此之后，很多重要奖项的评选开始高度依赖于这些高阶数据。用高阶数据来评判球星的历史地位也逐渐成为球迷圈层的一个热潮。最近读到了美国篮球评论赛道的自媒体博主Ben Taylor的一篇文章，他通过极为复杂的计算体系结合其对大量录像的观察，自己发明了一个被叫做CORP的评分系统，并且声称这个系统可以“客观地”平衡巅峰高度与职业长度，然后给出了他自己的历史排名（勒布朗第一、贾巴尔第二、乔丹第三）。

Around 2010, the NBA sphere underwent a so-called "advanced stats revolution." A wide array of advanced metrics—such as PER, BPM, RAPM, VORP, and WS—began to dominate how both the league and its fanbase evaluated star players. Following this shift, the voting for many major awards became highly dependent on these advanced stats. Using them to debate players' all-time historical standings also gradually became a widespread craze within fan circles.

I recently read an article by Ben Taylor, an independent American basketball analyst and content creator. Through an incredibly complex calculation system combined with extensive film study, he invented his own rating metric called CORP. He claims that this system can "objectively" balance a player's absolute peak with their career longevity, yielding his own all-time rankings (LeBron first, Kareem second, and MJ third).

当然，也有很多球迷并不买这些高阶数据的账。多数时候，他们会贬低那些喜欢用数据进行分析的人为“stats nerd”（数据怪咖），然后从“杀手本能”、“领袖精神”、“提升队友”、“关键球能力”等难以量化的角度来进行反驳，然后双方就变成了各说各话。对于这些人，本文从精神上赞同他们的大方向；但在具体的方法论上，本文却并不认同这种各自圈地、各自宣布胜利的对线模式。在多数时候，如果数学建模的结果和现实有偏差（比如高阶数据非常强的一些球员却总是缺乏赢球统治力），那是数学建模本身出了问题。只要有足够的建模洞察力，完全可以在这些stats nerds自己的领域把他们驳倒。

Of course, there are also many fans who do not buy into these advanced metrics. More often than not, they dismiss those who rely on statistical analysis as "stats nerds," countering with unquantifiable concepts like "killer instinct," "leadership," "elevating teammates," and "clutch ability." This usually results in both sides just talking past each other. While this article agrees with the general sentiment of these fans in spirit, it disagrees with their methodology—specifically, the debate style of retreating into separate camps and each declaring victory. In most cases, if the results of mathematical modeling deviate from reality (for instance, when players with phenomenal advanced stats consistently lack winning dominance), the flaw lies within the mathematical model itself. With sufficient insight into modeling, it is entirely possible to defeat these "stats nerds" at their own game.

为了分析各式各样的高阶数据的问题到底出在哪里，我们首先需要寻找这些高阶数据的共性。这看起来是一个让人望而却步的工作，因为搞清楚这些高阶数据怎么算的首先就挺麻烦了，更何况还有类似于CORP这种还要结合录像分析的数据。但实际上，大体了解一些这些高阶数据的核心逻辑，很快就会发现一个挺显著的事实：当前的几乎所有主要篮球高阶数据，在统计学上，都是在大样本的比赛数据的基础上进行的平均化处理。甚至Ben Taylor在CORP的文章中还明确指出，为了“避免赢球偏见”，他特意随机从大样本的录像数据中进行了抽样。在很多人的理解中，这种处理方式的确是“客观”的。

To analyze exactly what is wrong with this wide variety of advanced stats, we must first identify their commonalities. This might seem like a daunting task, as simply figuring out how these metrics are calculated is tedious enough, let alone dissecting something like CORP, which also incorporates film analysis. In reality, however, once you grasp the core logic behind these advanced stats, a rather striking fact quickly emerges: statistically speaking, nearly all major basketball advanced metrics are fundamentally based on averaging processes across large samples of game data. Ben Taylor even explicitly points out in his CORP article that, in order to "avoid winning bias," he deliberately took random samples from a massive volume of game footage. In the eyes of many, this methodology is indeed "objective."

但是，众所周知，率队获得总冠军、乃至连续率队获得总冠军，是一个超级巨星证明自己的最高方式。那一个球星怎样才能率队夺冠？他首先要在一个赛季的82场常规赛中取得优势的战绩，获得季后赛的席位；然后在季后赛中，要连续进行四次七场四胜制的系列赛，逐个击败当赛季的核心竞争对手，最终才能捧起那个奥布莱恩杯。而一旦深入季后赛的赛程（分区半决赛，分区决赛，总决赛），比赛强度、博弈策略都会发生突变。在常规赛中，球队会相对平均地遇到联盟的所有对手球队，赛程密集，往往会使用相对固定的战术进行作战；而一旦到了季后赛后半段，对手固定且实力强劲，两队都会针对对方的优缺点进行战术布置，比赛强度剧烈上升，很多常规赛战术往往无法打出来。

However, as is widely known, leading a team to a championship—or even consecutive championships—is the ultimate way for a superstar to prove their worth. So, what does it take for a star player to lead a team to a title? First, they must secure a winning record across an 82-game regular season to clinch a playoff berth. Then, in the postseason, they must survive four consecutive best-of-seven series, eliminating the season's premier contenders one by one, before finally lifting the Larry O'Brien Trophy. Once deep into a playoff run (the Conference Semifinals, Conference Finals, and NBA Finals), both the physical intensity and the strategic adjustments undergo a drastic shift. During the regular season, a team faces all opponents across the league relatively evenly; given the grueling schedule, teams typically rely on fairly standard, fixed tactics. But once in the later stages of the playoffs, the opponent is fixed and highly formidable. Both teams meticulously tailor their game plans to target each other's specific strengths and weaknesses. The intensity of the game skyrockets, and many offensive schemes that worked smoothly in the regular season are completely neutralized.

用数学建模的术语，在常规赛，一支球队面临的是一个随机决策问题：对手的分布已知且平均，你需要设计一个战术体系来最大化一个平均分布下的胜率。而要夺得总冠军，一支球队将会面临的是一系列的对抗性决策（或鲁棒决策）问题：对手将用尽全力来攻击你的薄弱点，最小化你的赢球概率。一个超级巨星想要获得总冠军，他当然需要有足够的实力和战术体系在常规赛中取得名次；但他更需要能够在联盟最强大的对手全力针对自己的弱点的前提下，依然通过更好的发挥把比赛赢下来。对于一个超级巨星而言，其在对抗性环境下依然能够打出良好表现的能力，比在平均情况下能够打出多少的表现，在分量上要重得多。

In mathematical modeling terms, a team faces a stochastic (or random) decision-making problem during the regular season: the distribution of opponents is known and even, requiring a tactical system designed to maximize win probability across this average distribution. To win a championship, however, a team faces a series of adversarial (or robust) decision-making problems: opponents will do everything in their power to exploit your vulnerabilities and minimize your chances of winning. For a superstar to win a title, they certainly need the requisite talent and systemic fit to secure a strong seed in the regular season. But more importantly, they must be able to outplay the opposition and secure wins even when the league's most formidable opponents are relentlessly targeting their weaknesses. For a superstar, the ability to maintain elite performance in highly adversarial environments carries far more weight than the baseline numbers they can produce under average conditions.

在这样的逻辑下，我们想要的答案一下子就非常明确了：在大样本的比赛数据的基础上进行平均化处理的那些高阶数据，全部都是在随机决策这个视域之下的数据，几乎完全不涉及到对抗性的鲁棒决策！甚至在CORP这样的历史排名系统中，作者还专门强调了他对比赛样本进行了无偏抽样；他可能以为这才能体现公平性，但他完全没有意识到：这样的处理方式恰恰就是用大量的常规赛数据去稀释了数量较少的高对抗性样本。在这样的逻辑下，这些高阶数据衡量出来的结果，和超级巨星率队夺冠所需要的真正的能力侧重，从最底层的计算哲学上就出现了严重偏差。

Under this logic, the answer we are seeking suddenly becomes incredibly clear: all those advanced metrics, calculated by averaging large samples of game data, exist entirely within the framework of stochastic decision-making. They involve almost zero adversarial, robust decision-making! Even in an all-time ranking system like CORP, the author specifically highlighted his use of unbiased sampling of game footage. He likely believed this was necessary to ensure fairness, completely failing to realize that this exact methodology uses a massive volume of regular-season data to dilute the much smaller sample of highly adversarial matchups. Following this logic, the results produced by these advanced stats severely deviate from the true skill sets required for a superstar to lead a team to a championship—a fundamental flaw rooted in the very foundational level of their calculation philosophy.

理解了上述论点，我们也就很容易为下面这几个老生常谈的问题提供一种严谨的、而非感性的解释。

Having understood these arguments, it becomes quite easy to provide a rigorous, rather than emotional, explanation for the following age-old questions.

一、科比的评价问题

NBA圈层一个很多人都关注到的现象便是：重视各种高阶数据的媒体对科比的评价，远远低于球员圈层对科比的评价。媒体圈层常常以低效、高阶数据不占优为由把科比排到历史第十之后（CORP的榜单中科比排第十三），但大量顶尖球星却认为科比至少是历史前三的顶级巨星。但从本文的视角来看，顶尖球星的观点虽然看起来更像是某种职业偏见，但反而具有更高的合理性：巅峰期科比强大的中距离投射能力、以及其极为丰富的技能包，恰恰就是在极端的对抗性决策环境之下的托底能力 -- 面对对手强硬的防守，在我方的战术配合难以发挥时，通过个人能力强行凿穿对方的防线。这就是所谓的“杀手本能”在量化建模层面的核心要义。

1. The Evaluation of Kobe Bryant

A widely noticed phenomenon within NBA circles is that media members who heavily weigh advanced stats evaluate Kobe Bryant far lower than his peers do. The media often ranks Kobe outside their all-time top ten (he ranks 13th on the CORP list), citing inefficiency and a lack of dominance in advanced metrics. Conversely, numerous top-tier players consider him at least a top-three superstar of all time.

From the perspective of this article, while the viewpoint of these top players might appear to be a form of professional bias, it actually holds far more logical validity. Peak Kobe's elite mid-range shooting and his incredibly deep bag of skills provided the exact floor-raising capability required in extreme adversarial decision-making environments. When facing suffocating defense and team offensive schemes are neutralized, he had the individual ability to forcefully dismantle the opponent's defense. At the level of quantitative modeling, this is the core essence of the so-called "killer instinct."

二、戈贝尔的评价问题

戈贝尔的防守能力可能是近年来NBA圈层中一个被广泛讨论的问题。在各种常规赛高阶数据模型中，戈贝尔都实现了毫无争议的统治，从而获得了四次DPOY（年度最佳防守球员）。然而，在季后赛的关键战中，戈贝尔竟然常常成为对方点名攻击的“防守漏勺”。这种割裂的表现在本文的视角下是不难理解的：面对常规赛的进攻手段，戈贝尔在篮下的确具有很强的统治力；但到了季后赛的高端局，对手根本就不在戈贝尔擅长的内线与其纠缠，而是将攻击范围向外拉，逼迫戈贝尔使用自己孱弱的移动能力来应对对方的攻击策略，从而使得戈贝尔瞬间从历史级的防守球员变成了一个突破口。

2. The Evaluation of Rudy Gobert

Gobert's defensive ability is perhaps one of the most widely debated topics in NBA circles in recent years. Across various regular-season advanced statistical models, Gobert has achieved unquestionable dominance, earning him four Defensive Player of the Year (DPOY) awards. However, in crucial playoff matchups, Gobert surprisingly often finds himself relentlessly head-hunted by opponents, turning into a "defensive liability."

This extreme dichotomy in performance is not difficult to understand from the perspective of this article. When facing standard regular-season offensive schemes, Gobert indeed possesses immense dominance around the rim. But in high-level playoff matchups, opponents simply refuse to engage with Gobert in the paint where he excels. Instead, they stretch the floor outward, forcing Gobert to rely on his weaker perimeter mobility to counter their offensive strategies. As a result, an all-time great defender is instantly turned into an exploitable weak link.

三、“魔球”问题

“魔球理论”是高阶数据在NBA影响力的一个重要注脚。在这个理论体系中，近战攻框和三分球的期望效率是更高的，而中距离出手却是低效的，因此“魔球战术”非常轻视中距离出手。然而，至今为止，魔球战术在总冠军层面都还没能取得突破。其原因也是显而易见的：虽然魔球战术确实可以提高球队面对平均防守环境的进攻效率，但它却相当于自己放弃了一种可能的进攻手段，使得季后赛的强大对手限制你的进攻的难度大大降低。一旦面临这种强力的对抗性限制，魔球的效率相比于其理论计算结果就会大幅下降。

3. The "Moreyball" Problem

The "Moreyball" theory serves as a major footnote to the influence of advanced stats in the NBA. In this theoretical framework, shots at the rim and three-pointers yield a higher expected efficiency, while mid-range attempts are deemed inefficient. Consequently, the "Moreyball" strategy heavily marginalizes the mid-range shot. However, to date, this strategy has yet to achieve a breakthrough at the championship level. The reason is quite obvious: while Moreyball does indeed elevate a team's offensive efficiency against average defensive environments, it essentially means voluntarily abandoning a viable offensive weapon. This drastically lowers the difficulty for elite playoff opponents to scheme against and contain your offense. Once confronted with this intense, adversarial defensive pressure, the actual efficiency of Moreyball plummets far below its theoretical calculations.

最后，回到本文最初的问题：高阶数据能够衡量NBA球星的排名吗？我们并不能给一个完全否定的答案，但必须记住的是：目前的这些主流的高阶数据，衡量的几乎都是“球星在平均的对抗环境之下的平均表现”；它并不能和一个所谓的历史地位划等号。用高阶数据来衡量球员的错误，并不在于“冰冷的数据并不能反映球员的全部实力”，而是在于这些高阶数据的底层计算哲学本就是片面的。至于有些“stats nerds”在甩出高阶数据的同时，还要声称自己是客观的、其他反驳他们的人都是不懂统计学，这不过是因为这些数据爱好者的数学与统计学功底还有待提高罢了。

Finally, returning to the original question of this article: can advanced stats accurately measure the all-time rankings of NBA stars? We cannot give a completely negative answer, but we must remember this: the current mainstream advanced metrics almost exclusively measure "a star's average performance under average adversarial conditions." They cannot simply be equated with a player's all-time historical standing. 

The flaw in using advanced stats to evaluate players does not lie in the cliché that "cold numbers cannot reflect a player's full abilities," but rather in the fact that the underlying computational philosophy of these advanced stats is inherently one-sided. As for those "stats nerds" who throw around advanced metrics while claiming absolute objectivity and dismissing their critics as statistically illiterate—it simply exposes that the mathematical and statistical foundations of these data enthusiasts still leave much to be desired.