Theory of aces: high score by skill or luck?

M. V. Simkin, Vwani Roychowdhury

Abstract

We studied the distribution of World War I fighter pilots by the number of victories they were credited with, along with casualty reports. Using the maximum entropy method, we obtained the underlying distribution of pilots by their skill. We find that the variance of this skill distribution is not very large, and that the top aces achieved their victory scores mostly by luck. For example, the ace of aces, Manfred von Richthofen, most likely had a skill in the top quarter of the active WWI German fighter pilots and was no more special than that. When combined with our recent study showing that fame grows exponentially with victory scores, these results show that both outstanding achievement records and resulting fame are mostly due to chance.

Introduction

In many fields, we tend to equate extreme achievement with extreme innate ability. We assume a "great" general or a high-scoring fighter ace must possess a level of genius far beyond the average practitioner. This paper investigates the "Theory of Aces"—the idea that a small number of individuals account for a disproportionate amount of success. Using historical data from World War I, we ask whether the distribution of aerial victories can be explained by simple probability or if it requires a model where some pilots are orders of magnitude more skilled than others.

Methodology

The study relied on historical records of German WWI fighter pilots, encompassing 6,759 credited victories and approximately 1,000 casualty reports. We defined a "recorded engagement" as a combat event resulting in either a credited victory or a pilot defeat (death, capture, or injury).

To analyze this data, we employed a mix of statistical modeling and Bayesian inference:

  • Maximum Likelihood Estimation (MLE): Used to find a distribution of "innate defeat rates" ($r$) that best fit the observed victory counts.
  • Maximum Entropy Principle: Combined with MLE to ensure the most unbiased distribution possible, helping us infer the underlying skill distribution of the entire pilot pool.
  • Bayesian Inference: Used to estimate the most likely innate defeat rate for specific famous aces based on their individual career records.

The Distribution of Skill

Our analysis reveals that the variance in innate skill among pilots is relatively narrow. The "skill" distribution is not nearly as wide as the "victory" distribution would suggest. The top aces were not necessarily ten times better than average pilots; they were simply the ones who survived long enough to stay on the winning side of repeated trials.

Inferred distribution of pilot skill (defeat rate) using maximum likelihood and maximum entropy methods.

As shown in the inferred distribution, most pilots cluster within a specific range of capability. The legendary status of top-scoring pilots is often a result of being in the right tail of this distribution and benefiting from a long string of favorable outcomes.

Selection Bias and the Experience Factor

A notable observation in the data is that the rate of defeat drops sharply after the first few fights. While this is often interpreted as "learning," we argue it is largely a result of selection bias. Unskilled pilots are filtered out early in their careers, leaving behind a pool of survivors who appear more skilled on average.

Observed defeat rate as a function of fight number, showing a sharp initial decline followed by a plateau.

This plateauing effect suggests that once a pilot survives the initial "filtering" phase, their probability of defeat remains relatively constant, further supporting the idea that long-term survival and high scores are driven by statistical luck within a pool of competent survivors.

Case Study: The Red Baron

Manfred von Richthofen, the "Red Baron," is the most famous ace of WWI with 80 credited victories. At first glance, the probability of winning 80 dogfights in a row by pure 50/50 chance is $10^{-24}$—essentially impossible. However, when calculating the actual average defeat rate of the German pilot pool (~10.7%), the probability that at least one pilot out of approximately 2,900 would reach 80 victories is 0.29.

In other words, the Red Baron’s record was statistically likely to happen to someone in that population. Bayesian inference suggests his innate skill (a defeat rate of ~2.5%) likely placed him in the top 27% of pilots—a high performer, but not a statistical miracle.

Conclusion

The predicted distribution of victories, using our inferred skill distribution, closely matches historical data. This supports the conclusion that chance is the primary driver of high scores. While achievement is largely luck, fame grows exponentially with that achievement, creating a "winner-take-all" scenario where a pilot who is slightly luckier than his peers becomes infinitely more famous. Both outstanding achievement records and the resulting fame are, in large part, products of chance.

Summary updated

Abstract

We studied the distribution of World War I fighter pilots by the number of victories they were credited with, along with casualty reports. Using the maximum entropy method we obtained the underlying distribution of pilots by their skill. We find that the variance of this skill distribution is not very large, and that the top aces achieved their victory scores mostly by luck. For example, the ace of aces, Manfred von Richthofen, most likely had a skill in the top quarter of the active WWI German fighter pilots and was no more special than that. When combined with our recent study (cond-mat/0310049), showing that fame grows exponentially with victory scores, these results (derived from real data) show that both outstanding achievement records and resulting fame are mostly due to chance.
BibTeX
@article{simkin2008theory,
  author = {M. V. Simkin and Vwani Roychowdhury},
  title = {{Theory of aces: high score by skill or luck?}},
  journal = {Journal of Mathematical Sociology},
  year = {2008},
  doi = {10.1080/00222500801932006},
}
Figure