At the end of last year, a review paper published in Science attracted a great deal of attention. The article by Arne Güllich and colleagues brings together research on the development of exceptional performance in domains such as elite sport, science, music and chess. It draws a striking conclusion: early high achievers are rarely the same individuals who later reach the absolute top. Early specialisation and rapid progress even appear to be negatively associated with later world-class performance.
The paper was widely shared. Not without reason. It synthesises an impressive body of data and raises questions that have long remained underexplored in research on expertise. At the same time, publication in a top journal is not a licence for unqualified conclusions. The authors themselves are aware of this, and that awareness is also reflected in the scientific commentaries that followed.
What the paper shows
The authors synthesise data from more than 34,000 international top performers, including Olympic champions, Nobel laureates, elite chess players and renowned composers. Across these domains, they identify a recurring pattern. Interestingly, those who belong to the absolute top at a young age are usually not the same individuals who perform at the highest level later in life. In many cases, later world-class performers even performed less well in their youth than peers who ultimately remained just below the very top.
The authors also distinguish between two different developmental pathways. Early high achievers are characterised by an early start, rapid progress, and extensive domain-specific practice. Later world-class performers, by contrast, more often show a slower start. They combine their main domain with other activities and specialise later.
This pattern is intriguing. It challenges the idea that early selection and maximum acceleration are self-evidently the best route to excellence.
What the authors explicitly qualify
Notably, the paper itself is already quite cautious in its claims, even though that caution was often lost in the way the findings were shared and discussed. This is a review of existing datasets, not new experimental research. The authors emphasise that their findings are descriptive. They show associations, not causal mechanisms. The fact that early specialisation is associated with a lower likelihood of later world-class performance does not automatically mean that early specialisation causes that outcome.
They also point out that the research necessarily focuses on domains where performance can be measured reliably and compared internationally. This makes the findings robust within sport, chess and certain academic contexts. However, it limits their generalisability to domains where performance is more diffuse or only becomes visible much later.
In addition, the authors stress that many later world-class performers performed above average in their youth. This is not a romantic story of mediocrity rising to the top. Instead, it is a story about subtle differences within an already highly selected group.
The commentaries: why caution remains necessary
Scientific commentary followed quickly. Not at the margins, but precisely on the point that received the most attention in public debate: the negative association between early performance and later top performance.
In a short but incisive analysis, Michel Nivard argues that this pattern can very likely be explained by collider bias, a form of selection bias also known as Berkson’s paradox. I was personally more familiar with a specific variant of this phenomenon, namely survivor bias. The problem is conceptually simple and methodologically treacherous.
When we look only at a strongly selected group, in this case, people who have reached the absolute top, and that selection depends on both early and later performance, a spurious negative association can arise within that group. This can happen even when early and later performance are positively related in the full population.
Nivard illustrates this using simulations. In the full population, early and later performance increase together in a straightforward way. But once one conditions on elite status, the relationship flips. In fact, within the selected group, it appears as if strong early performance undermines later performance, while this is in fact a mechanical consequence of the selection itself, not evidence for a different developmental pathway.
Implications
This has important implications. It means that the observation that later world-class performers, on average, stand out less in their youth may be descriptively correct within the elite group. However, it cannot automatically be interpreted as evidence for different causal mechanisms of talent development. The data show who remains after selection, not necessarily why selection takes the form it does. Other explanations may therefore also be at play.
The critique goes beyond a single correlation. Nivard shows that the same distortion also affects analyses of explanatory factors such as the amount of practice or the type of training. In elite samples, effects are attenuated, unstable or even reversed. Yet, those same factors show coherent relationships with performance in the full population. Without an explicit model of the selection process, such bias risks being mistaken for insight.
What this means for interpretation
Importantly, this critique does not invalidate the paper. It does not claim that the observed patterns are wrong, nor that early specialisation is necessarily the best route to excellence. What it does do is issue a clear warning against drawing strong conclusions about developmental mechanisms based solely on elite samples.
Ironically, this warning aligns closely with the authors’ own caution, which is often ignored in subsequent discussion. Their findings are robust as a description of what we observe at the very top. They become problematic when read as direct guidance for policy, selection or upbringing.
The real lesson, then, does not lie in the slogan that was distilled from the paper, but in the uncomfortable middle ground. Talent development cannot be reduced to early acceleration or late blooming. And anyone who looks only at those who made it, the survivor bias at the heart of this discussion, always runs the risk of misunderstanding why others did not.
That makes this Science paper not an endpoint, but a valuable starting point. Provided we are willing to take the methodological footnotes just as seriously as the figures that were so eagerly shared.
Image: https://commons.wikimedia.org/wiki/File:Chess_pieces_close_up.jpg
