Thomas Kuhn famously argued that ' Almost always the men who achieve these fundamental inventions of a new paradigm [or major breakthrough] have been either very young or very new to the field whose paradigm they change. … for obviously these are the men who, being little committed by prior practice to the traditional rules of normal science, are particularly likely to see that those rules no longer define a playable game and to conceive another set that can replace them' . Kuhn developed this hypothesis by studying a small sample of theoretical discoverers, mostly in early physics, like Einstein. But does this hypothesis hold up when we test it on science's major discoveries across fields? Indeed, Einstein was only 26 when he published his nobel-prize-winning paper on the law of the photoelectric effect in 1905, in the prestigious journal Annalen der Physik. He did this while working at the Swiss patent office. As Einstein himself boldly put it: ' A person who has not made his great contribution to science before the age of 30 will never do so' . Yet do the conditions at Einstein's time reflect science today? We uncover a striking finding: the golden age range of high productivity and impact in science is between 35 and 45 years of age, with exactly 50% of all Nobel laureates in science falling into this age range when uncovering their prize-winning discovery. The average age of discoverers is 39 years. This is the prime of most scientists' careers, with the odds drastically dropping as age increases afterwards. We find that only 7% of all nobel-prize discoveries and 15% of major non-nobel discoveries over the same period arise after 50. And more remarkably: only 1% and 3% after 60. So today, we can revise Einstein's claim and say: a person who has not made their great contribution to science before the age of 45 is much less likely to do so (as we uncover in Figure 4.2a). For the group of nobel-prize discoverers, we can partly explain this by the average life expectancy at present and the average 21-year gap between making the discovery and receiving the prize. Yet on average, Nobel laureates in science are 60 years old when they actually receive the prize. So why does the sweet spot of discovery seem to peak in a researcher's prime years? Younger, untenured researchers are often more motivated and ambitious to make their mark and achieve a breakthrough. Older researchers with already secure, tenured positions often do not face the same degree of external pressure—publish or perish. Younger researchers, those just entering a field, can have another advantage. These researchers are trained in the latest, up-to-date methods and technologies. They bring a fresh perspective to problems, without always accepting established assumptions, and can be more open to exploring new techniques. The evidence challenges the popular belief that greater age and experience are synonymous with innovation. In the past, some researchers uncovered groundbreaking discoveries very early in their careers. The Indian student Subrahmanyan Chandrasekhar, at just age 21, 36 37 40 40 41 0 10 20 30 40 econ/social physics chemistry astronomy medicine/biol Age at year of discovery - for all nobel-prize discoverers 0 5 10 15 Percentage of all discoverers 20 30 40 50 60 70 80 Age All major discoverers All nobel-prize discoverers Age distribution of scientists at time of discovery 22, Nash (Nash equilibrium) 24, Heisenberg (Developing quantum mechanics) 79, Weiss, Barish and Thorne (avg. age) (Developing LIGO detector and observing gravitational waves) Avg. age of nobel-prize discoverers Figure 4.2 The golden age range of high productivity and impact in science is between 35 and 45 years of age The data cover science's 761 major discoveries (including all nobel-prize discoveries) (Figure a)—and all 533 nobel-prize discoveries (Figure b). described the physical processes governing the evolution of stars, making him the youngest scientist ever to make a nobel-prize-winning discovery. Close behind him was the British student Brian Josephson, who created a framework for tunnelling supercurrents at 22, and the American John Nash, who introduced the game theory 108 T HE ENGINE OF SCIENTIFIC DISCOVERY concept of Nash equilibrium also at 22. The Swedish student Svante Arrhenius developed the electrolytic theory of dissociation at 24, and the British student Alan Turing devised the Turing machine also at 24, laying the theoretical groundwork for modern computing that later transformed our world. The German Werner Heisenberg, British Paul Dirac and Danish Niels Bohr made their major contributions to quantum mechanics at 24, 26 and 28, reshaping our understanding of reality. All these young researchers, except Turing, received a Nobel prize for these breakthroughs. (Tragically, Turing did not live to see the full impact of his work, dying mysteriously from cyanide poisoning at age 41, after being persecuted and chemically castrated by the British government for being homosexual). All these contributions by the youngest discoverers were theoretical. Why? The answer lies in the nature of research itself. Theoretical work often depends on the latest mathematical methods and modelling, along with little more than paper and a pen, to challenge theoretical assumptions from a fresh, new perspective. By contrast, experimental and methodological research relies on rigorous experimental designs, testing, observation and validation, all requiring significant training and access to lab space and advanced tools. It demands first mastering tools like x-ray devices, advanced statistical methods or electron microscopes. Today, it is extremely difficult to achieve a major discovery at such a young age. But why is that? Simply because acquiring the ever-expanding method training and knowledge to be able to discover something new takes longer—before we reach the research frontier. For our methods are more complex, and what we know is more vast. Yet the path to experimental and methodological breakthroughs is still often longer than for those that are theoretical. The low-hanging-fruit discoveries (the simpler ones) have largely been picked—so generating new breakthroughs today often requires solving more intricate puzzles.
universities that can help provide greater access to cutting-edge tools and resources What role can a scientist's university play in supporting the broader research environment—and accessing and developing new tools to spur breakthroughs? We uncover that only 30% of all nobel-prize discoverers are affiliated with a top-25ranked university, and also 30% of major non-nobel discoverers (the uncrowned laureates) over the same period—serving as an independent control group. Yet globally, less than 1% of researchers work at these elite institutions. Expanding the scope to the top 50 universities, we see that 38% of all nobel-prize discoverers and 34% of major non-nobel discoverers were at these institutions. So being at a top university is far from a prerequisite for making breakthroughs, providing hope for researchers not at top-tier institutions. The share of discoverers at a top 25 university is higher in astronomy (41%) and economics and social sciences (49%) (Figure 4.3). Why? We would expect this because in astronomy, the most advanced radio telescopes, laser interferometers D 0.00 0.00 0.17 0.090.09 0.32 0.030.03 0.24 0.070.09 0.80 0.28 0.33 0.95 0.36 0.45 0.99 0.28 0.50 1.00 0 .2 .4 .6 .8 1 Percentage of all discoverers <1600 1600–1699 1700–1799 1800–1899 1900–1949 1950–1999 2000–2022 Evolution of science: all major discoveries over time At top 25 university at discovery At top 50 university at discoveryAt a university at discovery 0.22 0.29 0.98 0.29 0.33 0.99 0.29 0.38 0.99 0.41 0.59 1.00 0.49 0.59 1.00 0 .2 .4 .6 .8 1 Percentage of all nobel-prize discoverers medicine/biol physics chemistry astronomy econ/social Contemporary science: all nobel-prize discoveries - by field At top 25 university at discovery At top 50 university at discovery At a university at discovery Figure 4.3 Only about one in three discoverers since 1950 worked at a top 25 university The data show science's 761 major discoveries (including all nobel-prize discoveries) (Figure a)—and all 533 nobel-prize discoveries (Figure b). The data reflect the discoverers' university affiliation at the time of the discovery, using the QS World University Rankings in 2021 as a common reference point. Most top universities have remained among the top over time. For earlier centuries, we should view data with caution: while most discoveries have been made while today's top 50 universities existed, some did not yet exist before the 1800s (Figure a). By combining all nobel-prize and major non-nobel discoveries over the same time period across these five fields, we find that, for example, the share of discoverers at a top 25 university is comparable: 22%, 28%, 29%, 42%, and 48%. (And beyond university location, we explore geographic location but also the religious background of discoverers in Appendix Figure 4.1.) 110 T HE ENGINE OF SCIENTIFIC DISCOVERY and space observatories needed to make breakthroughs are concentrated there—and because economics is the field most strongly centred in the US, where institutional reputation plays a dominant role. We find a similar pattern in other fields with some of the world's largest and most sophisticated instruments, like particle accelerators, electron microscopes and advanced x-ray technology needed to trigger discoveries, concentrated at better institutions. At Berkeley for example, a massive new particle accelerator, the Bevatron, built in 1954 enabled Emilio Segrè and Owen Chamberlain to run experiments and discover the antiproton in 1955. The breakthrough would not have otherwise been possible. Their groundbreaking article, Observation of antiprotons, was then published in the Physical Review. Yet many of our most common and important tools used to catalyse breakthroughs are remarkably inexpensive (as we uncover in Chapter 6). Contrary to popular belief, we find that most breakthroughs occur outside toptier universities—outside the traditional elite—especially across most fields that do not rely on those large, high-tech instruments often based there. Still, being at a top university can, depending on the field, give researchers a comparative advantage, not just through access to sophisticated instruments and lab facilities, but can also offer some researchers greater access to funding and networks of researchers, when necessary. Yet, for the share of discoverers at these universities, some of the more ambitious naturally gravitate towards and self-select these institutions in the first place. In short: discovery knows no institutional boundaries, with history showing that breakthroughs can happen anywhere. Exploring also gender disparities, we find that breakthrough science remains heavily biased towards men, with women continuing to face discrimination in science. Women represent only 5% of all scientists who made a major discovery and only 3% of all Nobel laureates. The pioneering Polish-French physicist Marie Curie was the first woman to win the prize. Her groundbreaking research on radium and polonium transformed our understanding of radioactivity. She dedicated her life to science, eventually dying of aplastic anaemia caused by exposure to radioactive materials she worked with. She has become arguably the most iconic woman in science, making history by winning two Nobel prizes—one in chemistry (by herself ) and another in physics (collaborating with Henri Becquerel and her husband Pierre Curie). But what is the role of collaboration in science? Science is a collective effort. Researchers within a community, working in cooperation and competition, need to inevitably build on the existing tools and research of others that contribute towards a breakthrough. Discovering DNA's double-helix structure is an example of a deeply collaborative effort. It was not just the work of Watson and Crick at Cambridge. But the remarkable breakthrough was only possible because of the pivotal x-ray work produced by Rosalind Franklin and her student Gosling—but as we highlighted, Franklin was discriminated for being a woman and was not fully recognised during her lifetime. Her trailblazing work relied on powerful x-ray crystallography methods developed by von Laue and the Braggs, who used x-radiation identified by Röntgen. The research also built on early work on DNA by Miescher, and was supported by parallel work on DNA by Wilkins and his group of colleagues at King's College London, along with others. The collaboration spanned across time and geography. Some scientists create the tools needed to carry out the research (like D von Laue's and the Braggs' x-ray methods), others may uncover the observational or experimental insights applying these tools (like Franklin's x-ray images), and others may then develop a theoretical explanation for the evidence (like Watson and Crick's theoretical model). Discovering the Higgs boson at CERN and unravelling evolution's multiple mechanisms are other classic examples of great collaboration—triumphs requiring the collective effort of hundreds of researchers working together over time. Larger teams can at times better leverage different methods, combine them and integrate more expertise. For relevant fields, they can help pool resources for more cutting-edge instruments and can have greater access to different technologies and lab equipment. While we focus on scientific superstars, these pioneers build on a foundation that was made before them and makes the last step towards discovery possible.
Ask scientists what drives discoveries, and the two most common factors stated are research funding and teams and the scientific community. Given the vast breakthroughs achieved in science, the role of governments after the Second World War changed: for the first time, public funding for science was spent strategically. With this shift, a global awareness emerged of the importance of science: the Chinese Academy of Sciences was founded in 1949, the National Science Foundation (NSF) in the US in 1950 and the German Research Foundation (DFG) in 1951. Before the war, science spending was negligible, making up only 0.1 billion US$ in 1930 in the US. After the war, the US arose as the most important player in science funding. Research and development (R&D) spending in the US surged nine-fold, increasing from 1.5 billion US$ in 1945 to 13.6 billion US$ in 1965. Strikingly, in the 1940s and early 1950s, over 80% of this spending was targeted to national defence alone. At the time, the number of PhDs awarded in the US also jumped 10-fold, from 1634 awarded in 1945 to 16,340 in 1965 (Figure 4.4). Yet science's biggest discoveries throughout history cannot simply be attributed to strategic science funding and a large, trained scientific community—because these factors largely only began after the Second World War. By 1945, more than half of science's major discoveries—almost 400 breakthroughs—had already been made, including the majority of major non-nobel discoveries since 1875. If strategic science funding had played a very strong role, we would see a very strong sustained spike in discoveries following the establishment of major science agencies and large-scale funding. Yet we do not see a sustained spike (Figure 4.4). Instead, we uncover a common pattern throughout the history of science: major discoveries consistently follow after the newly invented methods needed to make them. Remarkably, before the mid-1930s, scientists already developed many of the central methods and tools of science—and the discoveries they enabled. These include modern microscopes since 1873, particle detectors since 1911, x-ray crystallography 112 T HE ENGINE OF SCIENTIFIC DISCOVERY methods since 1913, the centrifuge in 1924, modern statistics in 1925, particle accelerators since 1929, electrophoresis in 1930 and chromatography in 1931. To create these powerful tools, large-scale research funding and a large scientific community were not the most important factors—or even an important factor—as they did not yet exist. These breakthrough tools were made at relatively low costs, and each then triggered multiple major discoveries. So while research funding fosters our system of science, if we trace the path of discovery up to the breakthrough itself, one key driver consistently emerges: the development of new methods and tools that directly spark new discoveries. We dig deeper and unpack this topic in Chapter 6. 0,0 20,0 40,0 60,0 80,0 100,0 120,0 140,0 160,0 180,0 0 10000 20000 30000 40000 50000 60000 1875 1895 1915 1935 1955 1975 1995 1915 Spending in Research and Development in US annually (bln $) Doctorates awarded annually in US Doctorates awarded in US for all universities and fields Spending in Research and Development in US (bln $) National Science Foundation grounded in 1950 382 doctorates granted in 1900 6,535 doctorates granted in 1950 1.9 billion US$ in 1950 0.1 billion US$ in 1930 0 2 4 6 8 10 12 14 16 Number of major discoveries annually Number of major discoveries annually Number of non-nobel discoveries annually 23 doctorates granted in 1875 Figure 4.4 Trends in science funding and the growing size of the scientific community in the US, in relation to major scientific discoveries The data reflect science's 620 major discoveries made since 1875—and the 90 major non-nobel discoveries since 1875. Other major discoveries before 1875 are not included. Once again, all data—throughout the book—reflect the year discoveries are made—not when the Nobel prizes are awarded. The data on R&D spending are from the Federal Reserve Economic Data, and total PhDs awarded are from the NSF. Science's major discoveries reveal that expensive tools are not always needed for most breakthroughs, beyond exceptional discoveries using instruments like large particle accelerators at CERN and space telescopes at NASA. Several of the ten most used central methods and tools of discovery are remarkably low-cost, including statistical and mathematical methods, light microscopes, electrophoresis, chromatography methods, centrifuges and thermometers (and later the PCR method and other assay D techniques). We can buy these new for less than a thousand or even few hundred dollars (Chapter 6). Statistical methods including fast-growing statistical programmes like R can be downloaded for free and some AI programmes have low costs, and play a critical role in analysing vast datasets. We see that nobel-prize discoverers in countries within the bottom two quintiles (the poorest 40%) have access and use common tools at similar rates as those in wealthier countries (Figure 4.5). Discoveries using common sophisticated instruments are not just concentrated in the richest countries. Many fields rely on affordable and easily accessible methods like advanced statistics and electrophoresis—and these, as expected, show little variation across income levels. Once a minimal income threshold is met, researchers in less affluent countries do not seem to face significantly greater constraints to triggering breakthroughs—though historically, discoveries have been concentrated in Europe and North America. Yet some countries' researchers live in—like their university—can influence access to more advanced instruments and greater funding, infrastructure and government support. Wealthier countries often house some of the more cutting-edge technological and computing facilities and specialised labs needed for some discoveries. Greater income per capita (more resources) and a larger research team (more researchers) can foster the basic conditions of science—but for most discoveries, they have not been needed (Chapter 6). Researchers are also not in science for the money, otherwise they would take their PhDs and go into industry with higher salaries. An exceptional and tragic case is 0 20 40 60 80 100 Percentage of discoveries Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 Use of sophisticated methods/instruments - from poorest to richest quintile electron microscope x-ray analysis electrophoresis chromatography centrifuge computer statistics Figure 4.5 Scientists in poorer and wealthier countries have comparable access and use of various sophisticated methods and tools, for nobel-prize discoveries since 1975 The data reflect the 125 nobel-prize discoveries since 1975. These methods and tools were first developed by 1950 and in wide use by 1975; to reflect this, these income quintiles are calculated using the broad set of countries in which nobel-prize discoverers lived from 1975 to 2022. The income data is adjusted for inflation, and we analyse the period from 1975 to 2022 to control for larger variations in income over time. 114 T HE ENGINE OF SCIENTIFIC DISCOVERY Nikola Tesla, the visionary discoverer who did pioneering work in electricity and wireless transmission, but died impoverished and living alone in a New York City hotel room despite his enormous contribution to science. Simply investing more money and people into science is not a focused strategy. Yet researchers' salaries and recurring overhead costs account for the vast majority of total science spending—it sustains the day-to-day running costs. The large increase in public science funding and the expansion of the scientific community since the 1950s have fuelled an extraordinary surge in publications—a surge that seems almost natural in an ever-growing academic community shaped by a publish-or-perish culture where frequent publishing is often prioritised over high-impact breakthroughs. Yet the surge has lacked a guiding framework—an evidence-based understanding of what triggers groundbreaking discoveries. Despite large funding bodies in place since the mid-20th century (like NSF and DFG) and early 21st century (like the European Research Council), no funding agencies or large research communities yet have large strategic programmes targeted to the development of transformative new methods and tools across science. Surprisingly, science's best tools happen to emerge as ad-hoc projects as a specific demand arises, like constructing a cutting-edge spectrometer or supercomputer (Chapter 6). Predicting new discoveries and new methods (regression results) After exploring how new methods drive discoveries, we turn to the broader question: what other factors support the discovery process and the needed methods? To analyse this, we take a different perspective here, using two logistic regression models—controlling for the common supporting factors we explored earlier. Model 1 controls for the time gap between the developed method or tool and the discovery it enabled, and for whether the discoverers conducted experiments, tested hypotheses and made observations; and it also controls for the common background factors of discoverers' discipline, age, the number of discoverers (collaborators), their gender and level of education (the independent variables). Model 2 extends these variables, controlling also for whether the discoverers had two or more different academic degrees, whether they were at a top-50-ranked university (a proxy for greater access to funding, networks and higher salaries) and broader country-level factors. These include the discoverers' geographic location, and the income per capita and population size of the country they lived in. (We explored these factors in previous sections.) We begin by exploring the question of what, if anything, distinguishes nobel-prize discoveries from major non-nobel discoveries over the same time period (the comparison group, or dependent variable)? That is, what factors may support nobel-prize discoveries? We find that nobel-prize discoveries are more likely to be made faster— within a shorter time span between the developed method or tool and the resulting breakthrough. They are also more likely to be made by highly educated researchers (PhDs or professors), by two or more researchers and through experimentation— while controlling for the set of supporting factors including discipline, age, gender, university ranking and geographic location. Being at a top university is not an ** *** ** ** *** *** *** *** ** *** ** ** ** *** *** *** ** ** Gap betw. method and discovery <12 years Experimentation conducted Hypothesis tested Observation used Chemistry (ref. group, physics) Medicine and biology Economics/social sciences Astronomy Age 35–45 (ref. group, 18–34) Age >46 Made by 2+ researchers (ref. group, 1) Male (ref. group, female) Professor (ref. group, Masters or <) PhD/Postdoc (including MD) Discoverer has two or more degrees Discoverer at top 50 university Population size, top tercile Income per capita, top tercile Country of residence in Europe Methodological approaches Discipline of discovery Traits of discoverers Educational and institutional traits Country-level factors (of discoverers) –0.50 –0.25 0.00 0.25 0.50 Marginal effects Model 1 Model 2 Figure 4.6 The enabling factors of nobel-prize discoveries The data reflect 616 major discoveries that include all nobel-prize discoveries compared to major non-nobel discoveries made over the same time period. –0.50 –0.25 0.00 0.25 0.50 Marginal effects Model 1 Model 2 ** ** ** *** ** *** ** ** *** ** Gap betw. method and discovery <12 years Experimentation conducted Hypothesis tested Observation used Chemistry (ref. group, physics) Medicine and biology Economics/social sciences Astronomy Age 35–45 (ref. group, 18–34) Age >46 Made by 2+ researchers (ref. group, 1) Male (ref. group, female) Professor (ref. group, Masters or <) PhD/Postdoc (including MD) Discoverer has two or more degrees Discoverer at top 50 university Population size, top tercile Income per capita, top tercile Country of residence in Europe Methodological approaches Discipline of discovery Traits of discoverers Educational and institutional traits Country-level factors (of discoverers) Figure 4.7 The enabling factors of method discoveries (major new methods and tools) The data reflect 697 major discoveries since 1575, including all nobel-prize discoveries. For interested readers, we provide the technical details and model specifications below Appendix Figure 4.2. 116 T HE ENGINE OF SCIENTIFIC DISCOVERY important predictor of making a nobel-winning discovery. Broad factors like income per capita and population size also play a more limited role in influencing nobel-prize breakthroughs (Figure 4.6). Next, we explore the essential role of methods and tools in triggering breakthroughs, with method discoveries—from the electron microscope to electrophoresis—responsible for 25% of science's major discoveries. The other 75% are empirical and theoretical discoveries. So what factors support these method discoveries? When we compare method discoveries to empirical and theoretical ones (the new dependent variable), we find that method discoveries, too, are most strongly supported by researchers with the highest levels of education (PhDs or professors) and through experimentation—while controlling for the same set of factors. The time span between the developed method or tool and the resulting breakthrough is very similar for method discoveries and empirical and theoretical discoveries, so the effect is not statistically significant. Additionally, method discoveries are least likely to emerge in medicine compared to physics (the reference group). And once again, we see a more limited role of broad factors such as income per capita and population size (Figure 4.7). Finally, we turn to the gap in years between developing the enabling method or tool and making the discovery using it. On average, the gap between the two is 14 years for all nobel-prize discoveries. But what shapes this gap? To explore this, we use linear regression to analyse the number of years between the new method and the discovery (the dependent variable). We find that the strongest factors influencing this gap are the specific discipline and time period in which the breakthrough occurred. Discoveries in medicine and biology tend to have a gap of about 7 years longer than those in physics (the reference group), while controlling for the same set of factors earlier and time periods. Strikingly, astronomical discoveries show the shortest time gaps, as we would expect: once new observational instruments are developed, they more immediately enable discovering new phenomena, providing clear evidence of the direct link between method and breakthrough. Other factors have little effect or are not statistically significant in shaping the timing of discoveries (Appendix Figure 4.2). The overall trends are consistent across the regression models. We unfortunately cannot explore psychological traits—like greater motivation and drive among some discoverers—in influencing the timing of discoveries. We cannot easily collect these traits simply because most discoverers have passed away. Yet we do highlight the remarkable stories of what inspired the inventors of the top ten most used and powerful tools in science (as we explore in the Boxes throughout the chapters and further in Chapter 6).