Understanding the Distribution of Scientific Productivity

When you look at any field of research, a curious pattern emerges. A small handful of scientists publish a huge volume of work, while the vast majority contribute just one or two papers and then disappear from the record. This is not random. Scientific productivity follows predictable mathematical patterns that researchers have studied for nearly a century. Understanding how this productivity is distributed, and the methods used to describe it, sits at the heart of informetrics and scientometrics. Let us break down how these distributions are organised and why they almost always take the same lopsided shape.

What does “distribution of scientific productivity” mean? 🔗

Productivity in this context refers to output, usually measured as the number of papers an author publishes, the number of citations a paper receives, or the number of articles a journal carries. When we collect this data and arrange it, we are studying a distribution: a description of how output is spread across the contributors who produce it.

The foundational idea comes from the concept of an Information Production Process (IPP), developed by Leo Egghe and Ronald Rousseau. An IPP is simply a system made up of sources that produce items. Authors (sources) produce papers (items). Journals (sources) publish articles (items). Papers (sources) attract citations (items). Once you frame productivity this way, you can measure and compare it across very different domains using the same tools.

The key insight is that this productivity is never spread evenly. A few sources do most of the producing, and most sources produce very little. The two standard methods for organising this data are the size-frequency approach and the rank-frequency approach.

Size-frequency vs rank-frequency: two ways to organise the same data 🔗

Both approaches describe the same underlying reality, but they arrange the numbers differently. Both convey the same information, yet each tells the story from a different angle. Choosing between them depends on the question you want to answer.

The size-frequency approach 🔗

The size-frequency method counts how many sources produce a given number of items. You group authors by output and count how many fall into each group. The size-frequency distribution is defined as the number of sources that have exactly n items, for n = 1, 2, 3, and so on.

For example, in a body of literature you might find 100 authors who wrote exactly one paper, 25 authors who wrote two papers, 11 who wrote three, and so on. The x-axis is the number of items (papers), and the y-axis is the number of sources (authors). This is the approach used by Lotka, who described the size-frequency distribution of papers over authors. The frequency approach relies on the direct counting of information sources, and it is the natural way to ask “how many authors are one-time contributors?”

The rank-frequency approach 🔗

The rank-frequency method takes a different route. Instead of grouping by output, it ranks every source from most productive to least productive and then plots output against that rank. The most prolific author is rank 1, the next is rank 2, and so on down the line.

This is the approach George Kingsley Zipf used in linguistics. Zipf arranged unique words in order of decreasing frequency and found that multiplying frequency by rank yields a roughly constant value across a text. In bibliometrics, the same logic identifies the “elite” contributors. As the rank-frequency model shows, a small number of dominant authors or journals sit at the top while the majority cluster at low output levels. Rank-frequency is the natural way to ask “who are the top producers, and how steeply does output drop off?”

Two reflections of the same law 🔗

It is important to understand that these are not competing theories. They are complementary reflections of the same underlying law and form. The frequency approach counts sources directly; the rank approach orders them by productivity. Mathematically, the two are closely connected. In fact, the rank-frequency and size-frequency forms are functional inverses of each other, meaning you can convert from one to the other by flipping the axes.

This relationship is most reliable when the data follows a clean power law. For distributions that follow Lotka’s law, the size- and rank-frequency presentations look similar, both showing a decreasing power law. For distributions that do not follow Lotka’s law, the two presentations can look quite different. This is why scientometricians choose their method carefully depending on the dataset and the question.

The reverse J-shaped distribution 🔗

Whichever method you use, the shape that emerges is remarkably consistent. Plot the size-frequency distribution and you get a curve that drops sharply and then flattens into a long tail. This is the reverse J-shape: a tall spike on the left representing the enormous number of low-productivity authors, falling away rapidly to a thin, extended tail on the right representing the rare, highly productive authors.

The reverse J-shape captures a fundamental truth about scholarly output. The majority of scientific contributors produce no more than one manuscript, while a small but active percentage of authors are disproportionately productive. The curve is steep on one side and long on the other precisely because of this imbalance.

Lotka’s inverse square law 🔗

The classic description of this shape is Lotka’s law, formulated by Alfred J. Lotka in 1926 in his paper “The Frequency Distribution of Scientific Productivity.” Lotka analysed publication counts from chemistry and physics abstracts and derived a simple relationship. Of any set of authors, roughly 60% produce one paper, about 25% produce two (1/2² of the first group), about 11% produce three (1/3²), about 6% produce four (1/4²), and so on.

Put formally, the number of authors making n contributions is about 1/n² of those making one contribution, and the proportion of all contributors who make a single contribution is about 60 percent. A striking consequence is that only around six percent of authors in a field produce more than ten articles. This inverse square relationship is exactly what produces the steep drop and long tail of the reverse J-shape.

Zipf’s law and the rank side 🔗

On the rank-frequency side, the same lopsided reality shows up through Zipf’s law. Zipf’s law is expressed as rf = c, where r is rank, f is frequency, and c is a constant. The top-ranked source has the highest output, the second-ranked has roughly half, the third roughly a third, and the decline continues hyperbolically. Plotted on a rank axis, this gives the steep early fall and gradual flattening that mirrors the reverse J on the size side.

Zipf’s law has a remarkably wide reach. It applies to word frequency, citation frequency, publication frequency, and even unrelated phenomena like income distribution, firm size, and biological species, and it is mathematically equivalent to the distributions of Yule, Lotka, Pareto, Bradford, and Price. This is why these laws are studied together: they are different faces of the same statistical regularity.

The link to the Pareto principle 🔗

If the reverse J-shape feels familiar, that is because you have likely met its cousin in everyday life. The economist Vilfredo Pareto observed that roughly 20% of people hold 80% of the wealth, and similarly about 80% of citations refer to a core 20% of titles. This 80/20 pattern is the same concentration that Lotka and Zipf described in scientific output. A small core does the bulk of the work, and a large periphery contributes the rest.

Why these distributions matter 🔗

These are not just mathematical curiosities. The distribution of scientific productivity has real consequences for how research is managed and evaluated.

First, it shapes collection development and journal selection. Because output is so concentrated, libraries and databases can cover most of a field’s important work by focusing on a relatively small core of prolific authors and key journals. This logic underpins the related principle of Bradford’s law of scattering, which is mathematically a cumulative version of Lotka’s distribution.

Second, it informs research evaluation. When you know the expected shape of productivity, you can test whether a particular field or journal behaves “normally” or deviates from the pattern. Researchers regularly apply goodness-of-fit tests such as the Kolmogorov-Smirnov test and the Chi-square test to check whether observed author productivity matches Lotka’s predicted curve. Some fields fit the law closely; others do not, and those deviations are themselves informative.

Third, it helps with prediction and policy. Understanding that a few contributors drive most output allows funding bodies and institutions to anticipate how productivity will scale and to identify the small elite who shape a discipline.

A note of caution on universality 🔗

While these laws are powerful, they are not iron rules. Empirical studies regularly find that Lotka’s exact inverse square value does not fit every dataset. The exponent often differs from the classic value of 2, and in some fields the law breaks down entirely. Even Zipf’s law, when applied to very large modern datasets, can show a downward-bending tail rather than the perfectly straight line the law predicts. The reverse J-shape is a robust tendency, but the precise parameters vary by discipline, time period, and how the data is counted.

Bringing it together 🔗

Scientific productivity is fundamentally unequal, and that inequality follows a predictable shape. The size-frequency approach counts how many authors produce each quantity of output, while the rank-frequency approach orders authors and plots their output against rank. Both describe the same reality and are mathematically linked. That reality is the reverse J-shaped distribution: a steep concentration of one-time contributors falling away to a long tail of prolific producers. Lotka’s law captures it on the size side, Zipf’s law on the rank side, and Pareto’s principle echoes it across society at large. For anyone working in libraries, research administration, or science policy, recognising this pattern is the first step toward managing information intelligently.

What do you think? If a tiny fraction of authors produces most of a field’s published work, should research evaluation focus on rewarding that elite core, or on supporting the wide base of one-time contributors who may become tomorrow’s prolific researchers? And when a discipline’s productivity does not fit Lotka’s law, what might that tell us about how that field actually works?

References

1. https://www.semanticscholar.org/paper/Introduction-to-Informetrics:-Quantitative-Methods-Egghe-Rousseau/51c378762d253800f8205c4e33cca6574e451740
2. https://www.sciencedirect.com/science/article/abs/pii/S0306457310000294
3. https://www.sciencedirect.com/science/article/pii/S2694610625000104
4. https://asistdl.onlinelibrary.wiley.com/doi/abs/10.1002/asi.4630330507
5. https://arxiv.org/pdf/1201.0676
6. https://www.redjournal.org/article/S0360-3016(10)02112-7/fulltext
7. https://arxiv.org/pdf/1601.04950
8. https://arxiv.org/pdf/2103.03738
9. https://arxiv.org/pdf/2102.09182
10. https://arxiv.org/pdf/1011.1533
11. https://www.nature.com/articles/srep00812