The Conditions and Accuracy of Bradford’s Law

Bradford’s Law is one of the foundational principles of bibliometrics. It tells us that if you arrange journals by how many articles they publish on a topic, a small core of journals carries a large share of the literature, while the rest is scattered across an ever-widening circle of less productive titles. The relationship is famously elegant: divide the literature into zones of equal article output, and the number of journals in each successive zone grows geometrically. Yet anyone who has tried to apply the law to real citation data quickly discovers that it does not always behave so neatly. The law holds beautifully under certain conditions and drifts away from the ideal pattern under others. Understanding exactly when Bradford’s Law works, and why it sometimes does not, is essential for anyone using it for collection development, journal selection, or research evaluation.

Key conditions for applicability 🔗

Bradford derived his observations from two real bibliographies he studied as Chief Librarian at the London Science Museum: the Current Bibliography of Applied Geophysics (1928-1931) and the Quarterly Bibliography of Lubrication (1931-1933). He ranked the journals in descending order of productivity and noticed a striking regularity in how articles distributed themselves. But that regularity does not appear automatically. For a bibliography to conform strictly to the law, three conditions need to be satisfied. As S. Naranan summarised in his interpretation of the law, the conditions for strict conformity are that the bibliography be complete, that it cover a limited time span, and that it refer to a well-defined subject.

A well-defined subject 🔗

The first requirement is that the bibliography should deal with a clearly bounded subject. Bradford’s Law describes how literature on a single, coherent topic scatters across journals. If the subject is too broad or vaguely defined, the dataset effectively merges several distinct literatures, each with its own core of journals. The result is a distorted distribution that no longer follows a clean geometric progression.

This sounds straightforward, but it hides a surprisingly deep problem. Birger Hjørland and Jeppe Nicolaisen pointed out that the meaning of “subject” was never explicitly examined in relation to Bradford’s Law, even though the entire law rests on it. They distinguished between lexical scattering, semantic scattering, and subject scattering, arguing that the way you define and search for a topic directly shapes the distribution you observe. In practice, this means two researchers studying the “same” subject can produce different Bradford distributions simply because they drew the boundaries of that subject differently. A well-defined subject is therefore not just a tidy starting point. It is the single most influential choice a researcher makes.

A complete bibliography 🔗

The second condition is completeness. Bradford’s Law assumes you have gathered all or nearly all the articles published on the subject within your scope. The law describes the full shape of how literature scatters, and that shape only emerges when the long tail of low-productivity journals is fully represented. If you stop collecting too early, or if your source database does not index the more obscure journals, you miss the very journals that define the outer zones.

Completeness is also the hardest condition to meet. No single database indexes every journal in any field, and the literature on most subjects keeps growing. As we will see, the failure to achieve completeness is one of the most common explanations for why real distributions bend away from the ideal.

A limited time span 🔗

The third condition is that the bibliography should cover a restricted period. A subject’s literature changes over time as new journals appear, old ones cease, and the focus of research shifts. If a bibliography stretches across too many decades, the core of “most productive” journals is not stable, because what was central in one era may be peripheral in another. Restricting the time span keeps the core meaningful and the distribution coherent.

Research has shown how strongly time matters. A study of journal output in twentieth-century logic and nineteenth-century mathematics found that the character of a Bradford distribution, including its core zone, depends on the stage of development of a scientific field and varies with the time span considered. A young, fast-growing field scatters differently from a mature, settled one.

The two formulations of the law 🔗

Before examining deviations, it helps to understand that Bradford expressed his law in two distinct ways, and the two do not always agree. The verbal formulation states the geometric progression of journals across equal-article zones. The graphical formulation plots the cumulative number of journals (on a logarithmic scale) against the cumulative number of articles, producing the well-known S-shaped curve with a rising head, a long straight middle, and a tail.

These two descriptions were long assumed to be equivalent, but they are not. B. C. Vickery noted in 1948 that the verbal and graphical formulations are not mathematically identical. This disparity matters because the two formulations can give different estimates of the core journals and different predictions about the total size of the literature. When a study reports that data “fits Bradford’s Law,” it is worth asking which formulation was tested, because a dataset can satisfy one while deviating from the other.

Deviations from Bradford’s Law 🔗

Real bibliographies rarely produce the textbook curve perfectly. The deviations are not random noise. They follow recognisable patterns, and each one tells us something about the data or the field being studied.

The Groos droop 🔗

The most famous deviation appears at the tail of the graph. Instead of continuing as a straight line, the curve begins to bend downward among the lowest-productivity journals. This downward bend is called the Groos droop, named after Ole V. Groos, who first described it in 1967. The droop appears consistently across many different datasets, which is why it has attracted so much attention.

The most widely accepted explanation, advanced by Bertram Brookes, is that the droop reflects the necessary incompleteness of the bibliography. The reasoning is that if the search were extended far enough to capture every last journal carrying a stray article, the line would straighten out again. In this view, the droop is essentially a signal that you have not yet reached completeness.

But this explanation is contested. Some researchers argue the droop is not merely an artefact of incomplete searching but an integral part of the scattering process itself, noting that many mathematical models of the law fail to reproduce it at all. Others have linked it to the merging of separate datasets, or to the discrete, whole-number nature of journal and article counts in the core. For students applying the law, the practical takeaway is that a droop in your graph is a useful diagnostic, often pointing to gaps in your data, but not always.

Anomalies in the nuclear zone 🔗

Deviations also occur at the opposite end of the curve. The first few highly productive journals, the nuclear zone, often do not sit neatly on the predicted line. The rising head of the curve tends to depart from the linear middle section. Brookes proposed a “hybrid” curve to handle this, suggesting that sociological factors influence the distribution of articles among the most productive journals, making the core behave differently from the rest. Editorial reputation, prestige, and author preference concentrate articles in a handful of leading titles in ways that a purely statistical model does not capture.

Concavity and the influence of the subject 🔗

The shape of the entire curve depends heavily on how homogeneous the subject is. When researchers compared social-science bibliographies, they found that the most homogeneous bibliography showed the classic straight-line form, while the most heterogeneous ones showed pronounced concavity. They also found that the lower the overall density of articles per journal, the greater the curvature. A related study of sixteen bibliographies showed that technical disciplines tend to produce less curvature than non-technical ones. In other words, a deviation from the ideal line is often a direct consequence of failing the “well-defined subject” condition. The geometry of the curve is, in a sense, a portrait of how tightly bounded the subject really is.

When neither formulation fits 🔗

Sometimes the data resists both formulations. In a large-scale study of human-computer interaction literature spanning 1987 to 2011, researchers found that neither the verbal nor the graphical formulation produced results consistent with the practical situation. The composition of the core journals kept shifting from one five-year block to the next, reflecting a field whose focus was continuously changing. This is a clear example of the limited-time-span condition breaking down across a long, dynamic period.

Practical implications of these deviations 🔗

Why does any of this matter to a working librarian or researcher? Because Bradford’s Law is not just a theoretical curiosity. It is a practical tool for deciding which journals to subscribe to, which to retain, and which form the indispensable core of a field. The deviations carry direct consequences for these decisions.

First, a deviation is diagnostic information, not a failure. A Groos droop may tell you your literature search needs to be extended. Strong concavity may tell you your subject definition is too broad and is mixing several literatures together. Reading the shape of the curve helps you refine your method.

Second, the law remains remarkably robust even when its conditions are not perfectly met. Naranan observed that the law seems to hold even when the conditions are not fully satisfied. This is reassuring for practical work: you will almost never have a perfectly complete bibliography on a perfectly bounded subject within a perfectly limited time span, yet the core-and-scatter pattern usually still emerges clearly enough to be useful.

Third, it is worth remembering that Bradford’s Law is a statistical regularity, not a precise mathematical certainty. It is best treated as a strong rule of thumb. Vickery also showed that the number of zones need not be limited to three; a bibliography can be divided into any number of zones each containing the same number of articles, which gives analysts flexibility in how they apply the law. The value lies in the pattern it reveals, not in a demand for exact conformity.

Finally, the deviations remind us that the law is sensitive to the choices of the analyst. The boundaries of the subject, the completeness of the search, and the time span are all decisions, not given facts. Two studies of the same field can reach different conclusions about the core simply because they made these choices differently. Good practice means stating these choices clearly so that results can be compared and trusted.

What do you think? If a Groos droop in your data might indicate either an incomplete bibliography or a natural feature of how literature scatters, how would you decide which explanation applies to your own study? And given that the definition of a “subject” so strongly shapes the resulting distribution, can Bradford’s Law ever be considered a truly objective tool for selecting core journals?

References

1. https://www.nature.com/articles/227631a0
2. https://www.academia.edu/4388892/Bradfords_Law_of_Scattering_Ambiguities_in_the_Concept_of_Subject
3. https://link.springer.com/article/10.1007/BF02458528
4. https://www.researchgate.net/publication/280218558_Bradford's_Empirical_Law
5. https://garfield.library.upenn.edu/essays/v4p476y1979-80.pdf
6. https://www.academia.edu/1307666/On_the_Theoretical_Foundations_of_Bradfords_Law
7. https://www.researchgate.net/publication/270492448_Application_of_Bradford's_Law_to_Human-Computer_Interaction_Research_Literature