Bibliometric Laws: Lotka's, Bradford's, and Zipf's Laws

Lotka's Law is empirical, not deterministic. It describes aggregate population patterns, not individual careers. Individual variation stems from career stage, field norms (experimental vs. theoretical), access to funding, team size, and luck. The law describes the statistical outcome across many researchers; it does not predict any single career. Possible causes of high productivity: senior career status, leadership of large labs, high funding, prolific writing style. Do not assume high productivity = high impact; citations and novelty are separate from publication count.

Bradford's Law still applies to traditional peer-reviewed journals. However, preprints (arXiv, bioRxiv) and open-access journals complicate the picture: papers may be available before formal peer review, and small specialized journals have global reach (not geographic barriers that existed in 1934). Modern application: identify 'core' sources (high-impact venues regardless of format), then audit your literature review to ensure you are not sampling only from one zone. The principle (small number of sources contain most literature) remains valid; the definition of 'source' has evolved.

Both laws describe stable equilibrium patterns in mature fields. They are predictive within the current field structure: if a field's Lotka exponent is 2.0, you can predict that the top 10% of authors will publish ~50% of papers. However, sudden changes (new funding program, emerging subdiscipline, methodological breakthrough) can shift distributions dramatically. Use laws for baseline predictions within current conditions; expect violations during periods of rapid change. For trend prediction, combine distributions with temporal analysis (compare distributions across time windows).

Research abstracts typically show Zipf exponents of 0.9–1.1. Domain-specific keywords (rare terms) may show steeper exponents (>1.2) because specialized vocabulary is concentrated. Abstract-derived keyword analysis often deviates more from Zipf than full-text analysis (abstracts are shorter, less redundant). If analyzing keyword frequencies (not full text), expect exponents 1.0–1.3. Always plot your data on a log-log scale and inspect fit; the exponent is less important than whether a power law fits at all.