Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
, 8 (6), e66938

Bibliometric Evidence for a Hierarchy of the Sciences

Affiliations

Bibliometric Evidence for a Hierarchy of the Sciences

Daniele Fanelli et al. PLoS One.

Abstract

The hypothesis of a Hierarchy of the Sciences, first formulated in the 19(th) century, predicts that, moving from simple and general phenomena (e.g. particle dynamics) to complex and particular (e.g. human behaviour), researchers lose ability to reach theoretical and methodological consensus. This hypothesis places each field of research along a continuum of complexity and "softness", with profound implications for our understanding of scientific knowledge. Today, however, the idea is still unproven and philosophically overlooked, too often confused with simplistic dichotomies that contrast natural and social sciences, or science and the humanities. Empirical tests of the hypothesis have usually compared few fields and this, combined with other limitations, makes their results contradictory and inconclusive. We verified whether discipline characteristics reflect a hierarchy, a dichotomy or neither, by sampling nearly 29,000 papers published contemporaneously in 12 disciplines and measuring a set of parameters hypothesised to reflect theoretical and methodological consensus. The biological sciences had in most cases intermediate values between the physical and the social, with bio-molecular disciplines appearing harder than zoology, botany or ecology. In multivariable analyses, most of these parameters were independent predictors of the hierarchy, even when mathematics and the humanities were included. These results support a "gradualist" view of scientific knowledge, suggesting that the Hierarchy of the Sciences provides the best rational framework to understand disciplines' diversity. A deeper grasp of the relationship between subject matter's complexity and consensus could have profound implications for how we interpret, publish, popularize and administer scientific research.

Conflict of interest statement

Competing Interests: The authors have declared that no competing interests exist.

Figures

Figure 1
Figure 1. Alternative predictions tested in this study, about how scientific domains should differ in any measure of hardness (or, equivalently, consensus. m =  mathematics; p  =  physical sciences; b  =  biological sc.; s  =  social sc.; h  =  humanities).
Predictions are explicit about empirical sciences, whilst mathematics and the humanities are tested secondarily.
Figure 2
Figure 2. Paper characteristics hypothesised to reflect the level of consensus, by scientific domain.
Domains are attributed based on journal, following the classifications of Essential Science Indicators and Arts & Humanities Science Citation Index: m  =  mathematics; p  =  physical sciences (Space Sc.+ Physics + Chemistry); bh  =  biological-hard sciences (Molecular Biology + Biology & Biochemistry); bs  =  biological-soft sciences (Plant and Animal Sc. + Environment/Ecology); s  =  social sciences (Psychiatry/Psychology + Economics & Business + Social Sciences, General); h  =  humanities. [Data sourced from Thomson Reuters Web of Knowledge].
Figure 3
Figure 3. Bibliographic coupling network of papers, partitioned by scientific domain (total N = 28,477; yellow  =  mathematics; blue  =  physical sciences; darker green  =  biological-hard sc.; lighter green  =  biological-soft sc.; red  =  social sc.; purple  =  humanities).
Panel A: probabilities to share a given number of references with any other paper in the sample, estimated by exponential random graph modelling. The model controlled for number of references cited by each paper, number of triangles and edges. Error bars are 95% Confidence Intervals, bh is the reference category, and has therefore all values set to zero. Panel B: network of shared references, in Yifan Hu Proportional layout. Panel C: network partitioned by domain, with average degree, modularity and average path length. [Data sourced from Thomson Reuters Web of Knowledge].
Figure 4
Figure 4. Estimates of regression analyses with, as dependent variable parameters hypothesised to reflect consensus and, as independent variables, scientific domains, weighted or corrected as described.
Bars are 95% confidence intervals, and lines are added to help visualize trends, with solid and dotted lines representing, respectively, main and extended test. Physical sciences are the reference category, and therefore have values set to zero. m = mathematics; bh =  hard-biological disciplines (Molecular Biology + Biology & Biochemistry); bs =  soft-biological disciplines (Plant and Animal Sciences + Environment/Ecology); s =  social sciences (Psychiatry/Psychology+Economics & Business+Social Sciences, general), h = humanities. See methods for further details on Methods section, and for the exact regression results with standard error see Table S1. [Data sourced from Thomson Reuters Web of Knowledge].
Figure 5
Figure 5. Paper characteristics hypothesised to reflect the level of consensus, by scientific discipline.
Classification is based on journal, following the systems of Essential Science Indicators and Arts & Humanities Science Citation Index: ma  =  mathematics; sp  =  Space Science; ph =  Physics; ch  =  Chemistry; mb  =  Molecular Biology; bb  =  Biology & Biochemistry; pa  =  Plant and Animal Sciences; ee  =  Environment/Ecology; pp  =  Psychiatry/Psychology; eb  =  Economics & Business; so  =  Social Sciences, General; ah  =  Arts & Humanities. [Data sourced from Thomson Reuters Web of Knowledge].
Figure 6
Figure 6. Bibliographic coupling networks, in Fruchterman-Reingold layout, with node size proportional to degree, and edge size and colour reflecting weight (i.e. number of shared references between any two papers: blue = 1; yellow =  ≥2; red ≥5).
Numbers report average degree, modularity and average path length. Classification based on journal, following the systems of Essential Science Indicators and Arts & Humanities Science Citation Index: ma  =  mathematics; sp  =  Space Science; ph =  Physics; ch  =  Chemistry; mb  =  Molecular Biology; bb  =  Biology & Biochemistry; pa  =  Plant and Animal Sciences; ee  =  Environment/Ecology; pp  =  Psychiatry/Psychology; eb  =  Economics & Business; so  =  Social Sciences, general; ah  =  Arts & Humanities. High-resolution images available from the authors. [Data sourced from Thomson Reuters Web of Knowledge].

Similar articles

See all similar articles

Cited by 13 PubMed Central articles

See all "Cited by" articles

References

    1. Lewes GH (1853) Comte's philosophy of the sciences. London: Henry G. Bohn.
    1. Comte A (1835) Cours de philosophie positive: Borrani et Droz.
    1. Feyerabend P (2002) Against method. London: Verso.
    1. Kuhn TS (1970) The structure of scientific revolutions. Chicago: The University of chicago Press.
    1. Dupre' J (1983) The disunity of science. Mind 17: 321–346.

Publication types

Grant support

DF was funded by a Leverhulme Early-Career fellowship. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Feedback