PaperlayerView exact revision on arXiv

Source-linked AI summary

The Research Space: using the career paths of scholars to predict the evolution of the research output of individuals, institutions, and nations

Miguel R. Guevara, Dominik Hartmann, Manuel Aristarán, Marcelo Mendoza, César A. Hidalgo

arXiv:1602.08409v3cs.DLcs.SIphysics.soc-ph

TL;DR

Citation-based science maps connect fields through knowledge flows, while scholars’ career paths offer another basis for predicting which fields they may enter. The paper maps fields by the probability that scholars publish in both and finds better predictions for individuals and organizations, but comparable accuracy for countries.

  • Problem

    The paper asks whether links based on scholars’ overlapping fields can predict future field entry more accurately than citation-based maps.

    Source: Introduction · p. 3

  • Method

    Using career data for more than 200,000 scholars, the authors link fields according to the probability that an author has published in both.

    Source: Introduction · p. 3

  • Results

    The research space predicts future field presence more accurately for individuals and organizations, while its accuracy is comparable to the UCSD citation map for countries.

    Source: Results · pp. 15–16

  • Takeaways & Limitations

    Career trajectories provide more predictive information than citation maps for fine-grained units, but not for country-level forecasts.

    Source: Introduction · p. 3, Discussion · p. 17

    The results apply only to career trajectories observable in Google Scholar, whose adoption varies across academic fields and age groups.

    Source: Data · p. 6

Abstract

from arXiv · show

In recent years scholars have built maps of science by connecting the academic fields that cite each other, are cited together, or that cite a similar literature. But since scholars cannot always publish in the fields they cite, or that cite them, these science maps are only rough proxies for the potential of a scholar, organization, or country, to enter a new academic field. Here we use a large dataset of scholarly publications disambiguated at the individual level to create a map of science-or research space-where links connect pairs of fields based on the probability that an individual has published in both of them. We find that the research space is a significantly more accurate predictor of the fields that individuals and organizations will enter in the future than citation based science maps. At the country level, however, the research space and citations based science maps are equally accurate. These findings show that data on career trajectories-the set of fields that individuals have previously published in-provide more accurate predictors of future research output for more focalized units-such as individuals or organizations-than citation based science maps.

Introduction

The section examines how the research space, based on scholars’ career paths across fields, predicts future research output for individuals, organizations, and countries compared with citation-based maps. It covers multiple transitions, with accuracy advantages for individuals and organizations and equal accuracy for countries.

Introduction

  • Among 215,390 cited scholars in the dataset, 99.8% published in at least two fields, 99.2% in at least three, and 97.5% in at least four, usually spanning related disciplines.For scholars publishing at least two papers in Molecular Biology, 46.6% also published in Clinical Biochemistry, compared with 0.95% in Economics and Econometrics.Source: Introduction · p. 2

Mapping Science through Knowledge Flows and Career Paths

  • Knowledge-flow maps connect fields through citations, whereas the research space connects fields through scholars’ joint participation and uses career trajectories to predict future research output.Source: Mapping Science through Knowledge… · pp. 5–6

Data

  • The Google Scholar profiles enable author-level disambiguation, but adoption varies across fields and age groups, so findings apply only to career trajectories observable in that dataset.Source: Data · p. 6

Constructing The Research Space

Using the research space to predict future research output

  • The prediction method classifies entities’ field presence using expected output and estimates development prospects from the fraction of connected fields where the entity is already developed.Fields in an initial state are ranked by density, with higher-density fields predicted to transition first.Source: Using the research space to predict… · pp. 11–13

Results

  • Individuals averaged AUC 0.8963 with the research space versus 0.8034 with UCSD; it also significantly outperformed UCSD for organizations, while countries showed equal accuracy.For individuals, the difference was highly significant (ANOVA p<0.001).Source: Figure 4 · p. 15, Results · p. 15
  • For nascent-to-developed transitions, organizations scored AUC_research_space=0.6927 versus AUC_UCSD_science_map=0.6696 (p<0.05), while countries were not significantly different (0.6387 versus 0.6239; p>0.1).Source: Figure 4 · p. 15, Results · p. 16
  • For intermediate-to-developed transitions, organizational AUC was 0.6390 with the research space versus 0.6164 with the UCSD science map (ANOVA p-value<0.01).Source: Results · p. 16
  • For intermediate-to-developed transitions among countries, the research space and UCSD science map were equally accurate.Source: Results · p. 16
  • For nascent-to-developed transitions, the research space’s accuracy increase also held for individuals and organizations.Source: Results · p. 16

Aggregation

The research space predicts future research output more accurately than the UCSD citation-based map for individuals and organizations, while the maps are comparably accurate for countries. The paper leaves field-specific research costs unaddressed and suggests that combining cost estimates with success-potential estimates could help policymakers plan research efforts.

Aggregation

  • The research space significantly outpredicts the UCSD citation-based map for individuals and organizations, while both maps are comparably accurate for countries.The aggregation table reports AUC values for organizations and countries, but the supplied passage does not identify the columns’ comparison methods.Source: Discussion · p. 17, Aggregation · p. 17

Discussion

  • The analysis leaves unanswered how much it costs to develop each research field, whose infrastructure needs can differ substantially.Examples include advanced facilities for biology and accelerators or reactors for particle or plasma physics.Source: Discussion · p. 17
  • A method combining field-specific success potential with the costs of advancing research could help policymakers strategize research development.Source: Discussion · p. 18

Supplementary Material

The supplementary-material contents list resources on raw data, research-field classifications, journal assignments, and overlay maps.

Supplementary Material

1 Analysis of the raw data

The analysis starts with Google Scholar publications from 1971–2014, then cleans and filters them to build the journal-classified dataset used in the study.

1 Analysis of the raw data

  • The final dataset contains 4,745,774 publications; the raw dataset had 12,445,334 records from 1971–2014.Filtering removed records with missing or spurious publication years and other missing, fake, or nondisambiguated data.Source: §1 · p. 22
  • The figures show raw publication counts, annual publications per author on a log-scaled y-axis, and the year distribution for Scimago-listed journals.Source: Figure 1 · p. 22, Figure 2 · p. 23, Figure 3 · p. 24
  • Only publications with a 100% journal-name match to the classification lists were retained for assigning research areas.Source: §1 · p. 23
  • Authors with more than 50 publications in a year were excluded because many such records were misassigned.Source: §1 · p. 23

2 List of scientific areas and categories

The section lists the scientific areas and detailed categories used to classify publications by their journals.

2 List of scientific areas and categories

  • SCImago’s Scopus classification assigns journal-based research categories across 27 main areas and 308 fine-grained categories.Source: §2 · p. 24
  • The listed areas include agricultural and biological sciences, arts and humanities, and social sciences.Source: §2 · pp. 24, 27
  • The detailed categories also cover computer science, decision sciences, medicine, and health professions.Source: §2 · pp. 25–26

3 Multiple assignment of journals into categories

SCImago journals can be assigned to multiple categories, so the study accounts for the number of categories attached to each journal.

3 Multiple assignment of journals into categories

  • Most journals are assigned to one category, while the maximum number of category assignments is 12.Source: §3 · p. 28
  • Scholar production in a category is normalized by the number of categories assigned to the journal.Source: §3 · p. 28
  • A journal’s multi-category assignments are shown as a histogram of category counts.Source: Figure 4 · p. 28, §3 · p. 28

4 List of areas and categories for UCSD classification

The UCSD classification provides the category scheme used to compare the research space with a citation-based science map.

4 List of areas and categories for UCSD classification

  • The UCSD classification contains 554 categories grouped into 13 main areas and is used for comparisons with the UCSD science map.Source: §4 · p. 28
  • The listed categories span biology, chemistry, and earth sciences.Source: §4 · pp. 29–30
  • The UCSD lists also include humanities, medical specialties, and social sciences.Source: §4 · pp. 31–32

5 Research Space in Scimago classification

The paper presents the research space using the Scimago classification. Its map encodes communities by color and degree centrality by node size, while Scimago uses Scopus alone and UCSD has roughly twice as many categories.

5 Research Space in Scimago classification

  • Scimago is based entirely on Scopus, whereas UCSD uses clustering over Web of Science and Scopus datasets and includes roughly twice as many categories.Source: §5 · p. 33
  • In the Scimago research-space map, node colors mark Infomap communities and node size is proportional to degree centrality.Source: Figure 5 · p. 33

6 Overlay maps for countries and institutions

The research space visualizes inactive, nascent, intermediate, and developed fields for countries, universities or research institutions, and scholars. The overlay-map examples include comparative advantages for India, the Netherlands, and Venezuela in 2008–2010.

6 Overlay maps for countries and institutions

  • The research space visualizes countries’, universities’ or research institutions’, and scholars’ fields as inactive, nascent, intermediate, or developed.Source: §6 · p. 34
  • The captions label Fig. 6 as comparative advantages of India, Fig. 7 of the Netherlands, and Fig. 8 of Venezuela, each covering 2008-2010.Source: Figure 6 · p. 34, Figure 7 · p. 35, Figure 8 · p. 35
Loading 1602.08409v3…