Bifrost®: social science education

v. dev (8f744784)

Niemi, Kristian. (2026, 18 juni). Bifrost®-analys: social science education. Karlstads universitet. https://bifrost.kau.se/forskning/amnesdidaktik/smh/social_science_education.html

1 928
2026*: 65
1 182
61% of total 2026*: 72%
+8.4%
Average annual growth rate: 1965–2025
25%
Among level-classified journals
2026*: 34%
1339
2026*: 62
178
2026*: 5
*Year may be incomplete
About key indicators

Key indicators summarise the report’s central metrics. All values are calculated from the underlying dataset and refer to the full period unless otherwise stated. Percentages (peer-reviewed, Open Access, international collaboration) are calculated as a share of total publications per year.

Percentage change is not shown when the base value is below 10 units, as small base values produce statistically unstable percentages (Hicks et al. (2015), principle 8; cf. CDC rule for n < 16). Absolute values are shown instead.

The following query was used:
(“samhällskunskap” OR (civics education) OR (civic education) OR “social studies education” OR (citizenship education)) NOT hsv:(Arbetsterapi) NOT hsv:(Lantbruksvetenskap, skogsbruk och fiske) NOT hsv:(Annan medicin och hälsovetenskap) NOT hsv:(Naturresursteknik) NOT hsv:(Klinisk medicin) NOT hsv:(Annan naturvetenskap) NOT hsv:(Elektroteknik och elektronik) NOT hsv:(Maskinteknik) NOT hsv:(Medicinska och farmaceutiska grundvetenskaper) NOT hsv:(Materialteknik) NOT hsv:(Kemi) NOT hsv:(Annan lantbruksvetenskap) NOT hsv:(Miljöbioteknik) NOT hsv:(Husdjursvetenskap) NOT hsv:(Veterinärmedicin) NOT hsv:(Nanoteknik) NOT hsv:(Nanoteknologi med applikationer på växter och djur) NOT hsv:(Annan fysik) NOT hsv:(Sannolikhetsteori och statistik) NOT hsv:(Folkhälsovetenskap global hälsa socialmedicin och epidemiologi) NOT hsv:(Mikrobiologi)
Database: SwePub
Data quality: poor
Publication activity
Publication types
Number of scientific publications per type over years
61%
Peer-reviewed
2026: 72%
97%
Scientific
2026: 98%
845
Unique journals
2026: 43
25%
Level 2 (Norwegian list)
2026: 34%
157 of 627 classified

Insights
The peer-reviewed share: 67.4 % recent decade (2017–2026), up from 58.9 % previous decade (2007–2016). Long-term trend (1965–2026): increasing. Note that 0% in the first year of the period may reflect incomplete metadata rather than an actual absence of peer review.

1867 publications (97%) scientific, 61 publications (3%) other.

Older years (1965–2018) aggregated for readability. Full timespan available in data export.

Publication types over time
Proportional view
Journals: peer reviewed and other scientific
Missing match in the Channel Register (HK-dir). Common causes: missing ISSN in source data, channels outside the register, conference series, or recently launched journals. Lack of classification does not necessarily mean the journal lacks peer review.
Publications by NPI level

The Norwegian Publication Indicator (NPI), also known as the Norwegian list, classifies publication channels into two levels. Level 2 (top ~20% per field) is considered the most prestigious channels. Level 1 covers other approved channels.

Method and limitations
Data source
SwePub
Time period
1965–2026
Counting method
NPI level (1 or 2) is retrieved from the Norwegian Channel Register (HK-dir) via ISSN/ISBN matching. Publications without a match are assigned level X.
Limitations
  • The channel register does not cover all scientific publishing. Publications outside the channel list lack an NPI level and are counted as level X.
  • Level-based indicators should be interpreted contextually and not used as the sole quality measure. Hicks et al. (2015)
  • NPI classification is sourced from the Norwegian channel register (HK-dir) and does not cover all scientific publishing. Publications outside the channel list have no NPI level.
  • Data reflects publishing activity retrieved from SwePub and may differ from the institution’s internal statistics.
25.0%
Level 2
2026: 34.5%
157
Level 2 (Count)
2026: 10
445
Level 1 (Count)
2026: 19
1301
Unclassified
67.5% of total

A high proportion of publications (≥10%) lack NPI classification. Common causes: missing ISSN in source data, channels outside the register (~40,000 journals), conference series, or recently launched journals. See the journal table for unclassified channels.

NPI level by year
Method and limitations
Data source
SwePub + Kanalregisteret (HK-dir)
Time period
1965–2026
Counting method
Full counting — each publication counted as one unit
Limitations
  • Volume measures count publications, not pages published or contribution size.
  • Conference papers may be underrepresented in the source database, particularly for older periods and certain disciplines.
  • Data reflects publishing activity retrieved from SwePub and may differ from the institution’s internal statistics.
  • NPI classification is sourced from the Norwegian channel register (HK-dir) and does not cover all scientific publishing. Publications outside the channel list have no NPI level.
DORA

DORA mode is enabled for this report. Bifrost evaluates the report against the principles of DORA (2012) and CoARA (2022).

The report contains elements that are not compatible with DORA principle 1: “Do not use journal-based metrics as a surrogate measure of the quality of individual research articles.”

NPI level classification (Level 1/2) is shown in the report. NPI is a Nordic classification system that ranks publication channels by academic standing, i.e. a channel-based ranking that DORA advises against using as a quality surrogate. The classification is presented here as descriptive information about publication patterns, not as a measure of individual article quality.

Researchers

Researchers are listed below, sorted by scientific productivity.

2 036
Unique researchers
2026: 200
8%
Top 10 researchers’ publication share
The 10 most productive (of 2 036 total) researchers’ share of all publications
2.0
Co-authors/pub.
2026: 3.2

Insights
The top 20% most productive researchers account for 39% of publications (Gini 0.22, scale: 0 = even, 1 = fully concentrated). Average number of co-authors: 2.4 recent decade (2017–2026), up from 1.8 (2007–2016).

Method and limitations
Data source
SwePub
Time period
1965–2026
Counting method
Full counting — each publication counted as one unit
Limitations
  • Lists show the most productive researchers by volume. Rankings reflect registered publishing activity, not scientific quality or impact.
  • This section is descriptive. Individual researchers are not evaluated; the measure is a group result at aggregate level.
  • Data reflects publishing activity retrieved from SwePub and may differ from the institution’s internal statistics.
Collaboration
Co-authorship

Insights
78 research groups of roughly equal size; no single cluster dominates. Each researcher collaborates with an average of 3.3 others (a moderately connected network). Clear cluster structure (Modularity 0.98); researchers primarily work within their own group. The network is sparse: only 1.0% of all researcher pairs have a direct collaboration link.

Methodology

Each node represents a researcher and each link a co-authorship. Colors indicate research groups (clusters) identified via modularity analysis. Node size reflects number of publications.

The co-authorship network is built from co-authored publications. Each node represents a researcher, and each edge is weighted by number of joint publications. Edge weights are normalized using association strength Van Eck et al. (2009) before clustering with the Louvain algorithm. Centrality measures: degree (number of collaborators), collaboration intensity (total co-authoring frequency), and bridge score (weighted betweenness using inverse weights) Newman (2004). Network density measures the proportion of realized vs. possible collaborations. Terminology: «Collaborators (avg)» = mean degree; «Clustering» = modularity Blondel et al. (2008).

3.3
Collaborators (avg)
1965–2026
0.98
Clustering
1965–2026
Degree distribution
Summary

Percentages are calculated on pairs where both authors have country data (2038 classified of 2040 total, coverage 100%). Of which 2 pairs where both authors lack institutional affiliation, 0 pairs where the institution could not be mapped to a country, and 0 pairs where one side lacks data.

Individual network statistics

‘Co-authored texts’ indicates the number of texts the author has written together with one or more co-authors.

Network statistics show central nodes in the collaboration network. Degree is the number of direct collaborations, while betweenness shows which authors act as bridges between different groups.

Most common co-authorships

The first co-author listed is the one the author has written with the most times. ‘Number’ indicates the number of co-authored texts with the author. Up to four additional co-authors are listed, in descending order of co-authorship.

Network of co-authors

Below is a visualization of the 78 different groupings in the dataset. The colors indicate different groups.

The network shows 335 of 2036 co-authors: those who share at least 2 publication with another.

Why are not all co-authors shown?

A co-author is only included in the network once they have co-authored at least 2 publication with another. Pairs who meet in only a single publication are therefore excluded. The threshold dampens noise so recurring collaborations stand out more clearly.

Publications with more than 25 authors are excluded when building the network. These are typically meta-studies, systematic reviews, and large consortium articles where the full author list is printed. Letting them in would connect nearly everyone to nearly everyone else.

Because the network is large, weak links have also been dropped via backbone filtering (335 nodes, 552 edges in the original graph). The groups above are computed on the filtered, sparser network.

Adaptive visualization (large): 335 nodes / 552 edges → 140 nodes / 232 edges after filtering.

The visualization was restored to the unfiltered source network: the adaptive filters would otherwise have reduced it below the threshold for a meaningful display.

Group membership

Author names to the right; group ID to the left. You can see the size of the groups and the most common keywords of the groups in the tables that follow. A combination of search and sorting can be used to further explore group membership.

Group size
Group keywords

The table is limited to a) groups with more than 3 members; b) groups with at least one keyword in any publication; c) the ten most used keywords per group.

Method and limitations
Data source
SwePub
Time period
1965–2026
Counting method
Full counting — each publication counted as one unit
Limitations
  • A link in the network means two researchers share at least one publication in the selection; link strength indicates the number of co-authored works. Informal collaborations and unpublished projects are not visible.
  • Short time periods or small research groups produce sparse networks. Isolated nodes indicate researchers with few registered collaborations in the selection, not absence of collaboration in general.
  • Pairs sharing fewer than 2 publications are not included in the network (makeCoauthorMinPubs).
  • Publications with more than 25 authors are excluded from the co-authorship analysis (makeMaxAuthorsPerPub).
  • For networks exceeding 200 nodes or 500 edges, adaptive edge reduction is applied to the visualization (disparity filter, Serrano et al. (2009), or quantile threshold depending on size). Network statistics (centrality, cluster membership) are always computed on the full graph.
  • Whole counting: each shared publication contributes weight 1 per pair. To give each publication equal total weight regardless of author count, enable makeFractionalCounting = TRUE (per-publication 1/(N−1) weighting following Perianes-Rodriguez et al. 2016). Perianes-Rodriguez et al. (2016)
  • Association strength: AS(i,j) = w_ij / (k_i × k_j / 2m). The normalization reduces dominance of high-degree nodes. Van Eck et al. (2009)
  • Degree indicates the number of unique collaboration partners (network topology). Strength indicates total co-publication intensity (edge weights). A researcher with high Degree but low Strength has many shallow collaborations; conversely, high Strength with low Degree indicates few but intensive collaborations. Opsahl et al. (2010)
  • Cluster colors are based on co-publication patterns, not organizational affiliation. A cluster’s composite research groups may cross organizational boundaries.
  • Data character classified as large. Backbone filter, node cap, and cluster targets adjusted automatically.
  • Parameters: backbone α = 0.10, min. co-publications = 2, node cap = 120.
  • Leiden clustering: resolution parameter γ = 1.20 (deviates from default 1.0; calibrated for current corpus size).
  • Quality gates triggered and adjusted parameters: QG4_fragmented; QG4_fragmented; QG4_fragmented.
  • International collaboration filter configured but not applied in this version (FM4 support requires per-author country data, planned for a future release).
  • Data reflects publishing activity retrieved from SwePub and may differ from the institution’s internal statistics.
Bibliometric network visualizations complement, rather than replace, expert judgment. Van Eck et al. (2014)

See also the supervisor and opponent network in the researcher section for an analysis of academic collaboration patterns beyond co-authorship.

Supervisor and opponent network

Unlike the co-authorship analysis, which maps collaboration through joint publications, this section reveals the academic networks that emerge through dissertation supervision and opposition. Supervisors and opponents active at multiple institutions form informal knowledge bridges between organizations — relationships rarely captured by traditional bibliometric measures but which can reveal important patterns in academic knowledge transfer.

Insights
106 researchers have supervised or opposed across institutional boundaries. Strongest connection: Dalarna University – Karlstad University (Connection strength: 6). Based on 80.9% of dissertations with identifiable supervisors.

Supervisions  Oppositions
Supervisor and opponent network, figure
Supervisor and opponent network, table
Method description

The supervisor/opponent network is separate from the international collaboration map. The map is based on co-authorship between author affiliations, while supervisor/opponent relations are shown in the network below.

The network is based on supervisor and opponent relationships extracted from SwePub records. Connection strength is calculated as (number of supervisions × 2) + (number of oppositions × 1). The weighting (2:1) is a Bifrost convention reflecting that supervision is a longer and deeper collaborative relationship than opposition. The method lacks established bibliometric practice; it was developed specifically for Bifrost.

The supervisor:opponent weighting (2:1) is a Bifrost convention to reflect the supervisor’s greater role in the dissertation process. This is not established bibliometric practice.

Higher education institutions
255
Institutions
30%
Top-3 share

Insights
245 institutions contribute. Stockholms universitet, Göteborgs universitet and Karlstads universitet account for 29 % — a broad distribution.

Method and limitations
Data source
SwePub
Time period
1965–2026
Counting method
Full counting of publication appearances
Limitations
  • The count shows co-author affiliation appearances, not unique publications. A publication with three co-authors from the same institution counts three times.
  • Institution names have been harmonised against an internal name list. Unmatched variants are shown separately or excluded.
  • A total of 60 entries in the raw data were excluded from the table: 50 country/city names (geographic entities, not institutions), 4 manually verified non-institutions (known_unmapped), 8 departments, faculties or centres (pattern-based filter). This is why the institution count may differ from the number of unique affiliations in the source data.
International collaboration

Overview of international collaboration based on co-authorship and affiliations in publications.

46
Collaboration countries
10 on average per year
11%
International collaboration

Insights
46 countries represented in collaborations. United Kingdom, Finland and Norway are most common. 9.8 % of publications involve international co-authors — up from 7 % (2007–2016) to 13 % (2017–2026).

Collaboration countries
Distribution per year
View:
Co-authorship by country

Based on co-author affiliation country.

Networks and publications, geographically

Insights:
The network comprises 98 institutions with 272 collaboration relationships. The strongest collaboration is between Stockholm University and Uppsala University (20 co-publications). Uppsala University has the most collaboration partners (20).

Period overview

The map primarily shows co-authorship between institutions. Supervisor/opponent links are shown as separate network relations and may be fewer, because only records with clear institutional affiliation can be included.

98
Institutions in network
1965–2026
272
Collaboration relationships
1965–2026
20
Co-publications: Stockholm University – Uppsala University
1965–2026
Method and limitations
Data source
SwePub + OpenStreetMap/Nominatim
Time period
1965–2026
Counting method
Full counting — each publication counted as one unit
Limitations
  • The international share is calculated only from author affiliation country. Conference location, publication country, and other metadata are not counted as international collaboration.
  • Country analysis is based on co-author affiliation country. Full counting means each country in a co-publication is counted once.
  • The network map shows collaboration relationships between institutions based on co-authored publications. Node size reflects the number of collaboration partners, edge width reflects collaboration strength (Salton’s cosine index).
  • Based on author affiliation data. Incomplete affiliation information may affect results.
Subject areas
Subject categories
Social Sciences (86%)
Dominant category
5
Subject areas (level 1)
0.32 / 1.00
Subject diversity (evenness)

Insights
Social Sciences dominates (83 %). Subject breadth has increased (1970–2026, H: 0.48 → 0.62). Moderate interdisciplinarity — research combines related subject areas. Rao-Stirling: 0.582 (where 0 = single discipline, 1 = maximum diversity). Based on 5 HSV main areas (Swedish classification). Rao-Stirling (Stirling, 2007)

Method: diversity indices

Shannon H (evenness index) measures how evenly publications are distributed across subject areas. A value of 1.00 means perfect evenness; lower values indicate dominance by individual areas. Rao-Stirling measures interdisciplinarity by weighing both the distribution and the taxonomic distance between subject areas according to the Swedish classification system. The scale ranges from 0 (all publications in one subject) to 1 (maximum spread across distant subject areas).

Level 1
Proportional view

Proportion of total publications per year (%). Note that a publication may belong to multiple categories.

Category frequency over time
Level 2

During 1965–2026, 7.6% (146 of 1,928) publications lack subject classification at this level.

Proportional view

Proportion of total publications per year (%). Note that a publication may belong to multiple categories.

Subject categories level 2 (table)
Method and limitations
Data source
SwePub
Time period
1965–2026
Classification system
HSV/UKÄ (5 nivåer), OECD FoS (3 nivåer)
Counting method
Full counting per category
Limitations
  • Explore which subject categories are represented in the dataset. Note that publications usually have several categories. Therefore, it is the rule, not the exception, that the percentages together constitute more than 100%. If x is 100% and y is 15%, it means that all publications have been categorized as x, and of them, 15% have also been categorized as y
  • A publication classified under multiple subjects is counted in each category. The sum therefore exceeds the publication count — this is correct, not an error.
  • Classification may vary in precision across institutions and periods. Comparisons should be made with caution.
  • The Rao-Stirling index is computed at portfolio level (subject code shares across the entire dataset), not as a mean of per-publication RS. Absolute RS values are not directly comparable to benchmarks based on per-article calculations or other classification systems.
Keywords
HSV subject categories have been filtered from keywords

barn; socialvetenskap; learning; pedagogical work; historieämnen; lärande; språkvetenskap; etnicitet; nordiska språk; genus; history of ideas; övrig samhällsvetenskap; religious studies and theology; ekonomi; mathematics; teknikhistoria; social psychology; socialpsykologi; vårdvetenskap; idrottsvetenskap; biblioteks- och informationsvetenskap; elektronik; socialmedicin; artificial intelligence (ai); teknikvetenskap; idrott

4 799
Unique keywords
2026: 239
samhällskunskap (12.4%)
Top keyword
2026: sweden

Insights
Broad keyword profile — no single term dominates (HHI: 0.0012 — Herfindahl-Hirschman Index, where 0 = perfectly even distribution, 1 = one term dominates entirely). Most common is “samhällskunskap” appearing in 12.4 % of publications, across a total of 4799.

Keywords that have been manually excluded

education

Colors indicate frequency quantiles within this dataset.

Red: Highest frequency (12.4-9.94%); Blue: High frequency (9.94-7.48%); Green: Medium frequency (7.48-5.02%); Orange: Low frequency (5.02-2.56%); Gray: Lowest frequency (2.56-0.1%)

Method and limitations
Data source
SwePub
Time period
1965–2026
Counting method
Full counting — each publication counted as one unit
Limitations
  • Keywords are a mix of author-assigned and automatically generated terms. Indexing consistency varies across sources and time periods.
  • English keywords often dominate. Publications in Swedish or other languages are therefore often underrepresented in frequency analyses.
  • Keywords occurring in more than 70% of all publications are automatically excluded to prevent generic terms from dominating the analysis.
  • Data reflects publishing activity retrieved from SwePub and may differ from the institution’s internal statistics.
Keyword Insights
Declining themes
Steady decline

The following keywords show a steady decline over the entire time period, without having had a distinct burst period previously:

A declining trend may reflect reduced research interest or a shift to newer terminology for the same research area.

citizenship ↓ (r = -0.64)

Early publishers (1999–2006):

Researchers: Olson, Maria (2), Almgren, Nina (1), Bergström (fd Boman), Ylva (1)

Most active period (2014–2016, 41 publications):

Researchers: Olson, Maria (11), Dahlstedt, Magnus (9), Fejes, Andreas (9)

Institutions: Linköpings universitet (27), Stockholms universitet (11), Uppsala universitet (7)

Co-varying keywords: education, democracy, adult education

Last 3 years: 20 publications

Note: A declining trend may indicate terminological shift rather than decreased research interest.

citizenship education ↓ (r = -0.72)

Early publishers (2000–2007):

Researchers: Roth, Klas (2), Bergström (fd Boman), Ylva (1), Boman, Ylva (1)

Most active period (2014–2016, 29 publications):

Researchers: Fejes, Andreas (4), Olson, Maria (4), Sandahl, Johan (4)

Institutions: Linköpings universitet (13), Stockholms universitet (11), Karlstads universitet (5)

Co-varying keywords: education, democracy, social studies

Last 3 years: 12 publications

Note: A declining trend may indicate terminological shift rather than decreased research interest.

civics ↓ (r = -0.78)

Early publishers (1985–1995):

Researchers: Bron Jr, Michal (6), Wojciechowska-Bron, Agnieszka (1)

Most active period (2009–2011, 18 publications):

Researchers: Eklund, Niklas (2), Larsson, Anna (2), Andersson, Ragnar (1)

Institutions: Karlstads universitet (8), Stockholms universitet (5), Umeå universitet (4)

Co-varying keywords: samhällskunskap, social studies, international education

Last 3 years: 15 publications

Note: A declining trend may indicate terminological shift rather than decreased research interest.

democracy ↓ (r = -0.54)

Early publishers (2000–2007):

Researchers: Olson, Maria (2), Boman, Ylva (1), Brock-Utne, Birgit (1)

Most active period (2013–2015, 14 publications):

Researchers: Olson, Maria (4), Dahlstedt, Magnus (3), Carpentier, Nico (2)

Institutions: Högskolan i Skövde (4), Linköpings universitet (4), Stockholms universitet (4)

Co-varying keywords: education, citizenship, citizenship education

Last 3 years: 9 publications

Note: A declining trend may indicate terminological shift rather than decreased research interest.

Historical Trends

The following keywords had periods of high activity in the past but have since declined. The analysis shows when they peaked, what drove the interest, and how activity has evolved since.

civics: historical trend (2007–2020)

Burst period: 2007–2020 (moderate burst)

Peak year: 2011 (10 pubs.)

Driving actors during period:

  • Researchers: Tväråna, Malin (7 pubs.), Bron Jr, Michal (6 pubs.), Larsson, Anna (3 pubs.)
  • Institutions: Stockholms universitet (21 pubs.), Karlstads universitet (13 pubs.), Södertörns högskola (8 pubs.)

Co-varying keywords: samhällskunskap, social studies, ämnesdidaktik

Current status: Declining

citizenship: historical trend (2012–2019)

Burst period: 2012–2019 (moderate burst)

Peak year: 2016 (17 pubs.)

Driving actors during period:

  • Researchers: Olson, Maria (19 pubs.), Fejes, Andreas (15 pubs.), Dahlstedt, Magnus (14 pubs.)
  • Institutions: Linköpings universitet (41 pubs.), Stockholms universitet (19 pubs.), Göteborgs universitet (11 pubs.)

Co-varying keywords: education, utbildning och lärande, adult education

Current status: Declining

citizenship education: historical trend (2013–2016)

Burst period: 2013–2016 (moderate burst)

Peak year: 2015 (14 pubs.)

Driving actors during period:

  • Researchers: Fejes, Andreas (5 pubs.), Olson, Maria (5 pubs.), Thornberg, Robert (5 pubs.)
  • Institutions: Linköpings universitet (16 pubs.), Stockholms universitet (12 pubs.), Karlstads universitet (6 pubs.)

Co-varying keywords: subject learning and teaching, ämnesdidaktik, citizenship

Current status: Declining

subject learning and teaching: historical trend (2013–2022)

Burst period: 2013–2022 (moderate burst)

Peak year: 2015 (11 pubs.)

Driving actors during period:

  • Researchers: Sandahl, Johan (13 pubs.), Olson, Maria (11 pubs.), Björklund, Mattias (6 pubs.)
  • Institutions: Stockholms universitet (50 pubs.), Linköpings universitet (13 pubs.), Högskolan i Skövde (7 pubs.)

Co-varying keywords: ämnesdidaktik, citizenship, samhällskunskap

Current status: Stable

social studies: historical trend (2018–2021)

Burst period: 2018–2021 (moderate burst)

Peak year: 2019 (17 pubs.)

Driving actors during period:

  • Researchers: Tväråna, Malin (8 pubs.), Björklund, Mattias (7 pubs.), Jägerskog, Ann-Sofie (4 pubs.)
  • Institutions: Stockholms universitet (20 pubs.), Göteborgs universitet (7 pubs.), Umeå universitet (6 pubs.)

Co-varying keywords: samhällskunskap, civics, ämnesdidaktik med inriktning mot de samhällsvetenskapliga ämnenas didaktik

Current status: Rising

Method and limitations
Data source
SwePub
Time period
1965–2026
Counting method
Burst detection uses an automaton model implemented via the bursts package. The method identifies periods of statistically significant increased occurrence of individual keywords. Kleinberg (2003)
Limitations
  • Burst detection requires at least 5–10 years of data for reliable results. Short time series may produce unstable or misleading burst periods.
  • Keywords are a mix of author-assigned and automatically generated terms. Indexing consistency varies across sources and time periods.
  • English keywords often dominate. Publications in Swedish or other languages are therefore often underrepresented in frequency analyses.
  • Keywords occurring in more than 70% of all publications are automatically excluded to prevent generic terms from dominating the analysis.
  • Data reflects publishing activity retrieved from SwePub and may differ from the institution’s internal statistics.
Keyword co-occurrence

The heatmap shows how often keywords co-occur in the same publications. Association strength Van Eck et al. (2009) normalizes co-occurrence by the product of the individual keyword frequencies. Red asterisks () in the upper-right corner of cells mark statistically significant co-occurrences (p < 0.05).

The heatmap shows co-occurrence strength for all pairwise combinations of the most frequent keywords, including weak relations. It is a complete N×N matrix. The keyword network below complements this view: it shows cluster structure through backbone filtering that hides weak edges to emphasize strong patterns — the two visualizations measure the same thing but illuminate different aspects.
Top co-occurring keyword pairs
Method and limitations
Data source
SwePub
Time period
1965–2026
Counting method
Co-occurrence matrix strength is computed using association strength: c_ij / (s_i × s_j / 2m), where s is document frequency and m is the sum of pairwise co-occurrences. Van Eck et al. (2009)
Limitations
  • Association strength is normalized following Van Eck & Waltman (2009): AS(i,j) = c_ij / (c_i × c_j / 2m), where c_ij is the document frequency for the pair, c_i and c_j are the document frequencies for the individual keywords, and m is the sum of co-occurrences over unique pairs (upper triangle of the co-occurrence matrix). Van Eck et al. (2009)
  • Co-occurrence is counted per document (whole counting). Fractional counting is not applied at the keyword level — a deliberate choice because the keywords are controlled terms (SwePub) or extracted concepts (OpenAlex), not free-text author names.
  • The heatmap displays the 15 most frequent keywords (ranked by document frequency). Keywords below the minimum frequency threshold are excluded. The count can be adjusted in the report configuration.
  • Statistical significance testing uses the hypergeometric distribution at the pair level. No correction for multiple testing is applied — the analysis is exploratory, not confirmatory.
  • The heatmap remains readable up to top_n ≤ 25. For larger datasets — see also the keyword network.
  • The heatmap color scale is clipped at the 95th percentile of observed association strengths. Association strength with 2m rescaling is mathematically unbounded: rare keyword pairs with low individual document frequencies can produce extreme values that otherwise dominate the scale and render other cells invisible. Actual maximum values are shown in the top-pairs table.
  • Graph-based visualizations degrade for large networks (Van Eck & Waltman, 2014, pp. 288–289). The heatmap is designed for top-N pairs; the network graph uses backbone filtering to handle larger networks. Van Eck et al. (2014)
  • Keywords are a mix of author-assigned and automatically generated terms. Indexing consistency varies across sources and time periods.
  • English keywords often dominate. Publications in Swedish or other languages are therefore often underrepresented in frequency analyses.
  • Keywords occurring in more than 70% of all publications are automatically excluded to prevent generic terms from dominating the analysis.
  • Data reflects publishing activity retrieved from SwePub and may differ from the institution’s internal statistics.
Network diagram, keywords

Keywords that frequently co-occur in the same publications form thematic clusters. The table below summarizes the clusters; the interactive graph shows the relationships visually.

The network shows cluster structure and network position: backbone filtering retains statistically significant edges and hides weak relations to emphasize strong patterns. The co-occurrence heatmap above complements this view: it shows the full N×N matrix for the most frequent keywords, including weak pairs not visible in the network.

The network diagram shows how keywords relate to each other based on co-occurrence in publications. Larger nodes mean more frequent keywords. Lines show co-occurrence. Colors indicate thematic clusters identified using the Leiden algorithm (Traag et al., 2019).

Cluster overview
Method and limitations
Data source
SwePub
Time period
1965–2026
Counting method
Co-occurrence matrix strength is computed using association strength: c_ij / (s_i × s_j / 2m), where s is document frequency and m is the sum of pairwise co-occurrences. Van Eck et al. (2009)
Limitations
  • Adaptive filtering (scale: large): backbone α=0.05, min co-occurrence=3. Nodes 40→35, edges 175→47, density=0.079.
  • The SDSM filter tests each keyword pair against a stochastic null model that controls for both keyword frequency and the number of keywords per publication. Edges that do not deviate significantly are removed. Neal (2022)
  • Clusters are detected using the Leiden algorithm Traag et al. (2019), which identifies thematic groups where keywords co-occur more strongly within the group than between groups.
  • The network is limited to the most frequent keywords (top N). Rare keywords are excluded, which may hide emerging topics.
  • Filter parameters are selected by corpus size. Small corpora (< 50 publications or < 20 unique keywords) get no backbone filter. Medium (50–299 publications with 20–149 keywords) use backbone α=0.15 and 60 keywords. Large (≥ 300 publications or, at ≥ 50 publications, ≥ 150 keywords) use α=0.05 and 40 keywords. As an exception, small publication sets with rich vocabulary (< 50 publications but ≥ 150 keywords — typically OpenAlex individual reports) use the medium operating point rather than large, because the aggressive filter otherwise fragments the network. Thresholds are empirical operating points and can be overridden per report.
  • Graph-based layouts degrade for large networks (Van Eck et al., 2014). Above 200 nodes, visual clarity depends on backbone filtering and node reduction. Statistics (centrality, cluster membership) are always computed on the complete graph.
  • The number of keywords in the network is adapted to the material: roughly 30 % of the unique keywords are included, with a floor of 60 and a ceiling of 250. Of these, roughly 65 % (at least 30, at most 70 nodes) are shown initially to keep the graph readable. The remaining keywords can be added via the slider below the graph. These limits prevent both sparse and overloaded networks.
  • Keywords are a mix of author-assigned and automatically generated terms. Indexing consistency varies across sources and time periods.
  • Keywords in Swedish and English are mixed without separation. Lemmatization is not applied: ‘learning’ and ‘learners’ are treated as separate keywords. This is a deliberate choice for controlled terms, not a deficiency — but it affects the concentration around English-language concepts.
  • Data reflects publishing activity retrieved from SwePub and may differ from the institution’s internal statistics.
Methodology: adaptive visualization
Data character
manual
Filter parameters
α=0.05, top_n=40, min_cooc=3
Before filtering
40 nodes, 175 edges
After filtering
35 nodes, 47 edges
Edge density
0.079
Quality gate
Not triggered
Word frequency

The frequency of individual words in the dataset as a whole. Words have been taken from title, abstract, and keywords. “Frequency” is total uses, including the number of mentions in the same text, while “publications” is the number of unique texts where the word appears.

25 931
Unique words
education (84.8%)
Top word

Insights
The 5 most common words (by share of publications) are: “education” (85%), “citizenship” (56%), “social” (42%), “school” (32%), “swedish” (31%). These patterns reflect the thematic core of the dataset. Note that geographic markers such as “swedish” are common in academic metadata and reflect the national affiliation of publications rather than the research topic.

Word frequency table

Notice: The dataset contains 25931 rows. For best performance, only the 8000 with the highest frequency are displayed in the table.

Word stems

Here is a summary of word stems, i.e. parts of words, in the dataset as a whole. Words have been taken from title, abstract, and keywords. “Frequency” is total uses, including the number of mentions in the same text, while “publications” is the number of unique texts where the word appears.

Method and limitations
Data source
SwePub
Time period
1965–2026
Counting method
Full counting — each publication counted as one unit
Limitations
  • Word frequencies are derived from titles and abstracts in the dataset. Swedish and English stop words are filtered before calculation.
  • The analysis is language-dependent and does not merge synonymous terms across languages. No stemming or lemmatization is applied — word forms are counted separately.
  • Data reflects publishing activity retrieved from SwePub and may differ from the institution’s internal statistics.
Word Frequency Trends

These trends are indicative and complement the keyword analysis above.

Methodological note: Word frequency analysis is based on individual words extracted from title, abstract, and keywords. Unlike author-selected keywords, individual words can be noisier and more ambiguous — for example, the word ‘system’ may appear in both technical and social science contexts, while the keyword ‘adaptive systems’ is more precise. Stricter thresholds are used (minimum 10 occurrences, correlation > 0.5) and academic stopwords are excluded.

Rising/declining shows trends over time (Spearman correlation). New/disappearing shows lifecycle — when words started or stopped being used.

No statistically significant trends were identified among the most common words.

Publications
Theses

The first doctoral thesis in the dataset is from 1965, Samhällskunskap som skolämne: målsättningar, kursinnehåll och arbetssätt på den grundläggande skolans högstadium by Bromsjö, Birger. From then until 2026, a total of 178 theses have been registered. Of these, 148 are doctoral theses and 30 are licentiate theses.

178
Theses
2026: 5
Supervisors
Opponents
Method and limitations
Data source
SwePub
Time period
1965–2026
Counting method
Full counting — each publication counted as one unit
Limitations
  • Thesis data is sourced from the selected data source. Information on type, supervisor, and opponent depends on how the registering institution has entered the data.
  • Coverage may be incomplete: theses not registered in the data source are not visible in the analysis. Historical theses are often underrepresented.
  • Data reflects publishing activity retrieved from SwePub and may differ from the institution’s internal statistics.
Publications

A complete list of the search results. Initially sorted by year (descending) and author (ascending). Change the order at the column header. Search can be done over all displayed fields.


Method references

Blondel, V. D., Guillaume, J.-L., Lambiotte, R., & Lefebvre, E (2008). Fast unfolding of communities in large networks. Journal of Statistical Mechanics: Theory and Experiment, P10008. https://doi.org/10.1088/1742-5468/2008/10/P10008

Bornmann, L., & Marx, W (2018). Critical rationalism and the search for standard (field-normalized) indicators in bibliometrics. Journal of Informetrics, 12(3), 598–604. https://doi.org/10.1016/j.joi.2018.05.002

CoARA (2022). Agreement on Reforming Research Assessment. https://coara.eu/agreement/the-agreement-full-text/

DORA (2012). San Francisco Declaration on Research Assessment. https://sfdora.org/read/

Hicks, D., Wouters, P., Waltman, L., de Rijcke, S., & Rafols, I (2015). Bibliometrics: The Leiden Manifesto for research metrics. Nature, 520(7548), 429–431. https://doi.org/10.1038/520429a

Kleinberg, J (2003). Bursty and hierarchical structure in streams. Data Mining and Knowledge Discovery, 7(4), 373–397. https://doi.org/10.1023/A:1024940629314

Mann, H. B (1945). Nonparametric tests against trend. Econometrica, 13(3), 245–259. https://doi.org/10.2307/1907187

Neal, Z. P (2022). backbone: An R package to extract network backbones. PLOS ONE, 17(5), e0269137. https://doi.org/10.1371/journal.pone.0269137

Newman, M. E. J (2004). Analysis of weighted networks. Physical Review E, 70(5), 056131. https://doi.org/10.1103/PhysRevE.70.056131

Opsahl, T., Agneessens, F., & Skvoretz, J (2010). Node centrality in weighted networks: Generalizing degree and shortest paths. Social Networks, 32(3), 245–251. https://doi.org/10.1016/j.socnet.2010.03.006

Perianes-Rodriguez, A., Waltman, L., & Van Eck, N. J (2016). Constructing bibliometric networks: A comparison between full and fractional counting. Journal of Informetrics, 10(4), 1178–1195. https://doi.org/10.1016/j.joi.2016.10.006

Sen, P. K (1968). Estimates of the regression coefficient based on Kendall’s tau. Journal of the American Statistical Association, 63(324), 1379–1389. https://doi.org/10.2307/2285891

Serrano, M. Á., Boguñá, M., & Vespignani, A (2009). Extracting the multiscale backbone of complex weighted networks. Proceedings of the National Academy of Sciences, 106(16), 6483–6488. https://doi.org/10.1073/pnas.0808904106

Stirling, A (2007). A general framework for analysing diversity in science, technology and society. Journal of The Royal Society Interface, 4(15), 707–719. https://doi.org/10.1098/rsif.2007.0213

Traag, V. A., Waltman, L., & Van Eck, N. J (2019). From Louvain to Leiden: guaranteeing well-connected communities. Scientific Reports, 9, 5233. https://doi.org/10.1038/s41598-019-41695-z

Van Eck, N. J., & Waltman, L (2009). How to normalize cooccurrence data? An analysis of some well-known similarity measures. Journal of the American Society for Information Science and Technology, 60(8), 1635–1651. https://doi.org/10.1002/asi.21075

Van Eck, N. J., & Waltman, L (2014). Visualizing bibliometric networks. In Ding, Y., Rousseau, R., & Wolfram, D. (Ed.), Measuring Scholarly Impact: Methods and Practice (pp. 285–320). Springer. https://doi.org/10.1007/978-3-319-10377-8_13