Data source

Article records are drawn from CrossRef using the Metadata API, queried by journal ISSN. For each journal, Pinakes fetches all articles with DOIs and records their publication date, title, authors, and abstract where available. A small number of journals are indexed via RSS feeds or web scraping when CrossRef coverage is unavailable; these records may lack structured dates.

Chart construction

Articles are grouped by publication year and journal. The stacked bar chart shows the count of articles published per year for each journal. Years with zero articles for a given journal appear as gaps in that journal’s color band. The “Show top 8” toggle filters to the eight journals with the most articles across all years, reducing visual clutter for trend analysis.

Interpretation and limitations

The timeline reflects indexed output, not total output. Journals added to the index more recently show data only from their first fetch onward, even if they published earlier. Journals that publish infrequently or that deposit metadata irregularly with CrossRef may show apparent gaps that reflect data availability rather than actual publication hiatuses. The date used is the CrossRef published-print or published-online date, whichever is earlier; some publishers backdate or batch-release content, which can create artificial spikes.

Tagging system

Pinakes uses a controlled vocabulary of 61 disciplinary topics developed to capture the major research areas in rhetoric and composition. Each article is tagged automatically by matching its title and abstract text against keyword patterns for each topic. An article can receive multiple tags. The tagging is rule-based (string matching with word boundaries), not machine-learned; it trades recall for precision, preferring to miss a marginal match rather than over-assign topics.

Co-occurrence matrix

The heatmap displays a symmetric matrix where each cell (i, j) contains the number of articles tagged with both topic i and topic j. Diagonal cells show the total count for each topic alone. The color intensity uses a sequential scale from white (zero co-occurrences) to dark brown (highest co-occurrence count in the matrix). Only topic pairs with at least one co-occurrence are colored.

Interpretation and limitations

High co-occurrence between two topics means they frequently appear on the same article, suggesting thematic overlap or a recognized sub-field at their intersection. However, topics with very high base rates (e.g., “pedagogy”) will naturally co-occur with many others simply because they are common — the matrix does not normalize for frequency. Articles lacking abstracts receive fewer tags, so journals indexed via RSS or scraping are systematically underrepresented. The 61-term vocabulary was designed for rhetoric and composition specifically and may not adequately capture interdisciplinary work that uses different terminology.

Data source

Author names are extracted from CrossRef metadata for each article. Co-authorship is inferred from shared bylines: if two authors appear on the same article, they share a co-authorship edge. Author names are normalized (case-folded, whitespace-trimmed) but not disambiguated — two authors with the same name are treated as one person, and the same author with variant name forms (e.g., with or without middle initial) may appear as separate nodes.

Graph construction

The visualization displays an undirected, weighted graph where each node is an author and each edge connects two authors who have co-authored at least one indexed article. Edge weight equals the number of co-authored articles. The node set is filtered to authors with at least n indexed publications (configurable via the “Min. publications” dropdown), then capped at the top k by total publication count. Only edges between nodes in the displayed set are shown.

Layout algorithm

The graph uses a force-directed layout (d3-force) where nodes repel each other via simulated charge and edges act as springs pulling co-authors together. Stronger co-authorship ties (more shared papers) produce shorter, thicker edges. The simulation iterates until it reaches a visually stable configuration, causing research communities to form visible clusters.

Interpretation and limitations

Clusters in the graph reflect collaborative communities within the indexed journals. However, the network reflects only publications within this index: scholars who collaborate extensively in books, edited collections, or journals not indexed here will appear less connected than their actual collaborative record warrants. Solo-authored work — still the dominant mode in rhetoric and composition — is invisible in this graph. Prolific solo authors appear as isolates (unconnected nodes), which is an accurate representation of their collaborative profile within the index but not necessarily of their broader influence.

Author name disambiguation remains an unsolved problem in bibliometrics. Common names may inflate connectivity, while name variants for the same author may understate it.

The Author Network shows collaborative relationships (co-authorship). For the field’s perceived intellectual relationships — which scholars are cited together regardless of whether they have collaborated — see Author Co-Citation.

What author co-citation measures

Article-level co-citation (shown in the Co-Citation tab) identifies which texts the field treats as intellectually related. Author co-citation identifies which scholars the field treats as intellectually related. The distinction matters because an author’s perceived intellectual position is not reducible to any single article. A scholar who has published twenty articles across three journals occupies a different position in the field’s mental map than any one of those articles does individually. Author co-citation captures that aggregate perception.

Two authors are co-cited when a third article includes at least one work by each in its reference list. The co-citation count for a pair (A, B) is the number of distinct articles in the index that cite at least one work by A and at least one work by B. A high count means that many subsequent authors, independently, have treated A and B as relevant to the same argument. This is a strong signal of perceived intellectual kinship — stronger than shared topic tags, shared journal venue, or even direct citation, because it reflects the collective judgment of the citing community rather than the choice of any single author.

The method was introduced by White & Griffith (1981), who used it to map the intellectual structure of information science. It has since become one of the standard tools in bibliometrics for identifying schools of thought, tracing the evolution of research fronts, and revealing how scholarly communities organise knowledge at the person level.

Relationship to other tools in this index

The Author Network tab shows co-authorship: who has written together. Co-authorship is a social relationship — it reflects collaboration, mentorship, and institutional proximity. Author co-citation is an intellectual relationship — it reflects how the citing community perceives two scholars’ work, regardless of whether those scholars have ever met, spoken, or collaborated.

The Co-Citation tab shows article-level co-citation. Author co-citation is a strict aggregation of the same underlying data: every article-level co-citation pair (A, B) is rolled up to the author level by mapping each article to its author(s). The author-level view is coarser but more stable — it smooths over individual articles and reveals structural patterns that are harder to see at the article level.

How co-citation pairs are counted

For each article C in the index that has outbound references (a reference list deposited with CrossRef), the analysis identifies all pairs of cited articles and maps each to its author(s). If article C cites article X by author A and article Y by author B, that produces one co-citation instance for the pair (A, B). If X has two authors (A₁, A₂) and Y has three authors (B₁, B₂, B₃), one co-citation of (X, Y) produces six author-level co-citation instances.

Author pairs who appear on the same cited article are excluded from the count. Without this correction, co-authors would be inflated: any time one of their joint articles is co-cited with anything, both authors would register as co-cited with the same targets. The exclusion ensures that co-citation reflects the citing community’s perception, not the authorship structure of cited works.

Graph construction

The visualisation displays an undirected, weighted graph where each node is an author and each edge connects two authors with a co-citation count above the configured threshold. Edge thickness is proportional to co-citation frequency. Node size reflects total co-citation strength: the sum of all edge weights for that node, measuring how deeply embedded the scholar is in the field’s co-citation structure. The graph is capped at a configurable number of nodes, ranked by total co-citation strength, to keep the visualisation responsive.

Temporal filtering

The year-range filter restricts which citing articles contribute co-citation instances. Filtering to 2015–2025 shows how the field has organised its intellectual map in the last decade — which may differ substantially from the all-time view, since disciplinary turns can reshape which scholars are cited together.

Limitations

Author co-citation reflects perception within this index only. Citations from books, dissertations, or journals not indexed here are invisible. Author name disambiguation is imperfect: common names may merge distinct scholars, while name variants may split counts. Multi-authored articles inflate co-citation counts for prolific co-authoring teams relative to solo authors. Co-citation is retrospective: recently active scholars may not yet appear prominently because their work has had less time to be co-cited.

References

• White, H. D., & Griffith, B. C. (1981). Author cocitation: A literature measure of intellectual structure. Journal of the American Society for Information Science, 32(3), 163–171.
• White, H. D., & McCain, K. W. (1998). Visualizing a discipline: An author co-citation analysis of information science, 1972–1995. Journal of the American Society for Information Science, 49(4), 327–355.
• Small, H. (1973). Co-citation in the scientific literature: A new measure of the relationship between two documents. Journal of the American Society for Information Science, 24(4), 265–269.

Data source

Citation counts are derived from reference lists deposited by publishers with CrossRef. When an indexed article includes another indexed article in its reference list, that counts as one internal citation. Citations from or to articles outside this index are not tracked. The cite_fetcher.py script matches DOIs in reference lists against DOIs in the articles table.

Counting method

Each article’s citation count reflects the number of distinct indexed articles that cite it. Self-citations (an article citing itself) are included if they appear in the CrossRef reference list. The list is ranked by descending citation count. Ties are broken alphabetically by title.

Filters

The year filters restrict which cited articles appear (by their publication year), not which citing articles contribute counts. The journal filter restricts cited articles to a single journal. The topic filter restricts to articles carrying a specific topic tag. All filters interact: selecting “From 2010” + “Journal: CCC” shows only CCC articles published in or after 2010, ranked by how often other indexed articles cite them.

Limitations

This is an internal citation count — it reflects influence within the journals indexed here, not total citations across all of scholarship. Many rhetoric and composition journals do not deposit reference lists with CrossRef, so their citing behavior is invisible. Older articles have had more time to accumulate citations, creating an inherent recency disadvantage. See the coverage panel below for per-journal reference deposit rates.

Data source

Citation counts come from reference lists deposited with CrossRef, as described in the Citations tab. Each article’s citation count is the number of other indexed articles that cite it.

Chart construction

Articles are grouped by publication year. For each year, the chart plots the mean internal citation count — the average number of times articles published that year are cited by other work in the index. The line chart uses year as the x-axis and average citation count as the y-axis.

The time-accumulation curve

The chart will almost always show a declining curve from left to right. This is not a sign that older scholarship was “better” — it reflects the simple fact that older articles have had more years to accumulate citations. An article published in 2000 has had 25+ years of opportunities to be cited, while one from 2024 has had less than two. This citation age effect is a well-known feature of bibliometric data. To compare across years on equal footing, the chart would need to be normalized by citation window (e.g., citations received within 5 years of publication), which this index does not currently support.

Journal filter

Selecting a journal restricts the chart to articles published in that journal, allowing comparison of citation trajectories across venues. Some journals will show higher average counts than others; this reflects both the journal’s internal influence within this index and the degree to which other journals cite it.

Limitations

Average citation counts can be volatile for years with few articles. Journals that entered the index recently will show a truncated history. Because only internal citations are counted, a journal that is frequently cited by work outside this index will appear less influential here than its true impact warrants.

Data source

The citation network is built from reference lists deposited by publishers with CrossRef. Each directed edge represents one article citing another. Only citations between articles both present in this index are shown — references to external work are excluded.

Graph construction

The graph is directed: an edge from article A to article B means A cites B. Nodes are filtered to articles receiving at least n internal citations (configurable via the slider), capped at 500 nodes ranked by citation count. Edges are retained only when both the citing and cited article appear in the filtered node set. Node size is proportional to internal citation count; node color is assigned by journal.

Layout algorithm

The visualization uses a force-directed layout (d3-force). Nodes repel each other via simulated electrostatic charge; citation edges act as spring forces pulling connected nodes together. The simulation runs until the layout stabilizes. Densely interconnected articles cluster together, while articles with few shared citations drift to the periphery.

Search and interaction

The search bar filters by title or author name, highlighting matching nodes. Hovering over a node shows article details and citation count. Clicking a node navigates to its full article page. The graph supports zoom (scroll), pan (drag background), and node drag (drag individual nodes).

Limitations

The network shows only internal citation relationships. An article heavily cited outside this index appears small or absent. Journals that do not deposit reference lists with CrossRef contribute no edges. The cap at 500 nodes means peripheral articles may be excluded even at low minimum-citation thresholds. The force layout is stochastic: the exact node positions change on each load, though the cluster structure remains stable.

Data source

The citation network is constructed from reference lists deposited by publishers with CrossRef. When an article in the index includes another indexed article in its reference list, that creates a directed edge from the citing article to the cited article. Only internal citations are represented — references to work outside this index are excluded. Not all journals deposit reference lists with CrossRef; journals that do not contribute no edges to this graph. See the coverage page for per-journal details.

Graph construction

The visualization displays a directed graph G = (V, E), where each node v ∈ V is an article and each directed edge (u, v) ∈ E means article u cites article v. Nodes are filtered to articles with at least n internal citations (configurable via the “Min. citations” slider), capped at 600 nodes ranked by citation count. Edges are included only when both endpoints are in the filtered node set.

Eigenvector centrality (PageRank)

Eigenvector centrality measures an article’s importance based not just on how many articles cite it, but on the importance of those citing articles. An article cited by five highly-cited articles scores higher than one cited by fifty peripheral ones. This captures the intuition that influence propagates through the network: being embedded in the most connected region of scholarship matters more than raw citation volume.

The implementation uses PageRank (Brin & Page, 1998), a damped variant of eigenvector centrality originally developed for web search. PageRank adds a “damping factor” (α = 0.85) that models a random walker who follows citation links 85% of the time and jumps to a random article 15% of the time. This handles a practical problem with classical eigenvector centrality: real citation networks contain disconnected components (clusters of articles with no citation links between them), and classical eigenvector centrality assigns zero to all nodes outside the largest component. PageRank’s damping factor ensures every node receives a nonzero score, producing meaningful rankings across the full graph.

Scores are normalized to a 0–100% scale relative to the highest-scoring article in the current filtered set.

Betweenness centrality

Betweenness centrality measures how often an article lies on the shortest path between two other articles in the network. Formally, for a node v, betweenness is the sum over all pairs (s, t) of the fraction of shortest paths from s to t that pass through v. Articles with high betweenness are bridge articles — they connect subcommunities of scholarship that would otherwise be more isolated from each other. Removing a high-betweenness article would increase the distance between clusters of the network.

Where eigenvector centrality rewards being embedded in a dense, highly-connected cluster, betweenness centrality rewards occupying a position between clusters. An article can have modest eigenvector centrality but high betweenness if it is the primary link between two otherwise separate conversations — for instance, a genre theory article that bridges writing studies and technical communication.

Scores are normalized to a 0–100% scale relative to the highest-scoring article in the current filtered set.

Visualization

The graph uses a force-directed layout (d3-force), where nodes repel each other and edges act as springs pulling connected nodes together. The simulation iterates until it reaches a stable configuration, producing a layout where clusters of densely-cited articles appear as visual groupings and bridge articles tend to sit between them. This is comparable to the ForceAtlas 2 algorithm used in Gephi-based bibliometric studies.

Node size can be toggled between three modes: eigenvector centrality (larger = more influential), betweenness centrality (larger = more bridging), or raw citation count. Node color can be toggled between journal identity (each journal a distinct color) and a heat-mapped gradient (cream → brown) representing the selected centrality metric.

Interpretation and limitations

These metrics reflect structural position within this index only. An article that is foundational to the broader field but rarely cited within the journals indexed here will not score highly. Conversely, an article that is heavily cited within a small cluster of indexed journals may score higher than its broader influence warrants. Citation data depends on publisher deposits with CrossRef; journals that do not deposit reference lists are effectively invisible to this analysis.

The network is also sensitive to time: older articles have had more time to accumulate citations, so they tend to dominate both metrics. Use the year-range filters to isolate a specific period and see which articles are structurally central within that window.

Scores change as the index grows and as more citation data is fetched. They should be understood as a snapshot of the network’s current state, not as fixed measures of an article’s permanent importance.

References

• Brin, S., & Page, L. (1998). The anatomy of a large-scale hypertextual web search engine. Computer Networks and ISDN Systems, 30(1–7), 107–117.
• Freeman, L. C. (1977). A set of measures of centrality based on betweenness. Sociometry, 40(1), 35–41.
• Pruitt, M. (2026). (Re)Considering centrality: A bibliometric network analysis of College Composition and Communication, 1950–2022. College Composition and Communication, 77(3). https://publicationsncte.org/content/journals/10.58680/ccc2026773430

What community detection measures

Community detection partitions a network into groups (communities) such that the density of edges within groups is significantly higher than the density of edges between groups. The measure used is modularity Q (Newman, 2006), which compares observed within-group edge density to what would be expected in a random network with the same degree sequence. Values of Q above 0.3 indicate meaningful community structure.

Algorithm

The Louvain method (Blondel et al., 2008) is a greedy agglomerative algorithm that iteratively moves nodes to neighbouring communities when doing so increases modularity. The resolution parameter controls granularity: values below 1.0 produce fewer, larger communities; values above 1.0 produce more, smaller ones.

Graph construction

The citation network is converted to an undirected weighted graph. If article A cites article B, an edge is created between them with weight 1; if the citation is reciprocal, the weight is 2. Community detection operates on this undirected representation because intellectual affinity is symmetric.

References

• Blondel, V. D., Guillaume, J.-L., Lambiotte, R., & Lefebvre, E. (2008). Fast unfolding of communities in large networks. Journal of Statistical Mechanics, 2008(10), P10008.
• Newman, M. E. J. (2006). Modularity and community structure in networks. PNAS, 103(23), 8577–8582.

What co-citation measures

Direct citation records intellectual debt: article A cites article B. Co-citation records perceived kinship: articles B and C are co-cited when article A cites both of them. The co-citation count for a pair (B, C) is the number of articles in the index that cite both B and C in their reference lists. A high count means many authors, independently, have treated these two articles as relevant to the same argument.

Co-citation analysis was introduced by Henry Small (1973) as a method for mapping the intellectual structure of scientific fields. It has since become one of the standard tools in bibliometrics and scientometrics, used to identify research fronts, trace the evolution of ideas, and reveal how scholarly communities organize knowledge.

Graph construction

The visualization displays an undirected, weighted graph. Each node is an article; each edge connects two articles that have been co-cited at least n times (configurable via the “Min. co-citations” slider). Edge thickness reflects co-citation frequency — thicker edges mean a pair is cited together more often. Node size reflects co-citation strength: the sum of all edge weights for that node, measuring how deeply embedded it is in the co-citation network overall.

The graph is capped at 400 nodes, ranked by co-citation strength, to keep the visualization responsive. Edges are included only when both endpoints appear in the filtered node set.

How to read the clusters

Because the force-directed layout pulls co-cited articles together, visual clusters represent intellectual groupings as perceived by the citing community. Articles that cluster tightly are routinely cited together; articles positioned between clusters bridge different conversations. The clusters often — but do not always — align with topical or methodological boundaries.

Unlike direct citation (which flows forward in time from cited to citing), co-citation is retrospective: it reflects how later scholarship has organized earlier work. An article published in 1985 and one from 2005 can be tightly co-cited if post-2005 scholarship routinely invokes both.

Differences from direct citation and centrality

The Citation Network tab shows who cites whom — directed edges representing intellectual influence. The Centrality tab computes structural importance within that directed network. Co-citation shows something different: the field’s collective perception of which articles belong together. Two articles can be heavily co-cited even if neither cites the other, as long as subsequent work treats them as part of the same conversation.

Limitations

Co-citation data depends on reference list deposits with CrossRef. Journals that do not deposit reference lists contribute no co-citation edges. The network reflects only citations within this index — if two articles are routinely co-cited in books or in journals not indexed here, that relationship is invisible. Co-citation also has a recency bias in reverse: very recent articles have had less time to be co-cited, even if they will become tightly paired in the future.

References

• Small, H. (1973). Co-citation in the scientific literature: A new measure of the relationship between two documents. Journal of the American Society for Information Science, 24(4), 265–269.
• White, H. D., & Griffith, B. C. (1981). Author cocitation: A literature measure of intellectual structure. Journal of the American Society for Information Science, 32(3), 163–171.

This tab shows article-level co-citation. For the author-level equivalent — which scholars the field treats as intellectually paired — see Author Co-Citation.

What bibliographic coupling measures

Bibliographic coupling was introduced by Kessler (1963) as one of the earliest methods for quantifying the intellectual relationship between scientific papers. Two articles are bibliographically coupled when they both cite the same third article. The coupling strength of a pair is the number of references they share. A high coupling strength means two articles are drawing heavily on the same prior literature — they are engaged with the same conversation, use the same theoretical foundations, or address the same body of empirical work.

Relationship to co-citation

Bibliographic coupling and co-citation are complementary. Co-citation asks: “Which articles does later scholarship treat as related?” Bibliographic coupling asks: “Which articles read the same things?” Co-citation is retrospective and changes over time as new citing articles appear. Bibliographic coupling is fixed at publication — an article’s reference list doesn’t change, so coupling relationships are stable from the moment of publication. This makes bibliographic coupling especially useful for mapping recent work that hasn’t had time to be co-cited yet.

Graph construction

The visualization displays an undirected, weighted graph. Each node is an article that cites other work in the index; each edge connects two articles that share at least n references (configurable via the “Min. shared refs” slider). Edge thickness reflects coupling strength — thicker edges mean more shared references. Node size reflects total coupling strength: the sum of all edge weights for that node, measuring how deeply embedded it is in the coupling network overall.

The graph is capped at 400 nodes, ranked by coupling strength, to keep the visualization responsive. Edges are included only when both endpoints appear in the filtered node set. Only references to articles within this index are counted — shared references to external work are invisible.

How to read the clusters

Clusters of tightly coupled articles represent groups of scholars reading the same foundational literature. These clusters often correspond to research specialisations, methodological schools, or topical communities. Unlike co-citation clusters (which reflect how the field receives earlier work), bibliographic coupling clusters reflect how authors position their own work relative to prior scholarship.

Limitations

Bibliographic coupling depends on reference list deposits with CrossRef. Journals that do not deposit reference lists are invisible to this analysis. Only references to articles within this index are matched, so two articles that share many references to books, dissertations, or external journals will appear less coupled than they actually are. Coupling strength is also influenced by reference list length: articles with long reference lists have more opportunities for coupling, biasing toward literature reviews and empirical studies with extensive citations.

References

• Kessler, M. M. (1963). Bibliographic coupling between scientific papers. American Documentation, 14(1), 10–25.

What a Sleeping Beauty is

The term “Sleeping Beauty in science” was introduced by van Raan (2004) to describe publications that receive little attention for a long period after publication, then suddenly attract a burst of citations. The metaphor: the article “sleeps” for years, then is “awakened” — often by a “prince” paper whose framing makes the earlier work newly relevant.

In rhetoric and composition, Sleeping Beauties often reflect disciplinary turns: an article on technology published in the 1990s may be rediscovered when digital rhetoric becomes a major research front. Foundational theoretical work can also exhibit this pattern when a new generation of scholars re-engages with older frameworks.

The Beauty Coefficient

This analysis uses the Beauty Coefficient (B) introduced by Ke, Ferrara, Radicchi, & Flammini (2015). For an article published in year t₀ with peak citations in year t_m:

B = Σ_t=t0^t_m [ c_tm · (t − t₀) / (t_m − t₀) − c_t ]

The coefficient measures the area between a straight line from the origin to the peak and the actual citation curve. An article that accumulates citations steadily has B ≈ 0. An article that sleeps (near-zero citations for years) and then wakes sharply has a large positive B. The longer the sleep and the sharper the awakening, the higher the score.

Citation timeline construction

For each article, Pinakes counts how many times it was cited per year, using the publication date of each citing article. This produces a year-by-year citation trajectory from the article’s publication year to the present. Years with zero internal citations appear as gaps in the timeline. The peak year (t_m) is the year with the most citations; peak citations is the count in that year.

Sleep duration and awakening

The awakening year is identified as the first year when citations rise above a threshold of sustained activity (20% of the way from the average sleep-period rate to the peak rate, sustained for at least two consecutive years). Sleep duration is the number of years between publication and awakening.

False-positive correction

A naïve application of the Beauty Coefficient to internal-only citation data would flag many genuinely influential articles as “sleeping.” For example, Flower & Hayes’s “A Cognitive Process Theory of Writing” (1981) has over 6,000 global citations but only 13 internal citations in this index — it was never sleeping, it was just being cited in books, education journals, and other venues not indexed here. To address this, the Beauty Coefficient is scaled by the ratio of internal citations to total citations (internal + global CrossRef count) when the global count substantially exceeds the internal count. This effectively removes articles whose apparent “sleep” is an artifact of index coverage rather than genuine delayed recognition.

Limitations

The Beauty Coefficient depends on having enough citation history. Articles published recently (last 5–10 years) have not had time to “sleep” and are unlikely to appear here. The analysis reflects only internal citations — references from journals that do not deposit reference lists with CrossRef are invisible. The false-positive correction relies on CrossRef’s global citation count, which is itself incomplete for older articles.

A high B score here means delayed recognition within this particular set of journals. The article listing shows both internal and global citation counts so readers can assess each case individually.

References

• Ke, Q., Ferrara, E., Radicchi, F., & Flammini, A. (2015). Defining and identifying Sleeping Beauties in science. Proceedings of the National Academy of Sciences, 112(24), 7426–7431.
• van Raan, A. F. J. (2004). Sleeping Beauties in science. Scientometrics, 59(3), 467–472.

What the chord diagram measures

Each cell in the underlying matrix counts the number of times articles published in journal A cite articles published in journal B. This is a journal-level aggregation of the article-level citation data — a macro view of the same citation network shown in other Explore tabs. The chord diagram makes the directionality and magnitude of inter-journal citation flows immediately visible.

Directionality

Each chord has two ends of potentially different widths. The width at the source journal’s arc indicates how many citations flow from that journal to the other; the width at the target end shows the reverse flow. When a chord is markedly wider at one end, the citation relationship is asymmetric — one journal cites the other more than it is cited in return. This can reflect status hierarchies, methodological borrowing, or cross-subdisciplinary engagement.

Self-citation

The diagonal of the matrix (journal citing itself) appears as a chord looping back to the same arc. High self-citation can indicate a coherent internal research conversation, editorial preferences for citing the journal’s own literature, or insularity. Most journals in the index exhibit significant self-citation because authors publishing in a venue often engage with the existing conversation in that venue.

Limitations

Citation counts are influenced by journal size: journals with more articles generate more citations. The raw counts shown here do not normalise for journal output volume. Journals that do not deposit reference lists with CrossRef are invisible as citing sources but may still appear as cited targets. Only citations between articles in this index are counted — citations to books, dissertations, or journals outside the index are not reflected.

References

• Leydesdorff, L. (2007). Betweenness centrality as an indicator of the interdisciplinarity of scientific journals. Journal of the American Society for Information Science and Technology, 58(9), 1303–1319.

What citation half-life measures

Citation half-life is a standard bibliometric indicator introduced by Garfield (1972) and reported annually by Journal Citation Reports. The citing half-life of a journal is the median age of the works it references — it measures how far back in time a journal’s authors typically reach. A short citing half-life (3–5 years) suggests a fast-moving research front where recent work dominates; a long citing half-life (10–15+ years) indicates a field that draws heavily on older theoretical or historical scholarship. The cited half-life measures the reverse: the median age at which a journal’s articles are cited by others. A short cited half-life means the journal’s work is consumed quickly; a long cited half-life means its articles continue to be referenced for decades.

How it is computed here

For each citation link where both the citing and cited article have a publication date, the citation age is computed as year_of_citing_article − year_of_cited_article. These ages are grouped by journal (by the citing journal for citing half-life, by the cited journal for cited half-life), and the median of each group is the half-life. The interquartile range (Q25–Q75) is shown as a shaded bar to indicate spread. Negative ages (where the cited article appears to post-date the citing article) are excluded.

Interpretation for rhetoric and composition

Humanities journals typically have longer half-lives than STEM journals because theoretical and historical scholarship retains relevance over decades. Within this index, rhetoric journals may show longer citing half-lives than composition studies journals if rhetoric scholars engage more with historical texts. Journals focused on technology (such as TCQ or Computers and Composition) might show shorter half-lives as the technological landscape changes rapidly.

Limitations

Only internal citations are counted — references to books, dissertations, or journals outside this index are invisible, which can bias the citing half-life shorter (since external references to older foundational texts are missing). Journals with few resolved citations produce unreliable medians. Journals indexed via RSS or scraping do not deposit reference lists, so they contribute no citing data and appear only on the cited side.

References

• Garfield, E. (1972). Citation analysis as a tool in journal evaluation. Science, 178(4060), 471–479.
• Moed, H. F. (2005). Citation analysis in research evaluation. Springer.

What main path analysis reveals

Introduced by Hummon & Doreian (1989) to study the development of DNA theory, main path analysis identifies the most significant citation chains in a research field. Where centrality scores individual articles, main path analysis scores edges (citation links) and traces the single most-traversed route through the network from its newest to its oldest articles.

DAG construction

The citation network is treated as a directed acyclic graph (DAG) with edges flowing from citing (newer) to cited (older) articles. Any cycles (rare, caused by publication lag) are broken by removing edges that point from older to newer work. Source nodes (in-degree 0) are the current frontier — articles that cite others but are not themselves cited. Sink nodes (out-degree 0) are foundational works that cite nothing else in the network.

Search Path Count (SPC)

Each edge (u, v) is weighted by SPC(u, v) = (number of source→u paths) × (number of v→sink paths). Edges on the backbone carry high SPC because they are traversed by many source-to-sink paths; peripheral edges carry low SPC. The main path is then extracted by greedily following the highest-SPC edge at each step.

References

• Hummon, N. P., & Doreian, P. (1989). Connectivity in a citation network: The development of DNA theory. Social Networks, 11(1), 39–63.
• Liu, J. S., & Lu, L. Y. Y. (2012). An integrated approach for main path analysis. JASIST, 63(3), 528–542.

Network construction

For each time window a subgraph is built from articles published within that window and the citation links among them. An edge appears in the window when the citing article is published. The cumulative view adds each window's articles and edges to a growing graph, showing total network size over time.

Metrics

Density is the fraction of possible edges that actually exist (higher = more interconnected). Clustering coefficient (transitivity) measures the tendency of articles to form tightly-knit groups. Giant component fraction is the share of nodes in the largest connected component — a high value indicates a cohesive network. Modularity measures how strongly the network decomposes into communities (Louvain algorithm, computed when ≥ 10 nodes). Average path length in the giant component indicates how many citation hops separate typical articles.

Window size

Annual windows (size 1) provide maximum temporal resolution but may produce noisy metrics for years with few articles. Wider windows (3 or 5 years) smooth the signal but obscure short-lived structural changes.

References

• Barabási, A.-L. (2016). Network Science. Cambridge University Press.
• de Solla Price, D. J. (1965). Networks of scientific papers. Science, 149(3683), 510–515.

Data source

Institutional affiliations are retrieved from OpenAlex, an open bibliographic database. For each article with a CrossRef DOI, Pinakes queries OpenAlex to retrieve author-level affiliation data, which maps each author to their institutional affiliation at the time of publication. Institutions are identified by their ROR (Research Organization Registry) identifiers when available.

Attribution method

An institution is credited with an article if any author on that article lists it as their affiliation. Multi-authored articles may credit multiple institutions. This is a whole-count method (each institution gets full credit for each article), not a fractional-count method (where credit would be divided among co-authors’ institutions). Whole-counting tends to favor institutions with prolific collaborators.

Charts

The bar chart shows the 25 institutions with the highest article counts across the full index. The line chart shows the top 10 institutions’ article counts over time (by publication year), allowing comparison of institutional trajectories. Both charts use the same whole-count attribution.

Limitations

OpenAlex affiliation data is imperfect: some authors lack affiliations, others have outdated or ambiguous entries. Institutional mergers, name changes, and multi-campus systems can split one institution’s output across multiple entries. Articles without CrossRef DOIs (e.g., those indexed via RSS) have no OpenAlex record and contribute no affiliation data, systematically undercounting institutions associated with those journals. Authors who changed institutions mid-career are credited to whichever institution OpenAlex records for each specific article.

Relationship sets

Backward citations are articles the seed cites (its reference list). Forward citations are articles that cite the seed. Co-citation neighbours are articles X where at least one article C cites both X and the seed — a high co-citation count means X and the seed are frequently treated as related by the citing literature. Bibliographic coupling neighbours are articles X that share at least one reference with the seed — the more shared references, the stronger the overlap.

Scoring

Each related article receives a composite relevance score: +2 for a direct citation link (backward or forward), +weight for co-citation (capped at 5), +weight for bibliographic coupling (capped at 5), +1 if it shares a topic tag with the seed, +1 if published in the same journal. Articles appearing in multiple relationship sets accumulate points, surfacing the most broadly connected works.

Graph visualisation

The force-directed graph places the seed at the centre. Nodes are coloured by relationship type: blue for backward citations, green for forward citations, orange for co-citation, purple for bibliographic coupling. Nodes appearing in multiple sets show multi-coloured borders. Node size reflects the composite relevance score. Edges are drawn for every identified relationship, with colour matching the relationship type.

Limitations

Co-citation and bibliographic coupling are limited to the top 20 neighbours each. The reading path only covers articles within Pinakes’ index — citations to articles outside the indexed journals are not included. Articles from journals that do not deposit reference lists with CrossRef will have no backward citation data and limited co-citation / coupling data.