Findings Summary

Date generated: 2026-05-24

Scope

This project currently analyzes CAPES-7 journal article output for 2017-2020. The focal field is LINGUÍSTICA E LITERATURA, compared with eleven other journal-heavy CAPES evaluation areas.

The current processed dataset contains:

63 CAPES-7 programs.
39,049 journal article rows.
12 discipline groups.
Publication language from CAPES DS_IDIOMA.
SJR quartile matching via ISSN.

The quartile source currently used is the accessible UKZN mirror of the 2024 Scimago/SJR quartile file. This is a proxy because Scimago’s official historical CSV endpoint was blocked from this environment. For publication, replace this proxy with historical SJR files for 2017-2020 if possible.

Areas Included

The figures use every area currently present in the processed article table:

Area	CAPES-7 programs	Article rows
Administração Pública e de Empresas, Ciências Contábeis e Turismo	3	2,021
Astronomia / Física	11	10,261
Ciência da Computação	7	2,766
Ciência Política e Relações Internacionais	2	754
Ciências Biológicas I	5	3,698
Economia	4	564
História	2	771
Linguística e Literatura	6	3,878
Matemática / Probabilidade e Estatística	7	2,199
Psicologia	3	1,105
Química	10	9,487
Sociologia	3	1,545

Political Science is therefore already included, under CAPES’s area label CIÊNCIA POLÍTICA E RELAÇÕES INTERNACIONAIS.

Main Result

Linguistics/Letras is an extreme outlier on three related outcomes:

Outcome	Linguistics/Letras	Comparison fields	Gap
Matched to SJR	12.2%	73.4%	-61.2 pp
Published in English	8.3%	82.2%	-73.9 pp
Q1, all articles denominator	1.9%	43.1%	-41.2 pp
Q1/Q2, all articles denominator	6.0%	62.3%	-56.3 pp

The indexing result should not be treated as a mere missing-data problem. In journal-based fields, SJR coverage is itself a meaningful signal of whether a journal belongs to the main international indexing ecosystem. Being indexed does not guarantee quality, but being absent from SJR is rarely consistent with being a top international journal in the field.

Discipline-Level Summary

Area	Articles	SJR match	English	Q1, all articles	Q1/Q2, all articles	Q1/Q2 among SJR-indexed
Linguística e Literatura	3,878	12.2%	8.3%	1.9%	6.0%	49.3%
História	771	22.7%	6.0%	4.2%	15.6%	68.6%
Sociologia	1,545	20.6%	10.6%	3.8%	15.7%	76.4%
Ciência Política e Relações Internacionais	754	29.0%	17.9%	8.4%	22.3%	76.7%
Administração Pública e de Empresas, Ciências Contábeis e Turismo	2,021	41.8%	45.9%	17.0%	25.1%	60.1%
Psicologia	1,105	56.7%	50.0%	17.2%	29.7%	52.3%
Economia	564	56.2%	72.5%	31.6%	41.5%	73.8%
Ciência da Computação	2,766	67.3%	91.6%	39.1%	58.1%	86.4%
Ciências Biológicas I	3,698	91.8%	92.6%	58.1%	81.0%	88.2%
Química	9,487	84.1%	93.4%	42.3%	68.3%	81.3%
Astronomia / Física	10,261	80.3%	95.0%	56.4%	73.2%	91.1%
Matemática / Probabilidade e Estatística	2,199	84.6%	96.4%	57.2%	78.6%	92.9%

The indexed-only Q1/Q2 column is useful, but it answers a different question: conditional on a journal being indexed in SJR, where does it rank? The all-article denominator is the stronger field-level measure because it preserves the indexing gap rather than discarding it.

Regression-Style Checks

Simple logit models with a year control give the same qualitative result:

Odds of SJR indexing for Linguistics/Letras articles are about 0.050 times the comparison fields.
Odds of being in English are about 0.020 times the comparison fields.
Odds of Q1/Q2 placement are about 0.039 times the comparison fields.

These are descriptive checks, not causal models. They show that the focal-field gap is not an artifact of small year-to-year changes from 2017 to 2020.

Language as a Mechanism

The data support the idea that language is part of the mechanism linking field norms to indexing and quartile placement.

Field group	Language	Articles	SJR indexed	Q1/Q2, all articles
Linguistics/Letras	English	322	36.0%	27.3%
Linguistics/Letras	Not English	3,556	10.0%	4.1%
Comparison fields	English	28,925	81.9%	71.4%
Comparison fields	Not English	6,246	34.4%	20.4%

Within Linguistics/Letras, English articles are much more likely to be indexed and much more likely to be Q1/Q2. This is consistent with the proposed pathway:

field norms / audience / language orientation
  -> Portuguese and Brazilian/local journal publication
  -> lower SJR indexing
  -> lower observed Q1/Q2 placement

Adding English to the logit models reduces, but does not eliminate, the Linguistics/Letras penalty:

For SJR indexing, the focal-field odds ratio moves from 0.050 to 0.188 after adding English. English itself has an odds ratio of 8.39.
For Q1/Q2 placement, the focal-field odds ratio moves from 0.039 to 0.160 after adding English. English itself has an odds ratio of 9.74.

This is a descriptive mediation pattern: English publication explains part of the gap, but not all of it.

Brazilian Journal Proxy

For articles matched to SJR, the 2024 SJR file includes journal country. This allows a partial proxy for Brazilian versus non-Brazilian journals, but only among indexed journals.

Field group	SJR country group	Indexed articles	English	Q1/Q2 among indexed
Linguistics/Letras	Brazilian SJR journal	307	13.4%	39.1%
Linguistics/Letras	Non-Brazilian SJR journal	166	45.2%	68.1%
Comparison fields	Brazilian SJR journal	2,133	49.4%	29.2%
Comparison fields	Non-Brazilian SJR journal	23,697	95.5%	89.9%

The pattern is consistent with the hypothesis that Brazilian/local journal publication is associated with lower international quartile placement. However, this proxy cannot classify the large number of non-indexed articles by journal country. A stronger version would require an external ISSN-to-country registry or a curated Brazilian-journal list.

Working Hypotheses

The current evidence supports these working hypotheses:

Linguistics/Letras CAPES-7 output is structurally less integrated into SJR’s international journal-indexing system than the comparison fields.
Low English-language publication is one major mechanism behind low indexing.
Brazilian/local journal publication likely mediates part of the relationship between language and SJR indexing.
The Q1/Q2 gap should be measured primarily with all articles in the denominator, because non-indexing is substantively informative rather than a random missingness process.
Conditional Q1/Q2 among indexed journals is useful as a secondary diagnostic: it asks whether indexed Linguistics/Letras journals are also lower-ranked after excluding non-indexed output.
The results support the claim that Linguistics/Letras publishes much less in international indexed journals. They do not by themselves prove that the underlying research is worse; they show that journal placement, language, and indexing ecology differ sharply.

Figures

Generated figures are in figures/:

pct_english_by_discipline.png
sjr_indexing_rate_by_discipline.png
sjr_quartile_distribution_by_discipline.png
q1q2_among_indexed_by_discipline.png
english_vs_q1q2_by_discipline.png
english_share_over_time.png
sjr_indexing_by_language_and_field.png
q1q2_by_language_and_field.png

Current Caveats

The SJR file is a 2024 proxy, not a historical 2017-2020 panel.
CAPES’s accessible datastore copy of the 2013-2016 article-detail table is incomplete, so processed analyses currently use 2017-2020 only.
Professor-level denominators are based on observed faculty authors because the CAPES docente datastore returned empty filtered files for the selected program codes.
Linguistics/Letras in CAPES combines linguistics and literature programs. The present summary treats the CAPES area as the focal unit; a later refinement could split linguistics-oriented and literature-oriented programs.

Adding Areas

To add another CAPES evaluation area, add its normalized CAPES area name to STUDY_AREAS in scripts/derive_capes7_programs.py, then rerun:

python3 scripts/derive_capes7_programs.py
python3 scripts/fetch_capes7_datastore.py
python3 scripts/build_database.py
python3 scripts/export_summaries.py
Rscript analysis.R

Political Science does not need to be added because it is already present as CIÊNCIA POLÍTICA E RELAÇÕES INTERNACIONAIS.