Gender Differences in Research Topic and Method Selection in Library and Information Science: Perspectives from Three Top Journals

Technical Summary: Gender Differences in Research Topic and Method Selection in Library and Information Science

Problem Statement
While existing literature suggests that gender influences research methodology selection—typically positing that women prefer qualitative methods and men prefer quantitative ones—these findings often fail to account for the mediating role of research topics. Furthermore, current studies frequently rely on broad binary classifications (qualitative vs. quantitative) rather than specific methodological techniques. There is a lack of granular statistical data regarding gender differences in the selection of specific methods (e.g., interviews, theoretical approaches) across different research topics. Additionally, traditional manual content analysis for method classification is labor-intensive and difficult to scale for large datasets. This study addresses these gaps by investigating gender differences in research topic and method selection within Library and Information Science (LIS), controlling for topic variation and utilizing an automated classification approach to handle a large corpus of full-text articles.

Methodology
The study analyzed 5,281 research articles published between 1990 and 2019 in three top-tier LIS journals: Journal of the Association for Information Science and Technology (JASIST), Journal of Documentation (JDoc), and Library and Information Science Research (LISR).

1. Topic Identification: The authors employed Top2Vec, a distributed topic vector model, to identify research topics from the full text of the articles. The model generated 17 distinct topics based on semantic coherence, which were then manually named using domain knowledge (e.g., "Information seeking behavior," "Bibliometrics/Scientometrics").
2. Method Classification (CogFT): To overcome the limitations of manual annotation, the study utilized CogFT (Cognize Full Texts), an automatic classification model based on full-text cognition.
   • Architecture: Inspired by CogLTX, the model segments full texts into blocks, uses a self-attention-based global encoder (SciBERT) to determine the relevance of each block to research methods, and extracts the most critical blocks.
   • Classification: A classification network using SciBERT and a Sigmoid layer classifies the extracted blocks into 16 specific research methods (e.g., interview, theoretical approach, experiment) based on the taxonomy established by Chu & Ke (2017).
   • Training & Validation: The model was trained on a corpus of 1,981 manually annotated articles. To address class imbalance for low-frequency methods, the training set was expanded using keyword searches and refined via confident learning to remove noisy data. The model outperformed several baselines, including MS, Ngram+Chi2+NB, and various SciBERT/BigBird architectures.
3. Gender Inference: Author gender was inferred by matching first names against three large datasets (US Social Security Administration, PUBMED authors, and World Gender Name Dictionary). A consensus approach (requiring matches from at least two datasets) was used to ensure high precision (approx. 98%). Authors not identified automatically were manually verified.

Key Results
The analysis focused on first authors, representing 99.84% of gender-identifiable authors (60.67% male, 39.33% female).

• Topic Selection: Significant gender disparities exist in topic selection. Female authors were overrepresented in topics related to people and social issues, such as "Health information/communities" and "School libraries/information literacy." Male authors were overrepresented in topics related to systems and algorithms, such as "Information retrieval/models and algorithms" and "Science evaluation."
• Method Selection (Overall): Even when controlling for the overall gender ratio, specific methods showed significant gender bias. Female authors were significantly more likely to use Interview, Questionnaire, and Observation. Male authors were significantly more likely to use Bibliometrics and Theoretical approach.
• Method Selection (Within Topics): Crucially, gender differences in method selection persist even within the same research topics.
  • Interview: Female authors showed a strong preference for using interviews across five different topics, including those typically dominated by men (e.g., "Information organization/information management").
  • Theoretical Approach: Male authors showed a stronger preference for theoretical approaches across four topics, even those dominated by women.
  • General Trend: Female authors demonstrated a wider range of significantly preferred methods compared to male authors.
• Journal Variations: While the distribution of topics and methods varied across JASIST, JDoc, and LISR, the core finding remained consistent: regardless of the journal, female authors significantly favored Interview, while male authors significantly favored Theoretical approach.

Significance and Contributions
The paper claims several contributions to the field of scientometrics and gender studies in academia:

1. Granular Insight into Method Selection: The study moves beyond the broad qualitative/quantitative dichotomy to reveal specific methodological preferences. It demonstrates that gender differences are not merely a function of topic choice; women tend to favor Interview and men favor Theoretical approach across a wide range of topics.
2. Theoretical Explanation: The findings support social epistemological theories suggesting that men and women may perceive the world and acquire knowledge differently. The preference for "Interview" aligns with an "associative cognitive style" (interaction, listening, trust), while "Theoretical approach" aligns with an "independent cognitive style" (reasoning, analysis, independent judgment).
3. Methodological Innovation (CogFT): The study introduces and validates CogFT, a model capable of classifying research methods from full-text data with high accuracy using a relatively small annotated corpus. This offers a scalable solution for analyzing research methodologies in large corpora, reducing the reliance on labor-intensive manual coding.
4. Implications for Academic Equity: The authors suggest that understanding these specific design processes can help promote gender inclusivity. They propose that encouraging mixed-methods research and cross-gender collaboration could leverage the diverse strengths of different cognitive styles, potentially narrowing the gender gap in scientific output and impact.

The study concludes with modest limitations, acknowledging that the findings are specific to three top LIS journals and that gender inference based on names is binary and may not capture the full spectrum of gender identities. Future work is suggested to include broader journal coverage and qualitative interviews with authors to further explore the underlying causes of these differences.