Manuscript Classification to Support the Analysis of Biases in Publication Opportunities

Area of Science:

• Biomedical research
• Publication bias
• Gender studies

Background:

• Diversity in the research workforce drives innovation and impactful discoveries.
• Biases can hinder publication opportunities for female investigators and diverse researchers.
• Understanding and mitigating these biases is crucial for scientific progress.

Purpose of the Study:

• To develop and evaluate strategies for classifying biomedical manuscripts across four key dimensions.
• To create a framework for investigating gender bias in publication acceptance.
• To adjust analyses for potential confounding factors like experimental design, medical specialty, sample size, and funding source.

Main Methods:

• Manual annotation of 56 biomedical manuscripts by two expert reviewers.
• Development of ad hoc classification strategies for four manuscript dimensions.
• Assessment of model performance against inter-annotator agreement.

Main Results:

• The proposed classification strategies demonstrated strong performance.
• Accuracy scores ranged from 75% for experimental designs to 82% for funding sources.
• Performance was comparable to expert inter-annotator agreement.

Conclusions:

• The developed classification methods are effective for analyzing biomedical manuscripts.
• This approach provides a foundation for large-scale analysis of publication bias in MEDLINE.
• Addressing gender bias in research publication is essential for a diverse and equitable scientific community.