Carolina Biliotti - Working Papers

Breaking New Ground, Reinforcing Old Gaps: Gender Disparities in Access to Emerging Research Frontiers

Carolina Biliotti, Luca Verginer, and Massimo Riccaboni. arXiv:2404.04707

Abstract: This study exploits COVID-19 as an exogenous shock in biomedical research to show how the emergence of an unexpected new research topic exacerbates gender bias in key authorship positions of scientific publications relevant to the new research topic (e.g. Vaccines, Epidemiology). We determine author's gender based on the names listed on their scientific publications and analyze the changes in the composition of the scientific teams after the COVID-19 outbreak. Using a Difference-in-Differences approach, we find that although the share of female authorship has increased overall, women are less likely to be first or last authors (the most prestigious positions) on COVID-19-related research papers and more likely to be found in middle author positions. Stay-at-home mandates, the journal importance and funding opportunities do not fully account for the decline of women in key author positions. The main difference in first authorship is due to the composition of the team and the experience of the lead authors in COVID-19 related research. First authorship by women declined after teams of novices emerged, where lead authors have no prior experience in COVID-related research. Discretionality in first-author appointments for newcomers, combined with high pressure to publish quickly, may have led to discriminatory biases. Conversely, there may also be differences in risk-taking attitudes in doing research in unfamiliar domains. Monitoring gender inequality in scientific production is crucial for reducing gender inequalities and for implementing timely policies that ensure equal access to emerging research topics.

Gender in the Production and Reception of Surprising Science

Carolina Biliotti, James Evans, Massimo Riccaboni, and Jeffrey W. Lockhart.

Abstract: Despite women's increased participation in science and the absence of evidence for gender differences in research ability, women remain underrepresented in the top levels of academic and scientific achievement. These differences in outcomes may result from gender differences in the kinds of work men and women do, or from differential treatment of similar work. Utilizing data from Openalex and Genderize.io from solo-authored papers from U.S. institutions, we seek to provide insights into the complex dynamics of gender, recognition, and innovation in the scientific community. Our findings suggest that women tend to create more innovative combinations of sources in their reference lists, often connections that are ahead of the curve for popular innovations. However, this kind of innovation is less likely to gain recognition in high status journals when published by women than men. Simultaneously, men are more likely to innovate by connecting disparate concepts, and also more likely to benefit from doing so. Moreover, men's work is more likely to be credited in citation networks, suggesting gender-based disparities in recognition. Women's papers are treated as more disruptive for science equally ahead of time, but they experience a significantly steeper decline in disruption for increasing adoption rates. Overall, our study sheds light on the complex interplay between gender, innovation, and recognition in science, emphasizing the need for a more equitable and inclusive research environment.

Do Contracts Matter? The Role of Pharmaceutical Licensing R& D Agreements on Products’ Development.

Carolina Biliotti, Filippo Chiarello, Giacomo Marzi, and Massimo Riccaboni.

Abstract: We investigate whether the document representations of license agreements generated by LLM can help predict the outcomes of collaborative product development. We collect high-quality, proprietary data on license agreements in the biopharmaceutical sector from the Evaluate Pharma database and gain access to unredacted documents from FOIA license agreements (2000-2023) through the BioSci database. We proxy the performance of an agreement based on the progress of drug development in clinical trials. A deal is a success if the product successfully completes Phase II clinical trials after the license agreement begins; it is a failure if it never progresses beyond Phase II. We generate state-of-the-art vector representations of the documents with GPT-3. The embedding of the documents contains the semantic signal in the contracts that is required for effective prediction of success. Using a Random Forest model, we achieve an F1 score of 0.64 and 0.67 Recall with stratified 3-fold cross-validation. Adding non-textual features does not improve out-of-sample prediction performance. We replicate the prediction with more interpretable textual features, such as social and psychological dimensions in texts created with Linguistic Inquiry and Word Count (LIWC) text analysis tools, or bag-of-words (BOW) representations of nouns, and obtain higher performance in terms of F1-score (0.65 with LIWC and 0.69 with BOW) and Recall (0.7 and 0.72) of the random forest model. Post-LASSO regression using principal components of LIWC variables shows that successful deals correlate positively with clout language and negatively with genuine, informal language.