Source-linked AI summary
Gender Disparities in Science? Dropout, Productivity, Collaborations and Success of Male and Female Computer Scientists
Mohsen Jadidi, Fariba Karimi, Haiko Lietz, Claudia Wagner
TL;DR
The paper examines how gender disparities in computer-science careers relate to collaboration and scientific success. It finds higher dropout among women and different average collaboration networks, while successful women exhibit the same collaboration patterns as successful men.
Problem
Gender inequality remains a concern in science, and the study examines how gender differences in collaboration and careers relate to scientific success.
Source: §3 · pp. 20, 22Method
The study analyzes productivity, dropout, collaboration practices and success across one scientific field over time, relating researchers’ collaboration patterns to career outcomes.
Source: §3 · pp. 21–22Results
Women have higher dropout and smaller, more closed, shorter-lived networks, but successful women show the same collaboration patterns as successful men; closure and brokerage are both associated with success.
Source: §3 · pp. 20–21Takeaways & Limitations
Network closure and brokerage are both associated with success, while women are on average less likely to adopt collaboration patterns related to success.
Source: §3 · p. 21The baseline assumes any computer scientists could collaborate, so the analysis cannot determine why women show greater same-gender collaboration; subfield composition may also contribute.
Source: §2.3 · p. 15
Abstract
from arXiv · showhide
Scientific collaborations shape ideas as well as innovations and are both the substrate for, and the outcome of, academic careers. Recent studies show that gender inequality is still present in many scientific practices ranging from hiring to peer-review processes and grant applications. In this work, we investigate gender-specific differences in collaboration patterns of more than one million computer scientists over the course of 47 years. We explore how these patterns change over years and career ages and how they impact scientific success. Our results highlight that successful male and female scientists reveal the same collaboration patterns: compared to scientists in the same career age, they tend to collaborate with more colleagues than other scientists, seek innovations as brokers and establish longer-lasting and more repetitive collaborations. However, women are on average less likely to adapt the collaboration patterns that are related with success, more likely to embed into ego networks devoid of structural holes, and they exhibit stronger gender homophily as well as a consistently higher dropout rate than men in all career ages.
1 Data
The study combines publication and citation records into a time-evolving computer-science collaboration network, infers author gender, and compares community representation and network participation over time.
Gender inference
- Gender inference combines name- and image-based services, achieving above 90% accuracy for most countries, but Asian-name cases are often unknown and excluded from gender-specific analyses.A manual evaluation of the Asian-name detector classified 88 of 100 sampled scientists correctly; authors listed only by initials were also excluded, potentially disproportionately affecting women in early decades, though the authors expect this difference to be small.Source: §1 · pp. 3–4
Data
- Between 1970 and 2016, the records cover 1,634,682 scientists, 3,085,544 publications and 7,849,398 citations; 717,471 papers (23%) receive an in-corpus citation.The study combines DBLP publication records with Aminer citation relations by matching publication titles.Source: §1 · pp. 3–5
Descriptive statistics
- 1.7 to 1.1: the men/women ratio fell from 1970 to 2015, while LCC participation rose from about 20%/10% (men/women) in 2000 to 85%/80% by 2015.Some LCC growth reflects cumulative graph construction and endogenous densification; DBLP recency bias may contribute, though the authors consider relative gender differences meaningful.Source: Figure 1 · p. 5, §1 · p. 6
Data
- The collaboration graph contains 1,634,682 nodes and 7,304,250 edges; 43% of authors are men, 14% women and 43% of unknown gender.Edges represent coauthorship and are labeled with the year or years of the corresponding publications.Source: §1 · p. 5
Career measures
- Career age is counted from a scientist’s first publication year, and career length runs from first to last publication year in DBLP.For example, publications in 1995, 2000 and 2005 yield a career length of 11 years and career ages of 1, 6 and 11 in those years.Source: §1 · p. 5
2 Results
The results examine gender differences in dropout, productivity, and collaboration across career ages, and how collaboration patterns relate to scientific success. They assess whether success-associated patterns differ by gender and how often women’s networks exhibit features associated with success.
2.1 Dropout
- 40% of men and 47% of women stop after their first publication, and women’s dropout rates remain 2 percentage points higher across career ages.Among those who continue, 32% of men and 31% of women stay up to 10 years, 25% and 20% up to 25 years, and 3% and 2% become senior researchers, respectively.Source: §2.1 · p. 7, Figure 2 · p. 7
2.2 Productivity
- At matched career ages, the productivity gap nearly vanishes: output rises for 20 years, averages 3.0–3.5 publications yearly for the next decade, then declines; calendar-year differences track seniority (r = 0.86).Men average more publications in every calendar year, with the gap widening after 2005.Source: §2.2 · pp. 8–9, Figure 3 · p. 8
2.3 Collaboration patterns
- Women’s early degree and k-core disadvantages diminish as networks grow, while men’s distributions remain dominant in significant comparisons and men become stronger brokers on average by 1994.Men’s degree and k-core gaps decrease over time; brokerage differences shift from women stronger in 1970–1982, to indistinguishable in 1983–1993, to men stronger in 1994–2015.Source: Figure 4 · p. 10, §2.3 · pp. 11–12, Table 2 · p. 11
- From the 1990s onward, women are more likely to have closed networks with fewer structural holes and shorter collaboration duration; each feature alone explains little variance.The regressions find no significant effects before 1990, and no significant gender difference in collaboration strength.Source: §2.3 · pp. 13–14, Table 3 · p. 13
- Women’s homophily exceeds randomized-network expectations more than men’s, but subfield composition may contribute, and the study does not distinguish choice from induced homophily.The baseline assumes any computer scientist could collaborate with any other; the authors leave the source of observed homophily for future work.Source: Figure 6 · p. 14, §2.3 · pp. 15–16
2.4 Success
- Success models link large, enduring networks and repeated collaborations to citation impact and h-index; brokerage also predicts success, especially h-index, where closure is significant.In the citation-impact models, degree, collaboration duration and collaboration strength are positively related to success; the 1-hop models also identify neighborhood features.Source: §2.4 · pp. 17–18, Table 5 · p. 18
- Degree, collaboration duration, efficiency, and strength relate sizeably to h-index; collaborating with successful and senior scientists benefits researchers, especially for h-index.For productivity, ties to highly connected scientists are also beneficial; strong ties predict success more strongly than brokerage, while coauthors with highly closed networks have a negative association.Source: §2.4 · pp. 18–19, Table 6 · p. 19
- No significant gender interactions with collaboration patterns were found, indicating that the patterns explaining success are the same for male and female computer scientists.Successful women therefore show the same collaboration patterns as successful men.Source: §2.4 · pp. 19–20
- Women’s networks are smaller, more closed, more devoid of structural holes, and shorter-lived, making them less likely on average to match collaboration patterns associated with success.Women who become successful computer scientists nevertheless show the same success-related collaboration patterns as successful men.Source: §2.4 · p. 20
- 20–30% of outcome variance is explained by the regression models, and gender has a minuscule significant effect on h-index but not citations in 1-hop models.The structural approach therefore misses central aspects of research practice.Source: §2.4 · p. 20
3 Discussion
Addressing persistent concern about gender disparities, this study tracks career-stage differences in computer scientists’ dropout, productivity, collaboration, and success. It finds higher female dropout, gendered differences in network structure, and no gender-specific effect of collaboration patterns on success.
3 Discussion
- The study follows dropout, productivity, same-sex association, collaboration, and success across careers in computer science, extending prior work with a field-wide temporal analysis.The authors describe it as the first study to analyze these outcomes for men and women across one entire scientific field over time.Source: §3 · pp. 20, 22
- Women’s dropout is consistently higher, while men’s higher publication output is explained by a larger fraction of senior authors, resolving the productivity puzzle after career differences are controlled.Women are less likely to continue after their first publication year, enter early- and mid-career stages, and become senior researchers.Source: §3 · p. 20
- Network closure and brokerage are codeterminants of citation impact and h-index, with no gender-specific effect of collaborative patterns on success.Successful scientists combine large networks and repeated collaborations with brokerage across structural holes.Source: §3 · p. 21
- Women’s ego networks are smaller, more closed, offer fewer brokerage opportunities, and are shorter-lived; men more often bridge structural holes.These differences persist when controlling for dynamic effects and career ages.Source: §3 · p. 21
- Women are on average less likely to adopt success-related collaboration patterns, but successful women show the same collaborative behavior as successful men.The authors offer this as a solution to the apparent tension between average network differences and the shared success relationship.Source: §3 · p. 21
- Gender homophily has increased over time, and women’s homophily exceeds men’s when network topology and size are controlled.Source: §3 · p. 21
- The relationships are not established as causal and may be mediated by unobserved skills and knowledge; the analysis is limited to observable network characteristics.The authors suggest incorporating field content, such as subfields, in future regression models.Source: §3 · p. 22
- The findings are limited to non-Asian computer scientists; future work could extend the analysis across fields and ethnic groups and improve gender inference for Asian names.Source: §3 · p. 22