Inequality in computer science results from the dynamics in a system that is dominantly male
Individual inequality increases, and gender inequality exists, regarding the number of papers by computer scientists. Both patterns vanish when only papers by first authors with durable careers are considered. Underneath lies the Matthew Effect, a mechanism of cumulative advantage that amplifies differences, including gender differences. These are the results from a computational study of 300,000 female and male authors in computer science who started their careers between 1970 and 2000. The study helps identify the systemic mechanisms by which gender inequality emerges.
Individuelle Ungleichheit nimmt zu, und Geschlechter-Ungleichheit existiert, was die Anzahl der Veröffentlichungen in der Informatik betrifft. Beide Muster verschwinden jedoch, wenn nur Arbeiten von Erstautor*innen mit dauerhaften Karrieren betrachtet werden. Dahinter verbirgt sich der Matthäus-Effekt, ein Mechanismus des kumulativen Vorteils, der die Unterschiede, auch zwischen den Geschlechtern, verstärkt. Dies sind die Ergebnisse einer computergestützten Untersuchung von 300.000 weiblichen und männlichen Autor*innen in der Informatik, die ihre Karriere zwischen 1970 und 2000 begonnen haben. Die Studie trägt dazu bei, die systemischen Mechanismen zu ermitteln, durch die Geschlechter-Ungleichheit entsteht.
DOI: 10.34879/gesisblog.2024.78
Is inequality unfair? On the one hand, inequality is necessary for society to function. Let us stick to the example of science. Not all ideas are equally worthy. If all scientists were equally cited, nobody could spot the paradigmatic ideas that structure behavior in a field of inquiry. For that reason, fields harbor a mechanism that generates inequality: The more an idea or knowledge product has gathered attention, the more it will gather. The sociologist Robert K. Merton called this the Matthew Effect, and it is a central mechanism that gives texture to fields, whether in science or other domains. On the other hand, there is also an unfair and dysfunctional side to inequality. This is when it is not based on merit but on ascribed characteristics like gender.
Studies of individual and gender inequality in science have a long history. How large is inequality? Does it change over time? What are its underlying mechanisms? Classically, these questions have been tackled using cohort designs, but those were typically restricted in the amount of data. Nowadays, the latter limits not only do not apply anymore, but many fields have since transformed from a scholar-centered “little science” to a globalized, team- and project-based “big science.” To answer old questions with new computational methods in new contexts, we have amassed data1 on the productivity (publications produced) and impact (citations received) of about 300,000 authors in computer science. These belong to cohorts from 1970 to 2000, are each tracked over careers of 15 years, and comprise a stable female minority of 20 percent. We have inferred an author’s gender from the given names and corresponding faces found in an image search. This method2 is more than 90 percent accurate for nations besides China and South Korea and yields a binary variable.
In a paper3 published in Quantitative Science Studies, we report that individual inequality in productivity increases over the 15 career ages but remains on a similar level on the historical time scale of cohorts. Individual inequality in impact, albeit larger, is stable across careers and cohorts. Regarding gender inequality, we confirm the so-called productivity puzzle that women systematically write fewer papers than men. However, the increase of individual inequality, as well as gender inequality, in productivity vanishes when we only consider first authors with durable careers. This reveals that both patterns result from considering the full workforce and disappear for authors with comparable achievements.
The Matthew Effect (ME) can explain the different levels of individual inequality in productivity and impact: the stronger the effect, the larger the inequality. The ME is equally strong during an average scholar’s career but has continuously increased historically: Whereas, in 1970, the effect of past on present achievement was already sizable for impact, there was no such effect for productivity. The productivity ME then fully unfolded until 2000. This resembles the imperative to “publish or perish” that is so intimately connected to today’s big science.
The earlier in a career an author produces publications or receives citations, the earlier the ME can act on these achievements. The mechanism then interacts with factors that affect achievement such as the various forms of social support that a scholar can enjoy. Hence, after diagnosing patterns of inequality and establishing evidence for the ME, we set out to predict whether an author drops out of computer science and her or his impact after 15 years. It turns out that being productive early on is strongly positively associated with staying in academia and acquiring citations. Other common factors are publishing in top journals or conference proceedings and co-authoring in teams. But there is one life-course variable that is particularly revealing: The more co-authors a scholar has early on, the more likely she or he is to continue a career and impact others.
In an earlier paper4, we have shown that female and male computer scientists differ in the kinds of social networks they embed into. Women’s networks are relatively small and cohesive, much like a family of trustful relationships. The networks of men are comparatively large and devoid of redundancy, just as if men are still hunting for ideas out in the world. However, it is the male kind of collaboration network that is associated with success, and successful women adopt the male way of networking. Hence, computer science is not only a male world because men are the majority. The system is also governed by male rules since we, again, find that continuity and impact are associated with plain network size.
Many programs aim at improving women’s networking skills in the academic world. If their goal is simply to help women build larger networks that would mean perpetuating a male system. One could just as well argue that men would benefit from training on how to build smaller and more cohesive networks. Or should the goal not be to change ways of network embedding at all since those are deeply rooted in gender identities? Instead of changing the differences that are then amplified by the Matthew Effect, the mechanisms that amplify differences, particularly in the early career, could be targeted. Our research, thus, supports the feminist argument that it is the science system – not its parts – that should change to achieve less dysfunctional inequality and more equality of opportunities.
References
- Lietz, Haiko (2023). Computer Science (1970-2014). GESIS, Köln. Datenfile Version 1.0.0, https://doi.org/10.7802/2642.
- Fariba Karimi, Claudia Wagner, Florian Lemmerich, Mohsen Jadidi, and Markus Strohmaier. 2016. Inferring Gender from Names on the Web: A Comparative Evaluation of Gender Detection Methods. In Proceedings of the 25th International Conference Companion on World Wide Web (WWW ’16 Companion). International World Wide Web Conferences Steering Committee, Republic and Canton of Geneva, CHE, 53–54. https://doi.org/10.1145/2872518.2889385
- Haiko Lietz, Mohsen Jadidi, Daniel Kostic, Milena Tsvetkova, Claudia Wagner; Individual and gender inequality in computer science: A career study of cohorts from 1970 to 2000. Quantitative Science Studies 2024; doi: https://doi.org/10.1162/qss_a_00283
- Mohsen Jadidi, Fariba Karimi, Haiko Lietz, and Claudia Wagner: Gender disparities in science? Dropout, productivity, collaborations and success of male and female computer scientists, in: Advances in Complex Systems, Vol. 21, No. 03n04, 1750011 (2018), https://doi.org/10.1142/S0219525917500114
Leave a Reply