I Tried to Gauge What Percentage of Cryptocurrency Developers are Female. The Answer: Not Enough.
I scraped over 1 million code commits to assess gender balance among contributors to top cryptocurrency projects. The results show a distinct lack of female developers.
(Note: This article was originally written for publication in the sadly now defunct BREAKERMAG.)
A systemic problem
After decades of gender imbalance, many fields of technology are starting to reckon with the fact that they skew disproportionately male.
Tech companies like Facebook, Twitter and Google now publish yearly diversity reports, which are largely underwhelming. Emerging industries like artificial intelligence are in full blown diversity crisis according to insider reports, resulting in biases built into the products they ship. Yet although these statistics can be discouraging, highlighting such systemic problems is a crucial starting point for making change.
In the blockchain industry, some data on diversity already exists. A 2018 report from LongHash found that 14.5 percent of staff in blockchain startups were women across all staff positions, while analysis by Quartz found that 85 percent of blockchain companies launched between 2012–2018 had an all-male founding team. Another commonly cited statistic, Coin.Dance’s bitcoin engagement by gender summary, is based on Google Analytics data, and usually hovers between 5 and 10 percent female from week to week.
But when it comes to the gender balance of developers working in the field there’s little in the way of reliable data — especially considering that blockchain projects are not always formal companies, and can instead be organized as open source projects to which developers contribute code on an ad hoc basis.
In an attempt to investigate, I assembled a new dataset of contributions to the top 100 cryptocurrency projects — over 1 million code commits in total — and attempted to infer the gender of contributors who had associated real names with their user accounts. I hope the findings contribute to discussion of the current state of crypto, and how it could be improved.
Methodology
On April 3rd 2019, I used Messari’s OnChainFX dashboard to gather a list of the top 100 crypto projects by market cap, and compiled a spreadsheet linking each project to the GitHub account where all of the repositories of the project’s code are hosted.
Projects were discarded if too many of the organization’s repositories were unrelated to the core project — for example, Basic Attention Token is hosted on an account that also includes the codebase for Brave Browser. Whenever a project was discarded from the initial list of 100, a replacement was drawn from items 101 onwards on the list, maintaining the overall number.
Using the GitHub API and a custom Python script, I queried each of the 100 organizations to obtain a list of every repository belonging to the organization, and the username of every developer who had made commits to the master branch of each repository. Since many cryptocurrency projects start as forks of other projects e.g. Litecoin as a fork of Bitcoin, I used the type=source parameter to omit forked repositories, without which commits from a forked codebase would have been counted multiple times in the dataset.
I then separately queried each user account to find users who had filled in the name field with a real name, as shown below.
With another script, I queried each real name against a database maintained by Genderize.io, a web service that attempts to determine the gender of a first name. For each name Genderize returns a predicted gender along with a probability estimate that the inferred gender is correct.
I then merged the dataset of genderized names with the dataset of usernames and commits, creating a new dataset of code commits with predicted gender grouped by organization, from which the conclusions in this article are drawn.
Before examining the conclusions it should be noted that gender is complicated: Having a name that is parsed as female does not equate to being a woman and vice versa. Increasingly the male/female binary is seen as a reductive way to examine gender, so these findings only give a rough proxy for the gender identity of blockchain developers.
Findings
In total I scraped data on 1,026,804 code commits across 100 projects. Of these, 691,134 were made by developers with male-identified names, and only 47,678 were made by people with female-identified names. The remaining 287,992 commits were made by users who did not provide a real name, or (in a smaller number of cases) provided a name that could not be parsed as a given gender.
In total, GitHub users with female names account for less than 5 percent of the commits to the top 100 cryptocurrency priojects — 4.64 percent to be exact.
GitHub users with male names make 67.3 percent of the commits to the top 100 cryptocurrency projects, with the remaining 28.05 percent made by developers of unknown gender.
There was also a discrepancy in the average number of commits made by each of the three groups. Male-named developers in the dataset made thirteen more commits to each project than female-named developers on average, and ten more commits than developers of unknown gender.
From commits to users
If we count the number of individual user accounts in the dataset rather than the number of commits the picture is similar, with 4.75 percent of the contributors having female-identified names. Using this metric, the number of developers whose gender could not be inferred rises to 37.9 percent.
In most cases these developers had simply not provided information for the name field of their GitHub profile; in a smaller number of cases, developers entered a name of ambiguous gender, or for which Genderize could not return a prediction.
Individual projects
The discrepancy in gender balance was consistent across projects with a few outliers — although these were almost all attributable to misclassification by Genderize or having a large numbers of contributors without listed real names, rather than high levels of participation from female developers.
By plotting the number of commits contributed by female-named developers against the number of commits by male-named developers, we can get a sense of how individual projects line up against one another.
In the dataset as a whole, 54 projects incorporated fewer than 100 code commits from female-named developers, and 31 of these projects incorporated fewer than 10 code commits. Organizations in the lower right corner of the chart have a proportionally higher ratio of female:male code contributions, while those near the top left are particularly male dominated.
The charts also identify certain projects that appear to have more contributions from female-named developers than male-named developers, such as Bytom, VeChain, Neo and Theta Token.
On closer inspection, data from these projects shows the weakness of the Genderize API, particularly with non-Western names. VeChain’s lead blockchain developer Bin Qian was incorrectly identified as a woman by Genderize; when this error is corrected, VeChain falls from the list. Similarly, Theta Token’s CTO Jieyi Long was also misclassified as a woman.
Bytom web developer Zhiting Lin was correctly identified as a woman, and made 940 commits to Bytom repositories overall. However, Bytom has an above average number of contributors with no listed real names, which may have skewed results.
Conclusions
First, some caveats. An open question is what we can assume about the gender of developers who choose not to enter a real name. Should we guess that the balance of men to women is 12:1, as in the gender-identified names, or might we speculate that in a male dominated field, women are more likely to obscure their real name when contributing to a collaborative project?
On this question, data from existing studies suggest that the gender of developers in the ‘unknown’ category is likely to align with the gender split that has been shown above.
In the largest study on the topic, conducted by GitHub through interviews with 5,500 open source developers, 95 percent of contributors to open source projects were found to be male — a figure in keeping with the data presented here. In light of this, making comparisons between identifiably male or female names while ignoring contributions in the ‘unknown’ category is unlikely to introduce significant bias, though we should keep in mind the potential margin of error.
Also, the dataset counts commits but not lines of code. Different developers and organizations commit code with varying frequency, so the raw number of commits does not capture the full value of contributions to a project — and as a recent Motherboard article explained, even judging by lines of code is a flawed method.
Clearly then, the methodology used here can only give an approximate picture of the gender divide in cryptocurrency projects, and has many limitations. Undoubtedly the contributions of many women have been overlooked through the choice not to include a real name in their profile, and due to time constraints it was not possible to check every entry for possible misclassification.
Still, though the data is fuzzy, it paints a clear picture of a field that is dominated by male developers across almost every major project. It’s worth examining what the broader causes of such a gender imbalance are, and what steps the industry could take to encourage more women to take up blockchain programming.
Research in other fields of business have suggested that diversity provides a performance boost to companies, and we should be asking what potential benefits are being lost when the designers and developers pushing the industry forward come from a mostly homogeneous group.
