Preliminary Thoughts on Improving Gender Diversity in Tech

This article is a discussion on the importance of gender diversity in tech, with a focus on software engineering. I clarify the current state of diversity relative to other industries and times, I provide some preliminary causal variance explanation. I describe specific actions individuals can take within and outside of the tech industry to alleviate associated problems and improve social welfare. Of note, I disagree that the current state of tech is misogynistic, and I point out that spreading this false narrative tends to reinforce problematic socio-cultural narratives.

The first section gives some reasons that diversity is important in any industry, and in the information technology industry in particular. The second section goes over the state of the industry in the US compared to other industries and times. The third section goes over the results of an analysis of the tech industry by Girls Who Code and Accenture. The fourth section provides conclusions, describes some solution anti-patterns, and describes some recommended solutions.

Why Care About Diversity?

1. Consumers demand diverse companies. Consumers will literally pay a premium for a similar product produced by a diverse company.
2. Diversity is linked to innovation. Innovation improves firm technical efficiency, growth rate, and resiliance across industries, and it is critical in the information technology industry in particular.
3. Employees report higher satisfaction at relatively diverse companies. This presents multiple value propositions to the firm. The ability to hire more easily and at a lower price, the ability to better retain talent, and general firm reputation are a few highlights.
4. A diverse labor force enables better customer empathy and understanding. Diverse labor supports dogfooding and understanding product use in a wide variety of use cases.
5. Diversity initiatives can contribute to solving social problems like poverty and prejudice.
   1. Some social problems disproportionately impact demographic minorities, and diversity initiatives in many ways disproportionately empower demographic minorities or underrepresented groups even when they are not a minority.
   2. For example, women are underrepresented in tech although it is dubious to call them a demographic minority.
6. Inequality, the diversity of humankind, and the relationship between those two is simply interesting. Diversity initiatives include a valuable component of research in this area.
   1. Studying inequality is interesting because it can help us uncover why some people and nations are wealthier and more productive than others. This can be used to prevent or alleviate poverty and suffering and to increase general welfare.
   2. Humans vary according to so many characteristics, that on statistics alone we ought to expect that human variation is in some ways linked to inequality, preference, and productivity.
   3. Research to date has confirmed the original intuition that such variation is important, so we have even more reason to think that continued investment into this sort of research will yield return.

The State of Diversity in US Tech

1. The US leads the world in gender diversity among professional software engineers.
   1. Source, Stack Overflow developer surveys 2011-2021, with emphasis on the most recent year of data.
   2. While software development is not particularly diverse, it is not among the least diverse industries and adjacent job families in the broader tech ecosystem are among the most diverse occupations (AdvisorSmith, 2021).
   3. I think we also lead in ethnic diversity too? I haven't analyzed the numbers in detail but the US is known for its ethnic diversity.
2. They gender pay gap for software engineers in the US is now (2020+) small to nonexistant
   1. Analysis of US developer Stack Overflow survey data from 2020 by Nnamdi Iregbulem shows a statistically insignificant 1.4% controlled pay gap.
   2. Hired data from 2021 indicates that \"companies offer women 2.5% less on average than men for the same roles,\" although this data is from their own platform. It is a mix of US and data from other countries. It also seems that these figures are less controlled compared to, but otherwise largely consistent with, findings from Nnamdi.
   3. Glassdoor data from 2019 shows an adjusted pay gap in the US of 4.9% and trending down, placing it in better half of advanced economies studied. This data is not specific to the software development job category, so there are reasons to think the number would be lower even in 2019, but again this is largely consistent with the other data points.
3. Programming Industry Composition by Gender Over Time
   1. The first programmers were female.
      1. Ada Lovelace is said to have written the first machine algorithm in the 1840s.
      2. The world’s first computer programmers, according to historian Nathan Ensmenger, author of The Computer Boys Take Over, were six women who ran one of the first electronic computers, an ENIAC machine, at the University of Pennsylvania in the early 1940s. By the 1960s, women made up 30% to 50% of all programmers (Fast Company).
      3. Grace Hopper, part of the team that developed ENIAC, invented the first compiler A-0 in 1951 and coined the term 'compiler'.
   2. According to Census data, in 1970 less than 20% of programmers were female.
      1. By the 1970s, a study revealed that the numbers of men and women who expressed an interest in coding as a career were equal (Elizabeth Le).
   3. According to Census data, in 2014, 20 percent of software developers and 22 percent of computer programmers were female.
      1. 20 percent of US computer programmers being female is consistent with Stack Overflow data.
      2. In 2021, I ran a survey of US adults where I found no gender effect on (equal interest between genders in) desire for a career in programming.
4. Gender Composition of Computer Science at US Universities Over Time
   1. The world's first computer science degree program, the Cambridge Diploma in Computer Science, began at the University of Cambridge Computer Laboratory in 1953.
   2. The first computer science department in the United States was formed at Purdue University in 1962.
   3. In 1960, about 6% of women had a bachelor's degree. About 8% did in 1970. As of 2014, more women than men have completed four years of college or more. About 38% of women have done so in 2020.
5. Reconciling Programming as an Industry and as a University Degree
   1. Female participation in the programming labor force preceded and exceeded their share of CS enrollment.
   2. World War II ended in 1945.
      1. Following the end of the war, men seem to have flooded into the labor force and universities.
      2. Male labor force participation was at an all time high around 1950 as the measure began being taken.
      3. The GI Bill was originally passed in 1944 and men disproportionately benefited from the included education benefits. An associated fact is that men disproportionately increased in enrollment into college until about 1984, making up the majority of CS students all the while.
   3. The female share of industry crashed from the end of WWII through 1970, which is unaccounted for by the NPR media-oriented story, although it's easy to see that media messaging was some kind of contributor. This story also ignores changing economic incentives and the coincidental slowdown in male college enrollment.
      1. I'm favorable to the thesis of aptitude test bias, combined with an explanation from post-war changing economic incentives. Ad targeting clearly has an effect, but why did ads target this way? A strong candidate explanation seems to be that it was already a predominantly male market.
   4. Today, K-12 Boys have higher value and interest in a CS degree compared to K-12 girls (Source 1, Source 2), although adults don't have a significant difference.
   5. Non-traditional paths (ie without CS degree) to the programming career have been gaining steam in recent years. 41% of bootcamp grads in 2020 were women.
   6. It seems like American culture needs to improve it's messaging to K-12 girls. I'm not sure why we aren't able to communicate the importance of programming, but perhaps part of the problem is an impotent attempt to describe the importance of general computer science, as opposed to marketing programming as a practical tool applicable in any industry.

Analyzing the Girls Who Code Study

Here's a landing page on Accenture's website that contains some summary info and contains a link to the pdf report itself.

Before we dive into the report itself, let's cover a bit of background on the Girls Who Code organization and the tech industry as a whole, within which the report is nested because reports tend to contain the results presupposed or desired by their funders and analysts:

1. Pros:
   1. I personally endorse Girls Who Code, although I think it's important to acknowledge certain biases and incentives around the organization.
   2. Girls Who Code is well-funded, large, and effective as a networking tool for US women interested in tech. It is one of the two largest US nonprofits focused on women in programming and has many reputable corporate partners.
   3. Girls Who Code has a 100/100 score on Charity Navigator.
2. Cons:
   1. The US tech industry leans extremely far left and has a strong partisan bent toward the Democratic Party (This is a diversity problem in itself that is often swept under the rug).
   2. Girls Who Code was founded by a Democratic political candidate who was literally inspired while on the campaign trail. It is an organization that was unabashedly conceived and incubated in a decidedly political context.
   3. (This is a con against Girls Who Code, but it's actually great news for tech) There does exist another option.
      1. I would put these organizations as equally good at this time, I'm not convinced either is strictly better across the board.
      2. Women Who Code is an established nonprofit focused in the same niche, but it is international organization, so I think it has reduced political incentives compared to Girls Who Code.
      3. Women Who Code is also larger in terms of network size and funding, but I don't know if their US-specific network is larger than Girls Who Code. Another reason that Women Who Code isn't clearly preferred is that they are currently unscored by Charity Navigator.

For further context before I break down the study, let me make some special notes for my TikTok audience:

1. To give away the ending: To my knowledge, none of these statistics are at odds with personal anecdotes of problematic situations, even many such anecdotes.
   1. If you feel that I have devalued your personal experience, that's not my intent and that's not my analytical conclusion either. Please read again and if you still think I have personally devalued you in any way, please contact me personally.
   2. Consider that statistics are often counterintuitive, so that you may intuitively feel there is a contradiction prima facie, but if we reflect further we will often see there is no real contradiction, just a counterintuitive statistical description.
2. TikTok users are not representative of the tech industry professional makeup. Not even the #TechTok niche and related niches.
3. There's a ton of research in particular on the fact that anonymous social media accounts tend to behave in offensive ways. The reasons are many but two big ones include:
   1. Anonymity removes \"skin in the game,\" or the cost for behaving inappropriately.
   2. Many accounts, these days even accounts that appear to be real people, are actually paid-for trolls, whether manually or automatically producing content.

With that context out of the way, I will now go through the analysis itself in three subsections:

1. Major findings summarized
2. Problems in the analysis
3. Proper and improper learnings, net of problems and constraints

Conclusions, Anti-Patterns, and Applications

Top 5 Applications:

1. Prospective tech workers: Dive in without fear! Tech is one of the best career paths to higher quality of life for many kinds of people including women and other underrepresented minorities!
2. Let's focus specifically on encouraging non-male participation in software engineering, as opposed to broad STEM participation. If the reason this is a good idea isn't obvious, this paper has a bunch of details.
3. Employers should drop the degree requirement and generally reduce the value attributed to the degree. Instead, prefer direct investigation of skills related to the particular role at hand.
4. Society should generally increase adoption of coding bootcamps.
   1. On the prospective student or consumption side, prefer prestigious coding bootcamps. A prestigious bootcamp has more than 400 ratings on CourseReport with an average rating over 4.25.
   2. On the employer side, consider formally partnering with one or more bootcamps to create a stable, skilled, and diverse labor pipeline.
5. Let's continue to engage in research and development around diversity, but let's allow other responses, more expressive options, more careful response grouping, and open-ended responses.
   1. I would also like to see a least one large-sample study of in-depth interviews with women. I wouldn't suggest this become the normal practice, but I think it would add significant value to the state-of-the-art understanding of women's issue in tech and in particular could help us understand how to optimize cultural messaging to attract more women to programming.

Also see: How Companies Should Support Women in Tech (TrustRadius 2021).

Top 5 Anti-Patterns:

1. Miscommunicating the reality of US Tech to outsiders in such as way as to scare them off or trigger them into acting in a way that is not optimal for them.
   1. Particularly, inappropriately communicating to young girls that tech as a career is particularly sexist or harmful to women.
   2. Lack of transperancy or misrepresentation of diversity figures and trends are also bad, of course, and would fall into this category of miscommunication.
2. Poorly thought out diversity targets.
   1. For example, targeting the demographics of a state or country with no clear business case (in particular if your target customer is a different population, or if the targeted demographic itself is insignificantly diverse).
   2. Another example would be attempts to maximize the number of women in the industry or construct barriers to industry exit. Many women leave voluntarily and they should be free to do so. The logical extreme of maximizing female labor force participation is female forced labor, or slavery, and this is a bad thing. So we don't want the maximum number of working women (or men, or members of any diversity program group target), but instead we want the optimal number.
   3. Only targeting gender and ethnicity for diversity is a problem
   4. Similarly problematic is thinking that gender and ethnic diversity are only valuable insofar as they correlate to intellectual diversity.
   5. Similarly problematic is the reverse case of focusing on \"skin-deep diversity\" without intellectual or other diversity.
   6. Consider thinking through diversity along some of the following margins in addition to the forementioned: age, socioeconomic status, religious, military service, geography, nationality, industrial background, educational background.
3. Data shaming, resisting measurement, throwing out existing data, claiming that relevant things can't be measured, or priorizing anecdote over comparable data for the same analytical ends.
   1. Similarly problematic is insistance on using raw averages and dismissal of statistical controls.
   2. Reflecting on their research collaboration with Girls Who Code, Accenture says \"Make it a Metric.\"
4. Qualifying candidates based on years of experience or formal education when appropriate direct-measure skill assessment is available.
5. Diversity initiatives that become toxically anti-normative.
   1. For instance, \"triggering dudes\" is not a good strategy for improving diversity. Most programmers and tech workers are men, so tech transformation is facilitated by male allyship not by antagonizing the majority.
   2. It's not cool or helpful when I'm made to feel uncomfortable for being a white male wearing a wedding ring at work or at a DE&I meeting.
   3. It's not cool or helpful when a DE&I meeting turns into an anti-male, anti-white, or anti-conservative rant fest / gossip session.

Misc Notes

1. Hired has some great gender pay gap data, though their platform serves outside of the US. 2021 data here and 2017-2019 data here.
2. Certainly, we know that \"Women in majority-male workplaces report higher rates of gender discrimination\" (Pew, 2018).
   1. Because programming is heavily male-dominated, we can expect a higher rate of male-dominated workplaces and higher rates of discrimination.
   2. We also note that the entire labor force is generally male-dominated, and historically was even further dominated, so there may also be some cultural historical inertia.
   3. So, there are arguments to be made but they just don't work out ultimately; extrapolating from Pew's results into tech we should expect that only a minority of women in tech believe \"Women are usually treated fairly in recruitment/hiring.\"
3. One of the best arguments I've heard for \"Tech is Misogynistic\" + \"Guys in Tech are Sexist\" is the observation that 72% of women in tech (63% in sofware engineering + IT) report having worked at a company where \"bro culture\" is pervasive.
   1. Per TrustRadius, 2021.
   2. \"Bro culture\" is a pretty ambiguous term, and the survey did not use any particular definition, but the TrustRadius report links to a discussion on the term which in turn cites the Wikipedia article.
      1. In terms of negatives, there is thought to be some association with frat culture and sexual harassment.
      2. In many cases the term is used in a neutral or positive way.
      3. Bro culture appears to mean different things in different places. For example, Wikipedia above specifies that it overlaps with surfer culture in California.
      4. Perhaps bro culture could be reduced by screening against candidates with a fraternity affiliation, a university degree, or even surfers. Perhaps CS degrees are particularly toxic in direct connection with bro culture. There's definitely plenty more here that could be interesting to study.
   3. Caveat: I'm not sure that all survey respondents are all US professionals and I'm not sure how the respondents were selected or recruited to participate.
   4. Note that this does not seem to indicate that most women have a bro culture at their current workplace, but rather that they have seen it at some time in their career.
      1. I think we can definitively say \"tech is misogynistic\" if most workplaces accept, practice, or allow misogyny or misogynistic behavior inclusing sexual harassment.
      2. I do not think bro culture is identical to misogyny, although a correlation or causal association is intuitive.
      3. I do not think that having ever seen bro culture in a person's career is a good measure of the percent of employers engaged in bro culture. The \"ever seen\" vs \"currently see\" problem may also plague the Girl Who Code study, but I am assuming that it does not at the moment.
   5. Note that 41% of men responded affirmatively to the same question, so:
      1. Men and women for the most part agree (eg .41/.72 > .5, and the share in common is even larger for SWEs in particular, estimated at .41/.63).
      2. We can state that most women have seen bro culture at some point in their career, but we cannot say that most tech workers agree that bro culture is pervasive across the industry. I don't know of an accepted criteria of industrial misogyny that this would satisfy.