Analyzing ‘Resetting Tech Culture’ by Accenture and Girls Who Code

This article analyzes the Resetting Tech Culture report from Accenture and Girls Who Code in four parts:

  1. 6 Selected Results
  2. 8 Selected Criticisms
  3. What it Doesn’t Say
  4. Related Notes

Here is the Girls Who Code press release from 2020. Here is the Accenture landing page for the report, and here is the actual report!

6 Selected Results

I think this study is overall really great! It gives us a limited but informative, useful, and actionable picture of tech culture today.

  1. Women face real issues with less inclusive employers, but more inclusive employers are equally common
    1. By investigating the top and bottom fifth of employers by inclusivity, the study essentially tells us by construction that less inclusive employers are no more numerous than more inclusive ones.
    2. Even among the most inclusive employers, there is room for improvement. 85% of women in these companies love their job, which is strong but improvable. 15% of women in top companies have heard inappropriate remarks, which seems too high in my book.
    3. There are many red flags among less inclusive employers, but also some results that surprised me because I thought they would be worse. In this bottom-fifth of tech employers, more than 1 in 4 women say they love their job. I would’ve expected it to be lower. More than half (54%) of women at these low-end employers say they have heard inappropriate remarks. This is a huge problem, but I don’t consider it surprising.
  2. Senior HR leaders are more than twice as likely as female employees to say that it is easy for women to thrive in tech.
    1. A negative reading is that leadership is disconnected from female employee experience, but a positive reading is that many junior female employees underestimate their ability to progress through their career! Probably both are true to varying extents.
    2. I think it is nice as a sign of positive intent from leadership. I think it seems preferable in contrast to the opposite scenario where female employees have high hopes but leadership doesn’t think they will do well.
  3. More inclusive company cultures could drop annual attrition rate of women in tech by 70 percent, leading to 3 million women in tech by 2030!
    1. Parental leave changes are the single biggest effect contributor, and making these policy changes is straightforward!
    2. Mentorship is another major gap here. I think this is a place where male allies could be a huge value-add?
  4. The diversity issue begins long before employment, with tracibility through college programs and into high school.
  5. Actionable items to improve college culture! “At the most-inclusive colleges, for example, at least 35% of the faculty in tech, engineering or math classes are women.” I love this tip, as I think CS programs are a big stumbling block for girls interesting in coding.
  6. The study shows many policy changes are win-win across ethnic, gender, orientation, and other lines.
    1. Specifically, improvements to inclusive networking, creativity and innovation, presence of role models, diverse leadership, and parental leave improvements are generally good for all people regardless of gender.
    2. I think identifying changes that benefit multiple groups is hugely valuable because this can drive practical firm adoption and social unity for all people across group lines, instead of reinforcing group boundaries.
    3. It seems to me that reinforcing collective identity boundaries, group-specific norms, group egoism, and similar values, tends to propogate sexism, racism, and other discrimination.

8 Selected Criticisms

  1. Tech is not the Bottom Fifth!
    1. By focusing on the top and bottom fifth we do get insight into what makes a more or less inclusive employer – love that, but we get essentially no insight into what is truly normal in tech.
    2. Compared to a standard exploratory data analysis of the whole data set, focusing on the top and bottom fifth is a dangerous practice that tends to exaggerate effects. Differences between the top and bottom fifth are much larger than differences between the average, normal, or median firm and firms in either the top or bottom fifth.
    3. I would love to have Accenture and Girls Who Code publish a de-identified data set for reanalysis. Looking at what is normal in tech among both men and women is a major goal here.
    4. For whatever reason, when these results were communicated to the public, the public narrative has very much become that tech is not inclusive, tech is misogynistic, and so on, essentially identifying tech norms as the findings in the botttom fifth! What an awful and inaccurate narrative!
  2. What do some of these figures actually mean?
    1. “Have heard or read inappropriate remarks or comments” – is this how the questionnaire was actually worded, maybe a simple yes/no? Can we see a copy of the questionnaire?
    2. Does this mean at the respondent’s current work environment, ever in their tech career, or was it left ambiguous?
    3. It’s a problem that 54% of respondents in bottom-fifth tech companies have heard or read inappropriate remarks, but it’s importantly different whether that is on a per-career or per-employer basis. It’s a much, much bigger problem if that is a per-employer figure as it would indicate inappropriate behavior is not only occassionally observed but it is in fact more common than not among such employers.
    4. What is the content of these remarks? I think in pop culture many people will state that these are sexist remarks. As a guy, I could also state that yes I have heard inappropriate remarks or comments, but only a fraction of these would be sexist in nature, and zero of the sexist remarks that I’ve heard would have had a direct impact on anything tangible. (Once I found out that a manager had slept with a coworker, which is shockingly inappropriate, but I only found out after they had both left the company).
  3. The report lumps together misogynistic and non-misogynistic cultural factors
    1. Sexual harassment and sexual assault are a plague on this industry. In contrast, women who decide to stop working so that they can have more time with their children is not something we want to limit! Why on earth would you lump these into the same category?!
    2. There is no “other” option (whether or not allowing fill-in-the-blank response). This is a real analytical problem and a source of potentially substantial omitted variable bias and foregone learnings.
    3. Blame the pop culture consumption and media if you want, but their response is partially due to the communication of results within the report itself.
      1. By lumping together so many things under the umbrella of “cultural issues,” of course it became the number one correlate!
      2. The report then framed cultural factors as a “cause” of women leaving tech by age 35, and women leaving by 35 was determined to be bad.
      3. A more responsible analysis would break cultural issues into at least three, potentially more, subgroups: SA/SH, unregretted attrition (specifically around WLB and child-rearing), and regretted attrition (possibly breaking out WLB as distinct from other cultural issues).
      4. We might think “unregretted attrition is zero!” Let’s be careful about the implication: Do we want a society where both parents are always working full-time? No? Then do we want to foster a stay-at-home dad norm?
      5. I think a mix is ideal. Some families have a stay at home mom, for some families it’s dad, for some families both parents work whether part of full time, but I think it’s bad if we optimize on “all families have both full-time parents.” Therefore, I don’t think URA of 0 is appropriate.
  4. They didn’t talk about the guys!
    1. Talking about non-females is necessary to identify female issues. Otherwise, you mistake general issues for female issues. Issue like work-life balance are major concerns for both genders, and shouldn’t be taken as a sign of misogyny.
    2. Thinking about male effects can be a useful tie-breaker to help priotizie among different things that are good for women.
    3. Thinking about male effects can be a useful motivator at some companies that may have many male employees. Practically, it can increase the benefit value of any suggestion that gets pitched to leadership. Strategically, let’s leverage our numerous male allies and create win-wins for both genders.
    4. “Likelihood of advancing to manager” appears prima facie bad because it’s less than half of women, but on second thought, do half of men advance to manager? How big is the actual problem here?
  5. There is no causal analysis
    1. Basically the results of this data include a large number of summary statistics and cross-tabulations by gender, but where is the causal analysis and the regressions? Causality is largely assumed.
  6. They miss out on suggesting some really good ideas.
    1. Their analysis shows the problem begins with college, so what about a recommendation that employers drop the degree requirement?
    2. Again, because college is a key issue here, what about improved utilization of coding bootcamps? According to Course Report, in 2020, 41% of bootcamp graduates and 49.5% of online coding bootcamp graduates were female.
  7. Show me the data: Where are the confidence intervals?
    1. They describe a cumulative effect, but it’s not clear what this means technically.
    2. Are they making a prediction? If so, how confident are they in this prediction and why?
    3. Is the prediction based on a regression model? What specification? What are the factor-level effects? Is there an omitted variable bias or other specification bias? Was there any causal analysis?
    4. It seems like they didn’t do causal analysis from a modelling perspective because they do not partial out any of the company culture effect, saying, “Women plan to leave tech for various reasons, but in our study we found that poor company culture accounts for 37% of those making it the number one cause.”
      1. It’s easy to imagine the omitted variable bias. As reported by Desiret News, “46 percent of the mothers whose husbands earned at least $250,000 a year don’t work for pay,” so age, income, marital status, level of education, and number of dependents are all important factors that would contribute to higher WLB demand, which manifests in the Girls Who Code study as a “cultural factor” reason for leaving the workforce.
  8. I would like to see the results broken down for software engineers specifically.
    1. STEM and tech are both overly-broad occupational umbrellas when we think about the gender pay gap.
    2. Computer jobs and engineering jobs are specifically where the representation problem is located (Pew 2021, see image below as well).
    3. Software engineering is the intersection of computer and engineering jobs. In addition, the pay for this occupation is much higher than other occupations, so getting more women into software engineering specifically is a huge win for diversity and closing the gender pay gap.
    4. [A tangent] Movements like the STEAM movement to add art into STEM are counterproductive. It falsely appears to be gender-inclusive because the share of women in the arts is high, but in reality this doesn’t improve diversity in the companies that need it, and there is a huge opportunity cost of lost dollars for women when we convince a girl that an occupation in the arts is remotely comparable to a career as a software engineer, so the gender pay gap remains either not treated or perhaps worsened.
The gender pay gap and diversity problems are specifically resolved by attracting girls to software engineering, not through generic STEM, or worse yet, STEAM advocacy

What it Doesn’t Say

  1. This study does not support the allegation that tech is misogynistic or that men in tech are sexist.
    1. In fact, the study finds that women that who have a “good image” of tech 27% more likely to look for a job in tech. This makes propogating the stereotype that tech is misogynistic a toxic and harmful action that drives down female participation in tech.
  2. This study finds that 50% of women in tech roles leave them by the age of 35, but it does not support the claim that this is a bad thing.
    1. We could be seeing a positive wealth effect. Women in tech tend to be upper class and marry upper class in the US in connection with assortive mating norms. Combine this with the fact that wealthy women tend to stay home more than other women, we should be unsurprised by the result that women in tech would choose to stay home relatively frequently because they can.
    2. Let’s work to seperate out negative cultural effects like misogyny and racism from gender-neutral or even female-positive and unregretted attrition related to child-rearing or other reasons like early retirement.

Related Notes

  1. This 2019 study from WITI and IDC has some interesting findings
    1. 48% of women reported experiencing “sexual harassment in the workplace,” but so did 39% of men.
    2. Note: I’m not sure whether the above response indicates having ever experienced harassment during their career at any workplace, whether it is a report about their current workplace, or whether the survey left that ambiguous. As far as I can tell it is the last case.
    3. Men in tech are about twice as likely to report sexual harassment as women. This is contra the “men shielding men” narrative. Notice that men are far more likely to report harassment (84%) compared to having been subjected to harassment (39%), therefore they must often be reporting the harassment of other people.
    4. Professional mentoring is strongly associated with mentee salaries. If I’m reading their data correctly it indicates that women in general have far less mentorship, but that women actually earn more per percentage use of professional mentorship compared to men.
      1. So, mentorship may not only benefit women but it may disproportionately benefit women
      2. Perhaps within some constraints; disporporationate gain by gender may not work as a general rule; one constraint might be that mentorship has diminishing return over a population or group, so that as the intensity of female mentorship participation grows the overall return equilibrates towards the level men currently receive.
    5. Interested finding albeit a bit nuanced/complicated:
      1. The study analyzes perception of 9 subcategories of diversity.
      2. Each subcategory is broken down by male or female respondent gender, and perceived diversity ratings are compared by gender.
      3. Gender was:
        1. Absolutely ranked by women as 5 out of 9, where lower rank is higher in diversity. It’s a middle-of-the-pack concern according to women.
        2. Relatively ranked 4 out of 9 in the gap compared to men, where lower rank means there is a smaller gap between the views of men and women.
        3. So yes: men think it’s less of a problem than women, but women don’t prioritze gender diversity as the central problem themselves, and men align fairly well on this form of diversity compared to others.
  •  
  •  
  •  
  •  

Leave a Comment