Top Stories
comments 29

The Google Diversity Memo: It’s still stereotyping—just not the way you think it is!

As academics who have collectively done a lot of research on gender, we have been following the discussion about James Damore’s memo about diversity at Google and the subsequent arguments for and against with a lot of interest. First of all, we have to applaud James Damore for actually reading some of the science on this topic. Second, like Scott Alexander and other scientists who have spoken out on the topic, we agree with most of what he says about the science. A lot of data and many peer-reviewed articles show that women in the population are indeed more people-oriented than men. For example, the graph below taken from Adams (2016) shows gender gaps in values (defined as average male values minus average female values) in the European Social Survey (ESS). In this data, as well as in the World Value Survey, women are on average more benevolent and universalism-oriented, traits associated with being people-oriented, than men.

While some of the gender gaps might be small, we still have a very poor understanding how small differences in traits may translate into outcomes. Thus, although we don’t like these findings any more than other women (or men), we can’t argue with the basic findings that gender differences exist in the population.

However, Damore was still stereotyping, but not the way many people seem to think he was stereotyping. Saying that on average all women are people-oriented is not stereotyping if one is indeed talking about the population of women. The statement is based on hard facts, after all. Where Damore made a mistake was when he applied concepts that are valid for the population to the sample of women and men in tech and, more directly, the women and men working for Google.

According to some numbers, Google rejects 99.5% of applicants. This means the men and women who work at Google are highly selected. They chose to enter tech, they chose to apply to Google and Google chose them. These people are not the “typical” member of the population. As such, population averages are unlikely to describe their characteristics very well. Furthermore, since scientists have no data on the traits of men and women at Google (but we’d love to get it!), we have not studied them. Generalizing characteristics that may be relevant for the population to a highly selected subsample of the population without any data to back up the generalization is, in our minds, a classic example of stereotyping.

Is Damore’s mistake big? Potentially, yes. In research on selected samples that we have done together and with other co-authors, we have found that women and men in selected sub-samples are not “typical”. “Typical” gender differences can even reverse in selected subpopulations. We found, for example, that female directors in Sweden are much less “people-oriented” than women in the population in Sweden. Female CFAs are much less tradition-oriented than male CFAs, which is the opposite of the findings for men and women in the population. And female MBAs and directors in the finance industry are, if anything, less risk-averse than their male counterparts in finance, in contrast to the standard findings that women are more risk-averse than men.

The problem with Damore’s memo is that he argues Google should tailor diversity policies towards men and women in the population. This makes no sense. What Google needs to do is to tailor its diversity policies towards the men and women who work at Google. If, like female directors in Sweden, women at Google are less people-oriented than the “typical” woman in the population, does Google still need diversity policies? Presumably yes. It is always hard to be in the minority.

So Damore was stereotyping. But, what is not clear is whether the people who hated his memo understood in what way he was stereotyping. His mistake is a mistake many people make.

Should Google have fired him? Somewhat ironically, one of the reasons companies care about diversity is because diverse opinions can generate better decision-making. On the other hand, people who work at Google are not supposed to be the type of people who make the mistakes many people make.

 

Further Reading

Adams, Renée B. (2016) “Women on Boards: The Superheroes of Tomorrow?Leadership Quarterly, 27 (3), 371-386

Adams, Renee B. and Barber, Brad M. and Odean, Terrance, Family, Values, and Women in Finance (September 1, 2016).

Adams, Renée B. and Ragunathan, Vanitha (2014) “Lehman Sisters

Adams, Renée B. and Funk, Patricia C. (2012) “Beyond the Glass Ceiling: Does Gender Matter?” Management Science 58(2), 219–235.

Damore, J., 2017, “Google’s Ideological Echo Chamber: How Bias Clouds our Thinking about Diversity and Inclusion”, Unpublished memo.

Renee Adams and Vanitha Ragunathan

Renee Adams and Vanitha Ragunathan

Renée Adams is Professor of Finance at the University of NSW. She is also the Director of UNSW Business School’s Women in Leadership Network. She holds an M.S. in Mathematics from Stanford University and a Ph.D. in Economics from the University of Chicago.

Vanitha Ragunathan is a Senior Lecturer in Finance at UQ Business School at The University of Queensland. She holds a PhD and MFin in Finance from the Royal Melbourne Institute of Technology, and a BSc in Statistics from La Trobe University.
Renee Adams and Vanitha Ragunathan

Latest posts by Renee Adams and Vanitha Ragunathan (see all)

Filed under: Top Stories

by

Renée Adams is Professor of Finance at the University of NSW. She is also the Director of UNSW Business School’s Women in Leadership Network. She holds an M.S. in Mathematics from Stanford University and a Ph.D. in Economics from the University of Chicago. Vanitha Ragunathan is a Senior Lecturer in Finance at UQ Business School at The University of Queensland. She holds a PhD and MFin in Finance from the Royal Melbourne Institute of Technology, and a BSc in Statistics from La Trobe University.

29 Comments

  1. You need to cite the relative number of those female directors and Female CFAs compared to men in the same respective fields to make your claim that Damore was stereotyping. Not having the relative number makes the data meaningless for counter intuitive trait frequency.

    Damore was questioning if bringing up the numbers of people based on superficial traits like gender was wise given that they may not be even, or predictable, for certain percentiles (Basically the same argument you are making). It was a short and careful paper touching on many things but the main point is that it is certainly not predictable that all jobs at all percentile skill levels should represent genders 50/50, and even more spurious to assume that if they don’t we can assume discrimination is taking place.

  2. Sean says

    “Where Damore made a mistake was when he applied concepts that are valid for the population to the sample of women and men in tech and, more directly, the women and men working for Google.”

    Did he say this? I think the implication is, instead, that the overall percentage of the population which will feed the Google pipeline simply won’t match the percentage of women in the population, unless the job description adapts to include a more broad set of traits (appealing to women who enjoy people, not just things). It’s perfectly reasonable to observe that the women who actually work at google want to be there and enjoy coding, but it’s a bit difficult to find women who meet that profile (because of the pipeline, not because women are being disproportionately turned away at hiring time). In fact, at a different tech company, that’s what I anecdotally observe.

    “Is Damore’s mistake big? Potentially, yes”
    It might be, if he’d made specific numerical predictions. His policy suggestions are really exploratory, not academic journal submission material. And, I don’t think your expectations of what

    “Google needs to do is to tailor its diversity policies towards the men and women who work at Google. ”
    Hiring policy, policy to improve morale and inclusion, and retention policy aren’t quite the same things. Your statement here makes sense for the latter two.

  3. Lee R. says

    Good article… My only quibble would be… Read Lee Jussim on stereotypes.

    “Saying that on average all women are people-oriented is not stereotyping if one is indeed talking about the population of women. […] a classic example of stereotyping”

    The classic definition of a “stereotype” is muddled and incoherent. It’s something like a belief about a group that is rigid, inflexible, inaccurate, and pernicious. If rigidity is a requirement, it’s not clear that many people actually have stereotypes. Even most bona-fide anti-black racists understand that not every single black person has the negative characteristics they attribute to blacks as a group. If accuracy is a requirement…IOW, if a true generalization is not a stereotype, than you have to empirically demonstrate that it is false, or you can’t say it’s a stereotype. By that standard, there are virtually no known stereotypes in the social psychological literature…only beliefs about groups that researchers assumed were stereotypes.

  4. Andre Bouchard says

    “According to some numbers, Google rejects 99.5% of applicants. This means the men and women who work at Google are highly selected. They chose to enter tech, they chose to apply to Google and Google chose them. These people are not the “typical” member of the population. As such, population averages are unlikely to describe their characteristics very well. ”

    I think Damore knew that and probably thought (rightly or wrongly, not me to say) that this not “typical” members of the population are precisely at the tail of a fat-tail curve where differences get bigger.

  5. Google may hire the best and brightest, that does not necessarily mean the internal politics don’t reflect extreme bias.

    I’ve read that in meetings it’s considered OK to ostracize people for their opinions, particularly if the target is male. Again, the nature of the hiring process doesn’t guarantee a benign culture. Managers at Google may have come to believe that appeasing Social Justice Warriors in the wider population is their most important task. I think that’s what Damore was at least predicting, even if it hasn’t got that bad already.

    Google is their own left wing bubble, ensconced in one of the most Progressive areas in the country – and we know now that they don’t listen to contrary opinion.

  6. It seems to me Damore’s analysis is reasonable. Consider a simplified model where

    1. Google hires anyone who demonstrates a certain level of proficiency in computer science.

    2. Proficiency in computer science is directly proportional to interest in things rather than people.

    3. Men are, on average, more interested in things and women are more interested in people.

    …then male google employees will be more interested in things; and female google employees will be more interested in people. Why? Because that is how Gaussians work, e.g., look at the chart at http://tinyurl.com/ybvn464g and replace “SAT score” with “interest in things” or “interest in people.”

    Now of course I understand that points 1 and especially point 2 are not quite accurate. The point here is to describe how population averages can still be used to make reasonable inferences about top X% of population.

    The pattern will still hold in more complicated model, say where proficiency in computer science is not purely a function of interest but also of some random factors (capturing environment, genetics, propitious events, etc).

    If we had data on what “working at google” self-selects for, this analysis would be moot. But we don’t, which is why thinking about google as “employees with top X% ability in computer science” is the best we can do at present.

    • You are making the same mistake. Gaussians are good general approximations of distributions but specific features of Gaussians (thin tails) should not be assumed true of populations without explicit testing. Read the article again but slower this time.

      • No Gaussian *assumption* is made. Rather, the vast majority of genetic effects are additive (many genes, each makes a small contribution, and each contributes linearly; see http://particle.physics.ucdavis.edu/seminars/lib/exe/fetch.php?media=2012:feb:hsu.pdf for details) and the central limit theorem tells us that additive effects translate into Gaussian distributions (and, in particular, thin tails). For this reason, a Gaussian model should be your first prior when you talk about the expected distribution of a complex phenotype.

        I’m normally happy to explain the clarify the underlying reasoning (only so much info can be crammed into a comment) but given your inability to respond without resorting to petty insults, don’t expect me to respond further…

        • I strongly encourage anyone scientifically minded to follow the link posted in pierremanrd128’s comment instead of assuming it is actually a defense as claimed. If that is the best evidence of your position…

          • If the paucity of that link is not convincing, here is a layman’s description of the problem with normal assumptions in genetic variance: https://biology.stackexchange.com/questions/37167/do-biological-phenomena-follow-gaussian-statistics

            “Many datasets show power-law like/skewed normal distributions and people often make the mistake of assuming them to be normal. An example (from my experience) is the expression of all the genes in a cell. Very few genes have high expression and many genes have low expression. This also applies for degree distribution of nodes in some real networks such as gene regulatory network.”

            In general, nonparametric tests should be used where possible and Gaussian assumptions should be treated with extreme skepticism without exceptionally strong a priori reason to believe them which genetics, especially intelligence genetics, lacks.

  7. Stuart says

    Yeah the damore memo was very plainly making the argument that the gender ratios for certain types may not be equally split. It was very clearly not making the argument that no individual will have traits atypical of their gender. I feel like he really laboured the point overall group trends are not conclusively predictive of an individual.

    If google are highly selective in an already narrow group of people, it is as reasonable to assume the natural gender frequency will skew away from women as it is to assume it will skew towards them.

    The central argument of the memo is that we have no basis to claim that given gender split must be caused by prejudice against women.

    The simple fact is we don’t know what the natural gender frequency for a specific job is, therefore specific policies may be attempting to address problems that don’t exist, have unintended consequences, or worse exacerbate the problem.

    Moreover some policies such as implicit bias training, stand a good chance of being the homeopathy of social policy. So I see no small irony in claiming “people at google aren’t supposed to make the kinds of mistakes ordinary people make” about damore in such circumstances while google spend millions on snake oil.

    • You put my own thoughts to words perfectly. While I was intrigued enough by an article from the ‘other side’ showing up on Quillete that I clicked in, this article, in line with the rest of the articles critical of the memo, seems to be attacking something other than the actual memo.

      At no point did Damore attribute any traits in general to the Google population, male or female. In fact, all he was suggesting was that that 99.5% selection that he obviously knows will select for a population different than the one at large has a different set of standards for different groups, and he proposes the _question_, is this fair?

  8. Reality Bites says

    The google ceo went the fully expected cowards way out, stifle conversation, smash any thought to the contrary, no discussion. Typical inept arrogant pen pusher who has never done a real job in its life.
    I would check behind any google employee’s ear from now on… there will most likely be an implant.

    The Borg mother can’t have her drones wandering off now can she?

  9. There’s also studies indicating for instance IQ has greater variability in men than women. Therefore if you were hiring to get the top 0.5% based on raw IQ you almost certainly wouldn’t end up with even numbers. But it would depend on what characteristics were desired/emphasized in new hires.

  10. Epson says

    Good article, thanks for sharing your findings that outliers of a gender can exhibit gender-reversed behaviour. This is strong evidence for the utility of ‘women in leadership’ committees and so forth — because these are pretty exceptional women in their own right.

    I still think Damore is correct though, because his main point was that it’s unrealistic to think that Google can achieve a 50 – 50 gender balanced workforce. And the research here seems to reinforce that, so I don’t think it’s fair to suggest he should have been fired for missing the point about gender atypicality.

    Also, I’m not an expert on the policies Google has tried, but we don’t know if different policies wouldn’t work to foster female talent unless they are tried. Maybe they wouldn’t work, but who knows. What they HAVE tried seems to have been a colossal waste of money. See here: https://www.axios.com/googles-diversity-efforts-are-making-little-progress-2470784457.html

    BUT maybe Google really don’t care about promoting diversity all that much and these policies are just insurance for gender discrimination class action law suits. Who knows.

    All I know is that I wouldn’t want to be a Google exec right now, what a clusterf*ck!

  11. Vier says

    Google employees are highly selected. If they are selected based on skill, they are pretty similar regardless of their gender, nationality or skin colour. So it makes sense to treat all these people as individuals regardless of their gender, nationality and skin colour. Right?

    What Damore was pointing out was, that Google was not doing it. They want to increase diversity, which means that some people are discriminated against based on their gender, nationality or skin colour. Some capable people will not be hired or promoted because someone else with the “proper” gender, nationality or skin colour is. People call it positive discrimination, but there’s really nothing positive about it.

    Damore was also explaining, why there are more men in tech. That affects what people in general are interested in. Because of this, there are way more boys and men, who do IT as a hobby, than girls and women. So, when it comes to select the best of the best, it is natural that more men will be in the top 1% than women. It’s just about numbers.

    You can see this everywhere: Most NBA stars are black. Just look at the talent pool. It doesn’t mean that White NBA basketball stars are inferior. It’s just that out of all people, who strive to become pros and put in all the hours, sweat and tears, white athletes are a minority. If we tried to fix that by trying to get to 50/50, many black basketball players would be discriminated agains. And actually, many white basketball players would actually be inferior.

  12. Regarding (paragraphs 3 and 4 above) “Where Damore made a mistake was when he applied concepts that are valid for the population to the sample of women and men in tech and, more directly, the women and men working for Google.” And “Generalizing characteristics that may be relevant for the population to a highly selected subsample of the population without any data to back up the generalization is, in our minds, a classic example of stereotyping.”

    I don’t think it’s clear at all that Damore is extrapolating values from the aggregate population (or better the population sample represented in the studies) to Google’s population. And by the way that would not be a generalization but the opposite since it would be going from the general to the specific, and yes it would be fallacious reasoning. To my recollection Damore’s memo makes no determination specific to Google’s female population.

    First, Damore specifies that his observations relate only to the Mountain View campus of Google. It’s highly unlikely that a former Harvard PhD candidate in biology would believe that this micropopulation would perfectly reflect the statistical samply of any one study.
    Secondly, he writes this: “” I’m simply stating that the distribution of preferences and abilities of men
    and women differ in part due to biological causes and that these differences may explain why
    we don’t see equal representation of women in tech and leadership. Many of these differences
    are small and there’s significant overlap between men and women, so you can’t say anything
    about an individual given these population level distributions.”

    He provides appropriate qualification (“may explain”) and he understands that we cannot extrapolate to the individual. By virtue of that, he could not extrapolate to the small female population ( c. 2000) at Mountain view.

    I noticed that you did not provide a specific quote to support your assertion that Damore was insisting that Mountain View’s female population is perfectly consistent with the population sample in specific studies.

    This omission makes your thesis highly suspect. You can of course amend.

    • He hyperlinked over to Wikipedia to support his bullet-points on gender differences, which in effect extrapolates from general pop. to MV campus.

      Looking at the quote you cited (the one starting with “I’m simply stating…”), it seems to actually blunt or muddle his own argument. If differences are small and there’s so much overlap that you can’t say anything about an individual, then by extension one cannot say much about small groups of individuals either.

      Damore might have gotten some ideas right but expressed them somewhat unclearly…a Rorschach test (of sorts) for the chattering folk.

  13. Where Damore made a mistake was when he applied concepts that are valid for the population to the sample of women and men in tech and, more directly, the women and men working for Google.

    Where does he do this? He questions the idea that women should represent 50% of the workforce at Google, he doesn’t question the suitability of those who have been selected.

  14. David Aitken says

    According to this chart, only 18% of the people with a Bachelor’s Degree in Computer and Information Science are women. Since that is probably the population of software developers who get hired at Google, I think it’s unlikely that you will achieve a 50/50 ratio of men to women any time soon, without seriously biasing your selection criteria either for women or against men.

  15. nicky says

    The firing or Damore is unconscionable and short-sighted. He gave a good critical analysis and ways to improve for free. Companies normally pay outsiders good money for such an audit.
    He should have received a fat bonus. Of course, Google appears kind of invulnerable now, but with clutzes like Pichai, firing the one he should thank, I have no confidence for the long run.

  16. You gotta love how ignorant bean counters express their opinions about sociology and biology.

  17. If this article is an example of their thought processes, Renee and Vanitha seem to believe that loose speculation and rigorous argument are one and the same thing. By speculating that Damore may not have taken the differences between Google’s women and women in the general population into account — without actually demonstrating that that is what Damore actually did — they illustrate the type of sloppy reasoning and poor logic for which women in the general population are “so unfairly and stereotypically” known.

  18. DiscoveredJoys says

    A key question for me is what the relative ratios of men/women (black/white, atheists/believers, etc.) *ought to be* if there was no unjustified discrimination. Then any discrepancy could be investigated and recruitment/promotion practices modified as necessary.

    Employment and promotion are discriminatory processes – the trick is to ensure that that discrimination is properly justified rather than resting on unconscious stereotypes or political beliefs. I believe Damore’s memo was arguing for a cool and rational investigation of the justifications used within Google.

  19. Last sentence: seriously? It’s OK to fire him because he made a mistake that people at Google are not supposed to make??

  20. Stuart says

    Presumably one of the goals of diversity policies is to increase the diversity of google employees (i.e. to hire more women).

    In that context it isn’t clear that the appropriate population for comparison is google employees rather than the general population (or participants the job market). As such, absent further argument I am not convinced of thus article’s thesis that Damore was stereotyping the Google employee population.

Leave a Reply