We need to talk about James

In which I gingerly skip through the warzone that is the Google Memo debate and try to make sensible points without getting blown up by a drone or spilling Pot Noodle down my front.

Disclaimer: we run a service that helps reduce hiring bias, but I don’t want that to influence your opinion of the point I’d like to make, so happy to say up front… please don’t use our service, we’re thick as fudge. The reason I’m writing this is because we’ve been looking at these problems and data across many top companies for a while now and hopefully have a slightly more weathered grasp on the issues than some.

In case you missed it, an engineer at Google (James Damore, JD, Jam-Dam) wrote an internal memo that claimed that women as a whole were biologically less likely to be in software jobs, and that positive discrimination is unfair. He included some evidence to justify his points. Everyone was arguing about it, and then Google fired him. Argument continued.

The memo touched on many areas, some scientific, all emotive, so it’s easy for people on both sides to find elements they agree with & ignore the parts that don’t confirm the view they’re most comfortable holding. In short it’s a recipe for a partisan flamewar. Tempers are high; and a lot of people are shouting past each other rather than listening, which makes writing this post a pretty dumb thing to do, but hey, if the shoe fits.


There are five points I want to amplify in this post; most of which apply to both sides of the debate but James and his memo are the focus.

  1. Treat people with respect
  2. Less rhetoric, more listening please
  3. Biology is a red herring in a fishy minefield
  4. Meritocracy is a myth
  5. Not all diversity programs are good

After that I guess I’ll try and sum up & add some recommendations.


1. Treat people with respect

This should really go without saying; but sadly it doesn’t always. The official reason James Damore was fired is that he promoted negative stereotypes about under-represented groups. This is true, and we’ll get to that. Sadly this was incendental to the point he was trying to make (i.e. criticism of positive discrimination, which is okay even if you disagree with him) so it needn’t have ended this way. As things stand, his actions were disrespectful to those groups and will certainly have impacted productivity within Google because of the understandable offence caused.

You might argue “free speech”, as James apparently has, but like Randall says the right not to be persecuted by your government for your opinions is not a right to be listened to without consequence. If you want to function in a team you need to learn to play nicely with others.

Woah there, Nelly

I can hear some of you itching to bound in with “BUT IT’S SCIENCE YOU CAN’T BE OFFENDED BY SCIENCE!”. You folks will have to hold your horses a bit, I promise I’ll get to that, in the meantime here’s a picture of a puppy.

It’s a metaphor. You’re the puppy.

Own it.


2. Less rhetoric, more listening please

This is a highly, highly, emotive topic so care must be taken. Yes, on both sides. The kind of access we have to the labour market strongly affects our prosperity, and hence the life chances of any progeny. It’s important. It’s very easy to trigger a ‘fight or flight’ response.

What’s more, confirmation bias tells us that people aren’t listening for facts that might change their mind. Taking a tone-deaf approach spouting ‘facts’, or worse an approach involving inflammatory rhetoric, is not going to be effective.

The only real approach here is to listen, and try to form a rapport & trust upon which you can then build a discussion without people feeling threatened. Try to empathise even if it’s hard, even if you really really disagree.

Rather than go line-for-line, I’m just going to highlight one area in James’s memo that failed on this score. In a list of Google’s “discriminatory practices”, James includes:

“Hiring practices which can effectively lower the bar for diversity candidates by decreasing the false negative rate”
Quick terminology check just in case: false negatives are when you mistakenly turn away a good candidate. Google has a famously high rate of false negatives, with stellar engineers often having to apply multiple times.

To break this out; James is saying that the program within Google that makes sure “diversity” candidates are considered carefully is “lowering the bar” by reducing the risk of false positives.

While a program like that could increase the chance of hiring a diversity candidate by reducing mistakes, but by definition cannot “lower the bar”, after all to be a false negative you must already by definition be good enough to work at Google. This is inaccurate and inflammatory rhetoric.

Data. It’s a… oh nevermind.

Since some of you are about to pop at this point, full of “ENOUGH TOUCHY FEELY CRAP OMG THIS IS ABOUT DATA!!” let’s address that now. I know, I get it. I’m a scientist, an engineer, and a spectrum kid, so my magic low-empathy underpants are just as abrasive as yours… but broadly speaking; if you want acceptance for your idea (in this case the idea that positive discrimination is bad) you have to go through what I call “people” in order to get it. To convince people that you’re right you need to get past their fight response, and then get past their confirmation bias. That’s a slow process and they need to trust you. If they suspect that you’d push them in the mud tomorrow because data said so, then you get no cookie.

James’s failure to manage that threat response was no passing comment either. It’s structural within James’s argument. He takes painstaking care to provide evidence to support his case of biological differences between genders, but then broadens his conclusions to criticise diversity programs as a whole. By not justifying this, he risks people assuming he’s performing a rhetorical sleight of hand for political reasons, we lose any hope of trust, and we’re back to Cortisol City.


3. Biology is a red herring in a fishy minefield

Lots of the argument centres on whether biology a significant contributor to the numbers of women in tech.

This is a red herring.

Even if there are relevant biological differences between genders, as claimed (and there’s reason to assume otherwise) that’s a chasm between that and claiming a ‘natural’ difference in representation.

Firstly, there is no model (let alone a model acceptable to all parties) that can take those biological gender differences and predict what proportion of women, or other under-represented groups should ‘naturally’ be expected at your company.

Secondly, there is no model for the influence of cultural factors, which are clearly huge and can change quickly and artificially, such as when female participation in software dropped in the USA but not everywhere. If there were such a model its consequences would need consideration, e.g. if the next two decades brought a trend for a particular demographic group to do something other than work in tech, should that model de-emphasis hiring them? For how long? That approach is both speculative and reactive.

So there is literally nothing here on which to base an actual hiring policy.

I’ll repeat that point. Without further study, even if his premise were true, there is no way to responsibly implement James’s proposals. He says that Google doesn’t need to achieve full 50/50 gender representation, but falls far short of being able to propose an alternative measure of whether hiring is fair.

When you invoke biological differences in hiring discussions, aside from prompting a threat response, you also promote a stereotype. This increases the risk of unconscious bias, which exacerbates problems in hiring, problems in performance evaluation, problems in collaboration, and problems in pay.

That’s not to say that conversations about gender (or even racial) differences cannot happen, but they require primary research, not memos and hyperlinks. Having them happen in a company, rather than rigorously in a research environment, says much more about politics than it does about science, and does more harm than good.


4. Meritocracy is a myth

As you’re all probably aware, much of the reaction against positive discrimination (or affirmative action) follows from the assumption that without interference hiring is done relatively fairly. Based on that assumption, when positive discrimination is added white men are therefore at a comparative disadvantage due to race/gender, making hiring racist/sexist.

In that worldview, discrepancies in representation are caused by historical problems that will correct themselves over time, or perhaps differences in interest, motivation or ability.

That assumption is uninformed; there’s a mountain of evidence for both individual unconscious biases, and unintentional biases on an institutional level.

We could go into more detail about the evidence for unconscious bias, but it’s sailing a little close to my vested interest, so I’m not going to labour the point too much, but to give you a taste;

  • The research on bias shows how unconscious bias affects call-back decisions, performance reviews, mathematics grading, parole judgements, jury decisions, behaviour in meetings, perception of leaders, perception of teachers, assessment of interview candidates, job offers, starting salary, the list goes on and on and on.
  • There’s also real-world evidence, eg. we ran a study on 700 candidates that uncovered (and resolved) unintentional bias against candidates from a low socio-economic status within a customer’s hiring process. We’re not alone with that type of finding.

Believing in meritocracy increases bias

Daniel Effron at LSE has demonstrated an effect called moral licensing, effectively a process in which the subconscious horse-trades good deeds with itself in order to balance morality with other goals.

When people feel unbiased, e.g. you remind them they voted for Obama, they licence themselves to make more biased decisions.

I’ll step that out again for emphasis; if you genuinely believe that your hiring process is beyond reproach you’re statistically more likely to unconsciously licence yourself to act in biased ways.


5. Not all diversity programs are good

This one may be a bitter pill for some, and amber nectar for others, but please accept that this criticism is made constructively. Nobody wants hiring to be unfair, there’s just disagreement about what fairness looks like because people are (predictably) starting from their own perspective.

Clearly the intentions of diversity programs are good, but they’re not always effective, and can have negative effects, see below. It’s absolutely critical to monitor those programs and their effectiveness, just like you would any other important infrastructure.

James made this point, and we should all agree; to make diversity programs as effective as possible (within the bounds of fairness) we need to be willing to look at them critically and measure what works.

Resentment over quotas

I talk about ‘quotas’ here but really this applies to any program that actively prioritises, short-lists, signal-boosts, or gives a leg up to some groups over others.

In any discussion of positive discrimination there’s a risk that the overrepresented group (usually white men) may feel threatened. Unsafe. People aren’t born aware of their comparative advantage or disadvantage, and sometimes never see it, so when other groups seem to be given a leg up it can feel unfair.

Dobbin & Kalev great HBR article “Why Diversity Programs Fail” shows the negative effect of diversity training, among others

Feelings of injustice may be built on a misconception, but they still exist and are natural. In a corporate environment they can cause the following.

  • Because there’s a perception of a “lower bar”, incoming team members are seen as inferior, and resentment means they receive less support than their neutrally or positively-regarded colleagues. That lack of support network reduces their productivity (in contrast a boosted support network is one of the key reasons hiring from your team’s network is so successful).
  • To compensate for that reduced network, companies with diversity hiring programs must also offer specialised on-boarding/mentoring, which can cause further resentment.
  • In extreme cases, this resentment bubbles up into open disputes like this one and sooner or later someone is pissing on everybody else’s chips.

Unconscious bias training

Not everybody is aware of (or believes) the evidence that unconscious bias is a thing. Regardless of whether or not it’s a real thing (I’m not going to toot that horn because of our vested interest in that area) the evidence that training employees to counter it is not great.

  • Evidence that training works is mixed
  • Training wears off. Aside from individuals forgetting or losing habits over time, employee churn means you need to constantly keep training teams in order to achieve an effect. Paluck & Green (2009) found that the positive effect of bias training only lasted a day or two.

It gets worse:

  • Mandatory training can cause resentment amongst people that aren’t predisposed to its objectives, see above.
  • As discussed above, moral licensing means that when people feel unbiased, such as when they’ve graduated from diversity training, they statistically licence themselves to make more biased decisions.

So what to do?

  • Don’t change diversity programs based on James’s memo. It’s unlikely we’ll see a compelling scientific model that can accurately model this stuff any time soon, so don’t hold your breath.
  • Read the book What Works, by the legend that is Iris Bohnet. This is the Bible for evidenced diversity policy.
  • Dobbin & Kalev’s great HBR article “Why Diversity Programs Fail” has data on what types of diversity program have actually worked.
  • If you do want to use unconscious bias training, monitor real outcomes to measure effectiveness. There’s some more advice on bias training on HBR
  • Think about bias reduction… making your hiring process (a) consistent and (b) structured will help make it fair. Standardised, structured, interviews are your friend. As are (c) work samples, and situational judgement tests.
  • Be aware of gendered language, codes of conduct, skewed working conditions, and sourcing candidates from multiple places to balance biases in your sourcing pool.
  • Whatever approach you use, even if you use no approach at all, monitor your equal opportunities data like you would a marketing funnel. Look for problems in your conversion rate at each stage.

For those who like illustrations of some of the ways bias can creep in here’s some fun reading/playing:

I’m off. Play nice.

Richard Marr is co-founder and CTO of Applied, a SaaS platform that increases hiring precision and reduces bias. While he talks big, he’s actually full of crap, and rather spuriously claims to have invented A/B testing.