What do hiring managers have in common with beetles?
There are three big questions that I’ve circled around for years, and finally I’ve found a model which might just bring them together and get some answers:
Why is that some people can perform the same (or better) than their colleagues but still fail to gain the momentum they see to build?
Why do we still, after decades of DEI, fail to see meaningful diversity in senior management positions?
Why do the people who most need to engage with DEI topics participate the least?
Affinity bias
Affinity bias = the unconscious tendency to gravitate to people similar to ourselves
As a student of UX I’ve always loved collecting cognitive biases. In this case, Affinity Bias seems like a great fit. It speaks to the ego in leadership - leaders can praise people who are like them, over inflating their worth. People who are different are then less likely to get the same recognition. (This explains why it’s also called mini-me syndrome, and we’ve all seen that dynamic play out - the leader with a protege about 10 years younger, who they consider a younger version of themselves to take under their wing.)
Affinity bias has limits though - it gives us an explanation for some types of bias we see on a case by case basis. It doesn’t necessarily account or help us picture for the wider patterns of inequality across an organisation.
Beetles and bias
We’re going back in time to the 1950s some scientists did a bunch of experiments with beetles. They put two species of beetle into a container. No limits on space or resources, just different species. The opposing species still went out of their way to eat the eggs of the other group. After a time there would only be one species remaining. It wasn’t always the same species in every test, and it was not always the most biologically fit for survival. The winners were the species that gained the foothold first.
The Promotion Chain Model
In the 2010’s some economists at MIT stumbled onto this study. They quickly took to this idea of competing groups and egg eaters who would tip the scales of each generation of beetle towards their own species victory. They wondered how they could apply this to academics. They created a model that demonstrated the same behaviour, but instead of beetles, it was academics favouring students who supported the same theories as them. Each ‘generation’ of researchers made it to professorship because they belonged to the same group, and researchers who held alternative beliefs didn’t make it to tenured professors.
This model was expanded and crafted to apply to organisations with multiple layers of hierarchy. They created a formula which demonstrates the consequences when managers keep power and privileges for those who fit the mold and who they consider being their peers. This model bridges the gap between the individual promotion decisions and the broader perspective of diversity across the organisation at all levels.
“the principles we find regarding science also apply to promotions in hierarchical
institutions with advancements up a promotion ladder. As with science, promotions biased to resemble those in higher rungs, can filter employees with lower productivity up the promotion ladder—with the highest concentration of such workers at the very top.”
Breaking down the model
The model uses inputs to build the output. There are a few different values that need to be set:
The number of levels of hierarchy in the organisation
The starting proportion of each group - we’ll use red and green for our examples
The denial rate for qualified greens. This is a the percentage chance that someone from the green group will be denied a promotion they are qualified for due to the bias.
The promotion rate for unqualified reds. This is the percentage chance that someone from the red group is promoted into a role they aren’t qualified for.
Evaluator bias. This is the rate at which the hiring manager favours their own group. That could be red or green.
When we put all this together we can see how the outcome starts to play out.
It’s possible for any organisation with any number of levels of hierarchy to experiment with the model and see how they can manipulate the values to mirror their real-world results.
Putting into practice
I wanted to test it - I looked at the split of men and women in an organisation I knew well, and I played around with the variables until I got a result that mirrored the real world use case. Green = women, Red = men. Here is the break down:
Levels of hierarchy:
Value: 8
Real-world Meaning: 8 levels of seniority within the organisation, from CEO at the top, to lowest professional level
Starting Proportion of Greens:
Value: 30% at Level 8.
Real-World Meaning: Women make up 30% of the workforce at the entry level, reflecting a gender imbalance that may already exist in the applicant pool or hiring practices.
Denial Rate for Qualified Greens (α\alphaα):
Value: 0.30 (30%).
Real-World Meaning: There is a 30% chance that a qualified woman is denied a promotion. This reflects systemic barriers, such as unconscious bias, higher performance expectations for women, or a lack of sponsorship/mentorship.
Promotion Rate for Unqualified Reds (β\betaβ):
Value: 0.40 (40%).
Real-World Meaning: There is a 40% chance that an unqualified man is promoted. This might represent tendencies such as overconfidence, favouritism, or relaxed performance standards applied to men in leadership evaluations.
Evaluator Bias (ϵ\epsilonϵ):
Value: 0.45 (45%).
Real-World Meaning: Evaluators are 45% more likely to favour candidates like themselves (e.g., men preferring men). This perpetuates existing gender disparities at higher levels.
And the results are:
The most disappointing thing is how I got to this accurate representation of a real gender mix - I had to just keep increasing the bias values so I could mirror the real world scenario. Let’s break this down:
30% chance a qualified woman is overlooked
40% chance underqualified man is promoted
45% either group apply bias to their own group in promotion decisions (but each level up is already made up of more men than the group below)
How can these keep happening, surely managers start to see the pattern and correct for it? Nope. Each manager will make their promotion decision having justified it to themselves and others, and fail to step back and see that impact on the wider organisation.
“Promotions at each individual level usually fail to take into account
choices between Greens and Reds at all higher levels; thus they usually also fail to take into account the overall mix of Greens and Reds in the organization as a whole.”
This is depressing, so what next?
Well the good news is that I’ve started to answer the questions I had at the start:
Why is that some people can perform the same (or better) than their colleagues but still fail to gain the momentum they see to build? - Affinity bias, both at single levels of promotion and over time.
Why do we still, after decades of DEI, fail to see meaningful diversity in senior management positions? - Because individual decisions fail to take into account the whole, and they compound at each level of management until one group has dominance.
This leaves us with the last question
3. Why do the people who most need to engage with DEI topics participate the least?
While working through this model it became more apparent than ever.
People won’t invest in dismantling a system that serves them.
If a person benefits from the status quo, they won’t be helping to change it. If their next promotion and pay rise comes from an inherent trait they’re favoured for, they won’t bit the hand that feeds them.
The people who need the DEI training won’t invest their time and energy in it. It’s human nature. Naturally it’s Not All People. But it’s evidently enough that the Promotion Chain Model is still able to reflect real-world scenarios to show lack of diversity.
The way I’ll face it is head on - I’ll talk about it. I’ll share the science behind it. And I’ll see what I can do open doors for the people who don’ fit it.
Experiment with the Promotion Chain Model
When I first learnt about the model I did not become a statistics expert overnight. I trained a GPT on the model and set it up to explain the theory, run scenarios and share results. It’s open to anyone who wants to play around with the concept.