How do we measure progress in the diversity/equity/inclusion movement?

Question

No Politics Zone! This is a sincere question around how we conceptualize progress in Diversity/Equity/Inclusion (DEI). It is not a question inviting debate on the value of DEI.

For purposes of this question, it is assumed that the sentiment and overall mission of DEI is valid, but it is complicated to measure.

I am going to set up some context so it is clear what I am asking. The TL;DR; version is

"Has philosophy studied the underlying assumptions of DEI progress measures? Is the assumption that all groups are equally qualified for all positions in the general population, regardless of stratification method, sound? Or do we need to also adjust for differences in access to qualifying experiences? Or is it a normative statement meant to highlight differences in opportunities?

---

DEI has been increasingly important within many corporations, who have started introducing metrics/KPIs around progress in these dimensions. During a recent DEI session at my company (and many of you can also relate) we hear the refrain “We have made great progress but still have a long way to go.”

This got me thinking about how we measure progress towards DEI, so as to better identify what works and what doesn't.

A common way I see things measured is based on a proportion metric π (e.g., % holding sr management positions) calculated from an evaluation population E (e.g., my company), under a relevant stratification σ (e.g., gender) and then comparing it to the proportion of each stratum in a reference population R (e.g., Continental US Population), to calculate the representation gap

The representation gap γ(π,σ,E,R) is equal to π(E,σ) - σ(R), which is a list of such differences indexed over the elements of σ

Where σ(R) is just the proportion of the population in R that is in each stratum (evaluated element wise)

Example

A good example is looking at women holding senior leadership positions, where we have:

• π = "% holding senior management position" [assume there is a rigorous definition of this]
• E = US-based companies employing more than 5,000 people
• σ = Gender
• R = All registered workers in the lower 48 US states

Let's say we run our study and get π(E,σ)=10% (women), 90% (men) and σ(R)= 45% (women), 55% (men).

Therefore, γ(π,σ,E,R) = [women: 10%, men:90%] - [women: 45%, men:55%] = [women: -35%, men:+35%]

The conclusion most would draw from above is that working women are very underrepresented in senior management positions and we should take action to make promotion and evaluation policies less biased against women.

While this was made up, I think most would intuitively agree with the general sense that there are "not enough" women in senior management positions right now. However, the degree to which this is an issue and where to direct policy and non-policy action depends heavily on what we take R to be and if σ is sufficiently rich to capture confounding variables.

These are largely statistical questions, but there is a deeper, more principled assumption that I think warrants philosophical discussion. I'll call it the Equiqualification Principle:

Equiqualification Principle: For any choice of E,R,π,σ we expect γ(π,σ,E,R) = 0 for every element of γ(π,σ,E,R).

Where we can model the sampling distribution for a given E using multinomial distribution assuming the proportions hold from R.

With that, here's my more nuanced version of the question from my TL;DR; above:

Should we take the equiqualification principle as a positive or normative principle (or neither, if we think it's misguided) for measuring DEI objectives?

Related: Do we need to restrict R and/or σ to maintain relevance?

Answer 1

It depends on what your objectives are. It's pretty clear that life is riddled with all kinds of bias and prejudice, which probably accounts for some of the skewed outcomes we see, such as the high proportion of men in senior management roles. The question is whether your aim is to do away with bias and prejudice or to ensure that any selection process leads to an outcome that's representative of the wider population, because those two aims are not necessarily the same.

For example, some people will tell you that part of the reason why women are under-represented in various roles is because they are not interested in them, so even if you did away with bias and prejudice among the selectors, you still have a kind of bias and prejudice within the applicant pool that skews the outcome.

You also have to consider cultural effects. I am inclined to believe- albeit on the basis of no scientific evidence whatsoever- that part of the reason for the different career preferences of men and women is that we are treated differently from birth onwards. The fashion and cosmetics industries, for example, spend $billions annually in perpetuating stereotypes of femininity, and perhaps that contributes to the self image of certain women which in turn disinclines them from particular career paths. So should we be trying to prevent the kind of systematic and widespread social conditioning that leads to engrained gender-stereotypical mindsets, or should we be celebrating the stereotypical differences between men and women and accepting that they lead to different career aspirations? The same question can be posed in relation to countless other socially and culturally perpetuated differences in outlook, belief and aspirations that lead to skewed outcomes in many selections processes.

Your ingenious equiqualification principle, if I understand it correctly, seems to assume that the result of any selection process should represent the population at large, which might not be your actual policy objective, so you might need to tweak it somewhat. That said, if there was no better measure, the EP would seem to be a good yardstick for several purposes.