In the seventeenth century, men in England believed themselves to be outnumbered by women. There was no real evidence for this view, but it came in part from an apparent growth in women’s influence and visibility – for example as they started to petition parliament and protest for an end to the civil war. It was simply a perception.
In the earliest known attempt to actually measure the proportions of men and women (the ‘sex ratio’) in 1662, the philosopher John Graunt undertook an impressive exercise in big data mining. He combined death registers with records of burials and christenings in London, to build a statistical picture of the sex ratio.
What he found surprised people at the time. He showed the sex ratio to be 107 males to every 100 women – so in fact the proportions were very close, and certainly didn’t appear to favour women.
Unfortunately, even publication of Graunt’s findings failed to convince some men. Much like today’s cries of ‘fake news’, the data was generally ignored.
It wasn’t until half a century later that the doubters were really proven wrong. The physician John Arbuthnot applied the first known formal test of statistical significance to the sex ratio, using over 80 years’ worth of birth records. He concluded that the slightly higher proportion of men could not have been due to chance alone. Statistically speaking, men were in fact more abundant. It remains a mystery why males outnumber females, but theories include the need for natural selection to compensate for higher death rates in males or for higher parental resource investments for females.
Cue a long and complex debate about sex ratios over time and across countries, but these early studies showed the importance of taking an objective, scientific approach, and finally overturned men’s misconceptions about being outnumbered. Today, we can draw some important lessons from these studies in our own measurements of diversity and in how we use data to drive our organisation’s action on diversity.
What gets measured, gets managed
Unlike the situation in the seventeenth century, most organisations would now accept you can’t assume that if women, for example, are more ‘visible’, that gender diversity issues are solved. And much like the studies on sex-ratios, it is impossible for an organisation to really understand how well it is doing unless it has decent data and is measuring the right thing.
Measuring more ‘traditional’ aspects of identity diversity, such as gender and ethnicity, is valuable as a starting point. But be sure to disaggregate enough to uncover potentially important differences. For example, a simple split between BAME vs. white employees may mask important differences in the representation of different nationalities or ethnicities. The Office for National Statistics found, for example, that in the UK, Chinese and Indian employees had higher median hourly pay than all other groups. The same study also suggests more granular categorisations for ethnic background that could be useful if you want to repeat a similar analysis in your organisation.
Consider other dimensions too, such as socioeconomic background and neurodiversity. These dimensions of diversity may be harder to conceptualise and measure but can reveal important inequalities that matter societally and may also be holding back organisational performance.
Academic research has shown that cognitive diversity – differences in how we think – can be an important driver of team performance. Having greater diversity in mental toolkits and approaches to problem solving may even trump having higher individual ability across a team, especially for complex, high dimensional problems— think “wicked” policy problems such as tackling obesity or responding to a pandemic.
It’s also important to build an 'end-to-end' analytical approach, so that the biggest diversity gaps and improvement opportunities can be pinpointed scientifically and prioritised for action.
A simple but powerful 'people science' approach is to measure diversity across the entire employee ‘life-cycle’, from job applications (How diverse is the applicant pool?) through to employee departures (Do colleagues from a particular background leave more frequently?). This makes it possible to see where diversity is being gained or lost across key moments in the talent lifecycle.
Using this approach, researchers at the European Central Bank found that career progression for women was slower than for men and that in large part this could be explained by lower application rates amongst women. A lifecycle approach enables an objective overview of exactly where an organisation may be gaining or losing diversity along the employee pathway, helping to focus action. It can also be used beyond diversity considerations, for example to see how talent is managed as employees move through the lifecycle.
Take a statistical approach
Arbuthnot’s approach to measuring sex ratios was revolutionary at the time, but today it should be standard practice. Instead of coming up with subjective measures of ‘what good looks like’, to truly understand whether you are doing well, apply tests of statistical significance to your diversity data.
Statistical significance is a way of quantifying whether a finding is just due to chance, or to some other factor. Tests of statistical significance take group size into account – in recognition of the fact you will see more ‘noise’ in smaller groups. This is because each person represents a much larger proportion of the whole in a small group, so if one person arrives or leaves, the proportion of men to women can jump dramatically. For example, if one man joins a group of 2 men and 2 women, the group average goes from 50% male to 60% male. If a man joins a group of 100 women and 100 men, the average goes from 50% men to 50.2%.
Statistical significance tests can be used at each stage of the employee lifecycle to diagnose any systematic bias. For example, if there are more male than female new recruits, when applications for those roles are balanced (as in the employee journey figure above), tests of statistical significance can be applied to the stages of the recruitment part of the lifecycle to see where the bias is introduced.
Statistical tests can also tell us how confident you can be that you are reaching our diversity targets. For example, if the target is to have women comprise 50% of all employees across the organisation, you can use a test known as the chi-square test (or Z test) to tell us how you’re doing. It gives a measure of statistical significance in the difference between the proportion of women and this 50% target.
Below, using the chi-square test, you can see that in a large group 40% of women is statistically significantly different from 50% women, meaning the target is not being met. But in a small group, the difference between 40% and 50% is not meaningful and is probably due to chance.
Applying statistical significance tests and adopting an employee lifecycle lens removes the subjectivity from an assessment of “how we’re doing” on diversity and where issues are arising. It can show leaders which areas to prioritise for action, by pinpointing objectively where a target is being missed (statistically speaking), where bias is being introduced along the talent pathway, and where things are going well.
Is your target well calibrated?
Setting targets has become commonplace and targets can be a powerful way of tracking the scale of the challenge and focusing efforts.
Graunt and Arbuthnot’s question was relatively straightforward – do we have more than 50% women or less? But what should we be asking of our organisations? What should our targets be, and how should we measure progress against them?
Diversity targets should be ambitious, to help demonstrate a commitment to deliver against them. Signatories to the Women in Finance Charter, for example, have targets for senior management ranging from parity (50% women) down to 5%. Those with lower targets are starting from a low baseline, but HM Treasury would like to see all signatories move to at least 33%.
For many organisations, ideal diversity targets will at least in part be driven by a chosen demographic they want to reflect – the customer base or the region the organisation operates in. The last UK Census (completed in 2011) contains demographic breakdowns by region but it is due for an update in March 2021.
In many cases, it will pay to have a target that is realistic, so that it is feasible to meet. Again, it will be important to have the correct data to help understand what is possible. Some areas of an organisation may see lower turnover rates than others, so achieving a shift in the number of diverse hires will be more challenging in the short-term for these roles.
The figure below shows an example where achieving gender balance in the senior leadership team would take time and commitment. Depending on start point and factors like expected balance in the underlying pool of applicants, something of a 'Goldilocks' target ('not too big, not too small') may be attractive, creating a goal that is a stretch, but also attainable.
If a goal that would be highly desirable appears unrealistic given the starting point, it may still be worth using this insight to consider 'what it would take'. Focusing people on something of a 'moonshot' target could stretch the organisation’s ambition levels on diversity, challenge ingrained ways of thinking, and potentially lead to a more radical step change. Suppose it really were non-negotiable to reach balanced female representation quickly from a sluggish base, how could this be achieved? What is possible given current turnover levels – to what extent are we seeing departures and able to make new hires? If expected recruitment possibilities won’t get us to the ambition (as in the case of the moonshot target below), what else could we consider to transform our diversity and how long should it take us? This kind of thought experiment may reveal a disconnect between aspirations and what is possible in the short term, which may be disheartening, but may be powerful for shifting thought processes and mindsets that are too incrementalist. Of course, there can be unintended consequences and these need to be thought through; where a target can be achieved only through radical change, will positive action become positive discrimination?
Are we on track? Achieving challenging targets may not be possible in the short term, but what would it take?
It will be worth considering whether to set different targets for diversity at different levels of seniority.
Senior teams (groups which tend to be fairly small) are disproportionately influential. Decisions taken can impact the entire organisation and representation of women and minority groups in senior roles may be particularly important for role-modelling, visibility and setting organisational tone and culture. So setting and achieving meaningful targets for boards and leadership is particularly important.
Relatedly high representation of women or minorities in junior roles can be problematic in its own way. Whilst it’s vital to have a good pipeline of diverse talent moving up through the organisation, high representation of at typically lower paid levels will contribute significantly to the average pay-gap (via `vertical inequality’) and risks sending a signal that these groups belong in more junior roles.
Not just a numbers game
The ideas above can help an organisation identify an unwanted imbalance, allow changes to be tracked over time and for corrective action to be taken when needed.
However, having a diverse workforce doesn’t guarantee that people feel included or have an equal impact on decision-making. Inclusion is vital and comes from fostering a culture of empowerment and psychological safety, where employees can bring their full selves to work and have their voices heard.
So, as well as reporting on representation of different demographics, organisations should consider ways to quantify inclusion.
A ‘people science’ approach might include some more objective measures of inclusion as a complement to self-reported survey scores. Examples might include the number of network connections people have, the level of participation in important meetings, even time spent collaborating with more senior colleagues. Data like this needs to be treated with caution as the science of measuring inclusion is still at a formative stage and any single metric could be quite misleading. It’s also important to avoid the potential downsides of measuring human behaviour in the workplace, such as the risk of gaming metrics and colleagues’ privacy concerns. There are mitigants here too; to address privacy concerns an organisation might only give people access to their own individual data with personalised statistics and recommendations and explicitly ask for consent for any other uses. Going beyond the usual staff surveys to a broader basket of indicators could give a more evidence-based feel for how well the organisation is doing on inclusion and where some of the issues and improvement opportunities may lie.
Nurturing talent and promoting employee wellbeing is a major priority. Integrating more robust analysis, leveraging data science and behavioural science effectively over the employee lifecycle can bring huge benefits by attracting diverse, high-performing colleagues, whilst enabling them to be at their best by understanding and improving organisational culture and internal processes. As noted in a previous Insight article, behavioural science is key to understanding how to build in the right processes to support a culture of diversity.
Finally, even once a good system for measuring diversity is in place it needs to be reviewed regularly so the most relevant metrics are being captured and reported.
And of course unexpected changes in ways of working or employment patterns (such as those seen with coronavirus (Covid-19)) may result in sudden shifts in diversity and inclusion. We all need to stay on our toes.