Recently I had the opportunity to work on a project examining why women are more likely to suffer chronic pain than men. My assumption going into the project was that the disparity was likely to be driven by attitudes and biases of healthcare professionals, relative economic disadvantage, and the disproportionate burden of caring for others that women bear.
All of that did, unsurprisingly, turn out to be the case. However, what I wasn’t expecting was the extent to which poor data contributes to the issue. The experience brought home to me how much our decisions to measure (or not) reflect our assumptions about value and impact, and how having the right data can make issues visible.
Invisibility
To begin with, women’s health issues are notoriously understudied. Research into non-gendered conditions, such as heart disease, frequently don’t include female participants, or fail to disaggregate data by sex. If you’d like to understand more about the scale of the problem I strongly recommend the work of Caroline Criado-Perez, whose book Invisible Women has been instrumental in bringing these issues into the public eye.
The data gap not only risks individual tragedies but also skews priorities
In the absence of formal data, anecdotal evidence becomes the primary source of information. But in a worldview built on the idea of objective, tested evidence, women’s reporting of their own conditions is frequently treated sceptically. If their condition falls outside the known patterns, medical professionals too often try to force it into place, or dismiss it as either incurable or ‘all in the head’.
The data gap not only risks individual tragedies but also skews priorities. As Criado-Perez points out, women become invisible to the health system. Funding for research is not prioritised, and treatment protocols are inappropriate or ineffective. Women’s conditions are not recognised by employers in benefits entitlements or even by their own communities and families.
Growing recognition of the data gap means these issues are beginning to be addressed. For example, the Australian Government has committed funding to research, education and treatment of endometriosis: a condition that affects one in nine Australian women but has historically been poorly understood and treated. Calls for paid menstrual leave are growing, and employees of the Victorian State Government have recently been granted five days of paid leave for symptoms relating to reproduction, including IVF treatments.
Culture of challenge
The pattern that emerged from this research – data gaps, unexamined assumptions, dismissing evidence that doesn't fit – might feel familiar. The promise that big data offers – of streamlined decision making and penetrating new insights – too often leaves us blinded to the drawbacks. The literature is full of awful examples: facial recognition technology that is more likely to misidentify racial minorities, or human resource management software that prioritises white, male applicants
Complexities can too easily be lost through overreliance on the one story being told by your systems
Perpetuating disadvantage is not the only potential risk of unexamined or non-existent data. The complexities of the markets, stakeholders and social and natural environments in which you operate can too easily be lost through an over-reliance on the one supposedly ‘objective’ story being told by your systems.
Take, for example, the 'optimised' timetable introduced by a rail transport provider based on analysis from a newly developed scheduling program. Within a week of a pilot scheme affecting one line, the timetable was in chaos because the scheduling program failed to take into account vital data regarding the state of the tracks limiting speeds, short platforms requiring longer stopping times, and congested level crossings. At the onset, drivers claimed that the changes were unworkable but their concerns were disregarded in favour of unreliable and incomplete data.
As automation and artificial intelligence continue to permeate decision-making processes, having a culture of challenge around the data which is fed into these processes is imperative. This requires discipline around maintaining the transparency of the datasets and assumptions you are relying on. You can then test the completeness and accuracy of your data and assumptions regularly, with a wide range of stakeholders. Pay particular attention to groups that are not usually involved in building those datasets but who may be impacted by them, such as customers, communities or the custodians of the natural environments in which you operate.
Challenge can be exhausting and disruptive, both for decision-makers and those making the challenge. The ease of decision automation has to be offset by the effort of maintaining humility and flexibility. Encourage, promote and above all listen to any indications that your decision-making processes may not be working.
