People Analytics: Is Your Data Biased?
People analytics is about adding quantitative measures to often qualitative data to improve decision making. This is a significant undertaking with implications that stretch far and wide. As we know, evidence-based decision making requires reliable and valid data. As people analytics serves the people, practitioners have a requisite to examine how unconscious bias can and does impact that data and can skew its reliability and validity.
Consider that unconscious bias is about understanding what are the unrealized tendencies that hinder our ability to interpret a truth authentically. These truths tend to come to us in the form of data. Whether it is making a judgement about someone we have just met based on visual or contextual data we rapidly gather, analyzing a report, or organizing an excel sheet for presentation, our brains are constantly looking for shortcuts. But data is only as good as its reliability and validity.
Effective people analytics requires data inputs, analysis, and then outputs. Some systems, like AI for example, learn from their data. So if the inputs include human bias, the results will mirror that bias. HBR said it best, “AI can certainly help identify and reduce the impact of human biases, but it can also make the problem worse by baking in and deploying biases at scale in sensitive application areas.” In fact, Amazon stopped using a hiring algorithm after finding it elevated applicants based on word usages like “executed” or “captured”, those most prevalent on the resumes of male applicants. Their algorithm was also found to downgrade resumes with women’s only universities. These examples of biases in AI and learning systems are not unique to Amazon. Studies have found that systems that analyze faces tend to exhibit higher rates of error amongst women and underrepresented groups. When the data sample doesn’t reflect the widest aperture (inclusive), we get a narrower view (exclusive).
Not all systems for data analysis learn through inputs. Some require interaction. Microsoft, for example, deployed a twitter based chatbot that over the course of numerous interactions with users that encouraged racism and misogyny, had to be shut down. The offensive interactions created a trove of data whose outputs resulted in wildly offensive tweets. Here we can see how the number of opportunities to let unconscious bias slip into data are exhausting and require diligence to keep asking those critical questions, is this valid? Is this reliable?
One of the best ways to mitigate unconscious bias is to expand your aperture. That is, explore cultures, ideas, people, experiences that are different from your own. The brilliance of algorithms should support this, but some of the most popular functions have not brought this to bear. Facebook, Instagram, Tik-Tok, all learn to make content display decisions with the intent of curation but in so doing, perpetuate unconscious biases. Think about the way they work. You like a post or engage with a post, then you start to see more like it. Instead of widening your aperture, you narrow your focus. This has its place, but when it comes to good data it only introduces complexity to mitigating unconscious bias.
As discussed, people analytics often looks at qualitative data which means the likelihood that bias is abound is highly probable. Where quantitative data tends to be designed to test a theory, qualitative data leans towards the generation of theory. In people analytics where we work up from the business problem, that becomes an important distinction.
To ensure that you have data you can trust there are a few things you can do. The first is to express curiosity in the process and an ability to explore meaning rather than rooting tightly to “cause and effect”. Qualitative data requires a degree of abstraction so being able to critically question approach and observe themes and patterns is crucial. Objectivity is also vital. Understanding how to examine the words and perceptions of others as well your own.
Expanding objectivity to reflexivity will also strengthen your approach. That is to say that the better you can recognize your own likeness (“insider status”) or dis-likeness (“outsider status”) as it relates to the data will be significant, as this will further support the mitigation of unconscious bias. Finally, an understanding of the concepts around quantitative data gathering and analysis are crucial to framing any qualitative undertakings.
People analytics should support better people decisions and so the space for investing in this arena is huge. It should not however be approached flippantly because as we have seen, biases can seep into data. Stay curious, stay critical, and maintain an eye for learning because people analytics is complex. In our data driven world there is a massive opportunity to leverage this type of data to better support individuals, teams, organizations, and communities alike. As long as we do not enter blindly, we can truly maximize our impact.
