The importance of fair business decisions

Fairness means ensuring that your analysis doesn't create or reinforce bias.

Ensuring that analysis does not create or reinforce bias requires using processes and systems that are fair and inclusive to everyone.

As a data analyst, it's your responsibility to make sure your analysis is fair and factors in the complicated social context that could create bias in your conclusions.

Eg. a company that's kind of notorious for being a boys club.

• There isn't much representation of other genders. This company wants to see which employees are doing well.
• The data shows that men are the only people succeeding at this company.
• Their conclusion? That they should hire more men. After all, they're doing really well here, right?
• But that's not a fair conclusion for a couple of reasons.
  • it doesn't even consider all of the available data on company culture, so it paints an incomplete picture.
  • it doesn't think about the other surrounding factors that impact the data, or in other words, the conclusion doesn't consider the difficulties that people of different gender identities have trying to navigate a toxic work environment.
• If the company only looks at this conclusion, they won't acknowledge and address how harmful their culture is and they won't understand why certain people are set up to fail within it.
• The conclusion that only men are succeeding at this company is true, but it ignores other systematic factors that are contributing to this problem.

Eg. A team of Harvard data scientists were developing a mobile platform to track patients at risk of cardiovascular disease in an area of the United States called the Stroke Belt.

• It's important to call out that there were a variety of reasons people living in this area might be more at risk.
• they teamed analysts with social scientists who could provide insights on human bias and the social context that created them.
• They also collected self reported data in a separate system to avoid the potential for racial bias, which might skew the results of their study and unfairly represent patients.
• To make sure this sample population was representative, they oversampled non-dominant groups to ensure the model was including them. ‣