How do you check for bias and errors in data management?

A discipline that can assist in bias reduction is the incorporation of ethics into the various data management activities. In data collection, ethical principles and stipulations can ensure fair, transparent, and consented data lands the business ecosystem. Those are ingredients to which bias is allergic. Ethics lays out accountabilities for data actors throughout the data lifecycle, from acquisition to disposal or archival. This can help minimize bias in reporting. With the help of ethics, an organization can use data to benefit not only their bottom-line, but also the wider ecosystem in which they operate, thus promoting sustainable strategies. This collective participatory logic can help eliminate a lot of bias in data and analytics.

Data quality checks that are embedded in the business process can be incredibly valuable. Engage business process owners in creating best practices that control the completeness, accuracy, consistency and validity of the data being produced by the process. This has the twofold result of smoothing the transactional processes upfront and providing better quality data for data consumers downstream. Some ways to improve data quality checks are implementing data input controls in the applications, well defined procedures, and employee training as needed. I've witnessed small changes like adding a dropdown with a list of acceptable inputs can produce a measurable increase in data quality.

An important stipulation here is to make quality standards mandatory. Perhaps a functional unit will take a short-term hit in their bottom-line, but it is significantly better for the organization in the long-term. Enforced standards help avoid data errors. Some standards based on ethical principles that can help reduce bias are: diversity in collected data, accountability in managing data, transparency in the data lifecycle.

Data bias is one phenomenon, data management (DM) bias is another. On the latter, bias occurs throughout the data lifecycle, from collection to analytical interpretation. Some ways to assess DM bias are checking the data source and collection methodologies, auditing how people wrangle and transform data, analyzing the data models built for analysis, and evaluating how data and information are presented for consumers.

This is the fulcrum, as with nearly every issue. Communication in this matter will succeed more if feedback loops are designed and introduced in the existing or new business processes. This way, involved actors have less reactive ways to address issues throughout the data lifecycle.