Historically - models, analyses, and data sets created without consideration of equity have perpetuated unequal societal structures and policy decisions. The MITRE Social Justice Platform’s (SJP) Guide for Integrating Equity into Data Analysis provides recommendations, resources, and key questions to ask when considering equity at each project stage. Read about a few key considerations for each step below, then visit the full Guide for Integrating Equity into Data Analysis for more information, to access helpful external links and resources, and to learn from real world examples.
When identifying your problem and planning your approach, consider the populations you are modeling and how to integrate community partners from underserved communities and those who this work may impact. Consider how your own background and bias may influence your work and seek to develop equitable research questions.
Consider how alternative data sources, including qualitative data and community/use-case specific data, may help answer your research question. Use stratified data where possible and applicable, seek to understand how measurements may differ for underserved communities, and seek to understand collection processes and limitations of second-hand data.
Consider alignment of datasets, and make sure that subgroup definitions align. Variable processing, data processing, how variables are used, and more reflect the experience and understanding of those creating the model. Embed your perspective on equity into the choices and assumptions made in your model.
Be aware of the biases inherent for different models and methodologies (see the full Guide for resources). Also be aware of the biases of data fed into your model – consider how your model could be used by an “adversary” and be wary of the implied or explicit assumption that white is the default position.
Consider accessibility (how the visualizations, interfaces, and other parts of your tool can and will be accessed) including factors like color-blind friendly and accessible color choices and 508 compliance. Also consider implications of other choices in your tool, including language, data labels, ordering, icons, and visualization approaches. Think about your audience and how different groups will interact with the tool.
Perform spot checks and validation with affected communities and use literature and alternative sources of qualitative feedback to verify work. Consider context when interpreting your data to ensure you don’t discount upstream factors that impact what you are measuring.
Consider your audience(s), an asset-framing lens, language choices / messaging, and visualization choices when designing your dissemination strategies. Clearly acknowledge limitations and biases of your work and bring in affected communities to assist in drafting and verifying language choices. Share your data and work so that other researchers can use it, which also minimizes the burden of multiple inquiries for information from the community being studied.
Explore the resources we used to generate this insight.
