The Temporary Assistance for Needy Families (TANF) Opportunity Act created pilot initiatives across Tennessee aimed at improving the economic mobility and self-sufficiency of eligible families across the state. To supplement ongoing evaluations of these pilots led by MEF Associates, the Urban Institute partnered with MEF and the Tennessee Department of Human Services (TDHS) to build a public data dashboard that offers broad insights into the circumstances of all Tennessee families and children.
Launched in May 2025 on TDHS’s website, the dashboard features more than 60 measures that shed light on the factors shaping well-being across Tennessee. These measures include demographic information, income and poverty rates, employment and education data, safety net caseloads, health care access and outcomes, and others drawn from both public data sources and aggregated statistics from the state’s administrative data system.
Here, we share the design, data, and development techniques we used to create a scalable dashboard that is accessible to users and easy for developers to maintain.
Creating a cohesive interface through user-centered design
We adopted a user-centered, iterative approach to design and develop the dashboard. We began with a series of scoping discussions and trade-off exercises with internal and external stakeholders to prioritize features, identify the most important questions the dashboard should answer, and inform initial wireframes. TDHS emphasized that the dashboard should serve a wide range of users — including the general public, journalists, legislators and their staff, internal TDHS teams, and community partners — by centralizing accessible, easy-to-navigate information about the populations TDHS serves and the broader context in which the pilot initiatives are operating.
These conversations surfaced three core questions users would want to explore for each metric:
· How does Tennessee compare with the national average?
· How does the metric vary across counties in Tennessee?
· How has the metric changed over time?
Stakeholders also emphasized the importance of enabling intended users to view data by the areas that Tennessee refers to as “grand divisions” (East, Middle, and West Tennessee) and to overlay state legislative districts to support policy conversations.
To cohesively present more than 60 metrics from different sources, we prioritized both flexible and scalable visualization types. Rather than tailoring visuals to each metric, we selected a small set of familiar, adaptable chart types (i.e., bar charts, line charts, and choropleth maps) that could accommodate variation in geographic coverage, subgroup availability, and update frequency. Like our Urban Institute colleagues wrote, we went from asking, “How can we best visualize each metric?” to “How can we best visualize all metrics?” in a consistent, intuitive interface.
Over the next several months, we refined the dashboard design through several rounds of wireframing and user testing. We also worked closely with TDHS to ensure the dashboard met both their substantive and technical requirements. In addition to gathering early feedback on static mockups, we tested a functional beta version of the dashboard with a diverse group of users, including Tennessee state employees, staff of community-based organizations, and policy researchers. In user testing, participants reported that the dashboard was informative and user-friendly, and offered helpful suggestions that we incorporated into the final product.
Creating data pipelines with long-term maintainability in mind
To pull and process the data that power the dashboard, we built data pipelines in R with automation and long-term maintainability in mind. The dashboard draws on dozens of public and administrative sources, each with its own structure, update frequency, and geographic scope. We designed the pipelines to flexibly ingest, clean, and harmonize these sources into a single, consistent file ready for Tableau.
We used GitHub to manage the codebase, track progress, and streamline code reviews. GitHub’s issues feature helped us coordinate moving pieces and gather input from subject matter experts — including those without R programming experience — by keeping conversations tied to specific metrics and tasks. This was essential for drawing on the expertise of Urban’s researchers, whose deep knowledge of social and economic policy was crucial to identify relevant metrics and sources, determine how data should be processed and reported, and verify that the measures presented in the dashboard were accurate and meaningful.
To make the pipelines resilient to changes, we adopted defensive programming best practices to anticipate and guard against unexpected changes. We used renv to manage dependencies and assertr to verify assumptions about the structure and content of input data. These safeguards allow us to proactively catch issues like renamed columns, missing values, or unexpected data types that will inevitably arise as we continue pulling data in the coming years.
We applied many of these same defensive programming techniques when accessing administrative data ensuring data security from the start. Within this secure environment, we developed SQL and R pipelines to aggregate records, perform data quality checks, and suppress statistics for small groups to protect privacy. Our code and outputs were reviewed by subject-matter experts and disclosure reviewers before being approved for public release through the dashboard. Additionally, all staff were required to comply with Tennessee’s data security measures.
We also used Quarto to generate version-controlled documentation for each dashboard metric. Quarto made it easy to keep documentation in sync with the codebase and track and review changes while allowing us to create polished, accessible versions in multiple formats (HTML, PDF, and Word) to support collaboration with internal teams and TDHS.
To develop the dashboard, we had to balance flexibility, performance, and sustainability
We developed the dashboard in Tableau with three priorities in mind: maximizing performance, minimizing ongoing maintenance, and promoting reproducibility.
Though Tableau is powerful and user-friendly, it has customization limitations. Fortunately, the Tableau community offers a wealth of creative workarounds. During development, we explored a range of built-in approaches (e.g., parameter actions, Dynamic Zone Visibility, sheet swapping, and cascading filters) as well as custom solutions to support interdependent user selections and the responsive filtering needed to display dozens of metrics without overwhelming users. While we utilized some simple workarounds in the dashboard, we found that more complex approaches often introduced latency and maintenance complexity. As a result, we decided to prioritize Tableau built-in and optimized features to balance flexibility with performance and long-term sustainability.
To streamline future updates, we kept nearly all measure-specific logic outside of Tableau and within the R-based data pipelines, including label formatting, notes, value types, and data availability indicators. For example, instead of configuring conditional formatting within Tableau, we preformatted values in R (e.g., with dollar signs or percent signs) and included flags that trigger dynamic labels and notes when certain data are unavailable, ensuring each data file is “dashboard ready” upon upload. This approach minimizes manual effort, facilitates version control, and allows us to track changes more systematically than would be possible with Tableau’s point-and-click development.
Looking ahead
These approaches helped us build a scalable dashboard that is accessible to users and easy for developers to maintain. Urban will continue updating the dashboard with new data through the end of 2026, at which point TDHS will maintain the tool moving forward.
We’re eager to see how Tennesseans use the dashboard to inform their work, and we look forward to applying many of these same design and development approaches to projects that help practitioners make evidence-based decisions.
