Shirley-Leung-portfolio

home page data viz examples critique by design final project I final project II final project III

The final data story

Click here for the Shorthand story

Changes made since Part II

Since completing Part II, the story shifted in a meaningful way. Originally, the analysis leaned on fatality rates (FARS) as the primary risk measure. Through user feedback and deeper exploration of the NAIC data, the focus moved to a more actuarially grounded risk factor — combining accident frequency and claim severity from the NAIC Auto Insurance Database. This change made the relationship between risk and premium far clearer and more defensible.

The fatality rate map was kept as a supplemental visual to show readers why fatalities alone don’t explain premium differences, as serving as a deliberate “plot twist” before introducing the frequency × severity risk factor.

A plain-language annotation explaining how Risk Factor is calculated was added directly to the chart.

Changed the cover page and visual used, and break down information with zoom in, as I think this helps the audience to not to be overwhelmed by the information being displayed.

The audience

The primary audience is people considering relocating across state lines who want to understand whether the insurance costs they’ll face reflect actual road risk.
A secondary audience is policymakers and regulators interested in whether state premium structures align with actuarial risk.

References

References are cited in full on the Shorthand story page. No additional sources were used in this writeup.

AI acknowledgements

Claude AI - Sonnet 4.6, was used to assist with drafting the user research protocol, GitHub writeup wording, chart annotation language, and citation formatting.
All analytical decisions, data sourcing, visualization design, and narrative direction reflects the author’s own work.

Final thoughts

The biggest challenge was resisting the urge to explain too much. Early drafts tried to account for every variable that drives premiums — litigation environments, fraud rates, state laws — which diluted the core finding. Cutting down to frequency × severity as the single risk metric made the story significantly stronger.