How does the proportional distribution of fatalities from major natural disasters (>1,000 deaths) shift across different historical eras, and which specific types of disasters are responsible for the vast majority of the human toll within isolated decades?
I chose a Coordinated Multiple View (CMV) layout to explore this relationship. The top chart is a timeline scatterplot encoding Year against Deaths on a logarithmic scale to handle extreme historical outliers. I noticed the initial chart suffered from heavy overplotting, so I applied a strict casualty threshold (filtering out events with fewer than 1,000 deaths) and added white strokes to the marks to clearly define individual events. I also removed the aggregate "All Natural Disasters" category to ensure clear comparative analysis.
For the bottom chart, I used a normalized stacked bar chart (a 100% proportion chart) rather than a standard vertical bar chart because it effectively displays the two variables while saving vertical space. The Tableau10 color palette was chosen to consistently encode the disaster category across both views.
This visualization was implemented using Vega-Lite. To address my question, I used the disasters.csv dataset, sourced dynamically from the official Vega-Datasets repository. The cross-filtering interaction was adapted from the Vega-Lite documentation for interactive brushing, and the proportional view utilizes Vega-Lite's stack: "normalize" transform.
I worked solo on this assignment. The most challenging aspect was managing the state between the two views and finding a clean way to represent the aggregates without visual clutter. This design deliberately obscures certain aspects of the data; by implementing a casualty threshold (>1,000 deaths), the impacts of thousands of smaller, localized events are hidden. However, I felt any noticeable macro-level observations would be much more visible by isolating the significant events.
Despite these tradeoffs, the brushing and linking interaction successfully communicates its central finding: allowing the user to dynamically calculate proportional compositions (e.g., comparing the proportion of earthquake deaths in the 1920s versus the 2000s) provides a much higher degree of analytical utility than looking at static, century-level aggregates. I also thought it was interesting to see specific spikes in decades like the 1920s, and wonder how much of that is due to reporting infrastructure versus actual ecological shifts.