← All Work 05 / 06
Data Visualization Spatial Analysis Solo Project

COVID-19 Spatial Data Dashboard

Three interconnected spatial visualizations using Carto and Spatial Markov Chains to predict future COVID-19 hotspots — bridging the gap between raw county-level data and actionable public health insight.

Role
Solo — Data Viz Designer & Researcher
Client
Prof. Chris Alen Sula · Pratt Institute
Duration
6 weeks · Fall 2020
Tools
Carto · OpenRefine · Microsoft Excel
COVID Dashboard Cover
01 — Problem

Two Gaps in Existing COVID Visualizations

The COVID-19 pandemic generated an enormous volume of data visualizations — but a survey of the landscape revealed two consistent and significant gaps across virtually all of them.

Gap 01
No visualization accounted for county population within the display. Using the full county area as a visual proxy is misleading — low-density counties with few cases visually dominate high-density counties with larger outbreaks.
Gap 02
None of the visualizations used predictive analysis to show future trends. Spatial predictive analysis can identify likely hotspots in advance — enabling proactive resource allocation rather than reactive response.

The solution to Gap 01 was to use point size as the variable representing county population — decoupling visual area from data values. The solution to Gap 02 was to apply Carto's Predict Trends and Volatility (PTV) tool, which uses Spatial Markov Chains to calculate probabilistic future states.

02 — Data Work

Cleaning & Preparing the Dataset

The raw data required significant preparation. COVID case data was organized in separate columns, while county latitude and longitude data lived in a separate zip-format table. Merging them required a multi-step process across OpenRefine and Carto.

Data Pipeline

  • Import the county Latitude/Longitude dataset into Carto and export as CSV
  • Import datasets into OpenRefine and merge lat/lon using cross-table functions:
    cell.cross("c_03mr20_1","fips").cells["lon"].value[0]
  • Perform Transpose operation to interchange rows and columns, then export the final dataset

For some of the more complex datasets, I used a combination of OpenRefine and Carto together to achieve the desired merge results — a workflow that revealed some meaningful limitations of each tool used in isolation.

03 — Visualizations

Three Interconnected Spatial Views

Visualization 1 — Total Cases Over Time

The first visualization highlights total COVID cases over time in every US county, with point size varying by county population (Jenks classification, 5 buckets, size range 2–15). A date widget allows filtering to specific time windows, enabling analysis of the virus's propagation across states.

Visualization 1 — Total Cases
Visualization 1 — Click to interact with the live Carto prototype

Visualization 2 — Predictive Analysis

The second visualization uses Carto's PTV (Predict Trends and Volatility) tool with Spatial Markov Chains to identify counties likely to become the next hotspots. Markov Chains calculate transition probabilities — the likelihood of a system moving from one state to another — making it possible to visualize directional trends even when outcomes are probabilistic rather than certain.

Key metrics displayed: trend_up (probability of increasing cases), trend_down (probability of decreasing cases), and volatility (degree of variation over time).

Visualization 2 — Predictive Analysis
Visualization 2 — Predictive analysis using Spatial Markov Chains. Click to interact.

Visualization 3 — Mask Use Prediction

The third visualization predicts case trends based on mask use data from The New York Times. PTV analysis was run on the original dataset filtered by mask use behavior (always vs. never), with interactive widgets letting users see projected trend direction by county based on their masking habits.

I used a consistent color scheme across all three visualizations — color-blind safe and subtle enough not to distract from the data. Since each visualization stands alone, the shared palette builds coherence across the series rather than causing cross-interpretation confusion.

— Design rationale
04 — UX Study

Testing with Real Users

I conducted a brief user study with 2 participants — including one with limited technological experience but deep familiarity with the COVID-19 pandemic. I used the Think Aloud method to capture both functional comprehension and aesthetic reactions simultaneously.

Study Goals

  • Can the user interact with the visualizations without feeling lost?
  • Is the purpose of each visualization and its widgets clear?
  • Can users determine counties with higher infection risk?
  • Are the aesthetics engaging rather than distracting?
Finding 01
Some points were too small and light against the background — low-population counties in pale areas were effectively invisible.
Finding 02
The date widget had an off-by-one behavior — clicking a date selected the following week rather than the intended week.
Finding 03
Terminology like "volatility," "trend_up," and "trend_down" was opaque — users interpreted it based on context clues rather than understanding the underlying concept.
Finding 04
Absence of state borders increased cognitive load — users had significant difficulty locating specific counties without geographic reference lines.

One particularly revealing moment: when a participant asked what "volatility" meant in context, I responded with "what do you think it means?" — yielding a much richer insight into the gap between designer intent and user mental model than a yes/no comprehension check would have produced.

05 — Outcome

Results & Reflections

Both participants found the visualizations more useful than any COVID map they had previously encountered. The point-size-for-population approach was consistently praised — users felt it helped them analyze data far more accurately than choropleth maps that use geographic area as a proxy.

3
Live Carto Prototypes
2
User Study Sessions
5
UX Recommendations
Solo
End-to-end Ownership

Recommendations from Study

  • Use a darker base map to increase contrast for lighter, smaller county points
  • Add state border overlays to all maps — significantly reduces geographic cognitive load
  • Add a small description/legend explaining how to read the map and interpret terminology

Carto is a powerful spatial tool — but the constraints revealed in this project (no text box overlays, fixed widget positioning, projection limitations for Alaska and Hawaii) clarify where purpose-built visualization tools like D3 or Mapbox would offer more design control for production-quality work.

View Final Report →