Alberto Cairo’s massive open online course on information graphics and data visualization is providing 2,000 data geeks from all over the world with ample inspiration to create interactive dashboards about interesting topics. This week’s challenge is to redesign The Guardian’s DataBlog (of which I am a huge fan) post about US state unemployment.

Here is my submission (posted from 30,000 feet somewhere over New Mexico):

Breaking it down:

  • Title: A little tribute to Hemingway, what can I say?
  • Lead-in: Calling attention to the recent elections, and how state unemployment correlated with votes for Obama. Also inviting the reader to explore other correlations – urbanization and education. GDP could also be added here.
  • Filters & Highlights: This viz uses a drop down to filter the data to one particular region, a map to highlight a particular state (minus Alaska and Hawaii – a definite limitation), and a legend to highlight a region. Clicking on the bubble also highlights the corresponding state in the other views. These interactions are designed to be as intuitive as possible.

The Correlations:

  • Votes for Obama – States with higher jobless rates didn’t punish the Obama administration for their unemployment woes – rather, they voted for Obama in higher than average percentages.
  • Urbanization – the greater the percentage of a state’s population that lives in urban areas, the higher the unemployment rate, in general. It wasn’t immediately clear to me that this would be the case. I have heard of plenty of employment struggles in rural areas of the country. It seems rural areas are below average in terms of unemployment.
  • Education levels – The more a state’s population is educated, the lower the unemployment rates, though these correlations are somewhat weaker. This seems more-or-less self evident and unsurprising.

Some possible improvements to the project:

  1. Correlation Coefficients: I’d prefer to show R-squared values for these scatterplots – something I’d like to see Tableau add to future versions.  Eye-balling the strength of the correlation is one thing. Seeing the exact “goodness-of-fit” is quite another.
  2. Bubble size legend: there really wasn’t a convenient place to put the legend showing population levels associated with bubble sizes, so I left out the legend and added a note to the lead-in paragraph. Not ideal, but it was better than the alternative, in my view – a large legend that shrunk the size of the more important views.

Thanks for stopping by. As always, your comments, questions, suggestions & criticism are more than welcomed!
Ben