A1-MarkHowison
From CS294-10 Visualization Fa07
[edit] Good Visualization
This graphic, from an article in the Oakland Tribune, shows 2006 poverty rates by state. The use of a data map makes salient geographical patterns in the data, such as the concentration of high poverty rates in the deep south, that would be more difficult to show using a table or bar chart. The one shortcoming is that it is difficult to infer the exact percentage for a given state, since the shades of red cover a range of percentages. The use of color, however, helps, since it is easy to distinguish the state boundaries. On an adjacent page in the newspaper, there was a similar but grayscale data map of child obesity rates by state, and the black state boundaries were hard to see in a cluster of dark gray states.
Deconstruction
Spatial position encodes a nominal variable, the state.
Color encodes a quantitative variable, the poverty rate.
Note that nothing encodes for the total population of a state. The size of the state encodes land mass, but not population.
[edit] Bad Visualization
This graphic comes from an educational psychology article ("A Time for Telling"; Schwartz and Bransford, 1999) that compares learning gains for students completing a summarization task vs. an analysis task prior to attending a psychology lecture on schema and encoding. The experiment uses a typical 2x2 factorial design with summarize/analyze and schema/encoding as the factors. Students were given a prediction task to test for conceptual understanding (a higher score indicates better understanding).
Deconstruction
Spatial: The x-axis encodes a nominal variable, the concept type. The y-axis encodes a quantitative variable, the average of students' scores on a prediction task for that concept type.
The shading of the data points encodes a nominal variable, the condition type.
The lines through the data points encode the range of prediction task scores for that condition.
The lines connecting data points link the condition types (which is already indicated by the shading of the data point).
Redesign
The problem with the graph is that the lines connecting the data points seem to imply it is a time series, when in reality the "Concept Type" variable is nominal, not ordinal or quantitative. Also, the key at the top contains redundant information, since the x-axis already distinguishes between schema and encoding. It would be clearer with the lines removed and the data points colored as either analyze or summarize, so that the higher two values were both black for analyze and the lower two values both white for summarize, like this:
This would more boldly highlight the authors' results that the analyze-lecture condition led to better conceptual understanding. There is a trade-off, though, since it now looks like there are four different groups of students, when in reality there were only two: encode-summarize/schema-analyze and encode-analyze/schema-summarize (as the connecting lines try to indicate in the graph above).
