2.5 Hall-of-Fame / Hall-of-Shame
Effective data visualizations provide insights and facilitate understanding.
The basic principles can guide your visualization design and consumption.
Be creative, but keep your data and your representations honest.
Be mindful of attempts to distort trends and conclusions with flashy visuals.
Data and code should be made available along with the displays.
2.5.1 Misleading Charts
[Fox News, February 20, 2012, mediamatters.org, simplystatistics.org, badgraphs.tumblr.com]
[AAA Gas Prices, found on mediamatters.org]
[Norwich North Electoral District, mediamatters.org, simplystatistics.org, badgraphs.tumblr.com]
Problems: disingenuous, selective and/or incompetent reporting
Solutions: Consistent scales and units of comparison Full time series No cherry picking the data range Cutting off -axis will exaggerate some effects Numbers must add up
Some methods yield visually striking, yet misleading, charts.
Be on the lookout for: tampering with axes and linear scales scaling effects, when representing data points as shapes or volumes cherry-picking by omitting certain data points
For low-dimensional datasets, a tabular display may provide as much information and be less likely to mislead.
Several ways to quantify the misleading level of a chart: Lie factor: ratio of size of the effect shown on the graph by the size of the effect in the data Data density: number of observations by chart area Chartjunk ratio: ratio of area required to convey the data insight by chart area
Typically, the lie factor and chartjunk ratios should be close to 1, while the data density should be “high” (within reason).