Feature Interpretation Skills
To decide on the best chart/ graph to
use, it is important to understand your chart viewers
graphical interpretation skills.
(1984) conducted experiments to measure these abilities. He found accuracy skills rank as follows (Robbins):
Position along a common scale
Position along identical, non aligned scales
hue - color saturation - density
How Does This Help Me with My Charting Work?
Cleveland's research shows that the choice of chart/graph
affects the charts viewers ability to interpret the information.
With a pie chart (criticized in several places in this site), we
use angles to compare the data, a relatively poor interpretation
skill for chart viewers. For stacked bar charts, we rely on bar
lengths for comparisons of internal bar segments that do not a
have a common axis, not as effective as position along
nonaligned of common scales.
By understanding your viewers' skills, you will have a better chance
selecting the chart format that they will be able to effectively
Data Visualization Perspective -
What's Wrong With Pie Charts?
The pie chart is a good example of how Cleveland's research
fits into data visualization . Many data visualization writers like
Edward Tufte, Stephen Few,
Naomi Robbins and Howard Wainer do not use Pie Charts. The
US Energy Information
Administration (EIA's) Guidelines for Statistical Graphs, a useful
resource on statistical charting, shares some thoughts on pie charts. Selected excerpts:
Edward Tufte, in The Visual Display of
Quantitative Data, wrote "the only worse design than a pie chart is
several of them."
Howard Wainer of the Educational Testing
Service stated in a 1987 Independent Expert Review of EIA Statistical Graphs
policies that "the use of pie charts is almost never justified" and that
they "ought not to be used." Wainer recommended to EIA that dot charts be
used instead of pie charts in EIA products.
William Eddy of Carnegie-Mellon University,
formerly vice chair of the American Statistical Association (ASA) Committee
on Energy Statistics, said of pie charts at the April 1988 ASA committee
meetings in a session on the EIA Standards Manual, "death to pie charts."
Cleveland's graphic interpretation research
helps to explain the poor quality of pie charts as a communication
are an excellent alternative to pie charts because they show data
position along a common scale rather than rely on pie chart angles.
Charting Multivariate Data
The number of variables that we
are working with affects the types of charts that we need to use. Most
data charting situations can be grouped into 3 conditions:
- Single variable (univariate)
- Two variables (bivariate)
- Three or more variables (multivariate)
Excel's univariate and bivariate charting capabilities are
effective as long as the User avoids chartjunk. Histograms, box
plots, dot plots can effectively summarize univariate data. Simple
bar/column, line/XY scatterplots can effectively summarize bivariate
Excel users pre-made multivariate charting options, however,
Excel panel charts, similar to Cleveland's trellis
display and Tufte's small multiples, present multivariate data
in a form that can more easily interpreted than
Excel's stacked/ clustered bar/ column charts.