Unformatted text preview: Eﬀective graphical displays
Jean-luc Doumont What? So what? Get your audience to
pay attention to, Information understand, Message Interpretation To optimize communication… (be able to) act upon
a maximum of messages, given constraints First law Adapt to your audience Second law Maximize the signal-to-noise ratio Third law Use eﬀective redundancy Planning the graph Patient Gender Weight Conc. [kg] Select the type of graph on the basis of
the intended message or research question
(comparison, distribution, correlation, evolution) [ng/ml] 1 female 75 31.3 2 male 81 16.5 3 female 52 11.8 1.00 © 2010 by Principiae the structure of the data set
(number and nature of variables) Discrete
variables Designing the graph One continuous variable Two or more Comparison
among data Distribution
of a variable Correlation
among variables Evolution
of a variable Select the basic design on the basis of the intended message
or research question and the number of continuous variables.
To render discrete variables (that is, when you are comparing
comparisons, distributions, correlations, or evolutions of data
among subsets), distinguish the subsets within one panel
or display them in as many juxtaposed panels (or combine
these two approaches to represent several discrete variables). Subsets displayed in juxtaposed panels
(with identical scales) t0 Faster diﬀusion 1.00 © 2010 by Principiae t1 t2 Subsets distinguished
within one panel Faster drift To compare data, consider a length representation Showing the entire data set, as points along a scale, (horizontal bars, starting necessarily from zero) or, is the most accurate way to convey its distribution. for closely grouped data, a position representation For large data sets, it may be useful to summarize (dots along a scale that need not start from zero). the distribution with a histogram or with a box plot, Both of these aﬀord more accuracy than a pie chart. showing ﬁve percentiles plus the outliers. Box plots
allow an easy comparison between subsets of data,
each summarized by one box, along the same scale. Population [millions] Germany 82.2 France 60.5 UK 58.8 Netherlands 15.9 Belgium 10.2 0 10 20 Life expectancy at birth [years]
10 25 50 75 90% France
UK Subset 1 Belgium Subset 2 78 79 Scatter plots are a powerful way to reveal correlation The evolution of one or more dependent variables or simply to explore bivariate data by mapping it out. versus an independent one is best shown by lines. For more than two variables, they can be combined Variables expressed in diﬀerent units must be drawn in arrays (one scatter plot for each pair of variables). in diﬀerent panels, with a common horizontal scale. Voltage Unemployment rate [%] 10 VL [V]
20 Spain Vtot 0 VR
VC Greece −10
Italy 15 0 50 100 µs 0 50 100 s France
Finland Females European Union 10
Belgium Current 100 [µA]
0 Germany −100
Portugal 5 Netherlands Sweden
Ireland 1.00 © 2010 by Principiae Lux. Voltage 10 [V]
5 0 0
Males 10 Constructing the graph 1.0 A poor graph measured Output power [W] 0.8 calculated The graph exhibits a very low signal-to-noise ratio,
with excessive tick marks and uncalled-for grid lines, 0.6 and comparatively little ink to represent the data.
0.4 The graph is not intuitive, for the separate legend
0.2 (a key to the symbols) prevents global processing.
In a sense, it is a graph to read, not a graph to view. 0.0
15 16 17 18 19 20 Frequency [GHz] Output power 0.8 A good graph [W] measured
0.6 The graph is plainer and therefore better contrasted: calculated the background no longer interferes with the data,
0.4 yet it provides suﬃcient information about them. 0.2 The graph is more intuitive: the labels, positioned
next to the data, provide the required clariﬁcation
where it is needed (when viewers look at the data). 0.0
15 16 17 18 19 20 Frequency [GHz] A better graph Output power measured 650 mW The graph shows the data and nothing but the data: calculated tick marks are relevant, not arbitrarily equidistant;
nondata lines are gray, to make the data prominent.
630 MHz The graph readily answers questions about the peak
(position, height, and full width at half maximum)
and about the range over which data were acquired. 0 16 17.2 GHz 19 Adapted from Jean-luc Doumont, Trees, maps, and theorems
(Principiae, 2009). © 2009 by Principiae. All rights reserved.
Can be downloaded from www.treesmapsandtheorems.com. ...
View Full Document
This note was uploaded on 02/24/2012 for the course MECHANICAL 2.019 taught by Professor Yumingliu during the Spring '11 term at MIT.
- Spring '11