Transcript 2-3-6-7 Example

Visual Analytics and the Geometry of Thought—
Spatial Intelligence through Sapient Interfaces
Alexander Klippel & Frank Hardisty
Department of Geography, GeoVISTA Center
&
e-Dutton Institute for Education
Penn State
Star Plots
How Shape Characteristics Influence
Classification Tasks
Alexander Klippel & Frank Hardisty
Department of Geography, GeoVISTA Center
&
e-Dutton Institute for Education
Penn State
Overview
 Multivariate data displays
 Experiment on the influence of shape
(of star plots) on the classification of data
 Design of a tool to administer grouping
experiments
 Design of a tool to analyze individual similarity
ratings
 Does shape matter?
 Conclusion and future work
Displaying Multivariate Data
 We encounter limitations in displaying multivariate
data in two dimensions
 As a response to these constraints several graphic
designs have been advised, for example





Andrews curves
Parallel plots
Chernoff faces
Star plots
Etc etc.
 The big question is
 Which visualization technique does “work” for which data
sets and which does not
Parallel Coordinate Plot
Chernoff Faces
Source: http://mapmaker.rutgers.edu/355/links.html
www.ncgia.ucsb.edu
www.ghastlyfop.com
Star Plots
Star Plots
Star Plots
GeoViz Toolkit: http://www.geovista.psu.edu/grants/cdcesda/software/
Question
 In their work on Chernoff faces Chernoff and
Rizvi (1975) found that varying the assignment
of variables to facial characteristics has an
influence on classification tasks
 Question
 For star plots the assumption is made that the
assignment of variables to rays does not matter,
but is that really the case?
Experiment: Car Data
1-3-5-7
2-3-6-7
Price
1
Interior space
Emissions
Weight
2
8
3 Safety rating (higher is better)
7
6
Acceleration
(higher is faster)
4
Maximum speed
5
MPG
Miles per Gallon
20 participants in each condition
Penn State undergraduates
Weight
1
Interior space
Emissions
2
8
3 Safety rating (higher is better)
7
6
MPG
Miles per Gallon
Price
4 Maximum speed
5
Acceleration
(higher is faster)
10085
6550
3015
10085
The Grouping Tool
1-3-5-7
81 icons
(4 variables, 3 levels
(high, medium, low))
The Grouping Tool
2-3-6-7
81 icons
(4 variables, 3 levels
(high, medium, low))
Example: All Low Values
=
1-3-5-7
2-3-6-7
Data




Number of groups
Time to complete
Similarity matrix
Linguistic labels
Some Results
 There is no statistically significant difference
in the number of groups created in 1-3-5-7
and 2-3-6-7 (t = .241, df = 38, p = .811)
 There is no statistical significant difference in
the time participants needed to complete the
task (t = -1.533, df = 38, p = .134)
 The similarity values in both similarity
matrices are correlated and the correlation is
statistically significant (r = .581, N = 3240, p < .0005)
Cluster Analysis
1-3-5-7
2-3-6-7
Ward’s method
MDS Plots
1-3-5-7
MDS Plots
2-3-6-7
Grouping Analysis
1-3-5-7
2-3-6-7
Improvise by Chris Weaver (http://www.personal.psu.edu/cew15/improvise/index.html)
Price
1
Interior space
Emissions
Weight
1
Weight
2
8
3 Safety rating (higher is better)
7
6
Acceleration
(higher is faster)
4
5
MPG
Miles per Gallon
1-3-5-7
Maximum speed
Interior space
Emissions
2
8
3 Safety rating (higher is better)
7
6
MPG
Miles per Gallon
Price
4 Maximum speed
5
Acceleration
(higher is faster)
2-3-6-7
1-3-5-7
2-3-6-7
Price
1
Interior space
Emissions
Weight
1
Weight
2
8
3 Safety rating (higher is better)
7
6
Acceleration
(higher is faster)
4
5
MPG
Miles per Gallon
1-3-5-7
Maximum speed
Interior space
Emissions
2
8
3 Safety rating (higher is better)
7
6
MPG
Miles per Gallon
Price
4 Maximum speed
5
Acceleration
(higher is faster)
2-3-6-7
1-3-5-7
2-3-6-7
Price
1
Interior space
Emissions
Weight
1
Weight
2
8
3 Safety rating (higher is better)
7
6
Acceleration
(higher is faster)
4
5
MPG
Miles per Gallon
1-3-5-7
Maximum speed
Interior space
Emissions
2
8
3 Safety rating (higher is better)
7
6
MPG
Miles per Gallon
Price
4 Maximum speed
5
Acceleration
(higher is faster)
2-3-6-7
1-3-5-7
2-3-6-7
Price
1
Interior space
Emissions
Weight
1
Weight
2
8
3 Safety rating (higher is better)
7
6
Acceleration
(higher is faster)
4
5
MPG
Miles per Gallon
1-3-5-7
Maximum speed
Interior space
Emissions
2
8
3 Safety rating (higher is better)
7
6
MPG
Miles per Gallon
Price
4 Maximum speed
5
Acceleration
(higher is faster)
2-3-6-7
1-3-5-7
2-3-6-7
Conclusion
 Shape does matter
 The assignment of variable to rays in a star plot influences
classification tasks (compare Chernoff faces)
 Characteristic shape features have an influence on rating
the similarity of the represented data
 The more characteristic the shape, the greater the influence
 It may therefore be that star plots are less suitable
for lay person exploratory analysis but more
effective in communication (if carefully chosen).
Outlook
 Quantifying data analysis
 Cluster validation methods
 E.g., Rand statistic, Jaccard coefficient
 Individual analysis of “shape families”
 Relation to linguistic labels
 Continue work on how should variables be assigned to
rays
 For example, is there a time advantage for salient shapes?
 Influence of contextual parameters
 Of a star plot as such (e.g. number of variables/rays)
 As a symbol in a map (e.g. spatial patterns, and first law or
geography).
 Star plots in comparison to other visualization
techniques
Thank you