Cognition and Spatial Visualizations

First: A Reference: Tog On Interface , Bruce Tognazinni, Chapter 6. $50m R&D by Apple (1989)

What is the task?

What is the visual cortex good at?

How do you structure the visualization?

What is the task?

Software engineering analogy: Use-cases

First part of a specification (hint, hint)
Talk with users!

But be careful -- expert users may not be able to think of solving a task any way except how they've done it before

Distinguish between domain specifics and generic knowledge worker tasks

Build a model of the data

The task

What is the user's objective?

Finding
Organizing
Making a decision

What kind of user is it?

Expert vs novice
Frequent vs sporadic

What resources are available?

User resources (time, attention)
Computer resources (cycles, graphics, audio)

The task: Knowledge crystalization

Foraging -- Finding information

Plowing -- Planning and organization

Farming -- Assembling info in that organization

Weeding -- Problem-solving

Harvest -- Choose key factor

Suppertime! -- Make a decision

Foraging

Gather information sources

Not just search -- May not have boolean exp.

Cf. Google search: "user study keyboard mouse speed"
May need to find and assemble archives
May want specific resources (e.g. articles)

Kinds of tools

Overview
Zoom
Filter
Browse

Plowing

Analogy: Preliminary class hierarchy

Choose or decide on schema

What data is irrelevant?
What kind of data do you have?

Quantitative, ordinal, or nominative?

Tools

Clustering
Classes
Pattern detection
Summaries

Farming

Instantiate a schema

Choose:

Subset of data
Attribute ranking
Visual mapping for each attribute

E.g., position, shape, color (more later)

Visual correlation for each attribute

Alignment: Attributes may be related

Co-incidence: Attributes are related

Orthogonality: Attributes are unrelated

Weeding

Redoing the instantiation

Change subset

Change ranking

Change attributes

Change correlations

Harvesting

Choose and extract the key factors

Goal: Simple trade-off

Example: Space shuttle O-rings

Suppertime!

Visual cortex is specialized

Position, position, position

Cleveland and McGill: empirical verification

Position
Length
Angle/Slope
Area
Volume
Color/Density

Automatic, continuous, unconscious processing

Cf. mapping: conscious, controlled, sporadic

What kind of data?

Quantitative

Numbers, concrete values
Value itself has intrinsic meaning (e.g., price, length, duration)

Ordinal

Absolute value is less important than relative rank
Bigger vs smaller
Watch out for partial orderings

Nominative

No intrinsic value
Abstract label

Ordinal ranking

Position, position, position

Density, saturation, hue, texture

Connection, containment

Length, angle, slope

Area, volume

Shape

Nominal ranking

Position, position, position

Hue, texture

Connection, containment

Density, saturation

Shape

Length

Angle, slope

Area, volume

Guidelines

Use more effective attributes for more important data

Position is uniformly the most effective attribute

Quantitative: Length, angle, slope

Qualitative: Density, saturation

Nominal: Hue, texture, connection, shape

Structuring the visualization: Level of use

Infosphere

Workspace

Tool

Object

Infosphere

Large collections of data; Outside user's normal working environment

E.g., the Web, a data warehouse

Workspace

Set of tools in current use

Multiple visualizations, multiple data sets

E.g., the desktop

Tool

Knowledge tool operating on a data set

"Wide widgets" -- control as well as view

Reveal patterns

E.g., isolines or isosurfaces, graphs

Help search for patterns

E.g., magic lenses

Visual calculations

E.g., Tufte's Challenger diagram

Object

Enhance a given object with visual annotations

Often real-world object

Examples:

Outline view, XML structure view
False-color images
Labels

Structuring the visualization: Purpose of graphical objects

Signs

Controls

Landscapes

Navigation

Pop-ups

Queries

Signs

Graphic object representing data

See hierarchy of perception tasks

Important data -- easily processed sign
Reduction or removal of signs for unimportant data

Tufte: Maximum information in minimal signage

Less ink on page means better visualization
Not always empirically correct (e.g., serifs) but a good start

Controls

User-controlled graphic object (i.e., widget)

Subsetting

Date slider
Magic lenses
Worlds Within Worlds: Feiner and Beshers

Summary

Expected return graph
Isosurface (drag to different level)

Landscapes

Background, context

E.g., axes and labels on a graph
E.g., Fisheye views

Both signs and controllers

And correlations between the two

Visual cues

Especially for all-important position
Also length, angle

More perceptually tied than you might think

Cf. astronauts judging sizes

Navigation

Moving through the landscape

Subsetting

Changing focus

Pop-ups

Interactive detail on data

Wright uses "brushing"
Current terminology also includes mouse-overs

User-directed

Restore elided data

E.g., web browser: URL when mouse-over a link
E.g., "tool tips"
E.g., message display in mailer when selecting subject

Queries

Search the landscape

Find other data

Integrate it

Create the signs

Visually correlate it

Controls and/or signs showing relationship with existing data

For next week

Finish reading Chapter 2 in CMS

Focus on abstract data: trees, networks

And first paper: after President's Day