Data Visualization Notes

Weber’s Law
Kahneman's Thinking
Stephen Few’s Perception - 3
Gestalt Principles - 6
Tufte’s Principles - 7
Munzner's Rules of Thumb - 6

L01 Introduction

  • Definition of Visualization:

    • Computer-based visualization systems provide visual representations of datasets to enhance task effectiveness.

  • Purpose of Visualization:

    • Augment human capabilities rather than replace decision-making with automated systems.
    • Useful when fully automatic solutions are either unavailable or not trusted.
    • Helps in exploratory analysis, presentation of known results, requirement assessment, and verification of automatic solutions.

  • Role of External Representations:

    • Replace cognitive processing with perceptual processing.
    • Visual representations take advantage of the high-bandwidth human visual system, which processes information in parallel and pre-attentively.
      • High-bandwidth channel to the brain, providing an overview with background processing.
      • Vision supports simultaneous perception of data.
      • Sound, touch, taste, and smell have lower bandwidth and less effective record/replay capacities compared to vision.

  • Importance of Representing All Data:

    • Summaries may obscure important details.
    • Visualizations help confirm expected patterns, discover unexpected ones, and assess statistical models.

  • Resource Limitations:

    • Computational: Time and system memory constraints.
    • Display: Limited pixels and the need to balance information density with visual clutter.
    • Human: Limitations in time, memory, and attention.

  • Why Analyze Visualization Designs:

    • Provides structure to a vast design space.
    • Helps systematically evaluate choices and design new solutions.
    • Analyzing existing visualizations can inform new design approaches.

L02 Nested Model

Here’s the three-part framework for visualization design:

  1. What:
    • Identify the data to be visualized.
    • Determine data types (e.g., categorical, quantitative) and attributes.
    • Example: For a sales dashboard, specify metrics like sales volume, revenue, and profit.
  2. Why:
    • Define the purpose of the visualization.
    • Understand the goals and needs of the end-users.
    • Example: If tracking sales performance, reasons might include monitoring team effectiveness or evaluating new product success.
  3. How:
    • Decide on design methods and choices for visualization.
    • Choose encoding techniques, interactivity, and layout.
    • Example: Use a bar chart for comparing sales volumes or a line graph for revenue trends.

  • Analysis Framework: Four Levels

    • Domain Situation
      • Who are the target users?
    • Abstraction
      • Data Abstraction: What is shown?
      • Task Abstraction: Why is the user looking at it?
    • Idiom
      • Visual Encoding Idiom: How to draw?
      • Interaction Idiom: How to manipulate?
    • Algorithm
      • Efficient computation.

  • Nested Model

    • Downstream: Cascading effects.
    • Upstream: Iterative refinement.

  • Validation Challenges

    • Different ways to get it wrong at each level.
    • Solution: Use methods from different fields at each level.

  • Avoiding Mismatches

    • Computational benchmarks may not confirm idiom design.
    • Lab studies may not confirm task abstraction.

L03 Data Abstraction

  • Data Types and Meaning
    • Data types include items (individual entities), attributes (properties measured), links (relationships between items), positions (spatial data), and grids (sampling strategies).
      • Items are discrete entities like patients or cars; attributes are measured properties like height or horsepower.
      • Links express relationships; positions denote spatial locations; grids are used for sampling continuous data.
  • Dataset Types

    • Flat tables organize data with one item per row and attributes as columns.

    • Multidimensional tables index data by multiple keys (e.g., genes, patients).

    • Networks/graphs connect nodes with links; trees are a special case without cycles.

      • Node-link diagrams

        –Force-directed Placement (Links = springs pull together, Nodes = magnets repulse apart)

        –Circular layouts

        –Arc diagrams

      • Adjacency matrix representation

      • Enclosure (specific to trees)

        –Treemaps

        –Sunburst

        –Icicle plot

    • Spatial

      • Fields (continuous) - Represent continuous attribute values associated with cells in a grid.

        • major concerns – sampling: where attributes are measured – interpolation: how to model attributes elsewhere – grid types
        • major divisions – attributes per cell: scalar (1), vector (2), tensor (many)

      • Geometry (spatial) - Represents the shape and spatial position of items.

        • Choropleth maps
        • Symbol maps

        • Cartograms – Continuous cartograms

          – Grid cartograms

        • Dot density maps

    • Collections

      • sets
      • lists
      • clusters
  • Data Abstraction
    • Involves translating domain-specific language into a generic visualization format.
    • Identifies dataset types, attribute types, and cardinality (number of items and attributes).
    • Considers data transformation based on task requirements.
  • Attribute Types
    • Categorical (nominal) for equality comparisons.
    • Ordered (ordinal) for meaningful order comparisons.
    • Ordered (quantitative) for measurable magnitude and arithmetic operations.
  • Data vs Conceptual Models
    • Data models are mathematical abstractions; conceptual models are mental constructions supporting reasoning. – 32.52, 54.06, -14.35, ...; temperature.
    • Data abstraction processes rely on conceptual models for data transformation.
  • Derived Attributes
    • Computed from original data through simple changes, additional data acquisition, or complex transformations.

L03 Task Abstraction

  • Identify tasks users need or perform.

  • Find or transform data types to support these tasks.

  • Task Abstraction: Actions and Targets:

    • Action Types:
      • Analyze: Consume (discover vs. present), Enjoy (newcomer vs. casual), Produce (annotate, record, derive).
      • Search: Lookup (e.g., dictionary), Locate (e.g., keys), Browse (e.g., bookstore), Explore (e.g., new city).
      • Query: Determine how much data matters (one, some, all).
    • Targets: What is being acted on.

L05 Marks & Channels

  • Visual Encoding:

    • Analyze idiom structure through marks and channels.
    • Marks: Represent items or links (e.g., points, lines, areas).
    • Channels: Control the appearance of marks based on attributes.

  • Marks for Items:

    • Basic geometric elements: 0D (point), 1D (line), 2D (area).
    • 3D marks (volume) are rarely used.

  • Marks for Links:

    • Links can be represented using lines or areas.

  • Channels:

    • Control the appearance of marks (e.g., size, color).
    • Channel properties differ in the amount and type of information conveyed.
    • Types of Channels:
      • Position: Vertical, horizontal.
      • Color Hue: Represents categorical differences.
      • Size (Area): Encodes quantitative information.

  • Redundant Encoding:

    • Using multiple channels can strengthen the message but consumes more channels.

  • Marks as Constraints:

    • Geometric primitives (points, lines, areas) impose constraints on how data can be encoded.

      – points: 0 constraints on size, can encode more attributes w/ size & shape – lines: 1 constraint on size (length), can still size code other way (width) – interlocking areas: 2 constraints on size (length/width), cannot size or shape code

  • Channel Effectiveness:

    • Accuracy: Precision in differentiating encoded items. (length is accurate)
    • Discriminability: Number of unique steps perceived.
    • Separability: Ability to use a channel without interference from others.
    • Popout: The ability for items to stand out visually.

  • Factors Affecting Accuracy:

    • Alignment, distractors, distance, common scale.

  • Relative vs. Absolute Judgements:

    • Perceptual system relies on relative judgments rather than absolute.
    • Accuracy improves with common frame/scale and alignment.
    • Weber’s Law:
      • Ratio of increment to background is constant, affecting how differences are perceived.
    • Relative Luminance and Color Judgements:
      • Luminance perception is contextual based on contrast.
      • Color constancy maintains perception across varying illumination conditions.

L06 Visual Thinking Process (Perception, Cognition, Attention)

Kahneman's Fast and Slow Thinking

  1. System 1:
    • Fast, unconscious, automatic
    • No self-awareness or control
    • 98% of our thinking
    • Intuitive, automatic functions like reading faces
  2. System 2:
    • Slow, conscious, deliberate
    • With self-awareness and control
    • 2% of our thinking
    • Analytical, logical functions like solving math problems

Perception Insights

  1. Stephen Few’s Visual Perception:
    • Selective; sensitive to contrast and change.
    • Drawn to familiar objects.
    • Limited short-term visual memory.
  2. Pre-attentive Perception:
    • Immediate, primal reaction.
  3. Post-attentive Perception:
    • Slower, conscious activity.

Gestalt Principles for Data Visualization

  1. Similarity:
    • Elements with shared visual properties are considered in the same group.
  2. Proximity:
    • Elements close to each other are grouped.
  3. Enclosure:
    • A visual element surrounds related elements.
  4. Common Fate:
    • Shapes moving in the same direction are grouped.
  5. Parallelism:
    • Lines with similar slopes are grouped.
  6. Connectedness:
    • Elements that are visually connected are grouped.

L09 Interactive Views

Complexity Handling in Data Visualization

  • Manipulate Views
    • Change Over Time: Adjust encoding, parameters, arrangement, and aggregation.
      • Re-encode
      • Change Parameters: Use widgets such as sliders, buttons, and checkboxes
      • Change Order/Arrangement: Reorder data to find extreme values or correlations
      • Change Alignment: Align bars for flexible comparison
      • Animated Transitions: Smooth transitions between states to aid in tracking
    • Selection: Basic operation with choices for click/tap vs. hover, multiple click types, and interaction semantics.
    • Highlighting: Change visual encoding to provide feedback on selection; use channels like color, size, or motion.
    • Navigate
      • Scrollytelling: Navigating by scrolling; familiar but can lack affordances and direct access.
      • Change Viewpoint/Visibility: Pan, zoom, rotate, and slice; adjust to show specific items or attributes.
      • Unconstrained vs. Constrained Navigation: Unconstrained is easier to implement but harder for users; constrained uses animations and computed trajectories for better control.

Interaction Benefits & Limitations

  • Benefits: Flexible, powerful, intuitive; supports exploratory data analysis and fluid task switching.
  • Limitations: Time cost, cognitive load, screen space, and potential for unplanned user interactions.
  • Multiple Views/ Facet

    • Partition data into views to compare; use linked highlighting for coordination.

    • Juxtapose

      • Overview-Detail Views/ Navigation: Display detailed and overview information with bidirectional or unidirectional linking.
        • Google Maps
      • Tooltips: Provide additional details on hover or click but do not support overview.
      • Small Multiples: Display multiple similar charts to compare data across attributes or time.

    • Partitioning & Recursive Subdivision

      • Partitioning: Split data by attributes or regions.
      • Recursive Subdivision: Divide data hierarchically, with variations in order and encoding to reveal patterns.

    • Superimpose Layers

      • Layer objects within the same view; use color and design choices to distinguish layers.
      • Static Visual Layering and Dynamic Visual Layering

L11 Principles of Effective Information Visualization

Tufte’s Principles

  1. Show Data
    • Focus on data itself, avoid unnecessary embellishments.
    • Examples: Clean charts, appropriate scales, minimal decoration.
  2. Maximize Data-Ink Ratio
    • Use ink only to represent data.
    • Remove non-data ink and redundant elements.
    • Formula: Data Ink Ratio = Data Ink / Total Ink Used
    • Mantra: Show the data, maximize data-ink ratio, erase non-data ink.
  3. Use Effective Data Density
    • π·π‘Žπ‘‘π‘Ž 𝐷𝑒𝑛𝑠𝑖𝑑𝑦 = π‘π‘’π‘šπ‘π‘’π‘Ÿ π‘œπ‘“ π‘’π‘›π‘‘π‘Ÿπ‘–π‘’π‘  𝑖𝑛 π‘‘π‘Žπ‘‘π‘Ž π‘šπ‘Žπ‘‘π‘Ÿπ‘–π‘₯ / π΄π‘Ÿπ‘’π‘Ž π‘œπ‘“ π‘‘π‘Žπ‘‘π‘Ž π‘”π‘Ÿπ‘Žπ‘β„Žπ‘–c
    • High data density displays more information efficiently.
    • Examples: Small multiples, sparklines.
    • Balance data density with clarity.
  4. Provide Context and Comparisons
    • Add context and benchmarks for better understanding.
    • Examples: Historical trends, comparative data.
  5. Ensure Integrity and Accuracy
    • Represent data faithfully; avoid distortion.
    • Lie Factor: Measures distortion (Size of effect shown in graphic / Size of effect in data).
    • 𝑆𝑖𝑧𝑒 π‘œπ‘“ E𝑓𝑓𝑒𝑐𝑑 = |π‘†π‘’π‘π‘œπ‘›π‘‘ π‘£π‘Žπ‘™π‘’π‘’ −πΉπ‘–π‘Ÿπ‘ π‘‘ π‘£π‘Žπ‘™π‘’π‘’| / πΉπ‘–π‘Ÿπ‘ π‘‘ π‘£π‘Žπ‘™π‘’e
  6. Encourage Exploration and Interactivity
    • Use interactive elements to allow deeper insights.
    • Examples: Interactive dashboards, zoomable charts.
  7. Design for Universal Accessibility
    • Make visualizations accessible to all, including those with disabilities.
    • Examples: Alternative text, color-blind friendly schemes.

Munzner's Rules of Thumb

  • 3D vs 2D: Use 2D unless 3D has clear justification.
    • Reasons: perspective distortion, occlusion, text legibility
  • Eyes vs Memory: Use side-by-side views for comparison.
  • Immersion vs Resolution: Resolution is more important than immersion for abstract data.
  • Overview First, Zoom and Filter, Details on Demand: Start with an overview, then zoom in for details.
  • Responsiveness is Required: Ensure visualizations respond quickly.
  • Function First, Form Next: Prioritize functionality over aesthetics; aesthetics can be refined later.

Comments

Post a Comment

Popular posts from this blog

Human Computer Interaction Notes

Image
Donald Norman’s model - 7 Abowd and Beale framework - 4 Nielsen’s (1994) - 10 Shnedierman and Plaisant’s (2010) - 8 Norman’s 7 Principles - 7 Universal design principles - 7 Lecture 01 What is HCI? An Academic Discipline : Studying people interacting with (computer) technology. A Design Discipline : Designing interventions for systems involving people and technology. Design - Three 'Use' Words Useful : Functional, performs tasks. Usable : Easy to use, performs the right tasks, enjoyable user experience. Used : Attractive, available, acceptable to the organization. Lecture 03 Principles and Foundations of HCI HCI combines three main research activities: Creating usability methods and Capturing user experience. Developing new systems, tools, and techniques. Modeling and understanding human behavior. Influence of Experimental Psychology Models human neuromuscular motor control and perceptual abilities. Focus on controlled experiment paradigm: Indepe...
Read more