DAVI Interaction - redo

---

Interaction in Visualization

Why Interaction?

• Interaction transforms static diagrams into tools for data exploration.
• Allows users to:
  • Adjust the visualization: Zoom, pan, filter, and query the data.
  • Issue visual queries: Select interesting clusters, see linked data across multiple views.
  • Enhance the visualization: Tooltips, brushing and linking, and interactivity for better data understanding.
  • Support diverse users and tasks: Interaction helps cater to different goals and user skill levels.

Visual Information Seeking Mantra (Shneiderman)

• Mantra: “Overview first, zoom and filter, then details on demand.”
  • Start with a broad overview of the entire dataset.
  • Zoom in or filter to focus on interesting subsets.
  • Retrieve detailed information (e.g., tooltips, labels) as needed.

Overview & Detail vs. Focus & Context

• Overview and Detail (Separated):
  • Two basic ways:
    1. Spatial Separation: Example: A main view plus a smaller “minimap.”
       • The overview might be a small thumbnail, the detail a large main panel.
    2. Temporal Separation (Zooming): The user zooms into a region. You only see one level at a time.
       • Panning moves your view at the same scale.
       • Zooming changes scale, showing more or less detail.
• Focus and Context (Combined):
  • Integrate both detailed focus and the broader context in a single continuous view.
  • Often achieved through “fisheye” or “lenses” that enlarge a focus region while compressing context around it.
  • Examples:
    • Fisheye Tree Views: Large font for the selected node, diminishing sizes for nodes farther away.
    • Table Lens: Compresses rows/columns not in focus into 1-pixel lines showing only color-encoded values.
    • Interactive Lenses: Over a map or graph, a lens that shows detail or computes something (like density or sampling) in the lens region.

Navigation & Space-Scale Diagrams

• Zooming & Panning:
  • Represented by “space-scale diagrams,” which show how a “viewing window” moves and changes scale.
  • Smooth Navigation:
    • Combines zooming and panning so the user never loses context (origin or destination remain visible).
    • Example: Navigating large graphs by zooming out, showing origin and destination, then zooming back in at the target.

Navigational Cues

• For Panning: Radar views that show off-screen elements at the edges, guiding where to pan.
• For Zooming:
  • Computed “heterogeneity” or “differences” at coarser scale to highlight where something interesting might appear if you zoom in.
  • Example: Using color bands (heterogeneity bands) to show potential hidden patterns in a line chart.

Multiple Coordinated Views (MCV)

• When to use MCV:
  • Rule of Diversity: Many attributes, tasks, or user groups.
  • Rule of Complementarity: Different views show different aspects, providing a richer understanding.
  • Rule of Decomposition: Break complex data into simpler sub-views.
  • Rule of Parsimony: Don’t add views if one suffices.
  • Rule of Space-Time Resource Optimization: Save effort and adapt to multiple tasks/users in one environment.
  • Rule of Self-Evidence: Make relationships between views clear (e.g., highlight linked items in all views).
  • Rule of Consistency: Keep interaction and encoding consistent across views.
  • Rule of Attention Management: Use highlighting, filtering, etc. to guide user’s focus.
• Brushing and Linking:
  • Selecting (brushing) data items in one view highlights the same items in other views.
  • Supports comparisons across multiple dimensions or attributes.

Tangible Interaction and Software Architecture

• (This part was mentioned to be handled next Monday, so only a brief note)
• Tangible Interaction:
  • Physical interaction devices or tokens that directly manipulate the visualization.
• Software Architecture:
  • Interaction often implemented by layering additional pipelines (e.g., lens pipelines parallel to main visualization pipelines).

---

Visualization Critiques and Examples Given

General Steps in Critiques:

1. Identify chart type and data being represented.
2. Consider if the visual encoding matches the data’s nature (expressiveness).
3. Check perceptual aspects: Is it easy to read? Are encodings effective? Is it colorblind safe?
4. Consider efficiency: Are there simpler ways to convey the same information? Is it overly complicated?

Critique Example 1: A Poor Chart of Republican Primaries (2016)

• Shown as a weird column chart with days of the month encoded as bars, and states vs. primaries over time.
• Problems:
  • Day of the month is treated like a quantitative variable (bars) while it’s actually an ID-like attribute. Inappropriate chart choice.
  • Months encoded with different hues, even though months are ordinal (should use ordered channel).
  • Hard to interpret tasks: The chart lacks a sensible purpose or meaningful comparison.
  • Better Approaches:
    • Use a calendar-based layout, a timeline, or well-structured bar/line charts where time is on a continuous axis.
    • Example from Washington Post: Stacked bars or timelines more clearly showing when certain states hold their primaries.

Critique Example 2: History of Pandemics Visualization

• Complex bubble timeline with 3D perspective.
• Each pandemic is a 3D fuzzy ball placed along a timeline, sized by death toll.
• Problems:
  • Perspective distortion introduces a huge lie factor. Bubbles farther away look smaller than they should.
  • Comparing past pandemics to modern ones is distorted by both scale and population changes over time.
  • Color usage is unclear. Uncertain what color encodes.
  • Some dates not aligned properly (some events placed incorrectly in time).
• Better Approaches:
  • Normalizing death toll by world population.
  • Using simple 2D charts: line graphs, bars, or properly scaled bubbles on a single plane.
  • Annotated timelines or interactive visualizations allowing scaling and filtering.

---

Closing Notes

• This lecture wraps up basic static visualization concepts and transitions into interactive visualization techniques.
• Next sessions will look at tangible interaction and software architecture, and then delve into design exercises.
• Students are encouraged to review examples, practice critiques, and be prepared for the upcoming hands-on design activities.

Interactive Quiz on Chart Types and Principles

After the project discussion, the professor gave a quiz to reinforce visualization concepts:

1. Chart Identification & Variables:
   • Example Question: Given a certain chart, identify whether it is a bar chart, column chart, grouped column chart, or histogram.
   • Key Point: A histogram visualizes the distribution of a quantitative variable divided into bins.
     • The bins represent continuous ranges of values, not categories.
     • Histograms ideally have no gaps between bars.
   • Common Mistake to Note: If the x-axis represents categories (like “Age Groups”) but the axis values are actually continuous numeric ranges, that indicates a histogram. If there are gaps or the bins are not equal, it’s a flawed histogram but still conceptually a histogram.
2. Banking to 45 Degrees (Line Charts):
   • Key Point: Line charts are best interpreted when slopes are around 45°. This principle, known as “banking to 45°”, comes from Cleveland & McGill’s perceptual studies.
   • Other listed “principles” like tilting to 35°, sloping to 25°, or binning to 55° are not real. The correct principle is “banking to 45°.”
3. Stacked Area Charts vs. Other Charts:
   • Key Point: Stacked area charts require data that sum up to a meaningful whole.
   • For example, if each layer represents a department’s sales, stacking them gives the company’s total sales over time.
   • Histograms, grouped bar charts, or grouped area charts (the latter doesn’t really exist as a standard term) do not require data that sum to a whole.
4. Connected Scatter Plot:
   • A connected scatter plot takes regular scatter plot points (each with x and y values) and connects them in the order of observation (often time).
   • Distinguishing features:
     • Shows a trajectory or path through time or another sequence.
     • Different from a line chart because both x and y can represent different quantitative attributes (not necessarily time on the x-axis).
5. Overcrowding and Overplotting in Scatterplots:
   • Key Point: For overplotting (points placed exactly on top of each other), use transparency or jittering.
   • For overcrowding (too many points making a “cloud”), use binning (hexbin plots, 2D histograms) or density maps (heatmaps) to show data distribution more clearly.

Quiz Summary:

• Histograms = continuous numeric bins.
• Banking line charts to 45° = optimal readability.
• Stacked area charts = data summation must be meaningful.
• Connected scatter plot = connect points in observed order.
• Overplotting vs. overcrowding = different techniques to handle them.