DAVI Basic Charts - redo

Lecture Notes on Basic Chart Types and Visualization Principles

Date: (From the provided transcript)
Topic: Basic Chart Types (Scatterplots, Line Charts, Area Charts, Bar Charts, Histograms, and Radial Variants)
Context: The professor discusses fundamental design considerations for various common visualization types. The lecture also includes a quiz on color scales and data types, and emphasizes careful selection of charts, proper scaling, aspect ratios, labeling, and ensuring that the representation of data is accurate and expressive.

---

Introduction to Basic Chart Types

Why focus on basic charts?

• Common and well-known: Bar charts, line charts, scatterplots, etc., form 90% of use cases.
• Widely understood and easily created in standard tools.
• These charts solve most visualization problems.
• Basic charts serve as building blocks for more complex visuals.

Charts Covered:

• Scatterplots
• Line charts (including step and area variants)
• Bar charts
• Histograms
• Radial variants (briefly)

Concept: These design principles (scaling, aspect ratio, ordering) apply not just to basic charts but to any chart type you may later consider.

---

Scatterplots

Key Advantages:

• Extremely versatile: show correlations, trends, outliers, clusters, nonlinear relationships.
• Position is a strong channel, making scatterplots very perceptually effective.
• Show multiple patterns at once.

When to Use a Scatterplot:

• For two quantitative variables.
• If discrete attributes (like number of cylinders in cars) create vertical lines, a scatterplot might not be best. Consider boxplots or alternative representations.

Key Design Decisions:

1. Scaling and Zoom:
   • Zoom level affects perceived correlation strength. Too zoomed out = stronger apparent trend; too zoomed in = less strong trend.
2. Aspect Ratio:
   • No single perfect ratio. Some research suggests using algorithms or heuristics to find an optimal aspect ratio for best pattern visibility.
3. Tick Marks and Gridlines:
   • Placing tick marks is non-trivial. Various algorithms (Heckbert, R-Pretty, Wilkinson) exist to find “nice” tick spacing.
   • Gridlines: Use only if exact reading of values is necessary. Otherwise, gridlines add clutter.
4. Helpers:
   • Quadrants / Average Lines: Adding reference lines (mean lines) helps chunk data, aiding interpretation.
   • Trend Lines: Add only if a trend is relevant. Make them distinct (dashed, lighter) from the data.
5. Dot Shape/Size:
   • Dots can be circles, squares, etc. For sensor data with uncertainty, squares indicating a confidence interval region can help communicate measurement accuracy.
6. Overplotting and Overcrowding:
   • Overplotting: Many points on top of each other.
     Solutions: Transparency, jittering (small random offsets), or combining both.
   • Overcrowding: Too many points everywhere.
     Solutions: Binning (hexbin or square bins), heatmaps, splitter plots (density contours plus outliers as points).

Spatial (Map) Analogy:

• Dot maps can be treated similarly: binning or heatmaps may require normalization by population density to avoid misleading patterns.

Conclusion on Scatterplots:

• They are a “multi-tool” of visualization.
• Some researchers argue scatterplots could handle most tasks if designed properly.

---

Line Charts

Typical Use Case:

• Often used for time series data.
• Show trends over time (or over a continuous variable like input size in benchmarking).

Does It Make Sense?

• Don’t use line charts for categorical data (e.g., gold, silver, bronze) that can be reordered arbitrarily.

Design Considerations:

1. Scaling the Y-Axis:
   • Show full range to avoid misleading impressions.
2. Aspect Ratio:
   • Cleveland & Cleveland (1980s research) suggest aiming for a ~45° slope for the main trend line for good readability.
3. Gridlines:
   • Same rule as scatterplots: only if exact reading is necessary.
4. Trend Lines & Interpolation:
   • Trend lines are helpful to show general direction but distinguish them from actual data.
   • Consider step lines if data represent discrete intervals.
5. Multiple Lines:
   • If many lines overlap (spaghetti plot), draw highly fluctuating lines first, smoother lines last, to reduce occlusion.

Noisy Data:

• Consider smoothing noisy line data. Show original data as faint points or as an envelope (min-max range) around the smoothed line.
• Decide between line or scatter: If very noisy, scatter may be better.

Connected Scatterplots:

• Combine two variables over time and connect points in chronological order, showing a trajectory through attribute space.

---

Area Charts

Definition:

• A line chart with the area under the line filled in.
• Often used for amounts that accumulate.

Stacked Area Charts:

• Each layer represents a portion of the total. The top line (silhouette) shows the total sum at any point in time.
• Only stack if layers add up to something meaningful.

Ordering Layers:

• Order layers to be as horizontal as possible, avoiding the “sine-wave” illusion.
• Incorrect order can create optical illusions and misinterpretations.

Examples:

• Choosing layer order is computationally non-trivial but essential for clarity.

---

Bar Charts (Column Charts)

Column vs. Bar:

• Vertical bars: column charts. Horizontal bars: bar charts.
• Bar charts are great for categorical comparisons.

Axis Considerations:

• Always show a zero baseline to avoid misleading.
• If data vary in orders of magnitude, consider perspective charts or other techniques instead of breaking the axis.

Aspect Ratio:

• A bar should be at least 10 times as tall as it is wide for clarity.
• White space between bars: ~50% of bar width is a good starting point.

Ordering Categories:

• Alphabetical if the task is identification (find a known category).
• Sort by value if task is comparison or finding max/min easily.

Bar Variants:

• Lollipop charts (a more elegant bar variant).
• Grouped and stacked bars for comparisons and sub-categories. Only stack if it represents meaningful parts of a whole.

---

Histograms vs. Bar Charts

Histogram:

• Quantitative data divided into equal-sized bins. No gaps between bins.
• Shows distribution of data across intervals.
• The average would appear as a vertical line.

Bar Chart (Categorical):

• Categories on x-axis, gaps between bars.
• The average of categories is a horizontal line.

Bin Size and Starting Point:

• Different bin widths or offsets drastically change how the histogram looks.
• Always use equal-sized bins, or properly normalize if bin sizes differ.

---

Radial Variants

When Useful:

• If data is cyclic (e.g., time of day, direction), radial plots might be natural.

Examples:

• Radial scatterplots (polar coordinates).
• Radial line charts (e.g., plotting daily patterns in a circle).
• Radar charts (area charts on radial axes).
• Radial bar charts (like coxcombs or specialized charts that bend bars into arcs).

Partial Circles:

• You can use just a quarter circle (e.g., a radial bar chart) to emphasize certain aspects.

---

Visualization Critiques (Examples)

Practice in Critiquing:

• Evaluate based on expressiveness (show the data truthfully), effectiveness (using the best perceptual channels), and efficiency (effort vs. insight).
• Examples reviewed in class:
  • Gun deaths in Florida (misused an inverted area chart that caused confusion).
  • Georgia Covid-19 data (bars sorted by value rather than time, misleading trend inference).

Key Takeaways from Critiques:

• Don’t reorder time data. Always keep chronological order.
• Normalize data when comparing populations with different sizes.
• Using negative mapping or metaphorical visuals can be okay if done carefully (as in New York Times “Iraq deaths” chart).

---

Next Lecture

• Next week’s topic: Interaction in Visualization.
• Reading assignment: About 15 pages from Chapter 4 of the recommended book, focusing on interaction principles.

---

Additional Resources

• The professor mentioned literature and references for algorithms (Wilkinson’s, Heckbert’s methods) for tick placement.
• Andy Kirk’s and other experts’ commentaries on specific charts were referenced.

---

End of Lecture Notes

Initial Remarks and Quizzes

• Lecture Start:
  • The professor welcomes everyone and commends those who made it through the rain.
  • The main topic: Basic Chart Types.
• Quick Quiz Before Lecture Topic:
  Students are given a code (Menti code) to join an online quiz.

Quiz Questions Recap

1. Color Scales:
   • Question: Which one is not a property of a good color scale?
     • Options: Perceptual Equidistance, Increasing Luminance, Contrast Convergence, Colorblind Safety.
   • Answer: Contrast convergence is not a real property. It was made up by the professor.

   Key Points about Good Color Scales:

   • Perceptual Equidistance: A change in data should reflect evenly as a change in color.
   • Increasing Luminance: Ensures the scale still makes sense in grayscale and avoids artificial artifacts.
   • Colorblind Safety: 8% of males in Western culture are colorblind. Make sure the scale works for them.
2. Data Types:
   • Question: Gold, silver, bronze medals: What data type are they?
     • Options: Ordinal, Step Categorical, Nominal, Discrete Quantitative.
   • Answer: Ordinal (they have a natural order but you cannot do arithmetic like averaging them).

   Key Points:

   • Ordinal data supports an ordering but no arithmetic differences.
   • Nominal is unordered categorical data.
   • Discrete data usually refers to countable quantitative values, not mere categories.
   • For medals (gold, silver, bronze), use a semantic color coding that matches their inherent categories (e.g., gold = gold color).
3. Color Scales (Bivariate Example):
   • Question: Shown color scale. Is it two-tone, bivariate, histogram-equalized, or divergent?
   • Answer: Bivariate color scale (using two hues to encode two different variables at once).

   Key Points:

   • Bivariate Scale: Two attributes mixed, producing combined colors.
   • Divergent Scale: Has a neutral midpoint and diverges into two directions with different colors (e.g., representing temperatures around zero).
   • Two-tone Pseudocolor: Uses blending of discretized colors to indicate exact values within bins.
   • Histogram Equalized Scale: Adjusts bins so that colors are distributed evenly over data distribution.
4. Marks Definition:
   • Question: Marks are defined by their…?
     Options included perceptual precedence, expressiveness, dimensionality, Gestalt principles.
   • Answer: Dimensionality.

   Key Points:

   • Marks = basic geometric elements (points, lines, areas, volumes) onto which we map data.
   • Channels (position, color, size, etc.) are used to represent data attributes.
   • Perceptual precedence and expressiveness refer to channels, not marks.
5. Non-anchored Points:
   • Question: Non-anchored points are… pre-attentive cues, labeling forms, optical illusions, or a problem of 3D visualizations?
   • Answer: A problem of 3D visualizations (lack of depth cues makes 3D scatterplot points “float” without reference).

   Key Points:

   • Non-anchored points cause difficulty in interpreting depth in 3D plots.
   • Pre-attentive cues are properties like color or shape that the eye processes instantly.
   • Labeling and optical illusions are different concepts.

Quiz Conclusion: Another quiz will come next week.

---