Visualizing Time-Oriented Data

CS-GY 6313 - Fall 2026

Claudio Silva

2026-11-20

Introduction

Based on materials by Enrico Bertini

Why Does Time Matter?

Daily NYC taxi trips in 2011, 2012, and 2013 aligned by calendar date, showing recurring weekly fluctuations and occasional sharp drops.

NYC taxi and for-hire trips from 2009 to 2025, showing changes in the mix of services and a steep drop in 2020 followed by a recovery.

What is Temporal Data?

Definition: Data in which values depend on time and time is explicitly recorded.

In other words: we track when things happened or what values were at specific points in time

Example: Temporal Dataset Structure

The DATE and TIME columns make each collision record temporal data.

Application Domains

  • Business
  • Natural Phenomena
  • Behaviors/Movement
  • Traffic/Mobility
  • Medical/Healthcare
  • Finance

Type 1: Event Data

Time + Object (Attributes)

“Something happened at time T”

Examples: Tweet, Email, Alarm

Type 2: Measurement Data

Time + Measure(s)

“This is the value at time T”

Examples: Temperature, revenue, stock value

Temporal Information Components

Time = Position + Interval + Granularity

  • Position: When did it occur? (specific timestamp)
  • Interval: For how long? (duration)
  • Granularity: At what scale? (seconds, hours, days, years)

Example: A meeting starts at 2:00 PM (position), lasts 90 minutes (interval), measured in minutes (granularity)

Time Structures

Three key structures for temporal data:

  1. Sequential - Linear progression over time
  2. Cyclic - Repeating patterns (daily, weekly, seasonal)
  3. Hierarchical - Nested time resolutions (year/month/day)

Key insight: Most real temporal data has ALL THREE structures simultaneously!

Sequential and Cyclic Patterns

Sequential

(Jan 1, … , Dec 31)

Cyclic

(S, M, T, W, T, F, S)

Hierarchical Time Resolutions

Multiple Resolutions

  • Years
  • Months
  • Weeks
  • Days
  • Hours

Linking Data Types to Time Structures

Event Data

Time + Object (Attributes)

  • Sequential: Transaction logs over years
  • Cyclic: User logins by day of week
  • Hierarchical: System alerts by hour/day/month

Measurement Data

Time + Measure(s)

  • Sequential: Stock prices over time (line charts)
  • Cyclic: Temperature patterns by season
  • Hierarchical: Revenue by year/quarter/month

Key Insight: Data type + time structure → visualization choice

Aggregation Trade-offs

Benefits of Aggregation

  • Reduces noise
  • Shows overall trends
  • Manageable data size
  • Clearer patterns

Example: Daily → Monthly averages reveals seasonal trends

Risks of Aggregation

  • Masks short-term events
  • Loses extreme values
  • Can hide critical peaks
  • Simpson’s Paradox

Example: A 1-hour server outage disappears in monthly view

Design principle: Choose granularity based on the questions you need to answer

Warning: This is Simpson’s Paradox for temporal data - aggregation can completely reverse or hide patterns!

Example: Sales Over Time - Sequential View

Sequential: “How did our sales change over the years?”

Example: Sales Over Time - Cyclic View

Cyclic: “How does the number of orders change by day of the week?”

Hierarchical Structure and Resolution

Combining multiple time resolutions in one view

Time Resolution: Sequential Views

The same sales by week (top) and by month (bottom): granularity changes what you see.

Time Resolution: Cyclic Views

The same orders folded by month of year (top) and by day of week (bottom).

Example: Nesting Cyclic (Day) within Sequential (Year)

One line per year (2013-2016) across the days of the week: a cyclic view nested in a sequential one.

Example: Nested Time Patterns (Monthly)

Sales by month of the year, one line per year (2013-2016).

Visualization Methods

Line charts, area charts, and their variations

Line Charts: Time + Quantity

The fundamental temporal chart: position encodes both time (x) and value (y).

Aspect Ratio

How chart dimensions affect perception

Definition: Aspect Ratio = Width / Height

Impact on Trend Visibility

The same data at three aspect ratios: the trend looks steeper or flatter.

Perceptual Principle: Slope Judgment

Key finding: Humans judge slopes most accurately at 45°

  • Too shallow (<< 45°): Hard to distinguish small differences
  • Too steep (>> 45°): Also difficult to compare
  • Optimal: ~45° - Maximum perceptual sensitivity

This principle guides aspect ratio selection

Banking to 45°

Method: set the aspect ratio so that the average slope of the line segments is 45°

  • Makes trends perceptually salient and comparable
  • Cleveland, McGill & McGill introduced it in this 1988 JASA paper (a required reading this week)

Example: Banking to 45° Comparison

Canadian lynx trappings: a square frame (left) vs. banked to 45° (right). Which shows the shape of each cycle?

Best Practice

Rule of thumb: Always test different aspect ratios to see which one best conveys your message

Multiple Line Charts: Adding Categories

Color encodes the category: three customer segments over time.

The “Spaghetti Plot” Problem

It does not scale: 17 sub-categories in one chart.

Solutions for Multiple Lines

Possible solutions:

  1. Grouping
  2. Filtering/Focus
  3. Highlighting

Comparison: Grouping, Filtering, Highlighting

Original: 17 sub-categories

Filtering: keep 3 sub-categories

Grouping: roll up to 3 categories

Highlighting: 3 in color, the rest faded

Small Multiples

Small Multiple Line Charts and Area Charts

Example: Small Multiple Line Charts

One panel per sub-category on a shared y-axis: each series is readable and comparable.

Example: Small Multiple Area Charts

The same panels as areas: filled area emphasizes magnitude.

Stacked Area Charts: Total and Parts

Stacked areas show the total and how it splits across four regions.

Stacked Area Charts: Limitations

“Baseline bias”: only the bottom layer sits on a common baseline.

The same four regions as lines (left) and stacked (right). Try reading East’s trend on the right.

Normalized Stacked Area Charts

Normalized to 100%: composition only, the total is gone.

Interaction Techniques

Critical for exploring temporal data

Why Interaction Matters for Temporal Data

Challenge: Time series data is often too large or complex to understand in a single static view

Solution: Interactive techniques allow users to:

  • Navigate through time (zoom, pan)
  • Focus on specific patterns or events
  • Link multiple views for comparison
  • Dynamically adjust aggregation levels

Zoom and Pan

Purpose: Navigate dense sequential time series

Semantic Zoom

  • Change level of detail
  • Hourly → Daily → Monthly
  • Preserve context

Geometric Zoom

  • Magnify visual space
  • See more detail
  • Focus + context techniques

Example: Stock chart with overview + detail panes

Filtering and Brushing

Filtering: Show/hide data based on criteria

  • Time range selection
  • Category filtering (e.g., show only certain products)
  • Threshold filtering (e.g., values > X)

Brushing: Select data in one view, highlight in others

  • Temporal brushing: Select time range, see corresponding events
  • Linked views: Brush on map → highlight in timeline

Brushing & Linking Example

View 1: Geographic map

User brushes (selects) a region

View 2: Time series

Corresponding temporal data highlights automatically

Power: Discover spatio-temporal patterns (e.g., “Sales peak in Region A during Q4”)

Dynamic Aggregation

Interactive granularity control

Users adjust time resolution on-the-fly:

  • Slider: “Show me data aggregated by: Hour / Day / Week / Month”
  • Drill-down: Click on a month → see daily breakdown
  • Roll-up: Aggregate noisy hourly data to daily averages

Advanced: Time-series bagging, dynamic binning for very long series

Time Warping and Distortion

Fisheye/distortion:

  • Focus region shows detail
  • Context regions compressed
  • Maintains overview while examining specifics

Time-warping:

  • Align periodic patterns despite phase shifts
  • Dynamic time warping (DTW) for comparing similar patterns
  • Useful for comparing multiple time series with lag

Best Practices for Interaction Design

  1. Provide overview first: Show full temporal extent
  2. Progressive disclosure: Start simple, reveal detail on demand
  3. Maintain context: Always show where you are in time
  4. Link multiple views: Connect temporal, spatial, and categorical views
  5. Smooth transitions: Animate changes to maintain mental model
  6. Direct manipulation: Let users interact with the visualization itself

Event Data Visualization

Visualizing discrete events and durations

Types of Event Data

Timestamp + Event Properties

  • Incidents
  • Activity Logs
  • Messages (Emails, Chats, etc.)

Example: Dot Plot for Events

Rows = cities, x = ship date; dot size encodes a count (legend 1-52).

Example: Proportional Symbols plus Color

The same dots colored by ship mode: size and color carry two event attributes.

Events with Duration

How do you visualize events that have duration?

Gantt Charts

Horizontal bars show each event’s start, duration, and end.

Example: Gantt Chart

The lifespan of an evening news anchor

  • One row per anchor, grouped by network (NBC, CBS, ABC), since 1948
  • A line from start to end shows each tenure
  • The annotation calls out the longest run: Jennings, 27 years

Example: System Activity Monitoring

Tape mounts per drive over a day and a half: concurrent activity at a glance.

Alternative Visualizations

Heat maps and real-world examples

Heat Maps for Temporal Data

Matrix representation: one axis is time, the other is categories, color shows the value

  • Think of it as a table with colored cells
  • Here: sub-category × day of month, color = sales
  • Compact; shows patterns along both axes

Example: WSJ Vaccine Impact Visualization

Measles Today: New U.S. Cases by Week, 2025-2026

Measles in the United States, 1928-2026

Heatmap of reported measles cases per 100,000 people across the 50 states and D.C. from 1928 through November 12, 2026. Rates generally fall after the 1963 vaccine introduction, with increases in some states in 2025 and 2026. Gray indicates missing data, including 2023.

Variation: Size Instead of Color

The sub-category × day-of-month grid again, with square size encoding sales.

Example: FT European Football Visualization

How to read it:

  • x: player’s age
  • Rows: one player each
  • Length: career span
  • Color: non-penalty goalscoring rate (darker = higher)
  • Square size: share of minutes played

Periodic Patterns

Calendars, radial layouts, and spirals

Understanding Periodicity

Often the focus is on investigating and presenting periodical patterns:

  • Yearly, seasonal, monthly
  • Weekly, daily, hourly

Alternative Approaches

One line per year across the week: what else could show this periodic pattern?

Calendar-Based Visualizations

Key difference from heat maps: time is laid out by day-of-week and month

  • Heat maps: linear time progression
  • Calendar views: explicit cyclic layout (7-day weeks, 12 months)
  • Better for weekly and seasonal patterns
  • Here: Divvy bike-share rides per day, 2014-2017

Calendar Variations

Month × hour of day, two ways: a grid of marks (left) and a continuous heat map (right).

Radial Layouts for Periodic Data

Periodic phenomena are cyclical. Radial layouts can reduce temporal discontinuities

Spiral Plots

Key insight: spirals show cycles and progression at the same time

  • Cycles: the same phase of each period sits at the same angle
  • Progression: each turn moves outward in time
  • Weber et al.: the same periodic series as a line (left) and a spiral (right)

Spiral Example: Temperature Data

  • Angle: month of the year
  • Radius: temperature anomaly vs. 1850-1900 (red rings at 1.5 °C and 2.0 °C)
  • Color: year, purple (1850) to yellow (2016)
  • Later loops sit further out: warming

Climate Spiral Visualization

The animated version draws one loop per year, month by month, from 1850 to 2017

  • This still is its first frame, before any data is drawn
  • Play the animation to watch the loops push toward the 1.5 °C ring

Caution: Radial Layout Limitations

Use circular layouts with care!

Research finding: Humans are significantly slower and less accurate at judging radial distances and angles compared to linear position

Comparing values radially is harder than in Cartesian space - use only when cyclical nature is essential

Comparison: Linear vs Radial

Arab Spring timeline, radial (left) vs. linear (right): which makes durations easier to compare?

Scalable Visualizations

Horizon charts and sparklines for dense temporal data

Horizon Charts

Maximum density in minimal height: dozens of time series compared at once.

Why Not Line Charts? Same Series, Same Height

The same series as plain line charts at this height: almost flat, nothing to read.

How Horizon Charts Work: Step 1

Start with a standard area chart.

Step 2: Discretizing into Bands

Divide values into equal-height bands (e.g., 0-1, 1-2, 2-3).

Step 3: Flipping Negative Values

Mirror negative values above the baseline, told apart by hue.

Step 4: Collapsing Bands

Overlay the bands in one strip (left to right): darker = larger magnitude.

Final Result: Horizon Chart

A compact view that keeps temporal patterns in minimal height.

Sparklines

“Small, intense, word-sized graphics with typographic resolution. Sparklines can be placed anywhere that words or numbers can be: in sentences, maps, graphics, tables.”

— Edward Tufte

Sparklines in Context

In a table (top) and a dashboard summary (bottom): the data becomes part of the text.

Sparkline Examples

On a watch face (left) and in Tufte’s mutual-fund table (right).

Alternative Encodings

Beyond position: animation, color, and trajectories

When Position is Already Used

All methods so far use space to represent time.

What if position is already taken (e.g., maps)?

Three Alternative Encodings

Animation

Show changes over time through motion

Color

Use color/saturation to encode time

Trajectories

Draw paths showing movement

Caution: Animation Limitations

While animation can be effective, it has significant drawbacks:

Cognitive Limitations:

  • Limits working memory - hard to remember previous frames
  • Difficult to compare non-adjacent time points
  • Requires replay to catch missed details

Best Practices:

  • Provide playback controls (pause, rewind, speed)
  • Keep animations short (< 10 seconds)
  • Combine with static “key frames”
  • Consider small multiples instead of animation

Summary

Key Topics Covered

  1. Structure of temporal data (linear, cyclic, hierarchical)
  2. Line charts and variations (area charts, small multiples)
  3. Heat maps/grids, calendars, circular layouts
  4. Event data (dot/bubble charts, gantt)
  5. Scalable visualizations: horizon, sparklines

Key Takeaways

Design guidelines:

  • Consider time structure (sequential, cyclic, hierarchical)
  • Be cautious with stacked designs (hard to read upper layers)
  • Test different aspect ratios for line charts
  • Use circular layouts carefully (harder to compare values)
  • Choose the right technique for the data density

Key References (1/2)

  • Taxonomy of timeline visualizations
  • The TimeViz Browser
  • Visualizing time-oriented data—A systematic view
  • Visualizing pedestrian data
  • Cloudlines visualization method for event data
  • Banking to 45 degrees explained
  • Banking to 45 degrees: Aspect ratios for time series plots
  • The Shape Parameter of a Two-Variable Graph (original paper on banking to 45 degrees)
  • An Empirical Model of Slope Ratio Comparisons (follow-up work on banking to 45 degrees)
  • Quantitative Displays for Combining Time-Series and Part-to-Whole Relationships
  • Stacked area chart vs. Line chart – The great debate
  • The Fine Line In a Gray Area: When to Use Line vs Area Charts
  • Climate spirals (great example of use of spirals)
  • Enhanced interactive spiral display
  • Spirals for periodic data
  • Interactive Visualization of Serial Periodic Data
  • Visualizing Time-Series on Spirals
  • Explanation of Horizon charts
  • User study on effectiveness of Horizon Charts

Key References (2/2)

  • Visualizing Time with the Double-Time Bar Chart | Doug McCune
  • Visualizing Cyclical Time – Hour of Day Charts | Doug McCune
  • Visual analytics techniques for time series data
  • On the Benefits and Drawbacks of Radial Diagrams | SpringerLink
  • Visualizing Time with the Infinity Hour Chart | Doug McCune

Next Class: Friday, December 4

No class on November 27 (Thanksgiving). Week 13: Visualizing Network Data

  • Node-link diagrams, force-directed layouts, adjacency matrices
  • Trees: treemaps, sunbursts, dendrograms; edge bundling

Required readings:

Mini-project 3 (network visualization): due date as announced in lab.

Thank You!