11a: Manipulate, facet, reduce

Ideas to improve visualizations

Recap: How to design viz idioms

Visualization Interactivity

A lot of the viz design techniques discussed below requires interactivity to be baked into the visualization idiom.

  • View Manipulation — Change / Selection / Navigation
  • Faceting — Juxtapose / Partition / Layering
  • Reduction — Filter / Aggregate / Embed

Reading: Taxonomy of interactive dynamics for visual analysis, Schneiderman, Heer

View manipulation

Why do view manipulation?

To help navigate the dataset.

To highlight specific data, draw attention to insights / trends, etc.

Change over time

Basically animation tweening / transitions over time.

Time is encoded as an channel (animation) in the viz to show changes.

  • Change the layout arrangement
  • Change dataset
  • Reorder the data
  • Change specific channel encodings (e.g. color schemes, symbol marks)
  • Even change the chart idiom entirely (e.g. line chart to pie chart)

Bivariate animation chart (in R, video), Nathan Yao

Han's Roslings 200 countries

Selection

Selecting a datapoint, or a subset / superset of data

  • Highlighting
  • Tooltipping
  • Brushing

D3 Graph Gallery: Interactivity, Brushing

ObservableHQ: Scatterplot Matrix, Brushing

ObservableHQ: Brushing Collection

Navigation (items)

Navigating levels (subsets and supersets of the current view), i.e. going up and down the ladder of abstraction. Techniques:

  • Zooming
  • Panning
  • Guided (constrained) navigation

We've gone through examples in the geospatial space, but this can also be applied to other types of charts.

Zoomable scatterplot Obeservable HQ, Mike Bostock

SG Covid Case clusters

Lifemap (Leaflet + raster tiles), University of Lyon

Scale of the Universe 2

Navigation (dimension reduction)

An intuitive common understanding is taking a 3D dataset and turning it into a a 2D visualization via the below techniques. Of course these can be applied to higher dimentsions.

  • Slicing
  • Cutting
  • Projecting

D3 projection transitions, Observable

3D slicer, free open source compute platform for slicing 3D images.

Visualizing quaternions, 3Blue1Brown (Ben Eater) - Amazing!

Multiform Visualizations

Showing the data (all or subset) through different chart encodings.

Usage of linked highlighting shows how one view is linked to another.

Probably higher likelihood that the visualization is custom because of all the interaction.

Covid Case Clusters

Covid19 Surveillance Dashboard, University of Virginia

View faceting

Why facet?

Multiple views juxtaposed side by side (instead of presented one after another over time!) allow us to compare two views simultaneously.

Drawback is of course display space and working memory constraints.

Small Multiples

Show entirely different partitions of the data through the same viz encoding.

Thank Edward Tuft for popularizing sparklines and small multiples.

How Pew Research Center uses small multiple charts, Peter Bell

Global Technology Tracker, Bloomberg (Sparklines!)

Scottish politics, The Guardian

Layering

Superimposing layers of data on top of one another.

e.g. having multiple lines on one line chart, points of interest on maps, etc.

Often works with filters to show / hide layers.

MarineTraffic

View reduction

Why reduce?

Often to reduce visual clutter in your visualizations and make it easier to understand.

Filtering (item or attribute)

Use an interface option to show/hide a data items or attributes.

Shipmap.org

Aggregate

Group data elements to create new derived attributes.

E.g. Histogram, density plots, beewswam, hexbin, ridgelines, etc.

Income in Each State, Adjusted for Cost of Living, Nathan Yau (histogram, faceted)

How many humans have ever lived, VisualCapitalist

Mapped: All the world's military personnel, VisualCapitalist

Dimension reduction

Dimension reduction is often necessary for dealing with extremely large datasets, breaking the data down into feature vectors.

Reading: How to Visualize Your Data with Dimension Reduction Techniques

Embed (focus + context)

Focus–plus–context enables viewers to see the object of primary interest presented in full detail (focus) while at the same time getting a overview–impression of all the surrounding information (context).

Elide - ObservableHQ: Focus + context, Mike Bostock

Distort - Fisheye distortion, Mike Bostock

Distort - Worldmapper.org, (geospatial distortions)

Questions?

Chi-Loong | V/R