[Data Visualization] Marks and Channels

Marks and Channels

• Complex visual encodings can be broken down into two components: marks and channels
• Marks are basic geometric elements that depict items or links.
  • points, lines, areas, …
• Channels control the appearance of marks to convey data.

Marks

• A mark is a basic graphical element in an image.
• Marks can be classified according to the number of spatial dimensions they require.
  • 0D: point
  • 1D: line
  • 2D: area
  • 3D: volume (not frequently used)

Channels

• A visual channel is a way to control the appearance of marks.
• Independent to the dimensionality of geometric primitives!

Marks and Channels

• You can use multiple visual channels to encode more data.
  • x, y, size, color, shape, motion, …
• However, the more you encode, the harder to interpret (less effective)
• You can use two or more channels to encode the same thing!
  • Redundant encoding
  • The attributes redundantly encoded will be easily perceived.

Redundant Encoding

Interaction between Marks and Channels

• A point mark only conveys position (x and y), so you can encode data to its size (width and height).
• A line mark only conveys position and length, so you can encode data only to its width (thickness) not its height (= length, already taken!).
• An area mark is fully constrained, so you cannot encode data to width or height.
  • Exception: Cartograms
  • They carefully alter the boundaries so that the borders remain contiguous and each area’s shape is preserved as much as possible.

Cartograms

• Cartograms intentionally distort geographical areas to encode data.

Channel Types

• Magnitude channels tell us how much of something there is.
  • Good for encoding ordered data (Q, O)
• Identity channels tell us information about what something is or where it is.
  • Good for encoding categorical data (N, sometimes O)

Mark Types

• So far, we have focused on table datasets where a mark always represents an item (a row in a table dataset or a node in a graph dataset).
• However, marks can be used to represent a link between two nodes in a graph dataset.

Expressiveness and Effectiveness

• There are so many visual channels! How can I choose one for my vis?
• You need to consider expressiveness and effectiveness Data

• Expressiveness principle: the visual encoding should express all of, and only, the information in the dataset attributes.
  • No less and no more
  • e.g., ordered data should be shown in a way that our perceptual system intrinsically senses as ordered.
  • Beginners tend to encode more data.
• Effective principle: the importance of the attribute should match the salience of the channel, i.e., noticeability.
  • Allocate more effective channels for more important attributes

Example

• Any issues in terms of expressiveness and effectiveness?

Measuring Channel Effectiveness

• How can we measure the effectiveness of visual channels?
• There can be many criteria…
  • Accuracy (how accurately can humans read the true value from a representation?)
  • Discriminability (how many different values can be encoded in a discriminable way?)
  • Separability (with what other channels can a channel be used together?)
  • The ability to provide visual popout
  • The ability to provide perceptual groupings

Accuracy

• The obvious way to quantify effectiveness is accuracy
• How close is human perceptual judgement to some objective measurement of the stimulus?
• Humans perceive different visual channels with different levels of accuracy.
• It is known that our responses to the sensory experience of magnitude are characterizable by power laws.

Stevens’s Power Law

$S = I^n$

• S: the perceived sensation
• I: the physical intensity
• If n = 1, the perceived sensation is exactly proportional to the physical intensity (most accurate).
• If n > 1, the sensation is magnified, and if n < 1, the sensation is compressed.

• A channel with n close to 1 can be considered effective.
• Length has an exponent of 1.0 (the most accurate).
• The other visual channels are not perceived as accurately.
• Area and brightness are compressed.
• Red gray saturation is magnified.

• Which channel is more effective?

Effectiveness Ranks

Results from Controlled Experiments

• Task: proportionality estimation
• Estimate what percentage the smaller value was of the larger
• T1-T3: position on common scale
• T4-T5: length

Effectiveness Ranks by Data Type

Discriminability

• If you encode data using a particular visual channel, are the differences between items perceptible to the human as intended?
• Quantify the number of bins that are available for a visual channel, where each bin is a distinguishable step or level from the other.

• How many different linewidths can you see?

• How many different levels of lightness can you see?

• How many different levels of lightness can you see?

• How many different hues can you see?

Separability

• Some channels have interactions with other channels!
• Be careful when you use more than two channels at once.
• How accurately can people access information encoded by each channel?

• The important idea is to match the characteristics of the channels to the information that is encoded.
• If the goal is to show the user two different data attributes, you need to use separable channels (e.g., position + hue)

Visual Popout

• Another important aspect for measuring effectiveness of a channel is whether it provides visual popout
• Sometimes, called preattentive processing or tunable detection
• For some visual channels, our low level visual system can process the information in a massively parallel manner.
  • Even without the need for the viewer to consciously directly attention to items one by one (preattentive).
  • Less than 200-250 ms
  • Involves only information available in a single glance

• How many ‘3’s?

• Find a red dot!

• It is hard when there are many distractors!
  • We do perform serial search

• Popout happens for a single visual channel!
• Most pairs of channels do not support popout , but a few pairs
  • Space + Color
  • Motion + Shape
• Popout is not possible for three or more channels.

Preattentive Channels

Grouping

• The last criterion is whether a channel supports perceptual grouping.
• The easiest way to show visual elements as groups is just to connect the elements!
• However, it is possible without using extra lines…

• Gestalt laws tell how humans naturally perceive objects as organized patterns and objects.

• Proximity: Things that are close together are perceptually grouped together.
• Similarity: Similar objects are perceptually grouped
• Connectedness: Connected objects are perceptually grouped
• Continuity
• Symmetry: Symmetric objects are perceptually grouped together.

Proximity

Similarity

Connectedness

Continuity

• Humans are more likely to construct visual entities out of visual elements that are smooth and continuous, rather than nes that contain abrupt changes in direction.

Symmetry

Relative versus Absolute Judgements

• The human perceptual system is fundamentally based on relative judgements.
  • the amount of length difference we can detect is a percentage of the object’s length.
• Weber’s Law

• I is a stimulus intensity.
• 𝐾 is a fixed percentage.

Summary : Marks and Channels

• Marks are basic geometric elements that depict items or links.
  • Points, lines, areas, …
• Channels control the appearance of marks to convey data.
  • Position, color (hue, luminance, saturation), size, shape, tilt, motion, …
• Expressiveness and effectiveness
• Channels can be ranked according to their effectiveness.
  • Accuracy, discriminability, separability, popout , and grouping
  • Visual popout or preattentive processing
• Humans are good at relative judgements.