DS 2003

Two kinds of visualizations

• exploratory data

  • plotting data and looking at it, preprocessing

• Explanatory

  • interpreting data, integrating data, has an argument, post processing data

  • Underlying model perhaps 

Why use computers

• Better than drawing bc can update easily 

• interactivity : can inspect different aspects of plot 

• Integration: with algorithms 

• Efficiency : can re use charts and methods for different datasets 

• Quality: precise data rendering 

• Storytelling: can use time 

What is data

• An elementary assertion of a fact. 

  • Bob has a height of six feet ✅

  • 6 feet ❌

  • Needs to be a sentence 

• Thing (row/column), category (row/column), value (cell). 

• Subject , predicate, object

• Row = item, cell = value, attribute = column 

Standard data table  / “tidy data” types

• Wide

  • Each column is attribute 

• long 

  • Attribute values are columns 

Attribute types

• Nominal example: list of fruits 

• Ordinal: quality of meat (grades), A, AA, AAA

• Q interval: dates, locations ,only differences can be calculated but they will be things like distances or spans 

• Q ratio:  length, mass . fixed zero. 0 means the absence of something. Can measure ratios or proportions

ratio data

• Zero is absolute and means a complete absence

• No negative numbers 

• Weight height length time duration 

• Can say that one variable is twice as much as the other

interval data

• can go below zero 

• Cannot say that one value is twice as much as the other 

• Temp in C or F

marks

• Aka visual object, graphical objects

• Basic graphical element in an image 

• Different dimensions 

  • 0D: point 

  • 1D: line/oath

  • 2D: an area

  • 3D: volume

Visual channel

• Aka visual variable 

• A way to control the appearance of marks

• Position, color, texture, shape, tilt, angle, rotation, curvature, size (length area volume)

• Color 

  • Hue (what color is it), saturation (how pure is the color, how much of the hue is in the color), value (brightness of the color)

• Types of channels

  • Magnitude channels: how much of something . position, length, saturation (numbers) / ordinal and quantitative data 

  • Identity channels: what and where. shape, color . nominal data

• The points are marks

• And they represent the values in the table

• x position is quantitative, something else that is encoded

• The bars are marks and the encode the values 

• The x axis is a mark with nominal data and it encodes the x position 

• The y axis is a mark with ratio data and it encodes the y position

• The points are a mark and they encode the items 

• The process input is an attribute and it is the x axis location 

• The quality characteristic xxx encodes quantitative data and the y axis position

• X position: encodes quantitative attribute 

• Y position: encodes quantitative attribute

• Sizes: encodes quantitative attribute

• Circles to encode the item: nominal attribute

• area: 

• Color hue

• Color value

• Rectangle mark 

• Rectangle position

Chart jargon

• items are mapped as MARKS (dots, etc)

• relationships can be mapped as MARKS (lines, etc)

• attributes (things about item) are encoded to visual channels

Examples

• Month (ordinal) mapped as horizontal position 

• Items being mapped as point mark (on line graph) 

• Lines mark/ connections between dots connect consecutive months , ordering of time

Dimensions of a scale

•  Domain: the data values to be mapped (actual attributes not the labels on the axis)

  • example all the months, example 0 to 20 degrees C

• Range: example x pixels horizontal position

• Type: sequential, diverging, linear, log, power, square root 

• Aggregator: how are the values being aggregated if a mark represents multiple items 

domain

he data values to be mapped (actual attributes, example all the months, example 0 to 20 degrees C)

range

• : the visual channel values derived from domain (example x pixels horizontal position) 

aspect ratio

• approximate proportion of the chart to match the depicted trend 

• If domain doesn't start at zero, should center y axis on data’s mean

different scales

• Linear (look at absolute changes) 

• log scale (emphasizes small fluctuations, smoosh together high values and separate lower values) . use when data is very skewed , or when looking at percent changes 

  • Need positive non zero values 

how to handle outliers

• Could remove the outlier 

• Could do a scale break but becomes another cognitive load 

Color scales

• Use the flow chart . Q and ordinal data is ordered, nominal is unordered 

• scales use hues and value / lightness 

• can be sequential for diverging color scale

• diverging color scale if have meaningful midpoint (like elevation)

Design criteria

• Lots of possible encodings with n data attributes and k visual channels 

Design criteria

• Expressiveness 

  • Visualizations express all the facts of the data and only the facts of the data 

  • Tell the truth and nothing but the truth

• Effectiveness 

  • One visualization may be more effective than another if one is more readily perceived 

  • position is the most effective way to encode Q, O, and N data

  • Use encodings that people decode better (fast or more accurate) 

Principle of consistency

•  properties of image should match properties of the data 

Principle of Importance Ordering

• Encode the most important information in the most effective way.

• Tufte’s Integrity Principles

change in data should be proportional to visual change in graph

• Show data variation, not design variation

• Size of the graphic effect should be directly proportional to the numerical quantities (“lie factor”)

color choice

We got them cones. One for each color, RGB 

• Usually want to go from light to dark or dark to light so that color blind people dont crash out .

• Dimensions of color can see different luminance in black and white, but cant differentiate saturation 

• change value instead of hue

perception

• dentification and interpretation of sensory information 

• From the physical stimulus to recognizing information 

• Shaped by learning, memory, expectation 

• What you notice first 

• Example hear someone speak . what we hear immediately 

• For visual system: comes from eye optical nerve, visual cortex, basic perception, first processings, not conscious, reflexive

• Importance: explains why certain designs succeed or fail 

cognition

• The processing of information, applying knowledge 

• Thinking more deeply 

• Example understanding the language 

• Recognizing objects

• Relations between objects

• Conclusion drawing

• Problem solving

• Learning

Vision and cognition

Vision is constructed top down from the input

Want to make the most important things easy to perceive automatically / pre attentive processing . most important things is the goal of the visualization 

discriminability/accuracy/popout/separability in channel effectiveness

• accuracy: how precisely can we tell the difference between encoded items?

•  discriminability: how many unique steps can we perceive? How easily can we see different bins 

  • How many different bins can be represent clearly 

  • Line width is hard 

•  separability: is our ability to use this channel affected by another one? 

  • Example position and hue are completely separable visual channels. Can use position and color differences at the same time . interpreting one does not limit interpretation of the other 

  • Some interference happens with size and hue . really small = harder to interpret 

  • Some significant interference: width and height . area also complicates things 

  • Major interference: red and green and mixing them together to see which one shave high green and high red 

  • Size/area and shape would be hard af 

•  popout: can things jump out using this channel?

  • Ways to make things easier detectable 

  • Happens before focused attrition, what do you see immediately 

  • Difference in hue is good , so is shape

accuracy

part of channel effectiveness

• how precisely can we tell the difference between encoded items?

•  discriminability: how many unique steps can we perceive? How easily can we see different bins 

  • How many different bins can be represent clearly 

  • Line width is hard 

seperability

part of channel effectiveness

•  is our ability to use this channel affected by another one? 

  • Example position and hue are completely separable visual channels. Can use position and color differences at the same time . interpreting one does not limit interpretation of the other 

  • Some interference happens with size and hue . really small = harder to interpret 

  • Some significant interference: width and height . area also complicates things 

  • Major interference: red and green and mixing them together to see which one shave high greenland high red 

  • Size/area and shape would be hard af

popout

aspect of channel effectiveness

• can things jump out using this channel?

  • Ways to make things easier detectable 

  • Happens before focused attrition, what do you see immediately 

  • Difference in hue is good , so is shape 

  • Not valid for combinations. Cant find that one shape that is not hte same color. no conjunction targets

grouping

part of channel effectiveness

• Another type of preattentive process

  • Easy to see groups, look for pattern 

  • Marks for grouping are containment marks and connection marks

  • Containment =  shapes around points. Can be nested 

  • Connection marks = lines between points 

  • Grouping can also be through proximity and similarity . proximity = they are next to each other. Similarity = same identify channels ie have the same color 

pre attentive processing

• very first thing noticed, at very first view

• Can be used to draw attention to areas of interest

• Can be used to express similarity/group memberships

• Conjunctions must be avoided

in channel effectivity

Gestalt principles

• Expanding on grouping 

• How do we make it obvious that things fall into groups 

• How to guide people to quickly see patterns 

• Gestalt laws 

  • Proximity

    • Objects that are close to each other are perceived as a group

  • Similarity

    • Objects are grouped together if they are similar to each other 

    • 
      

    • Good for 1-2 groups or else it doesn’t stand out anymore 

    • Or can modulate everything else but use sparingly

  • Symmetry 

    • When two symmetrical elements are unconnected the mind perceptually connects them to form a coherent shape see 3 groups, not four 

    • can easily see symmetry 

  • Connectedness 

    • Elements are visually connected are perceived as more related than elements with no connection . overrides proximity 

    • marks can be physically lines, containment mark, outline 

  • Continuity 

    • Human eye will follow the smoothest path when viewing lines, regardless of how the lines were actually drawn

  • Closure

    • Humans tend to perceive objects as complete rather than focusing on the gaps that the object might contain 

    • triangle and box 

    • Mind fills stuff in for us. We should remove chartjunk bc we don’t need it ie we can fill it in ourselves 

  • Common fate 

    • Humans tend to perceive elements moving in the same direction as being more similar 

    • This wind plot but imagine its moving

Webers law

• we judges based on relative, not absolute differences. I can see difference of .2 in in shorter lines, but not difference of .2 in lines when the lines are longer 

• 
  

  • In order for a difference to be noticeable, the amount something must be changed (JND) is a fixed proportion of the reference sensory level

• Stevens powers law

• We perceive different visual channels with difference levels of accuracy 

• We are good at length and not good at area 

Interested in shape/distribution/skew:

• Histogram, density, box plot 

Interested in trend / relationship / two different variables:

• Scatterplot and fitted line 

• curvy lines of best fit are also possible 

interested in Comparing averages/medians/most estimates

• Estimate and uncertainty, mean + 95 % CI or show distributions 

What makes a plot statistical 

• Start with raw data, create summary (marks show aggregates, histogram, box plot, mean +- CI (confidence internal)), and then add a model (a mark that shows a fitted estimate like a regression line) 

• Each step adds assumptions and hides raw data

Confidence interval 

• More data shrinks the CI, but not the SD. Say which one your error bars show in a label or caption. 

• If we repeated the sample many times, about of the intervals would contain the true value.