Univariate

Bivariate

Choices Matter…

• Every decision we make will determine success or failure.
• Every decision is a compromise.
• Prioritise
• Choose wisely
• Justify everything (Kirk 2012)
• Let’s consider some examples…

Choices Matter… Cont.

• People have difficulty assimilating the information in box plots (Biehler 1997) and may forgot about sample size (Pfannkuch 2006).

Choices Matter…Cont. 2

Choices Matter Cont. 3

• Cleverland, Diaconis, and McGill (1982) found that “zooming out” on a scatter plot increases the perceived strength of a correlation between two quantitative variables.

Different plots, different inferences

• Baglin and Grant (2016) found that using different plots to summarise the same data, can significantly change the inference people draw.

Different plots, different inferences Cont.

Histograms

• Histograms, which tally a quantitative variable into equal interval “bins”, are sensitive to sample size and the number of bins selected.

Barcharts

• Position is a more accurate visual variable than length
• Let’s look at this issues by proposing the following question:
• Can we predict a person’s eye colour based on their hair colour?
• We will look at hair and eye colour data from 592 college students.

Barcharts Cont.

• Unequal sample size and uncommon baselines make comparisons difficult.

Barcharts Cont. 2

• Filling out the stacks makes it easier to compare proportions, but the uncommon baselines are still an issue.

Barcharts Cont. 3

• We can adjust the position by dodging eye colour which fixes the baseline

Mosaic Plots

• Are Mosaic plots the best of both worlds?

Side-by-side Comparisons

• Side-by-side comparisons of distributions are a highly effective way to compare groups.

Side-by-side Comparisons Cont.

• Ordering makes it easier to rank vehicles.

Side-by-side Comparisons Cont.

• However, we know boxplots hide sample size…

Side-by-side Comparisons Cont. 2

• We can also add the mean

Side-by-side Comparisons Cont. 3

• And some 95% Confidence Intervals for the mean

Multivariate

• Count how many variables are contained in the following visualisation

Multivariate Cont.

• The world is not bivariate.
• Our world is far more interesting and complex.
• The aim of multivariate data visualisation is to see through this complexity
• However, multivariate data visualisation is very difficult.
• As a general rule, no more than three variables in one plot.

Multivariate Cont. 2

• This data visualisation has three variables.

Multivariate Cont. 3

• We are not good at thinking in 3 dimensions either…

Multivariate Cont. 4

• There are four general strategies for multivariate data visualisation
  • Mapping additional aesthetics (No more than three in one plot)
  • Faceting (No more than two variables)
  • Purpose built - Use with caution
    • Sankey diagrams
    • Parallel coordinates
    • Multivariate mosaic plots
  • Animation (adds a frame aesthetic)

Animation

• Animation essentially adds an extra aesthetic - frame
• Frame is excellent for conveying time.

Getting Interactive

• Interactive data visualisations take advantage of our inquisitive nature and desire to play (Murray 2013)
• There are heaps of interactive features we can use in data visualisation:
  • Zooming
  • Selecting variables
  • Hover effects
  • Filtering data
  • Highlighting
• I only ask you to keep one thing in mind:
  

Getting Interactive Cont.

Activity

• Working with a partner or in small groups, discuss how the following visualisation can be improved.

Quick Quiz 4

• Let’s do a quick revision quiz:
  • Visit https://b.socrative.com/login/student/
  • Room Name: DATAVIS
  • Enter a nickname
  • We will work through each question as a group