Data Visualization and Exploration

some questions

Chance encounters

Why we should always inspect raw data by making our own graphs and tables.

Figure 1: The choice of data import function can have a big impact on how the data is read.

Chance encounters

Figure 2: A line graph on sea ice coverage with alt text provided.

Figure 3: The choice of data import function can have a big impact on how the data is read.

What questions do you have about these?

Chance encounters

Both graphs are shared from the Instagram account “wehavethedata” but show original sources.

Figure 4: A line graph of consumer spending by income.

Figure 5: A stacked diverging bar chart of the perception of morals among peers.

What do you notice?

Let’s write some alternative text descriptions.

Your questions

• “Can we talk a little more about aesthetics? Expectations or taboos?”
• “Can we talk more about established evaluation methods or research studies that formally compare the effectiveness of different visualizations?”

Aesthetics

As with everything, opinions are mixed.

“Decoration” might make graphs more memorable¹, but at what cost(s)?

Figure 6: The Holmes “Monster Graph”.

Figure 7: A simple portrayal of the Holmes “Monster Graph” data.

1. “Useful Junk? The Effects of Visual Embellishment on Comprehension and Memorability of Charts”

The research

Bateman et al. (2010)¹

conducted an experiment that compared embellished charts with plain ones, and measured both interpretation accuracy and long-term recall

They were interested in the accuracy of interpretation and long-term recall. They found

• accuracy in describing the embellished charts was no worse than for plain charts, and
• their recall after a two-to-three-week gap was significantly better.

This calls into question an older “rule” against “chartjunk”.

1. “Useful Junk? The Effects of Visual Embellishment on Comprehension and Memorability of Charts”

Gaze tracking research on graphics

Used “gaze tracking” to determine how much time users spent looking at different components.

Figure 8: Coded areas on embellished graph.

Gaze tracking data (I)

Participants spent less time looking at “data” in the embellished graph.

Figure 9: Proportion of time spent on various visual tasks.

Is there anything we might change about this graph?

Gaze tracking data (II)

Participants spent less time looking at “data” in the embellished graph.

Figure 10: Qualitative responses differed between embellished and plain graph.

Is there anything we might change about this graph?

Other aesthetic considerations

Strong imagery is claimed to introduce bias in interpretation, but biases can be introduced in visually-unembellished charts as well (e.g., colors, orderings)

The authors even call for more research,

it seems clear that there is more to be learned about the effects of different types of visual embellishment in charts.

Other aesthetic recommendations

Some journals prohibit most table line separators (vertical or horizontal) aside from separating the table header.

To minimize distraction many experts recommend minimizing gridlines in graphics.

Some authors¹ recommend not connecting axes that don’t start at the origin.

Figure 11: The visual impact of style choices on an otherwise simple line graph.

What do you notice? What do you wonder?

1. vis4.net

What type of graph should I make?

How we make and interact with graphs is a subject of surprisingly active research¹.

Recent work has raised awareness about the need to replace bar graphs of continuous data with informative graphs showing the data distribution.

Bar graphs

• should display counts or proportions across discrete categories, and
• should not be used to present summary statistics for continuous data.

Datasets with many different distributions may have the same summary statistics.

Visit this link, though not strictly about bar graphs, to explore more.

1. Riedel et al. (2020)

Showing raw data

Figure 12: A bar graph shown as dot plots¹.

1. “Beyond Bar and Line Graphs”, Weissgerber et al, (2013)

Showing raw data

Figure 13: A bar graph shown as “slope graphs”¹.

1. “Beyond Bar and Line Graphs”, Weissgerber et al, (2013)

Showing raw data

Figure 14: Notice that raw data above the mean isn’t shown and bar space is occupied where data isn’t present¹.

1. “Beyond Bar and Line Graphs”, Weissgerber et al, (2017)

Lines are better than bar graphs?

Often bar graphs are used to show time series data, instead we should use line graphs.

Figure 15: A bar graph (with GUI) from salesforce¹.

1. Salesforce CRM Analytics

Literature against bar graphs

Bar graphs are critisized across recent literature and should really only be used for

• count or proportion data,
• with respect to distinct categories.

Otherwise consider

• line plots,
• boxplots,
• dot plots, or

at minimum overplot to show raw data superimposed on bars.

How do we assess the quality of a graph?

The answer to this one is tricky: “It depends.”

There are (as outlined above) certain very common practices that we should avoid.

Beyond that, generally, review examples from

• contemporary books,
• expert blogs or essays, or
• relevant academic or journalism articles.

Remaking graphs

We have seen a few “not so good” examples today.

Let’s attempt to extract data and remake the graphs.

Assessing graphs for accessibility

There are specific tools for assessing graphs with respect to accessibility.

This is an important, different, and technically very precise assessment.

We can look at Chartability for one (important) perspective.

while a highly trained auditor may be able to casually evaluate an artifact in as little as 30 minutes or even hold heuristics in mind as they are doing their own creative work, those new to auditing may take anywhere between 2 and 8 hours to complete a full pass of Chartability.