2.4 Communicating Analysis Results
The crucial element of data presentations is that they need to help convey the insight (or the message); they should be clear, engaging, and (more importantly) readable. Our ability to think of questions (and to answer them) is in some sense limited by what we can visualize.
There is always a risk that if certain types of visualization techniques dominate in evidence presentations, the kinds of questions that are particularly well-suited to providing data for these techniques will come to dominate the landscape, which will then affect data collection techniques, data availability, future interest, and so forth.
Selecting a Chart Type
The choice of visualization methods is strongly dependent on the analysis objective, that is, on the questions that need to be answered. Presentation methods should not be selected randomly (or simply from a list of easily-produced templates).
In Figure 2.7 below, F. Ruys suggests various types of visual displays that can be used, depending on the objective:
who is involved?
where is the situation taking place?
when is it happening?
what is it about?
how/why does it work?
A general dashboard should at least be able to produce the following types of display:
charts – comparison and relation (scatterplots, bubble charts, parallel coordinate charts, decision trees, cluster plots, trend plots)
choropleth maps (heat maps, classification maps)
network diagrams and connection maps (association rule networks, phrase nets)
univariate diagrams (word clouds, box plots, histograms)
Check the data: outliers, spikes, anomalies
Explain encoding: don’t assume the reader knows what everything means
Label axes: knowing the scale is important
Include units: eliminate the need for guesswork
Keep your geometry in check: circles and 2D shape are sized by area, bar charts by length
- Include your sources: protect yourself, and let those who want to dig deeper do so
- Consider your audience: a poster can be wordy, a presentation should be minimalist
Is the point getting across? Integrated data helps convey the message.
In Semiology of Graphics, Bertin suggests that not all retinal variables are equally effective when it comes to convey or represent information. You may need to experiment to find the optimal choice for the given context.
Adding design elements can enhance our understanding of the data.
How we spot patterns affect what we get out of data presentations.
Data displays are not just about picking a random visualization method. The result varies depending on the structure of the data and the (combinations of) questions.