Sending meaningful visual messages in R

Graphics like tables and figures are an important aspect of communicating information because primates are highly visual animals. Well-designed graphics make it easier for us to us to communicate the “sense” of data far better than sentences or tables alone.

There are rules for making successful graphics. Some of these are best practices that make good use of the abilities our eyes and brain has, but others are conventions depending on the kind of data that we are describing.

Some ground rules: types of data

First of all, know that the word “data” is a plural word. The singular form is “datum,” meaning one data point. Your data come in a few different varieties or levels:

Type of data	Level	Example
Categorical	Nominal—names only Ordinal—there is a magnitude difference, but not a numerical one	names of species small/medium/large
Numerical	Interval—measured on a scale with equal units, but without a 0 on the scale Ratio—measurable on a scale with a 0	temperature weight, counts, length, etc.

Additionally, numerical data can either be discrete (whole numbers only) or continuous (can be any real number).

The table above describes the levels of data from low to high—meaning that as you go down the rows, the data have more inherent information in them. It’s aways possible to “downshift” data from a higher level to a lower one. For example, you could weigh beetles using a scale (ratio level) but later call them “small or large” (ordinal level). You’d lose some information, but you’ be justified in your conclusions.

However, you can’t go the other way, so that you’re “inventing” information. If I think of my students as “tall vs. short,” I can’t later proclaim that all “tall” students are 72 in. tall while all “short” students are 48 in. tall. That would “invent” information, and that is unjustified.

So, best practice when you’re collecting your own data is to always try to gather them at as high a level as possible.

Levels of data and graphics

Why talk about this now? Because the graphics that you make send a message to your viewer about the kind of data that you observed. This video describes different kinds of graphics appropriate for various types of data. Some take-home messages:

• When describing frequency distributions of samples, categorial data are described with bar charts but continuous data should be shown with histograms.
• Relationships between two numerical variables are best shown by a scatterplot (regular XY), line graph (if a trend is shown over time), or a map (for trends with space or geographic area).
• Relationships between two categorical variables could be shown with a grouped bar chart or a mosaic plot (which has the advantage of expressing the difference in sample size and proportion).
• Relationships between one numerical and one categorical variable are described well by strip charts, box plot (also called box-and-whisker plots), multiple (stacked) histograms, or a cumulative frequency distribution.