Jon Udell has an excellent blog post showing why you should not assume that images are providing an accurate picture.

He mentions reading two of Edward Tufte’s books over Christmas. Coincidentally, I was also reading some of Tufte’s work during December: ‘The Visual Display of Quantitive Information‘ and ‘The Cognitive Style of PowerPoint‘. Both are excellent reads if you are interested in doing a better job of presenting information, as well as how to spot misleading visualisations.

Here is one snippet that should be observed by all those (mostly Microsofties, at the moment) who are swooning over the new visualisation features in PowerPoint 12 and Excel 12:

The number of variables depicted should not exceed the number of dimensions in the data. The use of 2 (or 3) varying dimensions to show one-dimensional data is a weak and inefficient technique, capable of handling only very small data sets, often with error in design and ambiguity in perception.

Anybody who uses 3d bar charts needs to consider this point carefully. Here’s a simple example:

The data presented in these two charts is identical, but it is much clearer and easier to analyse on the right. The shaded area in the 3D version on the left does not add any information and makes it harder to compare the data values.

On a related subject (I had a bit of a reading splurge during December) the book Freakonomics provides some great examples that demonstrate why we should not jump to conclusions and assume cause and effect when we see correlation between two data sets. Correlation only indicates a relationship between two elements, it does not prove that one causes the other. One of the most common abuses of statistics is to present indicators as causes.

As the technologies to store and analyse large quantities of information improve, it is important that we also improve our abilities to correctly interpret and present the information if we are to avoid poor decisions and the resulting consequences.

Related posts:

, , , ,
%d bloggers like this: