Moonplots visualize associations between categorical variables from a Correspondence Analysis, Principal Components Biplot or Linear Discriminant Analysis output. It is simpler and easier to interpret than a scatterplot.
Method
The chart below shows a correspondence analysis moon plot of the mobile phone data (see How to Do Traditional Correspondence Analysis for the data and the traditional scatter plot). The columns are now portrayed equidistant from the center of the map and the size of the font indicates the strength of the column category as a discriminator (i.e., it represents the information previously shown in the green font in the previous chart).
If using a moon plot for understanding brand positioning, it would be interpreted as follows:
- Generally, brands are shown in the middle (swap the rows and columns of your table to achieve this). The actual position of the brand is the center of the text label.
- The closer to brands are together, the more similar their positioning.
- The further a brand is from the middle, the more different it is from the “average”. Big brands are often in the middle because they define the average.
- Attributes are shown on the perimeter.
- The larger the font size of the attribute, the greater the level of discrimination of brands on the attribute (note: discrimination does not mean importance).
- The closer a brand to an attribute, the greater the association between the two.
Note, different versions of Microsoft PowerPoint are not entirely compatible – if creating a moon plot in one version for use in another, it will often be advisable to convert it into a picture in PowerPoint.
When researchers use scatterplots to display correspondence analysis, occasionally they place circles on the maps to show clusters of similar brands. While this particular practice is generally not a great idea (because it encourages people to misread the maps, as it incorrectly implies associations between things that are nearby on the map), the equivalent method for grouping things on a moon plot is using wedges – see Figure 103 (and, these wedges are also more appropriate for a traditional correspondence analysis plot, unless one is wanting to emphasize that the things in the middle are undifferentiated). A particular benefit of using wedges is that the width of the edge (i.e., the angle at the minimum of the group), gives insight into the cohesiveness of any brand groups. We can see, for example, that based on the data shown in the map, Coke and Pepsi are nearly identical in their positioning. By contrast, a wedge that included Pepsi Max and Diet Pepsi would take up almost half the map, thus emphasizing how different the brands are. As a general rule, it is inadvisable to have a wedge of 90 degrees or larger, as, technically, any statements that are 90 degrees or more apart (as measured from the angle at the center of the chart) are unrelated (i.e., will have an expected correlation of zero), so such a large wedge will inevitably be spanning quite different positioning.
