Space Graphs are a visual way of illustrating data relationships between two groups of variables. For example, between brands and image attributes or between brands and consumer groups. Space graphs use Correspondence Analysis to decompose the data table into separate dimensions for plotting.
Space graphs are only available to view on screen. Use these in online stories and dashboards.
In this article
1. Correspondence Analysis
Correspondence analysis is a good method for summarizing data tables and focusing attention on the most consequential relationships between variables. It identifies similarities and differences in any data table so the relationships can be brought to life through visualizations such as Space Graphs.
It is common to see correspondence analysis results plotted in a two-dimensional graph. In Harmoni, Space graphs display correspondence analysis in three dimensions simultaneously. You can rotate Harmoni Space Graphs using your mouse to clearly see relationships between all the variables in the plot.
Harmoni performs the correspondence analysis and generates the Space Graph Visualisation with a single click.
Because correspondence analysis highlights similarities and differences, it will not simply show you the brands that score the highest in a category. Instead, Space Graphs position variables, such as brands and attributes, according to relativities across all brands and attributes. Brand A may show a stronger association to an attribute relative to Brand B, but when you look at the weighted table, Brand B may, in fact, have a higher percentage for that attribute.
2. Space Graph
Tables of data can be challenging to interpret without visual aids. Visualizing tables with a Space Graph in Harmoni shines light on relationships that matter.
In a table of brands by attributes or demographic groupings, for example, brands that have a similar profile are positioned closer together on the chart, and the groups they most closely relate to are positioned nearby. In this way, Space Graphs show you which brands are competing, and which groups they are competing for.
A Space Graph positions row variables according to how similar or different they are to each other. It then separately does the same for column variables. This means columns and rows are mapped onto Space graphs separately (with potentially different scales), so you cannot interpret simple linear distance between row and column variables as showing differentiation. Instead, items such as brands and attributes are closely related in a Space Graph when they are on the same or similar lines radiating out from the center of the plot.
Therefore, to see how closely related column and row variables are in a Space Graph you need a line from the center of the graph out to the variables you are interested in. You can then see which are closest in terms of the angle between the lines (more about this below).
When evaluating the space graph, consider these important points:
The origin (center) of the graph is the weighted average of the row and column clouds (where the x and y values are both zero)
- Variables near the origin are very similar to the overall average profile and are therefore less distinct
- Variables further from the origin are relatively better discriminators
- The size of the bubbles indicates the relative size of each group
To produce your space graph, you first need to create a table using variables with two or more elements in the down and across dimensions. Then create the graph by clicking the space graph in the Visualize menu.
-
- Create the table
- Click on the Visualize menu
- Select the Space Graph
- Expand the Variance Explained panel to see the results of the dimensions
- You can rotate the graph to provide a 3D view of the data
Let's take a closer look at this example
Here is the analysis table showing Source of Awareness data for five brands.
We can see that TV Local is high across all the brands, while Billboards and then Radio are next.
Now let's look at the Space Graph
- Firstly, you can say that TV Local is not a differentiating source of awareness as it sits close to the origin. All brands share TV Local in common and there is little variation from the mean and the data point sites.
- We can see that Billboard stands out and although Orange Swing is not directly close to it, the angle between the two is small and there is some proximity.
- We can also see that Golden Nectar and Newspaper are similarly aligned although both are closer to the origin.
Keep in mind that you can rotate the space graph on the screen in Harmoni which allows to take a closer look at the relationships and positioning.
Drawing a line
You can draw a line from the origin to a data point by clicking on each bubble. Harmoni draws the line and includes the unit distance from the origin to the bubble. The distance is a relative measure and is a numeric representation of the distance from the bubble to the origin, as a proportion of the distance from the furthest data point to the origin. This can be helpful when bubbles are similar distances from the origin and can make it easier to see the angle between two data points which is relevant for assessing the similarities.
-
- Create the space graph
- Click the bubble to draw the line
- Rotate the graph to assess the angles of the lines
Variance Explained
The Variance Explained panel shows the percentage of overall variance that is represented across the dimensions in the space graph. You can click the three-dots to expand the panel and display all the dimensions.
The distance between data points along each dimension reflects the differences or similarities between them. Variance helps to identify the extent of variability in the data. A higher variance suggests greater diversity or differentiation among the data points, while a lower variance indicates a high degree of similarity or clustering.
Let's look at the variance shown below. Dimension 1 explains 86% of the variance in the data, dimension 2 explains 9%, and dimension 3 only 4%. You can infer the relative amount explained by each dimension on a well-drawn map. That is, we can see on this map that the points vary much more along dimension 1 than across the other dimensions. We can also see that the total variance is 100% which means that all the data is summarized in the space graph.
In the analysis, it's possible that not all the data is summarized, and when this happens., the total will not add to 100%.