Similarity matrix

Jump to section

The similarity matrix helps you identify strong card pairings and potential groupings in your open and hybrid card sorts. 

It shows you the percentage of participants who grouped two cards together, and clusters the most closely related pairings along the right edge of the triangle visualization. The darker the blue and the larger the cluster, the higher the agreement between participants on which cards go together.

You could use the similarity matrix to:

  • draft a potential website structure based on the darkest clusters seen on the right-hand edge
  • quickly see which card pairings are the most common and therefore probably belong together on your website
  • quickly see which cards are very rarely paired together so you don’t need to waste time thinking they might.

To explore the similarity matrix, hover over any square to highlight the two cards and see the exact number of participants who paired them together. 

The matrix below shows a number of strong clusters along the right edge, which tells us many people agreed about which cards belong together. A glance tells us this immediately, before we’ve even looked at the detail:

When we look more closely, we can find out which cards are paired together the most often:

And which cards are rarely paired together rarely, if at all:

With agreement levels like the ones in the darkest blue clusters, one option for us is to draft a set of categories based on these. And although this isn’t an exact science, we found this draft incredibly useful:

For cards that don’t have a strong cluster along the right hand side, you might want to look at rewording these topics and running another hybrid or closed card sort to see if there is any clearer clustering.

Competing mental models

Sometimes when you look at a similarity matrix you see dark blue areas along the right hand edge, as expected, but you also see dark blue areas in the body of the triangle for some of the same cards. What this is telling you is that your participants had two, or more, different overall approaches to sorting the cards that were presented: there are probably two competing mental models at play. =

In the example shown here, ‘Community funding’ and ‘Housing support’ have been grouped together a high number of times , but each of the cards has been grouped with other cards, as shown in the lower block of blue.  

Roughly comparing the percentages in the card pairings  – or comparing the shades of blue – will give an indication of how much the grouping on the right-hand edge was more prevalent than the other grouping, but this is not an exact science. Use what you can see here, and by checking the Categories tab and the PCA tab, to see if you can identify the different mental models. What was it about the card labels that led to two different approaches? If you changed the label names would this fix things?

Don’t forget that you can apply filters when viewing the similarity matrix. You might find that the two different approaches to sorting the cards relate to two different user audiences. This is why we’d recommend adding some pre-test survey questions to your study, so that you have something to slice and dice your data by.