How do I interpret the Dendrogram in Card Sorting?
In this help, we'll discuss how to interpret the Dendrogram in open and hybrid card sort analysis. The dendrograms are calculated using two methods:
- Actual Agreement Method (AAM) - "How many people agreed with this exact group?" - Works best with 30 and more respondents
- Best Match Method (BMM) - "How many people agreed with parts of this group?" - Helps you extract as much information as possible from less data (less than 30 respondents)
Read further to learn about:
- Interpreting the AAM dendrogram
- Interpreting the BMM dendrogram
- Main difference between AAM and BMM dendrograms
- Why are dendrograms not used in closed card sort
- PDF Export of Dendrograms
The AAM (Actual Agreement Method) dendrogram organizes cards into exact groups created by the respondents, selected so the dendrogram represents groupings that were agreed upon by the highest number of respondents. The Actual Agreement Method brings the best results when it's being used with data from 30 and more respondents. The score of a group tells you that "X% of respondents agreed with this exact group".
The AAM algorithm first takes all categories created by all respondents and assigns them a base score of 1. Then, for each category that is a superset of another category, the score of the subset category is incremented by 1 (a.k.a. with two categories [card1, card2] and [card1, card2, card3], the score of [card1, card2] would be 2 and the score of [card1, card2, card3] would be 1). Categories are then queued to be added into the dendrogram by high score. When a category is added into the dendrogram, all categories that are still in the queue and would contradict it are eliminated.
When choosing which category from a list of contradicting categories of same score should be used in the dendrogram, UXtweak Card Sorting uses the category from a respondent that it intelligently judges to have been the 'best'. The best respondent is someone who paid good attention to the card sort and then completed it efficiently.
The BMM dendrogram uses a different approach to looking at the card sorting data in order to extract as much information as possible from a limited number of results (less than 30 respondents). Unlike the AAM which tells you what were the popular groupings and what percentage of respondents agreed with them, BMM considers which were the most popular card pairings and uses this information to make assumptions about the bigger picture card clustering. The score of a BMM group tells you that "X% of respondents agreed with parts of this group". Such extrapolation is quite useful when working with fewer or incomplete results.
The BMM algorithm splits all categories from all respondents into sets of their internal pairings (a.k.a. [card1, card2, card3] gets turned into [card1, card2], [card1, card3] and [card2, card3] ). All pairings are then scored based on how many times they're found in all respondents' categories and put into a queue based on their score. Cards are added into the dendrogram by taking card pairings from the queue. If neither card from a pair is in the dendrogram yet, the two cards form a new cluster. If one card from a pair is already in the dendrogram and the other one isn't, the new card is attached to the cluster that contains the other card. If both cards from a pair are already in the dendrogram but in different clusters, the clusters are joined.
Let's say you're an online food retailer and your open card sort inludes apple, watermelon and cucumber. If ten respondents sorted these cards [apple, watermelon], [cucumber] and other ten respondents sorted them [apple], [watermelon, cucumber]
- AAM would tell you that [apple, watermelon, cucumber] is a bad group because no respondents placed them in the same group
- BMM would tell you that [apple, watermelon, cucumber] is a good group because all respondents agreed at least with a part of it
The dendrograms are most useful as a tool for exploring how respondents group the concepts behind the cards into groups by their own logic. Since closed card sorts already have a pre-existing set of categories binding the respondents to a cartain way of thinking, there isn't really anything to explore in regards to how respondents conceptualize the content on their own. One could hypothetically create dendrograms based on how the respondents sorted cards into the pre-existing categories, however these would carry the bias of those categories as well. The pre-existing categories would be clearly visible in the dendrogram, in one way or another.
For simple explanation, let's say we have cards that represent an assortment of furniture and home accessories that we want run a card sort with to help us organize the content on our website. In an open card sort, the respondents might group the items in various ways - the room in the house where the items belong (living room, kitchen), the type of the item (chair, table), the brand, the price and so on. The dendrograms will help you simplify such complex data and help you find ideas for groups that aggregate the various ways of thinking.
However, if we did a closed card sort with pre-defined categories that already state that the cards should be sorted by room (living room, bathroom, kitchen etc.), that's the only way of thinking that your respondents are going to use. This unified way of thinking will lead to relatively high levels of agreement within clusters, which would correspond with the pre-defined categories.
You can create a PDF report of all the information listed above, exactly like you can see it in the web interface. To create an export, click PDF Export in the upper right corner of the tab contents.
If you wish to create a report using more than one section of your study's results, go to the Export tab