top of page

Unofficial Guide to Creating Collective Mental Models with 3CM and Anthropac

Step 5: Generate Clusters

Now that we have the MDS map of terms, it’s time to determine the clusters of terms for our composite map.

  1. Reopen the MDS map. You should be able to open it using the Items MDS tab at the bottom of the Anthropac screen. Alternatively you can open it from Items > Multi-dimensional Scaling. If the term numbers are displayed instead of labels, you can change to the labels using the labels button in the upper-right section of the screen (change Attribute from id to label and click OK).

  2. Next, go to Items > Cluster Analysis. In the window that appears, select your clustering method. I used Average link. Then enter the Max. clusters to show – while you are in the learning stage, I suggest entering  a large number that includes the maximum number of clusters your data set can support. In this case the sample data set has 57 terms, so the maximum numbers of clusters is 57 (each cluster would contain one term). Click OK.


  3. Upon clicking OK, all of the terms in your map are enclosed in a polygon and at the bottom of the right sidebar, a Clustering section appears with _partition1 displayed. The single polygon and _partition1 are indicating that the terms are currently set within a single cluster.


  4. Click the drop-down arrow next to _partition1. This displays all of the available cluster divisions. You'll see there are a total of 43 partitions available. 

    Now select _partition2. This selection tells Anthropac to use the aggregate proximity matrix to determine how the terms, which were enclosed in a single polygon or cluster, should be divided to create two clusters containing with terms that are least related to one another. As shown below, for this data set, the division creates one cluster of 56 terms and one cluster of 1 term (Community Zoning). 


  5. Now select _partition3. Anthropac divides the large cluster into two smaller cluster of terms containing those terms that are least related based on the aggregate proximity matrix.


  6. Click through several partition levels to see how successively smaller clusters are created at each level. _partition43 is the final partition level. You may be curious as to why there are not 57 partitions. The reason is that at some partition levels more than one new cluster is added because the aggregate proximity matrix indicates a tie regarding where the division should occur.

  7. The next step in creating the composite map is determining which partition level to use (or how many clusters the map should contain). This is where art meets science, because there is no right or wrong answer -- you just have to determine what number of clusters is most useful for your analysis. In other words, at what partition are the clusters small enough that you can make sense of what they contain and evaluate the relationships between terms? I like to go through the partitions one by one to see what changes with each partition. At the point where I’m mostly getting clusters with just a few terms in them, I stop. Then, after I get a little further in the analysis, I’ll go back and look at the partitions again to see if a change would help progress my research. For this data set, I personally like _partition6, _partition9, or _partition12 but your analysis may differ. Again, this is where art meets science.

Step 6: Name the Clusters

bottom of page