Successful test of use of a cluster analysis app for the analysis of MSC stories

[rick davies]

Hi everyone

Over the last few weeks I have developed, then tweaked, the design of a cluster analysis app that can work with text materials and generate a nested classification of those text documents in the form of an annotated tree diagram. And do other things as well...

Today I successfully tested it out on a subsample of 16 MSC stories collected, and selected, in Bangladesh in November 1994. [ I still have the original Bangla text, and English written translations of these, in a filing cabinet in our garage]

Some highlights:

1. Using a customised prompt, the app identified 10 significant differences between the 16 storylines, which you can see attached in a graphic image, which will be readable if you click on it.
2. The app then did reverse translation type of exercise, to see how reliably each of the texts could be identified as one of those different types of change. The overall level of agreement between the original categorisation and the reverse translation was 91%
3. The app then developed a dendrogram /tree diagram showing how each of the stories was related to each other in terms of relative similarity of attributes (using the 10 differences that were identified). The second attached file shows you a graphic image of that tree structure and one of the tooltips (indicated by the diamonds)  which shows how one particular branch is different from the next one in terms of the attributes of that story versus those on the next branch.
4. I then used the query function, to ask for the identification of which of the storylines described negative events (result or unresolved), and was surprised to find there were six of them. I also ask the app to generate a descriptive headline describing how the first of the two major branches of storylines were different from the second, which it did very well. You can see the full details right here:

 Propose a short, distinctive headline (max 8 words) for each cluster, and say what sets it apart from the others.

Claude

## Cluster Headlines and Distinguishing Features --- ### Cluster 1 **Headline: "Individual Lives Transformed Through Personal Determination"** **What sets it apart:** The stories are anchored in **named individuals or families** whose personal circumstances change, usually through a combination of CCDB loan access and their own hard work. The narrative arc is essentially biographical — we follow one person's journey. Even where collective structures (shomitis) appear, they are backdrop rather than protagonist. The most distinctive thread is **personal agency against adverse circumstances**, including abandonment, gender oppression, and poverty. --- ### Cluster 2 **Headline: "Collective Institutions Growing Into Self-Governance"** **What sets it apart:** The subjects here are **groups, forums, and shomitis acting collectively** — buying land for forum centres, establishing schools, managing donor visits, recovering bad loans, achieving financial independence from CCDB. The individual is largely absent. What is being celebrated is **institutional maturity and collective decision-making**, particularly the shift from dependence on CCDB toward autonomous action. Several texts explicitly mark the significance of the forum or shomiti doing something **that previously required CCDB's initiative**. --- ### Core Distinction Between the Two Clusters | | Cluster 1 | Cluster 2 | |--|-----------|-----------| | **Hero** | Named individual | The group/institution | | **Change type** | Personal welfare | Collective capacity | | **Dependency theme** | Breaking from poverty | Breaking from CCDB |

If any of you are as excited about this analytic capacity as I am, please feel free to contact me, to find out more how you and/or I could do similar analyses of any MSC story collections that you have access to (and of course permission to analyse them)

With best wishes, Rick