I am running a structural topic model on a set of qualitative survey responses. Let’s say that in each response people list a variety of reasons why company A’s product might be considered better than company B’s.

I want to get an understanding of what characteristics people are focusing on in their responses (e.g. price, build quality, aesthetics, ethics) etc.

Obviously people may raise a number of these issues in their responses – not just one. To my (admittedly basic) understanding of the principles of STM, this is fine – documents can contain more than one topic.

However, I’m struggling to reconcile this with the ‘Expected topic proportions’ produced by plot.STM summary. My understanding was that these proportions reflected the proportion of documents that contain each topic. However, they sum to 100% – which would suggest that each document is associated with only one topic.

Am I missing something here?

thanks in advance!

1. List item