I’ve always been intrigued with mapping and networks, given my background in applied mathematics. It’s refreshing to see, instead of mapping sets of real numbers to prime numbers (cue the groaning of all the non-STEM readers) this information being put to more practical use. (I hope my Axiomatic Set Theory professor doesn’t get annoyed I said that!) I really liked using Palladio to create a network, as I find networks inherently more interesting than plain maps. Maps are great for one-dimensional representations–to show relationships, however, I think it’s necessary to use networks.
We used the same data we’ve used for the past few exercises, which is metadata from interviews conducted by government employees with former slaves from 1936-1937. I found this exercise highlighted the most interesting revelations with the data. With mapping, I said that it mostly eliminates grunt work, but it doesn’t really reveal anything new. I guess the same could be true for network analysis, but I don’t think so. Human beings are terrible at grasping relationships between two sets. It’s why we need phrases like “correlation does not equal causation” but also why people can’t even believe correlation might mean something. (Cough, data relating to climate change leading to bad things.) Sifting through metadata may allow me to grasp a basic understanding of how big the Roman Empire was, but I can’t grasp that same understanding with the Republic of Letters, no matter how many times or how long I could stare at it. Therefore, the network analysis allowed me to see things I had never seen before, even with the text mining and the mapping.
We used Palladio for this exercise, which I found to be easy and intuitive. Although many of the projects we reviewed used Gephi, we used Palladio because Dr. Robertson instructed us to. Easy enough decision, then! The setup, I’ll admit, was a bit confusing, as it involved dragging and dropping .cvs files into the Palladio webpage. Then I had no idea what everything meant, but all that confusion was easily translated into networks when I hit the “graph” button. I have no idea how it translated all the metadata into a network–and I’m sure I’d have to get a degree in computer science to grasp it fully–but it translated the data beautifully, and it allowed me to choose relationships that I wanted to highlight. Just as with comparing any two sets of data, some were more useful than others. When both the source data and the target data contained large quantities, the network was too large to make any sense of. When one group was smaller, limited to say two or three categories, the relationships were plain to see. And relationships varied greatly. You could make a graph to show the relationship between where a particular former slave had been enslaved and where that same former slave was interviewed–effectively showing you where he/she was and where he/she is (at least when the interview was conducted). So that was a relationship across time, for the same person. But you could also graph relationships between people, for instance, between the interviewers and the interviewees. Which people interviewed former slaves the most? Were any former slaves interviewed twice? Suddenly these questions had easy-to-see answers. Then, to really get into what the former slaves talked about, you could graph what males talked about vs. what females talked about. However, just as Dr. Weingart said, just because you can network a relationship doesn’t mean you should. Some graphs were more visually informative than others, and others were just complete messes to both academics and laypeople. The relationships that best benefited from network analysis didn’t involve too many categories–even sorting what slaves talked about by age was a little too much. One-to-one is usually the best, but that doesn’t mean overlap isn’t important–in fact, overlap shows where common interests/places/people exist.
In conclusion, perhaps I loved this exercise just because I like networking, but even all the love of networking and graphing couldn’t make up for terrible software. I really liked Palladio, and I would gladly use it again.