I have been working with that wonder of ungovernable data, the Twitter feed. One of the unavoidable hypotheses about it is that users respond to other users, not necessarily to discrete content or events in the outside world. Predicting the content of tweets about coronovirus in Hawaii from incidence levels was not impossible, but R-squares less than .02 are not very convincing.
Running an ill-thought-out MCA on the user identifiers tweets entities mentions, tweets enitites hastags, users descriptions and user names presented a very much more dramatic picture--attached.
Unfortunately, I cannot use this in any kind of presentation, because I cannot disentangle the labels. Smaller type and shorter titles will not really do the job. How can I get into the file that generates the figure and extract the X, Y coordinates, and the label text? If I can do that, I could edit the label text, shorten words, and suppress near-duplicates so the picture can be interpreted. It is clear that the sources come from around the country--and the blue numerals indicate that incidence levels are associated with some of the Twitter entries (irrespective of the content of the text--the association is with the entry itself!)
I would really like some insight into the technology. It would appear that MCA is in fact a device for analysis of potentially sparse data that does not require the generation of n-user x n-user matrices. I hope I am right in this supposition.
