I put Tove Jansson's Moomin character the Muddler, Sw. Rådd-djuret, into a presentation. It's about multivariate statistics for archaeologists, and I accompany the picture with the following quotation.
How could you forget about the Muddler when you launched the ship, Sniff said accusingly. Did he ever get his button collection back into order?
Oh yes, many times, said Moominpappa. He came up with new button systems all the time. Sorted them according to colour or size or shape or material, or according to how much he liked them.
Amazing, Sniff said dreamily.
--- Tove Jansson 1968, The Exploits Of Moominpappa, Ch. 3
The reason, for any Dear Reader who makes do in their daily life without correspondence analysis, is that archaeologists have a constant need to look at many different traits in our source material at the same time without going cross-eyed. We need to sort stuff -- objects, closed finds, entire sites -- according to colour, size, shape, material AND according to how much we like them, simultaneously.
What would you suggest one do with a correspondence analysis in which one site and one type come out isolated on their own. I tried to run a cluster analysis or discriminant analysis to check for statistical significance, but rather obviously found they would not work.
I'd delete the site or type from the dataset and comment in the text that this outlier site or type is so deviant that it can't usefully be analysed together with the rest. Also I'd explain what traits it deviates in.