Positive but not perfect correlations in novel-writing and psychometrics

My favorite podcast is BBC Bookclub. I’ve been working my way through the archives, listening when I’m on my bike, and unfortunately I’m almost done. Today I was listening to Colm Toibin answer questions about his book, The Master. As is often the case, I’ve never read anything by the author, but the discussion is still interesting. Several times these interviews have motivated me to go the library and read the book being talked about.

And lots of interesting things come up in these conversations. Here’s something Toibin said:

You’re building a character. And how you’re doing that is you’re doing it with detail. A novel is a thousand details. It may be two thousand. And that the best way of describing this is to look at how Cezanne built up a painting. In tiny little daubs, each daub having a tiny tiny little tone different from the previous one. . . .

This makes a lot of sense!

In a short story or in a genre novel, characters can be one-dimensional. But if you want to create a three-dimensional character for a novel, you have to present that character in many ways; you can’t just tell the same story over and over. So, each daub is different. But they can’t be too different; they all have to be “in character.”

To draw an even closer connection to visual art, consider a long-form bande dessinée such as L’Arabe du Futur, where each character is drawn a zillion times, but each version is slightly different. You wouldn’t want this sort of subtlety in an Agatha Christie mystery, but in a novel by John Updike or Richard Ford or Meg Wolizer or whatever, you want that kaleidoscopic depth.

A connection to statistics

There’s a principle in psychometrics, or in statistical measurement more generally, that if you’re creating a composite score, you want to build it from components that are positively correlated with each other, but not perfectly correlated. If two components have perfect correlation, there’s no point in measuring both of them. If their correlation is zero, then they’re not measuring the same thing and they probably don’t belong in the composite score.

So there’s this interesting thing that you want different items to be positively correlated, but not too much so.

This seems very similar to Toibin’s idea that the different depictions of a character in a novel will each tell you something new, but they will be similar to each other because these are all describing the same person.