As I wrote here, Errol Morris has for the last few years been writing the occasional brilliant series of articles in the NY Times, and he’s got another one going now; this is the first installment, and I commend it to the attention of anyone interested in the influence on our thinking of things that you wouldn’t think should influence it. I kind of hate to reveal even as much as I do here, but it’s only a spoiler for the first section and the fun stuff comes afterwards; he links to this earlier piece, focused on a quiz about optimism and pessimism, then says:
Here is my confession. My quiz wasn’t really a test of the optimism or pessimism of the reader. There was a hidden agenda. It was a test of the effect of fonts on truth. Or to be precise, the effect on credulity. Are there certain fonts that compel a belief that the sentences they are written in are true?
For the exciting answer, read the essay (and I assure statistically minded folks in advance that the p-value for the main result is 0.0068). Oh, and there’s a nice bit on the Crimean War, too. Thanks, Nick!
It’s interesting, but I can’t help feeling that a certain amount of data-massaging was needed to arrive at the desired result, with Baskerville on top and Comic Sans at the bottom. It’s good statistical practice to decide on how to analyse results before the analysis and not after.
Thanks for that comment, Athel. I suspected that there was a lot of accomodating, change-argumentative-tactics-in-midstream stuff going on, otherwise known as hand-waving. The familiar principle, in your words: “decide how to analyse results before the analysis and not after” – is the one most commonly offended against in popular presentations of statistical analysis, the only kind I usually see. That has been my impression, in any case.
I don’t understand why these popular presenters bother at all. The layman can’t make out what is going on – not understanding the special, technical sense of “confidence”, for example. David Dunning, the analyst, puts crucial notions in what seem to be scare quotes, as if he were not sure what to make of them, or – more likely – how to explain them without writing a book. So one has to ask oneself, what is the point of it all:
Any student of the graphic arts, and especially of typography, knows that shapes and colors affect the viewer’s or reader’s emotional response.
Here’s another, more humdrum explanation for the phenomenon: a few more people (5) took part in the test when they read the text in Baskerville, simply because it put them in a better mood to take part.
In other words, the font did not influence their beliefs, but their readiness to engage with yet another questionnaire. There are a lot of people who like the feeling that their opinion is being sought, especially – it seems – when it is sought in a well-fonted manner.
Scare– and emphasis quotation marks are fun! Just the day before yesterday, I had something described as
> Grilled “salmon”
at the local restaurant. Whatever it was, it tasted convincingly like salmon.
English is missing a word for “the day before yesterday”? http://en.wiktionary.org/wiki/day_before_yesterday#Translations
leoboiko: I don’t understand the complaint. At your link I find the familar word “nudiustertian”, which I use all the time. That’s only an adjective, though. What’s missing is merely a noun form, for which I propose “nudiusterdyday”.
Or perhaps “nudeturdday”.
I find it reassuring that apparently hardly anyone feels there is much worth saying about the “statistical analysis” that is the subject of this post.
I came away from the article, thankful as always for the gift of Morris’s careful explorations but still wondering what someone less ignorant about statistical data than I am, say Mark Liberman, would have to say about the conclusions so far. My doubt sensors were tingling mightily.
I’m neither optimistic or pessimistic about the propriety of the statistical analysis, but I am surprised that they don’t at all mention that looking at the effect of these fonts when used for a small passage surrounded by whatever the NYTimes usually uses is not the same as looking at the effect of the fonts when used throughout a document.