# The Font of Wisdom

on August 10, 2012 at 3:26 pm

1 ed August 10, 2012 at 6:08 pm

Something about this piece sets of my BS detector. Can we really reliably distinguish the “effects” of these fonts, especially the ones that look fairly similar (i.e. not Comic Sans)? What, exactly, is the hypothesis being tested?

I’d like to see what Andrew Gelman would say about this:
http://andrewgelman.com/2007/11/statistical_cha/

2 Jason August 10, 2012 at 7:00 pm

The weighting factor ratio 1:3:5 is much larger than the expected variance of the strong vs slight 1:1.5:1.4 so ends up skewing what seem to be random fluctuations.

I’m not sure from the article what specifically the given p-value is testing. To my math all of the responses are well within chance — and the subsequent strong vs slight measurement errors are correlated with the agree-disagree measurement errors.

Since Baskerville happens to have the one of the highest number of agrees (but still within chance) and one of the lowest number of disagrees (but still within chance) and that the strongly agree/disagree are going to be correlated with those statistical errors, hence Baskerville has the highest number of strongly agrees and lowest number of strongly disagrees — and then multiplying that effect by 5! — you’re going to get it at the top and bottom.

Basically all the agrees should fall within 4628 +/- 80 (which is 4548 to 4707) 68% of the time and only two results violate this bound by +8 and -23 which are <0.5% of the mean.

I'd say this is a case of improper statistics in breaking out the strongs vs slights.

3 mkt August 11, 2012 at 7:18 pm

Right, this is a dataset which cries out for ordinal statistics, not the ANOVAs based on arbitrary scales that Errol Morris’s statistical consultants performed. (A Mann-Whitney-Wilcoxon test perhaps?) I don’t have time to do the calculations but just eyeballing the raw numbers, those look like small differences to me. “Small” both in the sense of quite possibly lacking statistical significance as Jason calculated, and also in Andrew Gelman’s sense, as cited by ed above.

4 Adrian Ratnapala August 10, 2012 at 6:10 pm

The article says that “Georgia” is the font used by the NYT. This on an NYT website. Do you think the sample population might be a little bit skewed?

5 Enrique August 11, 2012 at 12:11 am

For what it’s worth, I noticed that MR uses a number of different fonts. Is this a hedging strategy, or just random noise?

6 Bill August 11, 2012 at 3:42 am

+1, but can you believe this comment because it was written in sans serif?

7 WL August 11, 2012 at 12:13 am

Secret weapon of the Chicago school via the JPE.

8 Sonia August 11, 2012 at 8:08 am

How do they know its the font of wisdom/truth and not the font of optimism? (Or something else relateed to the 1 question he used)

9 Baphomet August 12, 2012 at 7:59 am

You mean typeface, not “font.”

10 Shortsighted August 13, 2012 at 1:16 am

It may just be that for a given font size in points, Baskerville looks smaller and is definitely harder to read. Does the extra time or effort needed to read it result in readers accepting the message more readily?

Comments on this entry are closed.

Previous post: