Image: Skinny-leg jeans. Not my legs. Or my jeans. © Claude Truong-Ngoc CC BY-SA 3.0, via wikimedia.org
I went shopping for jeans last week, and came home frustrated. (As usual, yes, I’m eventually heading somewhere.) I have calves of considerable circumference, and the fashion in men’s jeans now seems to be for a very narrow-cut leg. I took pair after pair into the fitting room, only to discover I couldn’t even force my leg through the available hole. I know, hold the presses – I’m old and I don’t like today’s fashion; and while we’re at it, all you kids get off my lawn!
But from my (admittedly weird) utilitarian point of view, I just don’t understand skinny-leg jeans. Here’s why. If you make a pair of skinny-leg jeans, they can be used by a skinny-leg person, but not – not even a little bit – by a non-skinny-leg person. If you make a pair of wide-leg jeans, they accommodate both. There’s a fundamental asymmetry in usefulness that makes it seem obvious, to me, how jeans ought to be sewn.
The same asymmetry is why I teach students to report exact P-values, not just “P<0.05” or “P>0.05”.* There are two ways one can think about P-values in statistical hypothesis testing. You can see a P-value as a continuous measure of strength-of-evidence, or you can see a P-value as an absolute line-in-the-sand measure in which all that matters is which side of a pre-set alpha level (conventionally, 0.05) it lands on. Neither view is silly (despite the derision heaped on phrases like “nearly significant” that emerge from the first view). Here’s the asymmetry: if you hold the continualist view and report an exact P-value, an absolutist reader can easily discard the unwanted extra information and pay attention only to “Is P>0.05?”. But if you’re an absolutist and report only “P>0.05”, a continualist reader is hooped: the extra information (for them) is lost and unrecoverable. Exact P-values are wide-leg jeans; “P>0.05” is a pair of skinny-leg ones. So, whether you want them or not, please always report exact P-value.
And because P-values aren’t controversial enough, how many spaces after a period? It’s remarkable how heated people get on this topic, especially when the answer is, obviously, two**. Now, I can hear the furious gnashing of teeth from here, so let me explain. We can debate the functional superiority of one space or two (very conveniently, there seems to be plenty of opinion and almost no data bearing on the question; if you really must, you can make your case in the Replies). But the same asymmetry in flexible use that condemns skinny-leg jeans and “P>0.05” weights in against the single space. If I use two spaces in my manuscript, and a journal or, lord help me, a co-author*** prefers one, well, that’s a quick global search-and-replace. But if you use one space and I prefer two, there’s no similar tool. The search-and-replace strategies fails, because there are lots of other places where period-space occurs****, and so we get oddities like vs.-space-space and Dr.-space-space and the like. A single period is a pair of skinny-leg jeans.
So: jeans, P-values, and spaces after a period. These things may seem unrelated; but they’re united by the principle of asymmetric usability. And by the fact that I’m nerdy enough to stitch them together.
© Stephen Heard January 7, 2019
*^By “exact” I don’t mean “to eight decimal places”. That level of exactness serves no purpose – although gosh, it’s common in student papers, submitted manuscripts, and other places it ought not to be. This and a lot more about how best to write statistical results in this post.
**^Before you count and declare me a hypocrite, I do put in the second space. WordPress then strips it out again. Which, as you’re about to discover, is exactly the point.
***^You know who you are.
****^This is, of course, exactly the point of using two spaces after a sentence: to distinguish the sentence-ending full stop from other uses of the period. But wait, I rolled my eyes at the notion of us actually debating the functional merits… Oh well, too late.