The Sandwich Argument

by Chris_Leong1 min read9th Apr 20204 comments



Today I saw a very interesting argument pattern.

Normally I'd expect some evidence for X. I don't think there was any evidence for X apart from Y. So in conclusion, there was no evidence for X.

It's well known that people tend to remember the start and end more than the middle. This pattern is optimised so people to believe there is no evidence for X.

The middle part is where it gets interesting. If the statement had just said that there was no evidence for X then the author would have looked like they were ignorant of Y. It would have left them vulnerable to a comment saying "What about Y?".

Mentioning Y indicates they aren't ill-informed and blocks this obvious reply. But rather than addressing Y, they have merely used some sleight-of-hand. Their argument doesn't really account for Y; it merely pretends to.

Have you seen this pattern before? If so, please comment below.

See also: Motte and Bailey, Shit Sandwich

4 comments, sorted by Highlighting new comments since Today at 6:56 PM
New Comment

Here's my steelman of this abstract argument:

My prior probability for not-X is very high, and the evidence for X is so weak and scant, it might well just be coincidence, cherry-picking, or a data artifact. In the interest of brevity, I'm going to round this off to "no evidence." By phrasing it this way, I transmit my confidence to others in order to avoid a stupid debate over a non-issue.

In real terms, this is exactly the form of argument I think is appropriate against claims of homeopathy and psychic powers.

In my original version I justified phrasing it as "no evidence" by saying "The research of Philip Tetlock shows that forecasters achieve better Brier scores when they exaggerate their confidence." I no longer endorse this.

The research of Philip Tetlock shows that forecasters achieve better Brier scores when they exaggerate their confidence.

They showed that it's good to extremise the predictions of teams, when combining predictions that agree with each other, but I don't think that individual forecasters were systematically underconfident.

Ah, interesting! Thanks for the catch.

really hard to generalize this. When I've seen it in small-group discussions among people I respect, and when I've used it, it is usually the case that Y is something we've separately discussed and dismissed as not very good support of X, but we have a few pedants who'll bring it up every damn time if we don't remind them that we've addressed it.

When used persuasively or in large-group announcements (propaganda), I suspect the implication is "and Y is irrelevant to X", which may or may not be true. Objections to the argument should be of the form "Y is sufficient evidence of X to invalidate your argument" (or "you're still ignoring Z, and the combination of Z and Y is sufficient to support X").

Best example I know: