No true Scotsman

The disqualified cousin

Dana: Nobody who grew up in Minnesota complains about the cold.

Lee: My cousin grew up in Duluth, and she complains all winter.

Dana: Then she’s not a real Minnesotan.

Lee’s cousin meets the ordinary meaning of the group Dana named: she grew up in Minnesota. The only thing that disqualifies her is the very behavior the claim was about. If “real Minnesotan” means “a Minnesotan who doesn’t complain about the cold”, Dana’s claim has become “Minnesotans who don’t complain about the cold don’t complain about the cold”. This is the clear-cut case.

“Proper” code review

Sam: Teams that do proper code review don’t ship this kind of bug.

Jordan: The payments team reviews every change, and they shipped exactly this bug last month.

Sam: Then whatever they were doing wasn’t proper code review.

This sounds more reasonable than the first example, because “proper” code review really can mean something specific, and a team can go through the motions without doing it. But Sam hasn’t said what the payments team skipped. The only evidence offered that their review wasn’t proper is that the bug got through. If the criterion had been stated in advance (two reviewers, tests required, say) and the team had failed it, the reply would be fair. Stated only afterward, and defined by the outcome, it makes Sam’s claim impossible to test.

No “true” required

Morgan: Nobody in marketing understands how the product actually works.

Riley: Priya’s in marketing, and she wrote the clearest technical spec we had last quarter.

Morgan: Priya’s basically an engineer, though.

The word “true” never appears, but the move is the same: the counterexample is relabeled out of the group. Priya is still in marketing. Morgan’s reply keeps the generalization by deciding, after the fact, that anyone in marketing who understands the product doesn’t count.

• The explicit form: adding “true”, “real” or “proper” to the group’s name after a counterexample. “No true fan would leave early.”
• Relabeling the counterexample: moving the inconvenient member into another category without changing the group’s name. “He doesn’t count, he only moved here for work.”
• Monster-barring: the philosopher Imre Lakatos’s name for a move in mathematics: handling a counterexample to a conjecture by redefining the objects the conjecture is about, so that the counterexample no longer qualifies. Andrew Aberdein gives the no true Scotsman as a familiar example of the same kind of technique outside mathematics.

The name comes from an example the philosopher Antony Flew used in Thinking About Thinking (1975), in which a generalization about Scotsmen is saved from a counterexample by retreating to “no true Scotsman”. Maheshi Gunawardane argues that Susan Stebbing had already described the pattern in 1934, without the name: someone who claims that no musicians nowadays admire Wagner, and, when challenged, replies “Well, no true musician does.”

Is it a kind of special pleading? The two are close, but this site doesn’t file one under the other, because the references it relies on don’t. Bradley Dowden’s fallacy list in the Internet Encyclopedia of Philosophy classes the no true Scotsman as a kind of ad hoc rescue (an assumption added only to save a claim from refutation), and treats Special pleading as a form of inconsistency. They differ in what gets protected. Special pleading exempts a case from a rule or standard that should apply to it. A no true Scotsman removes a case from a group so that a generalization about the group survives. The two overlap when the redefinition is used to exempt a favored group from a standard, as in “no real member of our club would do that, so the complaint doesn’t concern us.”

Saying “that isn’t really an X” is often correct. It becomes a no true Scotsman only when the definition is adjusted to absorb a counterexample.

• The definition was stated first, and not tailored to the claim. If a group has an established or announced definition, pointing out that someone doesn’t meet it is fair, whenever it comes up.
• Membership can be checked independently of the behavior in question. A definition that doesn’t mention the behavior the claim is about can’t be doing the claim’s work for it.
• The claim was definitional all along. “Bachelors are unmarried” isn’t refuted by a married man who calls himself a bachelor, and nobody pretended it was a finding about bachelors.
• The generalization is openly revised. “Fair point. I should have said most people who grew up there” concedes the counterexample instead of defining it away.
• The definition itself is argued for. Disagreements about how to define a word are real disagreements. Pruś and Aberdein, following Douglas Walton, suggest treating a definition offered in a dispute as an argument that can be evaluated like any other. What matters is whether reasons are given for the narrower definition, apart from the fact that it saves the claim.

The test: could you tell whether someone belongs to the group without first checking whether they do the thing the claim is about?

The vegan who eats meat

Alex: My coworker says he’s vegan, but he has a steak for lunch most days.

Casey: Then he isn’t vegan, whatever he calls himself.

Casey is redefining nothing. “Vegan” had a public meaning before this conversation (someone who doesn’t eat animal products), and it isn’t tailored to rescue some empirical generalization about vegans. The coworker simply doesn’t meet it. This is the definition stated first condition: the claim that vegans don’t eat meat is true by definition, and everyone knew that before the counterexample showed up.

A policy written down in advance

A team’s written policy, adopted in January, says a change counts as reviewed only if two people approve it and the tests pass. In December someone says: “No reviewed change caused an outage this year.” A colleague points to the March outage. The records show that change was merged with one approval under an emergency override. “Then it wasn’t a reviewed change under our policy.”

This sounds like the “proper code review” example above, but the criterion existed before the outage and can be checked in the records without knowing whether the change caused problems. The team isn’t redefining “reviewed” to protect its claim; it’s applying the definition it already had. That’s the checked independently condition.

The generalization being defended may have been a Hasty generalization to begin with, which is why a counterexample turns up so easily. And generalizations about groups often aren’t held as ordinary factual claims. They express pride in a group you belong to or admire, or suspicion of one you don’t, so a counterexample feels like an attack on the group rather than a correction to a claim. Tuomas Manninen describes the arguer as presuming “to be the authority on determining what it takes to be a member of a certain group”.

The move is also easy to miss because words like “true”, “real” and “proper” sound like clarifications. Pruś and Aberdein give the no true Scotsman as an example of a persuasive definition: adding an evaluative word such as “true” to a term changes what the term covers, while the change presents itself as merely making the meaning more precise. The speaker may not notice either. Adjusting a definition feels much less costly than admitting a claim was too broad, and the result (a claim that can no longer lose) can feel like a win.

It has the same structure as the ad hoc rescues described under Unfalsifiability: the saving assumption is added only after the failure, and can’t be checked apart from the failure it explains.

• Ask what the narrower group means, in terms you could check without looking at the behavior in question. “What would make someone a real Minnesotan, apart from not complaining?”
• Ask whether that definition would have been given before the counterexample. If it would have, the reply may be fair after all.
• Offer the honest revision. “So is the claim that most people who grew up there don’t complain?” A narrower claim that can still be wrong is often worth keeping.
• In yourself, notice when “real” or “true” appears right after someone has given you a counterexample, and ask whether you would have used that word before.