The main point of the context is that it's one of four models he presents.

It's perfectly fine to model every cow as being spherical. There's nothing irrational about saying that you have a model where every cow is spherical. It's also not bad writing.

Different models have different usefulness.

I'm not particularly convinced that dentist anxiety would be any better in a world where yelling at your dentist for hurting you were considered socially acceptable, though.

In that issue anxiety is produced when your system I considers going to the dentist bad because he will hurt you but your system II drags you to the dentist. Simply yelling at the dentist doesn't resolve the dilenma.

To actually release the anxiety you need to reconcile your system I and system II. Depending on how good you can do that, your system I can also shut down the pain response so that the dentist doesn't have to give you anesthesia.

Surely in many cases, anxiety is a direct result of perceived danger, or of anticipating or being confronted with scary things.

The standard word for the emotion that people usually fear as result of direct perceived danger is fear.

Apart from that you miss the deeper point. If you look at a person getting afraid in an elevator there something that distinguishes them from other people who don't get afraid in an elevator. People don't randomly develop claustrophibia out of nothing.

A person who constantly suppresses his emotions is more likely to develop claustrophibia. Therapeutically it's useful to work on the topic of expressing one's emotions instead of being nice to overcome the issue. There are more direct and faster ways to cure claustrophibia but that doesn't mean that the hidden emotion model isn't applicable. That's why it's in David Burns book.

There we are at: "Scientists who write a claim that a lay person disagrees with are bad writers because they obviously don't mean what they claim."