The way I like to think of it is this example. For clarity of explanation, let's assume these ones are mutually exclusive. (IRL, they may overlap but need not be). Let's assume the user wanting to increase privacy/security/anonymity is the 'sender', and the receiver is your threat model.

Privacy - you receive an unlabelled box. Upon opening the box, you can't exactly tell what is in there, only something. There's a receipt, but you can't read it.

Security - you receive a box from a sex toys store. You try to open it, but it takes way too much effort, so you don't bother.

Anonymity - you receive a box from a sex toy store. Upon opening it, you only find a sex toy, but don't find a receipt.

Bad practice would be to receive a box, clearly labelled as from a sex toys store (and it also is a sex toy). It's easily opened and you find a receipt telling you who ordered it.

Good practice is that you receive an unlabelled box, can't open it, and if you somehow managed to open it, you don't find a receipt. At most, you'd find there's something in there, but what?

Would that be a good illustration?