The practical reason: https://3v4l.org/LEar1 (No SIV in that list.) It's also not present in libsodium or (shudder) mcrypt.

What property are you hoping to extract out of AES-SIV in particular? (And would AEZ, HS1-SIV, AES-SIV-GCM, etc. solve your problem?)

From what I understand, SIV mode still uses a nonce, but it doesn't explode if you reuse it for different messages. Are you aiming for nonce-less encryption?

The end goal that was being worked towards in the article (Argon2-derived Bloom filters for partial searching) is a little more useful than nonce-less encryption.

I was specifically after a deterministic encryption scheme that would allow for exact-match comparisons and whose deployment was fairly foolproof (so a non-expert couldn't screw it up too badly), which AES-SIV appears to provide nicely. Another benefit in my case is that I'm using a NoSQL database where separating the encrypted value from its indexed value is not always practical, so having a single value that serves both purposes is highly desirable.

I'm not familiar with the other schemes you mentioned, but the practical reason for using AES-SIV is the converse of yours for not using it: there's an implementation for JavaScript. :) It's unlikely there are implementations for the other schemes you listed.

I agree that the Bloom filters are a nice natural outcome of your approach, though it would also be possible to combine AES-SIV for primary encryption with a truncated HMAC for partial match Bloom filters.

Here's the killer feature of my approach: Each filter has its own HMAC/Argon2 key, so it becomes less useful for what I like to call crossword puzzle attacks. (I hope that's easier to understand than a wordy explanation of how such an attack would work.)