I've used Pandas a bit and it certainly works well, but there are a lot of library functions to learn while J seems to basically be learn the two uses for every symbol and know how to do all math stuff ;).

I'm always impressed to see derivative and both matrix inversion & transpose as elementary operations instead of googling for a library that is most likely poorly documented. Furthermore, since the vector is the basic data type, things like matrices crop up a lot more in J & APL. In Python they are for scientific pursuits. In J they will often crop up even if you're writing a blackjack game. I think this is why it doesn't appeal to many programmers without a strong mathematics background, but it's very appealing to those who have that kind of background and grow tired of writing incessant loops.

In all honesty it's more a comfort thing: I'm so used to vanilla Python that the extra stuff in Pandas messes with my head. It's just easier for me to work in a completely different language.

Also J is more fun as a puzzle language, so I use it for personal projects. If I was writing stuff for production I'd probably (if begrudgingly) use Pandas instead.