No hard and fast rule here. I developed my own workflow similar to Audacity's, but simplified... most of my records are in pretty good shape, so I don't bother with HF noise reduction, and I only use "click removal" between tracks, preferring the result from "repair" with the few clicks I usually find... but I digress.
My arm/cartridge resonance is around 8 Hz. I'd guess 90% of records I've copied excite this to some extent. Sometimes it's 30 dB below the music, sometimes much higher.
I base my choices on how much musical LF info is present vs. the strength of the unwanted rumble. As noted, most rock/pop records don't have much below 40 Hz, but a few with synth bass lines might. Pipe organs can go down to 16 Hz (some huge ones even lower) but such frequencies are extremely rare on LP. Tympani in orchestral music may reach down to 30-ish... for records I know well, I can usually tell by ear, having played them on a number of systems.
For others, the spectrum analyzer tool is your friend. It takes a little practice to interpret, but by selecting sections of audio with a lot of perceived low bass, you can get a pretty good idea where the lowest musical tones are (and see how much infrasonic info is present.)
So if the music goes down to 30 Hz, I may select 22 or 24 Hz cutoff and 36 dB/octave. If nothing much below 50 Hz, I might use a higher freq, 36-42 Hz but at 24 dB/.
Theoretically, response is 3 dB down at the cutoff freq. An octave down is half that freq, so for 24 Hz at 36 dB/, 12 Hz will be down 39 dB, 6 Hz 75 dB, etc.
The reason for the different slopes is to minimize effects above cutoff where you can hear them. In particular, smearing of bass transients may occur, although those may not be particularly noticeable. Sometimes though, filtering may counterintuitively *increase* peak signal level, which can cause clipping even if your original recording level didn't.
You basically want to use as little as you can to minimize the problem, while retaining as much of the desired audio as possible.
Sorry if this is long...