Precisely. BUT it's interesting. If I make 6 versions of a track, each 1dB apart (intensity, not loudness, note), any comparison of two adjacent (in level) tracks will almost always favor the higher level track.
***BUT*** Comparing the highest level to the lowest level may very well (not always, of course) have the listener pick the LOWER level track.
It's not a transitive property.
Yup, I *almost* hedged 'louder' with the phrase 'to a degree' but I was unsure what that degree is. Has it ever been studied? I'd guess it to have some content dependency too.