Thanks, I've read this piece before
. However, this for headphones and IEM, so not quite the same. Note that despite these preferences, Harman's IEM studies were still able to predict listener preference to a remarkably high degree (91 percent, I think). With headphones, there's more variation in the bass department because of people trying to replicate the tangibility of speakers, I think.
Harman's generalized speaker performance model was able to predict which speakers (of 70 tested) would be preferred over others with 86 percent accuracy. In an earlier, smaller test with more controlled parameters (13 speakers), they were able to predict preference with basically 100 percent accuracy. In both cases, flat on axis, smooth off axis, and bass extension were the three biggest contributing factors to preference.