I do multi-unit electrophysiology, and although I agree that many different
analyses are necessary to analyze ensemble properties, the difficulty is in the
simple ignorance of the important response variables. The functional anatomy
for any specific neuron is rarely well known, and the particular response
properties for that neuron can only barely be sampled. So while little activity
in the end may be noise, much of it is noise to the researcher.
Defining the engram/pathway used in a given task often starts with some form of
whole brain evaluation technique, whether cellular labeling or fMRI. Both high
resolution-low volume and low resolution-high volume techniques have a long way
to go in development terms, so I wouldn't argue at this time which is more
capable. Ultimately, I believe they will work hand in hand.