Relate loglike_unit and loglik_pop with contraction, containment, compromise?

[hyunjimoon]

automatic explanations to the ECDF plots

• Perhaps replacing this with left/right tail thin/heavy and then having a separate descriptor for crossing the bound near the center as "mean over/under estimated" would be better?
• detecting local spikes and lack of data, as these may be highly suspicious, but not bring the ECDF out of the bounds.

i) contraction: unit_layer (likelihood) is more informative, but consistent with population_layer (prior)
loglik_unit << loglik_pop
ii) containment: population_layer (prior) is more informative, but consistent with unit_layer (likelihood)
loglik_unit >> loglik_pop
iii) compromise: tension btw information encoded in population and unit
loglik_unit ~ loglik_pop

https://discourse.mc-stan.org/t/how-to-identify-which-datapoints-have-the-most-influence-on-the-posterior/29988/12