LL's comments on features of DL+PLC (Diagonal Line, with Power Constrained Limits)
and F-C (Feldman Cousins)

Here is my list of features of the 2 approaches to the simple Gaussian problem. (The last 2
items are for possible extensions of the problem.)  I have tried to be factual, and would 
appreciate clarifications/corrections. Of course, not all items are of equal significance, but
ranking them is a matter of opinion.


                                             DL+PCL                                                        F-C

Basis of method            Ad hoc (resembles van Dyk)                Likelihood ratio (Kendall + Stuart)

Arbitrary features          Choice of cut-off

Upper Limit                    For x not too negative, tighter              Weaker UL because 2-sided
                                        than F-C (by 0.3 sigma at large x)     

x=5                                 UL= 6.3                                                   mu = 4.4 --> 6.6

Negative x                     UL constant  for x < cut-off                    Decreases slowly as x decreases

Exclude mu = 0 ?          Never. Needs alternative method        Unified approach to UL, 2-sided interval
                                       (p_0 excludes at same x as F-C)         and discovery


Ease of calculation      Yes                                                          Harder

Usage                           Common (with variant of PCL)            Less so at present

Systematics                          ?                                                           ?

Identifiable subsets     None?                                                     None?

Coverage                     Overcovers at small mu                        Exact coverage

Multi-D data                 Not usable (except via L-ratio)            Extendable, provided not too many

Poisson problems       OK?                                                   Typical overcoverage for discrete data