Re: Riess' prior

From: Don Groom (deg@panisse.lbl.gov)
Date: Thu Feb 27 2003 - 13:00:46 PST

  • Next message: Robert A. Knop Jr.: "Re: Riess' prior"

    Please forgive me for thumpiing an old tub, but the E(B-V) prior business
    stirs up some deep religious matters for me...

    There is a big difference between a physical quantity and the value we
    measure, and they have to be kept separate. There is a "true" E*(B-V), and
    our measured value E(V-B). One can say that E*(B-V)<0, but it is incorrect
    to say that E(B-V) < 0. For example, suppose that E*(B-V) = 0. Then our
    measured values are negative as often as positive, distributing about 0
    as per our measurment uncertaintly. If we impose the requirement E(B-V)<0
    on these experimental numbers, then the average is *always* postive, in
    other words, badly biased. That was at least what was happening in some of
    the early arguments we've had on this. If somebody introduces a prior
    on E based on physics, he's pretending that E* = E. That's useful of
    course,
    for saying what might happen in future experiments, but not in
    interpreting data.

    The usual example is the square of the neutrino mass, m^2. If somebody
    thinks the measured value should be positive, she will be sadly misled.
    If she says that m^2 < 0.5 eV^2 at the 90% CL based on either Bayesian or
    frequentists arguments, then there's a useful limit to use in designing
    future experiments. But in using that result, the experimental number m^2
    = 0.10 +/- 0.15 eV^2 (or whatever it is) must be used.

    The halfway house is to think of the experimental distribution of E(B-V)
    as the physical E*(B-V) distribution convoluted with the experimental
    resolution. But kicking out negative results is dangerous nonsense.

    Praises be to Neyman...

    Don

    |-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|
    Don Groom (Particle Data Group, Supernova Cosmology Project)
    DEGroom(at)lbl.gov www-ccd.lbl.gov Voice: 510/486-6788 FAX: 510/486-4799
    Analog: 50-308//Berkeley Lab//Berkeley, CA 94720



    This archive was generated by hypermail 2.1.4 : Thu Feb 27 2003 - 13:01:05 PST