Re: Comments on today's draft. --Here they are.

From: Robert A. Knop Jr. (robert.a.knop@vanderbilt.edu)
Date: Mon May 19 2003 - 05:28:42 PDT

  • Next message: Lifan Wang: "Re: SN 1998ax"

    Some thoughts --

    P. 3, second comment. I don't think we *need* to distinguish from the
    priored extinction correction here. We have "unbiased" in there, which
    is all that's necessary. We beat that to death later. I wouldn't
    change anything in that sentence.

    P. 13: I put that in because Greg was confused about the blue side. IT
    looks like it didn't work, because you are also confused. Here's the
    deal: error bars are on points. Reddening means that points can be
    redder than the fit, or (equivalently) that the fit can be bluer than
    the points. Thus, you want more freedom on the error bar that's on the
    *blue* side of the point, *not* on the red side of the point. The
    latest iteration didn't use an error bar of a different size on each
    side, but did use the "stretchy" error bar. I can just describe that
    and not use the word extendable if you don't like it.

    P. 18, right column. I agree with Greg. We don't need to take credit
    for the Bayesian prior; not only don't we want to, but that's a bit
    disingenuous because the *only* reason it was in that paper was in
    reaction to Riess. Let's leave this the way it is, where we say Riess
    did it, and mention Fit E in P99 later. Re: a paragraph describing why
    the Hatano prior isn't crazy, I can't write that, because I have a
    gut-feeling belef it *is* crazy. I haven't read Hatano, but just
    looking at my E(B-V) vs. s distribution for the low-redshift supernovae
    shows me that there are lots of supernovae at relatively high
    extinction, more so than I would intuitively expect from looking at the
    Hatano prior. (I would have to look more carefully; probably something
    with a huge peak at 0 and a long tail does make sense, but I haven't
    thought enough to know if I believe that that tail is high enough.)
    Moreover, as Don is always reminding us, using a prior like that is
    *always* crazy because you're losing track of the difference between
    measurement uncertainties and the intrinsic distribution.

    In any event, even if the prior itself may be likely to describe a true
    E(B-V) distribution, I would really rather not spend any doing anything
    that sounds like defending use of this prior in E(B-V) corrections. If
    we are going to put something in defending the use of a low-extinction
    set as likely to be unreddened due to the Hatano information, let's put
    that in a different place to clearly distinguish that approach from
    something that might look like it's defending the Riess 1998 approach.

    Page 21: I'd sort of like to leave a reference to the Phillips RB in
    there, because I believe that a lower RB is almost certainly going to
    figure into the future of extinction corrected supernovae. We decided
    not to use it here to avoid getting in trouble with this paper, but
    there should be little nudges as often as possible to hopefully get the
    community used to the idea that maybe it's not crazy.

    P. 28: 95% isn't really consistent with paragraph 3 below since they're
    such different things; one is the SNe alone, the other is the SNe
    combined with other things, and the shape of the regions are extremely
    different in each case. People who *don't* understand that will get all
    hot and bothered by the 95% CL's, since for the primary set you end up
    with a 95% limit somewhere w<-1 for the supernovae alone, but much
    higher when combined with the other two. "But wait, how can adding more
    constraints make your data *less* limiting!?!" is what those who don't
    really think through the probabilities will say. (I was listening to a
    neutrino talk a month or two ago where the guy was pointing out exactly
    the same sort of complaints having been raised with some limit or
    another, when future measurements moved confidence regions outside
    former confidence regions, and he spent time talking about probability
    and such. People *will* be confused on this issue if there's an
    apparent-but-not-really sore thumb like that stikcing out.) Quoting a
    99% limit here both gives us a more stringent limit on what just the SNe
    alone set, and don't obviously open that Pandora's box. Antoher
    alternative would be to quote the SN upper limit with some w prior,
    e.g. w>-1 or w>-2. (The extinction corrected 99% confidence region
    extends somewhere below -80-- that's as far down as I've fit it.) I can
    do whatever you want here, but I wanted to make sure you know what
    you're going to see before it shows up in the next draft.

    -Rob

    -- 
    --Prof. Robert Knop
      Department of Physics & Astronomy, Vanderbilt University
      robert.a.knop@vanderbilt.edu
    


    This archive was generated by hypermail 2.1.4 : Mon May 19 2003 - 05:28:45 PDT