The "right" Alpha and Alpha fitting method

From: Robert A. Knop Jr. (robert.a.knop@vanderbilt.edu)
Date: Wed Mar 26 2003 - 14:27:23 PST

  • Next message: Greg Aldering: "HST final refs refered to in today's telecon"

    My current leaning is to do the fits by allowing the alpha used in error
    analysis to vary right along with the alpha being tested. This is the
    cleanest and "most obvious" way to do it.

    For P99, we calculated errors using a pre-fixed version of alpha, which
    then turned out to be larger than the most probably version of alpha.
    The rationale was twofold. First, the fits tended to like to have
    larger and larger alpha because it expanded error bars. Second, there
    is a built-in bias: you're more likely to find high stretch supernovae
    in greater extinction, which would tend to *flatten out* your stretch
    distribution, thereby giving you a falsely low alpha; however, you'd
    expect the stretch uncertainty to propogate into a magnitude uncertainty
    via the true slope.

    In my (current) opinion, the procedure is cleaner if you just let error
    alpha track the real alpha-- that way, you're really propogating your
    errors as directly as possible. This opinion, however, complete ignores
    the aforementioned systematic effect. The question is, though, how does
    that systematic effect compare in the high and low redshift supernovae?
    Things get pretty complicated if you really try to fix an alpha to avoid
    this systematic. Additionally, I'm not sure I'm even seeing that
    systematic effect.

    I've fit just the low redshift supernovae, trying to figure out what
    they think is the best value of alpha. See

       http://brahms.phy.vanderbilt.edu/~rknop/scp/hst/#whatisalpha

    In particular, when I throw out the worst outlers, I get a very similar
    alpha with and without extinction corrections. If the aforementioned
    systematic effect were biasing alpha, this shouldn't be the case.

    Note that this alpha I get is very similar to the alpha I get when I fit
    the "primary set" including the high-redshift supernovae, allowing the
    error alpha to track the testing alpha. (Unsprising: you'd expect the
    low redshift supernovae, with their better stretch measurements, to
    domiante this.) Since alpha isn't balooning out of control, and since I
    get what seems to be a "reasonable" value of alpha, this approach
    doesn't seem loony. And, given that it's the simplest one to explain
    (i.e. we did what was obvious, rather than something non-obvious based
    on a systematic bias that requires explanation), it's easier just to do
    this.

    To see how letting erroralpha slide affects the cosmology, compare line
    B-4 on the hst web page to either line B-1, or to line A-1 (filled) or
    line A-5 (stroked). It makes enough of a difference-- and it's not *so*
    obvious which really is exactly the right way to do it-- that we have a
    fairly sizeable "fit method" systematic along the direction of the major
    axis of the confidence ellipse.

    Comment if you have a strong opinion on the matter.

    -Rob

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


    This archive was generated by hypermail 2.1.4 : Wed Mar 26 2003 - 14:27:24 PST