Re: Low-z stability test

From: Alex Conley (aconley@panisse.lbl.gov)
Date: Fri Mar 14 2003 - 10:25:02 PST

  • Next message: Saul Perlmutter: "Re: "Error floor""

    Using chisquare as your decision rule can actually be a bit dangerous.
    The problem is that for very high quality data, the chisquare is always
    awful because our template just isn't good enough. The other issue
    is that the accuracy of our template is really a function of epoch --
    it's more poorly constrained towards the tail, better near the peak, etc.
    What we really need is some sort of error-snake like Adam Riess -- except
    done correctly of course!

    Using the chisquare directly as your decision rule can result in fits
    which are completely driven by a bunch of tail data and miss the peak
    horribly. This is a bad thing because what we care about, and the part
    of the template we kind of trust, is near the peak. If we understood
    the template errors better we could formalize this, but as it is I you
    can harm yourself greatly by trusting the chisquare to be meaningful.

    Alex



    This archive was generated by hypermail 2.1.4 : Fri Mar 14 2003 - 10:25:04 PST