From: Saul Perlmutter (saul@LBL.gov)
Date: Sun May 18 2003 - 12:49:00 PDT
I think I agree with what both you guys are saying here. I'll also go back and
look at the hstpaper archive email that I missed to see if anything there changes
my opinion.
Greg Aldering wrote:
> >From robert.a.knop@vanderbilt.edu Sun May 18 11:04:04 2003
> >To: Greg Aldering <aldering@panisse.lbl.gov>, Saul Perlmutter <saul@lbl.gov>
> >
> >On Sun, May 18, 2003 at 12:47:16PM -0500, Robert A. Knop Jr. wrote:
> >> > Given that, we have to reconsider what purpose the low-extinction subset
> >> > serves. We are using it because models tell us that there should be
> >> > a ridgeline of low-extinction. Therefore, we a supposing that as long
> >> > as we throw out extincted SNe, nature is guarenteeing similarity in
> >> > whatever small residual amount of extinction that remains. We have
> >> > tested whether this assumption holds for the low-extinction subset,
> >> > and we find that within our ability to measure, it does hold.
> >>
> >> Unless we want to put in some sort of prior assumption on the intrinsic
> >> extinction distribution for purposes of evaluating the systematic, then
> >> we can't do any better than that uncertainty of 0.015 or 0.025
> >> magnitudes-- which is going to give us something like an 0.1 systematic
> >> uncertainty in the flat-universe value of Omega_M on the low-extinction
> >> subset. I'd like to avoid putting in a prior assumption, after we spend
> >> all that time in the paper trying to discourage that sort of thing when
> >> doing statistical E(B-V) corrections.
> >
> >OK, thinking about this more--
> >
> >Of course, as has been previously noted, we *are* using a prior on our
> >low-extinction subset, that is E(B-V)=0+-0. This prior has the
> >advantage of being unbiased even if your error bars are different at low
> >and high redshift. (Sort of; in fact, there is an implicit bias,
> >because if one set has worse error bars, it will keep more mildly
> >extincted supernove than the other set. The prior doesn't *impose* a
> >bias the way the Riess one does)
> >
> >If we really want to be self-consistent and run with this-- basically
> >doing what you say, we've tested this assumption and it sure seems to
> >hold-- then we should use *no* host galaxy extinction systematic
> >whatsoever on the low-extinction subset. If somebody cares about
> >extinction, then they look at Fit 6.
> >
> >This approach may actually make the most sense.
> >
> >Thoughts?
> >
>
> Yes, I've been mulling this over too, and I think the concept you have
> outlined is the way to go. The key aspect that is different than what
> Riess et al does is that we don't need to know the functional form of
> the prior, and as you say, we don't bias our fits - at least not in the
> same way.
>
> Elaborating further, I would also say that our assumption is even more
> generous than E(B-V)=0+-0. It is more like:
>
> | <P(A_B,z=low | E(B-V) < XX)> - <P(A_B,z=high | E(B-V) < XX)> | << stat error
>
> where P(A_B) is the extinction distribution, and XX indicates the reddening
> cut that makes the low-extinction subsample, and < > indicates the mean of
> the distribution. I put in the "<< stat error" to indicate the general case
> that people really only worry about systematics if they think the size of the
> systematic is becoming appreciable relation to the statistical error.
>
> Let's see if others are on board with this.
>
> - Greg
This archive was generated by hypermail 2.1.4 : Sun May 18 2003 - 12:48:26 PDT