From: Alexander Conley (AJConley@lbl.gov)
Date: Wed Nov 03 2004 - 13:06:08 PST
Hi Greg,
I have tried to look into the question of the 'running sigma' in the
resid versus resid plot. Yesterday I told you that doing the ML fit
had bumped the significance up to about 2 sigma. I noticed
that this was heavily influenced by a few points with really large
errors. Since I don't take the errors into account (and in fact am
not sure how I would), this was a little troubling. So I placed
a cut on the errors of 0.4 mags (removing only 3 SNe) and
the significance dropped back to 1.5 sigma. This is probably
a more believable result.
But 1.5 sigma is still a detection, and is still worth trying to
understand. Another thing I did was to break down the sample
by redshift and try to figure out what was driving the running
of the sigma. It seems to be mostly driven by the low redshift
sample, which has a non-zero slope at the 1.5 sigma level.
The high redshift sample shows a much smaller slope, but
the significance is less than 1 sigma.
To be more precise -- if the sigma is written
sigma_0 + a*x
where x is the Bbv resid, then
complete sample: a = -0.099 +- 0.063
low-z : a = -0.141 +- 0.110
high-z : a = -0.074 +- 0.077
What does this mean? Well, besides confirming that positive residual
means dimmer, I still don't really have any good ideas. If anything
I would expect a weak effect in the other direction -- dim SNe should
have larger photometric errors, and hence scatter slightly more.
I guess in some sense this is telling us that intrinsically dim SNe are
more homogenous, even after stretch correction -- at about the 1.5 sigma
level of detection.
As a reminder, we have a SCP collaboration meeting on Saturday.
The first item on the agenda is me unblinding. I'd rather not have
to ask Tony and/or Saul to remove me from the agenda, so I need
your responses.
Alex
This archive was generated by hypermail 2.1.4 : Wed Nov 03 2004 - 13:06:04 PST