Re: Baseline Optimization Memo

Hi Josh,

Thanks for taking a look.

I agree that it is an over simplification to treat systematic and statistical
errors on the same footing. In particular systematic errors with energy
dependence, like Li9, are especially poorly suited for this kind of treatment.
But this is effectivly how the Huber result is derived, and I still have not
seen a credible treatment of correlated isotope decays in the context of shape
analysis. The main point is that the baseline optimization is very sensitive
to the assumptions that you make about systematic errors, so it importtant to
study that scaling in detail.

I'm a little confused by your statments that you'd "rather be statistically
limited" and "build the biggest experiment you can afford." These two points
of view seem to be inconsistant, so perhaps I did not make myself clear in the
memo. The statistics limited case is where you can gain in a simple counting
experiment by increased statistics. If you are in the systematics limited case
then it does you no good to take more data, at leat in the context of a rate
only analysis.

Thanks,

Jon

Josh R Klein wrote:

> Hi, Jon,
> The memo looks very nice. One comment I have is that
> we probably should not treat systematics and statistics on an
> equal footing---for the same total uncertainty, I'd rather be
> statistically limited, since we know how those uncertainties are
> distributed, and we usually don't know for systematics. In other
> words, build the biggest experiment you think you can afford, and
> place it at a baseline which minimizes the absolute systematic
> uncertainty, not the ratio to statistical uncertainty.
>
> In any case, I suppose I would argue for the slightly closer baseline if
> that allows a better rate+shape analysis. I think that although the sources
> of uncertainty are larger for such an analysis (relative energy
> non-linearities and background shapes matter much more), the additional
> information should provide an overall reduction in the total systematic
> uncertainty (not to mention, be a more believable measurement in general).
>
> Also, Bayes would say that dM2 is likely to go up from where it is,
> not down, since that is where it came from...though I suppose I wouldn't be
> willing to bet too much money on it either way.
>
> Thanks,
> Josh
Received on Fri May 14 13:45:08 2004