RMS uncertainty on a data point

From: Don Groom (deg@lbl.gov)
Date: Thu Sep 30 2004 - 12:18:28 PDT

  • Next message: Tony Spadafora: "Fwd: Texas at Stanford Symposium"

    The following is a conjecture that I would like to look at further before
    implementing in any analysis. The idea arises in connection with the pdf
    note I recently distributed on the inappropriateness of fitting to
    weighted-average points.

    Suppose we have several images of an object obtained on the same night,
    all of which have the same exposure time.

    The usual deal is to take the variance as the number of electrons in
    the object (SN aperture, or whatever). The intensity is the number of
    electrons minus sky. [Never mind all the conversions and corrections for
    now.]

    However, this is not the true variance but our best estimator of the
    variance based on a single image. Can we come up with a better estimator?

    For convenience define the weight for the jth image on this night as
    w_j = 1/\sigma_j^2. The the mean w for the night is \overbar w =
    \sum w_j.

    I suspect that \overline w is a better estimator of the weight than w_j.

    If so, we are still making the approximation that the sky doesn't change
    throughout the interval in which the images are obtained. But we can't
    generalize to other nights, when the sky is likely to be different.

    To be continued,

    Don

    |-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|-+|
    Don Groom (Particle Data Group, Supernova Cosmology Project)
    DEGroom(at)lbl(dot)gov www-ccd.lbl.gov 510/486-6788 FAX: 510/486-4799
    Analog: 50R6008//1 Cyclotron Road//Berkeley Lab//Berkeley, CA 94720-8166



    This archive was generated by hypermail 2.1.4 : Thu Sep 30 2004 - 12:20:11 PDT