Tj. Loredo et Im. Wasserman, INFERRING THE SPATIAL AND ENERGY-DISTRIBUTION OF GAMMA-RAY BURST SOURCES .1. METHODOLOGY, The Astrophysical journal. Supplement series, 96(1), 1995, pp. 261-301
We describe Bayesian methods for analyzing the distribution of gamma-r
ay burst peak photon fluxes and directions. These methods fit the diff
erential distribution, and have the following advantages over rival me
thods: (1) they do not destroy information by binning or averaging the
data (as do, say, (2)(chi), [V/V-max], and angular moment analyses);
(2) they straightforwardly handle uncertainties in the measured quanti
ties; (3) they analyze the strength and direction information jointly;
(4) they use information available about nondetections; and (5) they
automatically identify and account for biases and selection effects gi
ven a precise description of the experiment. In these methods the most
important information needed about the instrument threshold is not it
s value at the times of burst triggers, as is used in [V/V-max] analys
es, but rather the value of the threshold at times when no trigger occ
urred. We show that this information can be summarized as an average d
etection efficiency that is similar to the product of the exposure and
efficiency reported in the First BATSE Burst (1B) Catalog, but signif
icantly different from it at low fluxes. We also quantify an important
bias that results from estimating the peak flux by scanning the burst
to find the peak number of counts in a window of specified duration,
as was done for the 1B Catalog. When the duration of the peak of the l
ight curve is longer than the window duration, a simple nux estimate b
ased on the peak counts significantly overestimates the peak flux in a
nonlinear fashion that distorts the shape of the log (N)-log (P) dist
ribution. This distortion also corrupts analyses of the V/V-max distri
bution that use ratios of counts above background to estimate V/V-max.
The Bayesian calculation specifies how to account for this bias. Impl
ementation of the Bayesian approach requires some changes in the way b
urst data are reported that we describe in detail. Subsequent papers w
ill report analyses of the 1B Catalog data using the methods described
here.