Astrostatistics
Researchers in observational astronomy often encounter situations where
the scientific goals require a statistical interpretation of rather
complex
data. These problems cover a vast range of statistical issues, such
as:
 When does the spatial distribution of galaxies in a cluster
indicate
subclustering?
When do groupings of photons on an Xray detector indicate a source is
extended? These questions lie in the purview of the well established
statistical
field of spatial point processes.
 How does one establish luminosity functions, compare
samples, or search
for correlations in data subject to flux limits? Flux limits cause
statistical
censoring (i.e. non detections of known objects) or truncation (i.e.
faint
objects are missing entirely).
 How can one best discriminate stars from galaxies in
digital optical
surveys?
What is the most valid classification scheme for spiral galaxy
morphologies
or emission line AGN? These are questions in multivariate analysis, a
huge
field that is rarely tapped to solve astronomical problems.
 How can one interpret quasiperiodic oscillations in Xray
sources, or
best establish periodicities in sparse unevenly spaced data? These are
problems in time series analysis.
 What are the best estimate of Hubble's constant and the age
of the
Universe
derivable from cosmic distance scale studies? This depends on a
thorough
understanding of linear regression techniques, including those treating
measurement errors.
For the past decade, Dr. Babu has worked with astronomers at Penn State
and NASA to address such questions. Some efforts have been made to
develop
new techniques specifically for astrostatistics. But much can be
learned
by applying methods already known to statisticians to astronomical
problems.
Dr. Babu and his colleague Dr. E. Feigelson from Astronomy &
Astrophysics
department have been leaders in the advancement of statistical
methodology
for astronomical research.
Feigelson and Babu organized international conferences called
Statistical
Challenges in Modern Astronomy (SCMA) held at Penn State. The purpose
is
to enhance the dialog between astronomers and statisticians on
important
research issues. Proceedings of the first
SCMA conference of 1991, the second
SCMA conference of 1996 and the third SCMA
conference of 2001 are
available. In 1996, they also wrote a
monograph called Astrostatistics
which introduces astronomer and statistician readers to the other
field,
and provides the first overview of astrostatistical issues. An
extensive
bibliography and index assists readers in pursuing topics introduced in
the book. Drs. Feigelson and Babu have spoken at many
meetings
in both fields, promoting improvements in statistical methodology for
observational
astronomy, with an emphasis on time series analysis and multivariate
analysis.
In the latter context, they recently published a multivariate study of gammaray
bursts which reports the discovery of a new class of
GRBs.
A popular Web metasite called StatCodes
is being maintained which provides hypertext links to many statistical
codes and services on the Web. Among many codes available from
StatCodes
are three developed at Penn State: ASURV
implementing a suite of survival methods treating censored data (i.e.
with
nondetections), SLOPES
providing linear regressions with analytical and bootstrap error
analysis,
and CENS_TAU
for estimating correlation in multiplycensored multivariate
datasets.
