- #1
wolram
Gold Member
Dearly Missed
- 4,446
- 558
http://nedwww.ipac.caltech.edu/level5/Sept03/Feigelson/paper.pdf
PhyStat 2003: Statistical Problems in Particle Physics, Astrophysics, and Cosmology
Statistical Challenges in Modern Astronomy
E. D. Feigelson
Department of Astronomy & Astrophysics, Penn State University, University Park PA 16802, USA
G. J. Babu
Department of Statistics, Penn State University, University Park PA 16802, USA
Despite centuries of close association, statistics andastronomyare surprisingly distant today. Most observational
astronomical research relies on an inadequate toolbox of methodological tools. Yet the needs are substantial:
astronomy encounters sophisticated problems involving sampling theory, survival analysis, multivariate classification and analysis, time series analysis, wavelet analysis, spatial point processes, nonlinear regression, bootstrap
resampling and model selection. We review the recent resurgence of astrostatistical research, and outline new
challenges raised by the emerging Virtual Observatory. Our essay ends with a list of research challenges and
arXiv:astro-ph/0401404 v1 20 Jan 2004
infrastructure for astrostatistics in the coming decade.
1. The glorious history of astronomy and statistics
Astronomy is perhaps the oldest observational science1 . The effort to understand the mysterious luminous objects in the sky has been an important element of human culture for at least 104years.Quantitative
measurements of celestial phenomena were carried out
by many ancient civilizations. The classical Greeks
were not active observers but were unusually creative
in the applications of mathematical principles to as
tronomy. The geometric models ofthe Platonists with
crystalline spheres spinning around the static Earth
were elaborated in detail, and this model endured in
Europe for 15 centuries. But it was another Greek
natural philosopher, Hipparchus, who made one of the
first applications of mathematical principles that we
now consider to be in the realm of statistics. Finding
scatter in Bablylonian measurements of the length of
a year, defined as the time between solstices, he took
the middle of the range – rather than the mean or
median – for the best value.
This is but one of many discussions of statistical issues in the history of astronomy. Ptolemy estimated
parameters of a non-linear cosmological model using a
minimax goodness-of-fit method. Al-Biruni discussed
the dangers of propagating errors from inaccurate in
struments and inattentive observers. While some Medieval scholars advised against the acquisition of repeated measurements, fearing that errors would compound rather than compensate for each other, the usefulnes of the mean to increase precision was demonstrated with great success by Tycho Brahe.
During the 19thcentury, several elements of modern
mathematical statistics were developed in the context
1The historical relationship between astronomy and statistics is described in references [15], [38] and elsewhere. Our
Astrostatistics monograph gives more detail and contemporary
examples of astrostatistical problems [3].
of celestial mechanics, where the application of Newtonian theory to solar system phenomena gave astonishingly precise and self-consistent quantitative inferences. Legendre developed L2 least squares parameter estimation to model cometary orbits. The least
squares method became an instant success in European astronomy and geodesy.Other astronomers and
physicists contributed to statistics: Huygens wrote a
book on probability in games of chance; Newton developed an interpolation procedure; Halley laid foundations of actuarial science; Quetelet worked on statistical approaches to social sciences; Bessel first used the
concept of ”probable error”; and Airy wrote a volume
on the theory of errors.
But the two fields diverged in the late-19thand 20th
centuries. Astronomy leaped onto the advances of
physics – electromagnetism, thermodynamics, quantum mechanics and general relativity – to understand
the physical nature of stars, galaxies and the Universe
as a whole. A subfield called “statistical astronomy”
was still present but concentrated on rather narrow issues involving star counts and Galactic structure [30].
Statistics concentrated on analytical approaches. It
found its principle applications in socialsciences, biometrical sciences and in practical industries (e.g., Sir
R. A. Fisher’s employment by the British agricultural
service).
2. Statistical needs of astronomy today
Contemporary astronomy abounds in questions of
a statistical nature. In addition to exploratory data
analysis and simple heuristic (usually linear)modeling
common in other fields, astronomers also often interpret data in terms of complicated non-linear models
based on deterministic astrophysical processes. The phenomena studied must obey known behaviors of atomic and nuclear physics, gravitation and mechanics, thermodynamics and radiative processes, and so forth. ‘Modeling’ data may thus involves both the selection of a model family based on an astrophysical--------------------
PhyStat 2003: Statistical Problems in Particle Physics, Astrophysics, and Cosmology
Statistical Challenges in Modern Astronomy
E. D. Feigelson
Department of Astronomy & Astrophysics, Penn State University, University Park PA 16802, USA
G. J. Babu
Department of Statistics, Penn State University, University Park PA 16802, USA
Despite centuries of close association, statistics andastronomyare surprisingly distant today. Most observational
astronomical research relies on an inadequate toolbox of methodological tools. Yet the needs are substantial:
astronomy encounters sophisticated problems involving sampling theory, survival analysis, multivariate classification and analysis, time series analysis, wavelet analysis, spatial point processes, nonlinear regression, bootstrap
resampling and model selection. We review the recent resurgence of astrostatistical research, and outline new
challenges raised by the emerging Virtual Observatory. Our essay ends with a list of research challenges and
arXiv:astro-ph/0401404 v1 20 Jan 2004
infrastructure for astrostatistics in the coming decade.
1. The glorious history of astronomy and statistics
Astronomy is perhaps the oldest observational science1 . The effort to understand the mysterious luminous objects in the sky has been an important element of human culture for at least 104years.Quantitative
measurements of celestial phenomena were carried out
by many ancient civilizations. The classical Greeks
were not active observers but were unusually creative
in the applications of mathematical principles to as
tronomy. The geometric models ofthe Platonists with
crystalline spheres spinning around the static Earth
were elaborated in detail, and this model endured in
Europe for 15 centuries. But it was another Greek
natural philosopher, Hipparchus, who made one of the
first applications of mathematical principles that we
now consider to be in the realm of statistics. Finding
scatter in Bablylonian measurements of the length of
a year, defined as the time between solstices, he took
the middle of the range – rather than the mean or
median – for the best value.
This is but one of many discussions of statistical issues in the history of astronomy. Ptolemy estimated
parameters of a non-linear cosmological model using a
minimax goodness-of-fit method. Al-Biruni discussed
the dangers of propagating errors from inaccurate in
struments and inattentive observers. While some Medieval scholars advised against the acquisition of repeated measurements, fearing that errors would compound rather than compensate for each other, the usefulnes of the mean to increase precision was demonstrated with great success by Tycho Brahe.
During the 19thcentury, several elements of modern
mathematical statistics were developed in the context
1The historical relationship between astronomy and statistics is described in references [15], [38] and elsewhere. Our
Astrostatistics monograph gives more detail and contemporary
examples of astrostatistical problems [3].
of celestial mechanics, where the application of Newtonian theory to solar system phenomena gave astonishingly precise and self-consistent quantitative inferences. Legendre developed L2 least squares parameter estimation to model cometary orbits. The least
squares method became an instant success in European astronomy and geodesy.Other astronomers and
physicists contributed to statistics: Huygens wrote a
book on probability in games of chance; Newton developed an interpolation procedure; Halley laid foundations of actuarial science; Quetelet worked on statistical approaches to social sciences; Bessel first used the
concept of ”probable error”; and Airy wrote a volume
on the theory of errors.
But the two fields diverged in the late-19thand 20th
centuries. Astronomy leaped onto the advances of
physics – electromagnetism, thermodynamics, quantum mechanics and general relativity – to understand
the physical nature of stars, galaxies and the Universe
as a whole. A subfield called “statistical astronomy”
was still present but concentrated on rather narrow issues involving star counts and Galactic structure [30].
Statistics concentrated on analytical approaches. It
found its principle applications in socialsciences, biometrical sciences and in practical industries (e.g., Sir
R. A. Fisher’s employment by the British agricultural
service).
2. Statistical needs of astronomy today
Contemporary astronomy abounds in questions of
a statistical nature. In addition to exploratory data
analysis and simple heuristic (usually linear)modeling
common in other fields, astronomers also often interpret data in terms of complicated non-linear models
based on deterministic astrophysical processes. The phenomena studied must obey known behaviors of atomic and nuclear physics, gravitation and mechanics, thermodynamics and radiative processes, and so forth. ‘Modeling’ data may thus involves both the selection of a model family based on an astrophysical--------------------