The current quality of statistical analyses in astronomical research often begs for improvment. There is both inadequate research on important new challenges (Section 5) and inadequate application of known advanced methods to astronomical problems (Section 3). Astronomy clearly needs needs a strong and rapid surge of energy in statistical expertise. Three types of activities should be promoted:
In the U.S., the typical curriculum leading to a career in astronomical research requires zero or one course in statistics at the undergraduate level, and zero at the graduate level. Analogously, the curriculum of statisticians includes virtually no coursework in astronomy or other physical science. While statisticians can learn basics from "Astronomy 101" courses given at all universities, the statistical training of astronomers is not as easily accomplished. New curricular products summarizing the applicable statistical subfields, short training workshops for graduate students and young scientists, and effective statistical consulting are all needed.
While several astrostatistical research groups are making exciting progress (Section 3), the total effort is too small to impact the bulk of astronomical research. Very roughly, astrostatistical funding is currently $1M of the $1B spent annually on astronomical research. This fraction is far below that spent in biomedical or other non-physical-science fields. Though top academic leaders of statistics have expressed great enthusiasm for astronomy and astrostatistics, we can not pull them away from biostatistics and business applications without a major increase in funding. We might seek, for example, 10 - 20 cross-disciplinary research groups active at any one time at the end of a decade's growth.
For various policy and cultural reasons, astronomers rarely purchase the large commercial statistical software packages, preferring to write their own software as needs arise. This approach has contributed to the narrow methodological scope of astronomical research. Avenues for improving this situation are emerging. R is a large statistical software package with the flexible command-line interface preferred by astronomers that has recently emerged (http://www.r-project.org). A wide variety of specialized packages and codes are also available on-line (http://www.astro.psu.edu/statcodes). The new Web services concept being developed within the context of a Virtual Observatory permits coordinated access to heterogeneous software developed specifically for astronomical applications.
At Penn State, we are in the early stages of developing a Center for Astrostatistics to help attain these goals (http://www.astrostatistics.psu.edu). This is an inter-disciplinary Center to serve the astronomy and statistics communities around the nation and worldwide, seeking to bring advances in statistics into the toolbox of astronomy and astrophysics. The Center's Web site will maintain the popular StatCodes, build an instructional library of R programs, coordinate with the nascent VOStat Web service, and develop an archive of annotated links to selected statistical literature applicable to astronomy (and vice versa). The site is also planned to include tutorial handbooks and curricular products developed specifically for astrostatistical needs.
We thank the referee, Peter Shawhan (Caltech), for very helpful improvements to the paper. This work was supported by NSF grant DMS-0101360.