Astrostatistics in a Nutshell
In the 19th century, the astronomers and the statisticians were the same people: Legendre, Laplace and Gauss developed least squares regression based on the normal (Gaussian) error distribution to solve problems in celestial mechanics. But for most of the 20th century, astronomers and statisticians had little contact. Astronomers thus largely missed the move from least squares to the more sophisticated maximum likelihood estimation. They were mostly unaware of stochastic autoregressive models to treat aperiodic variations in time series, spatial statistics to treat pattens of points in space, local regression for data smoothing, survival analysis to treat nondetections, and many other innovations in applied statistics. Some modern techniques – bootstrap resampling, wavelet analysis, and particularly Bayesian inference – have become popular in portions of the astronomical community. But the majority of astronomers continued to use a narrow suite of familiar statistical procedures, or develop ad hoc methodology for specific purposes.
Since the1990s, this situation has been changing. Astronomers have become more knowledgeable about modern statlstical methodology, and statisticians have learned about the fascinating challenges presented by astronomical data. Cross-disciplinary research groups emerged, specialized conferences brought experts together, short Summer Schools educated graduate students, and textbooks have been published. Dozens of astronomical papers annually use advanced modern methods to considerable effect in addressing important research problems. More rarely, new methods are developed specifically to address astronomical issues. Thus, the field of astrostatistics has risen from a century of neglect into a youthful, vibrant enterprise.
Prior to the computer, statistical practice relied on tests and procedures that could be calculated analytically, often based on restrictive assumptions such as linear relationships with normal error distributions. But in recent decades, these assumptions have been relaxed as computationally intensive procedures have been developed. Heavy computations are perforce necessary when megadatasets are involved. Thus, astrostatistics has a close relationship to the emerging field of astroinformatics which emphasizes algorithm development and computational sophistication. Astronomers, statisticians, computer scientists and applied mathematicians thus work together to bring the most reliable and effective methods to astronomical data and science analysis.
These fields are now creating organizational entities to further progress in cross-disciplinary research and to promulgate methodology developed by the vanguard to the wider astronomical community. The International Statistical Institute (sister organization to the International Astronomical Union) created an Astrostatistics Committee. At its August 2012 meeting, the IAU will consider a proposal to create a Working Group in Astrostatistics and Astroinformatics. A Web portal (http://asaip.psu.edu), a multifaceted `social medium’, has recently been created to serve all of these organizations and the wider cross-disciplinary communities. The public domain R statistical computing environment, a huge software package with >60,000 pre-coded functionalities, is being taught to hundreds of graduate students from dozens of universities worldwide. Astrostatistics and astroinformatics sessions are organized at general U.S. (AAS, JSM) and international meetings (IAU, ISI), and specialized conferences are now common.
The 220th meeting of the AAS in Anchorage is thus an auspicious time for the AAS to create a Working Group in Astroinformatics and Astrostatistics, to provide focus for the energetic work of North American astronomers in these new fields and to amplify the beneficial effects of their work for the wider astronomical community.