Abstract
This “perspective” chapter briefly surveys (1) past growth in the use of Bayesian methods in astrophysics; (2) current misconceptions about both frequentist and Bayesian statistical inference that hinder wider adoption of Bayesian methods by astronomers; and (3) multilevel (hierarchical) Bayesian modeling as a major future direction for research in Bayesian astrostatistics.
This volume contains presentations from the first invited session on astrostatistics to be held at an International Statistical Institute (ISI) World Statistics Congress. This session was a major milestone for astrostatistics as an emerging cross-disciplinary research area. It was the first such session organized by the ISI Astrostatistics Committee, whose formation in 2010 marked formal international recognition of the importance and potential of astrostatistics by one of its information science parent disciplines. It was also a significant milestone for Bayesian astrostatistics, as this research area was chosen as a (non-exclusive) focus for the session.
As an early (and elder!) proponent of Bayesian methods in astronomy, I have been asked to provide a “perspective piece” on the history and status of Bayesian astrostatistics. I begin by briefly documenting the rapid rise in use of the Bayesian approach by astrostatistics researchers over the past two decades. Next, I describe three misconceptions about both frequentist and Bayesian methods that hinder wider adoption of the Bayesian approach across the broader community of astronomer data analysts. Then I highlight the emerging role of multilevel (hierarchical) Bayesian modeling in astrostatistics as a major future direction for research in Bayesian astrostatistics. I end with a provocative recommendation for survey data reporting, motivated by the multilevel Bayesian perspective on modeling cosmic populations.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
Notably, Sturrock [3] earlier introduced astronomers to the use of Bayesian probabilities for “bookkeeping” of subjective beliefs about astrophysical hypotheses, but he did not discuss statistical modeling of measurements per se.
- 2.
A third paper [11] had some Bayesian content but focused on frequentist evaluation criteria, even for the one Bayesian procedure considered; these three presentations, with discussion, comprised the Bayesian session.
- 3.
Roberto Trotta and Martin Hendry have shown similar plots in various venues, helpfully noting that the recent rate of growth apparent in Fig. 2.1 is much greater than the rate of growth in the number of all publications; i.e., not just the amount but also the prevalence of Bayesian work is rapidly rising.
- 4.
I am not providing references to publications exhibiting the problem for diplomatic reasons and for a more pragmatic and frustrating reason: In the field where I have repeatedly encountered the problem—analysis of infrared exoplanet transit data—authors routinely fail to describe their analysis methods with sufficient detail to know what was done, let alone to enable readers to verify or duplicate the analysis. While there are clear signs of statistical impropriety in many of the papers, I only know the details from personal communications with exoplanet transit scientists.
- 5.
Efron [16] describes some such cases by saying the frequentist result can be accurate but not correct. Put another way, the performance claim is valid, but the long-run performance can be irrelevant to the case-at-hand, e.g., due to the existence of so-called recognizable subsets in the sample space (see [9] and [16] for elaboration of this notion). This is a further example of how nontrivial the relationship between variability and uncertainty can be.
- 6.
There are theorems linking single-case Bayesian probabilities and long-run performance in some general settings, e.g., establishing that, for fixed-dimension parametric inference, Bayesian credible regions with probability P have frequentist coverage close to P (the rate of convergence is \(o(1/\sqrt N )\) for flat priors, and faster for so-called reference priors). But the theorems do not apply in some interesting classes of problems, e.g., nonparametric problems.
- 7.
The convention is to reserve the term for models with three or more levels of nodes, i.e., two or more levels of edges, or two or more levels of nodes for uncertain variables (i.e., unshaded nodes). The model depicted in panel (d) would be called a two-level model.
- 8.
http://www.mrc-bsu.cam.ac.uk/bugs/
- 9.
It is worth pointing out that this is not a uniquely Bayesian insight. Eddington, Malmquist, and Lutz and Kelker used frequentist arguments to justify their corrections; Eddington even offered adaptive corrections. The large and influential statistics literature on shrinkage estimators leads to similar conclusions; see [22] for further discussion and references.
- 10.
I am tempted to recommend that, even in this regime, the likelihood summary be chosen so as to deter misuse as an estimate, say by tabulating the + 1σ and −2σ points rather than means and standard deviations. I am only partly facetious about this!
References
Gull, S.F., Daniell, G.J.: Image reconstruction from incomplete and noisy data. Nature 272, 686–690 (1978). DOI 10.1038/272686a0
Gull, S.F., Daniell, G.J.: The Maximum Entropy Method (invited Paper). In: van Schooneveld, C. (ed.) IAU Colloq. 49: Image Formation from Coherence Functions in Astronomy, Astrophysics and Space Science Library, D. Reidel Publishing Company, Dordrecht vol. 76, p. 219 (1979)
Sturrock, P.A.: Evaluation of Astrophysical Hypotheses. Astrophysical Journal 182, 569–580 (1973). DOI 10.1086/152165
Jeffreys, H.: Theory of probability. Third edition. Clarendon Press, Oxford (1961)
Jaynes, E.T.: Probability theory in science and engineering. Colloquium lectures in pure and applied science. Socony Mobil Oil Co. Field Research Laboratory (1959). http://books.google.com/books?id=Ft4-AAAAIAAJ
Ripley, B.D.: Bayesian methods of deconvolution and shape classification. In: Feigelson, E. D., Babu, G. J. (eds.) Statistical Challenges in Modern Astronomy, Springer, New York pp. 329–346 (1992)
Readhead, A.C.S., Lawrence, C.R.: Observations of the isotropy of the cosmic microwave background radiation. Annual Review of Astronomy and Astrophysics 30, 653–703 (1992). DOI 10.1146/annurev.aa.30.090192.003253
Feigelson, E.D., Babu, G.J. (eds.): Statistical Challenges in Modern Astronomy. Springer (1992)
Loredo, T.J.: Promise of Bayesian inference for astrophysics. In: Feigelson, E. D., Babu, G. J. (eds.) Statistical Challenges in Modern Astronomy, pp. 275–306 (1992)
Loredo, T.J.: The promise of bayesian inference for astrophysics (unabridged). Tech. rep., Department of Astronomy, Cornell University (1992). http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.56.1842CiteSeer DOI 10.1.1.56.1842
Nousek, J.A.: Source existence and parameter fitting when few counts are available. In: Feigelson, E. D., Babu, G. J. (eds.) Statistical Challenges in Modern Astronomy, pp. 307–327 (1992)
Lampton, M., Margon, B., Bowyer, S.: Parameter estimation in X-ray astronomy. Astrophysical Journal 208, 177–190 (1976). DOI 10.1086/154592
Loredo, T.J.: Computational Technology for Bayesian Inference. In: Mehringer, D. M., Plante, R. L., Roberts, D. A. (eds.) Astronomical Data Analysis Software and Systems VIII, Astronomical Society of the Pacific Conference Series, Astronomical Society of the Pacific, San Francisco vol. 172, p. 297 (1999)
Sims, C: Understanding non-bayesians. Unpublished chapter, Department of Economics, Princeton University (2010). URL http://www.princeton.edu/~sims/#UndstndngNnBsns
Kuhn, T.S.: The structure of scientific revolutions. Second edition. University of Chicago Press, Chicago (1970)
Efron, B.: Bayesians, frequentists, and physicists. In: L. Lyons (ed.) PHYSTAT2003: Statistical Problems in Particle Physics, Astrophysics, and Cosmology, SLAC, Stanford CA, pp. 17–24 (2003)
Bayarri, M.J., Berger, J.O.: The interplay of Bayesian and frequentist analysis. Statist. Sci. 19(1), 58–80 (2004). DOI 10.1214/088342304000000116. URL http://dx.doi.org/10.1214/088342304000000116
Gelman, A., Carlin, J.B., Stern, H.S., Rubin, D.B.: Bayesian data analysis, second edn. Texts in Statistical Science Series. Chapman and Hall/CRC, Boca Raton, FL (2004)
Little, R.J.: Calibrated Bayes: a Bayes/frequentist roadmap. Amer. Statist. 60(3), 213–223 (2006). DOI 10.1198/000313006X117837. URL http://dx.doi.org/10.1198/ 000313006X117837
Loredo, T.J.: Statistical foundations and statistical practice (contribution to a panel discussion on the future of astrostatistics). In: Feigelson, E. D., Babu, G. J. (eds.) Statistical Challenges in Modern Astronomy, p. 7pp. Springer (2012) (in press)
West, M.: Commentary. In: Feigelson, E. D., Babu, G. J. (eds.) Statistical Challenges in Modern Astronomy, p. 328ff (1992)
Loredo, T.J.: Analyzing Data from Astronomical Surveys: Issues and Directions. In: Babu, G. J., Feigelson, E. D. (eds.) Statistical Challenges in Modern Astronomy IV, Astronomical Society of the Pacific Conference Series, Astronomical Society of the Pacific, San Francisco vol. 371, p. 121 (2007)
Loredo, T.J., Hendry, M.A.: Bayesian multilevel modelling of cosmological populations. In: Hobson, M. P., Jaffe, A. H., Liddle, A. R., Mukeherjee, P., Parkinson, D. (eds.) Bayesian Methods in Cosmology, p. 245. Cambridge University Press, Cambridge (2010)
Loredo, T.J., Lamb, D.Q.: Bayesian analysis of neutrinos observed from supernova SN 1987A. Physical Review D 65(6), 063002 (2002). DOI 10.1103/PhysRevD.65.063002
Loredo, T.J., Wasserman, I.M.: Inferring the Spatial and Energy Distribution of Gamma-Ray Burst Sources. II. Isotropic Models. Astrophysical Journal 502, 75 (1998). DOI 10.1086/ 305870
Petit, J.M., Kavelaars, J.J., Gladman, B., Loredo, T.: Size Distribution of Multikilometer Transneptunian Objects. In: Barucci, M. A., Boehnhardt, H., Cruikshank, D. P., Morbidelli, A., Dotson, R. (eds.) The Solar System Beyond Neptune, pp. 71–87. University of Arizona Press Tucson (2008)
Kelly, B.C.: Some Aspects of Measurement Error in Linear Regression of Astronomical Data. Astrophysical Journal 665, 1489–1506 (2007). DOI 10.1086/519947
Feigelson, E.D., Babu, G.J. (eds.): Statistical Challenges in Modern Astronomy V. Springer (2012)
Luo, S., Loredo, T., Wasserman, I.: Likelihood analysis of GRB repetition. In: Kouveliotou, C., Briggs, M. F., Fishman, G. J. (eds.) American Institute of Physics Conference Series, American Institute of Physics Conference Series, AIP, Woodbury vol. 384, pp. 477–481 (1996). DOI 10.1063/1.51706
Graziani, C., Lamb, D.Q.: Likelihood methods and classical burster repetition. In: Rothschild, R. E., Lingenfelter, R. E. (eds.) High Velocity Neutron Stars, American Institute of Physics Conference Series, AIP Woodbury vol. 366, pp. 196–200 (1996). DOI 10.1063/1.50246
Budavári, T., Szalay, A.S.: Probabilistic Cross-Identification of Astronomical Sources. Astro-physical Journal 679, 301–309 (2008). DOI 10.1086/587156
Loredo, T.J.: Commentary on Bayesian coincidence assessment (cross-matching). In: Feigelson, E. D., Babu, G. J. (eds.) Statistical Challenges in Modern Astronomy, p. 6pp. Springer (2012) (in press)
Bovy, J., Hogg, D.W., Roweis, S.T.: Extreme deconvolution: Inferring complete distribution functions from noisy, heterogeneous and incomplete observations. Ann. Appl. Stat. 5(2B), 1657–1677 (2011)
van Dyk, D.A., Connors, A., Kashyap, V.L., Siemiginowska, A.: Analysis of Energy Spectra with Low Photon Counts via Bayesian Posterior Simulation. Astrophysical Journal 548, 224– 243 (2001). DOI 10.1086/318656
Esch, D.N., Connors, A., Karovska, M., van Dyk, D.A.: An Image Restoration Technique with Error Estimates. Astrophysical Journal 610, 1213–1227 (2004). DOI 10.1086/421761
van Dyk, D.A., DeGennaro, S., Stein, N., Jefferys, W.H., von Hippel, T.: Statistical analysis of stellar evolution. Ann. Appl. Stat. 3(1), 117–143 (2009). DOI 10.1214/08-AOAS219. URL http://dx.doi.org/10.1214/08-AOAS219
Dobigeon, N., Tourneret, J.Y., Scargle, J.D.: Joint segmentation of multivariate astronomical time series: Bayesian sampling with a hierarchical model. IEEE Trans. Signal Process. 55(2), 414–423 (2007). DOI 10.1109/TSP.2006.885768.
Loredo, T.J.: The return of the prodigal: Bayesian inference For astrophysics. In: Bernardo, J. M. Berger, J. O., Dawid, A. P., Smith, A. F. M. (eds.) Bayesian Statistics 5 Preliminary Proceedings, volume distributed to participants of the 5th Valencia Meeting on Bayesian Statistics (1995).http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.55.3616. CiteSeer DOI 10.1.1.55.3616
Wandelt, B.D., Larson, D.L., Lakshminarayanan, A.: Global, exact cosmic microwave background data analysis using Gibbs sampling. Physical Review D 70(8), 083511 (2004). DOI 10.1103/PhysRevD.70.083511
Kitaura, F.S., Enßlin, T.A.: Bayesian reconstruction of the cosmological large-scale structure: methodology, inverse algorithms and numerical optimization. Mon. Not. Roy. Astron. Soc. 389, 497–544 (2008). DOI 10.1111/j.1365-2966.2008.13341.x
Loredo, T.J.: Commentary on Bayesian analysis across astronomy. In: Feigelson, E. D., Babu, G. J. (eds.) Statistical Challenges in Modern Astronomy, p. 12pp. Springer (2012) (in press)
Mandel, K.S., Narayan, G., Kirshner, R.P.: Type Ia Supernova Light Curve Inference: Hierarchical Models in the Optical and Near-infrared. Astrophysical Journal 731, 120 (2011). DOI 10.1088/0004-637X/731/2/120
Jasche, J., Wandelt, B.D.: Bayesian inference from photometric redshift surveys. ArXiv/1106.2757 (2011)
Carroll, R.J., Ruppert, D., Stefanski, L.A., Crainiceanu, C.M.: Measurement error in nonlinear models, Monographs on Statistics and Applied Probability, vol. 105, second edn. Chapman and Hall/CRC, Boca Raton, FL (2006). DOI 10.1201/9781420010138. URL http://dx.doi. org/10.1201/9781420010138. A modern perspective
Kunz, M., Bassett, B.A., Hlozek, R.A.: Bayesian estimation applied to multiple species. Physical Review D 75(10), 103508 (2007). DOI 10.1103/PhysRevD.75.103508
Goel, P.K., DeGroot, M.H.: Information about hyperparameters in hierarchical models. J. Amer. Statist. Assoc. 76(373), 140–147 (1981).
Hadjicostas, P., Berry, S.M.: Improper and proper posteriors with improper priors in a Poisson-gamma hierarchical model. Test 8(1), 147–166 (1999). DOI 10.1007/BF02595867.
Sinharay, S., Stern, H.S.: Posterior predictive model checking in hierarchical models. J. Statist. Plann. Inference 111(1–2), 209–221 (2003). DOI 10.1016/S0378-3758(02)00303-8.
Gelman, A.: Prior distributions for variance parameters in hierarchical models (comment on article by Browne and Draper). Bayesian Anal. 1(3), 515–533 (electronic) (2006)
Bayarri, M.J., Castellanos, M.E.: Bayesian checking of the second levels of hierarchical models. Statist. Sci. 22(3), 322–343 (2007). DOI 10.1214/07-STS235.
Patanchon, G., et al.: Submillimeter Number Counts from Statistical Analysis of BLAST Maps. Astrophysical Journal 707, 1750–1765 (2009). DOI 10.1088/0004-637X/707/2/1750
Hogg, D.W., Lang, D.: Telescopes don’t make catalogues! In: EAS Publications Series, EAS Publications Series, EDP Sciences, Les Ulis vol. 45, pp. 351–358 (2011).DOI 10.1051/ eas/1045059
Jaynes, E.T.: A Backward Look to the Future. In: Grandy, W. T., Jr., Milonni, P. W. (eds.) Physics and Probability, Cambridge University Press, Cambridge pp. 261–276 (1993)
Acknowledgments
I gratefully acknowledge NSF and NASA for support of current research underlying this commentary, via grants AST-0908439, NNX09AK60G and NNX09AD03G. I thank Martin Weinberg for helpful discussions on information propagation within multilevel models. Students of Ed Jaynes’s writings on probability theory in physics may recognize the last part of my title, borrowed from a commentary by Jaynes on the history of Bayesian and maximum entropy ideas in the physical sciences [53]. This bit of plagiarism is intended as a homage to Jaynes’s influence on this area — and on my own research and thinking.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2013 Springer Science+Business Media New York
About this chapter
Cite this chapter
Loredo, T.J. (2013). Bayesian Astrostatistics: A Backward Look to the Future. In: Hilbe, J. (eds) Astrostatistical Challenges for the New Astronomy. Springer Series in Astrostatistics, vol 1. Springer, New York, NY. https://doi.org/10.1007/978-1-4614-3508-2_2
Download citation
DOI: https://doi.org/10.1007/978-1-4614-3508-2_2
Published:
Publisher Name: Springer, New York, NY
Print ISBN: 978-1-4614-3507-5
Online ISBN: 978-1-4614-3508-2
eBook Packages: Mathematics and StatisticsMathematics and Statistics (R0)