326
PETER REICHERT, MARK SCHERVISH, AND MITCHELL J. SMALL
REFERENCES
function evaluations needed for posterior maximization com-
mon to all techniques) for the present example. A comparison
of rows 2 and 3 indicates that in the current example, using
the low-discrepancy sequence instead of random samples con-
tributed more to the improvement than the exible shape of
the distribution allowed by the copula representation. This
conclusion is not generalizable however, the opposite result
can be expected for posteriors with shapes that deviate more
signi cantly from the normal. The Markov chain technique is
not competitive at such a small sample size, but does lead to
good results if more simulations can be performed.
¨
¨
Abbaspour, K. C., Schulin, R., Schlappi, E., and Fluhler, H. (1996), “A
Bayesian Approach for Incorporating Uncertainty and Data Worth in Envi-
ronmental Projects,” Environmental Modeling and Assessment, 1, 151–158.
Berger, J. O. (1984), “The Robust Bayesian Viewpoint,” in Robustness of
Bayesian Analyses, ed. J. Kadane., Amsterdam: North-Holland, pp. 63–124.
Brand, K. P., and Small, M. J. (1995), “Updating Uncertainty in an Integrated
Risk Assessment: Conceptual Framework and Methods,” Risk Analysis, 15,
719–731.
Bratley, P., and Fox, B. L. (1988), “ALGORITHM 659: Implementing Sobol’s
Quasi-Random Sequence Generator,” ACM Transactions on Mathematical
Software, 14, 88–100.
Chaloner, K. M. (1996), “The Elicitation of Prior Distributions,” in Bayesian
Biostatistics, eds. D. A. Barry and D. K. Stangl, New York, Marcel Dekker,
pp. 141–156.
Clemen, R. T., and Reilly, T. (1999), “Correlations and Copulas for Decision
and Risk Analysis,” Management Science, 45, 208–224.
5. CONCLUSIONS
Dakins, M. E., Toll, J. E., Small, M. J., and Brand, K. P. (1996), “Risk-
Based Environmental Remediation: Bayesian Monte Carlo Analysis and the
Expected Value of Sample Information,” Risk Analysis, 16, 67–79.
Dennis, B. (1996), “Discussion: Should Ecologists Become Bayesians?,” Eco-
logical Applications, 6, 1095–1103.
Dilks, D. W., Canale, R. P., and Meier, P. G. (1992), “Development of
Bayesian Monte Carlo Techniques for Water Quality Model Uncertainty,”
Ecological Modelling, 62, 149–162.
In importance sampling for Bayesian inference, the poste-
rior distribution of model parameters is approximated by a
sample from a multivariate normal distribution or from another
analytically tractable distribution, and the sample is then cor-
rected with weights to improve the approximation. The pro-
cedure proposed in this article extends this approach by using Ellison, A. M. (1996), “An introduction to Bayesion Inference for Ecological
Research and Environmental Decision-Making,” Ecological Applications,
6, 1036–1046.
multivariate distributions constructed from arbitrary marginals
and given pairwise correlation coef cients with the aid of a
copula and by using low-discrepancy sequences for sampling.
The rst of these extensions makes the approach more exible
in improving the approximation of the shape of the posterior
by the sampling distribution, and the second leads to a better
approximation of the distribution by the sample. The result-
ing estimates of the marginals and correlation coef cients can
then be used to improve the sampling distribution for the next
iteration of the procedure, and the weights can be used to
diagnose the quality of the approximation. This combination
of techniques can be expected to lead to good estimates of
properties of the posterior distribution with a minimum of sim-
ulation runs. The advantages of the proposed procedure are
most relevant for problems concerning inference of a small
number of parameters (less than about 10), because analysis of
a large number of marginal distributions can become cumber-
some, the chosen form of the copula may limit the degree of
approximation of high-dimensional distributions, and the low-
discrepancy sequences used in this article need many points
to cover a high-dimensional distribution. Still, the proposed
procedure can lead to signi cant ef ciency gains for compu-
tationally demanding models with a relatively small number
of in uential parameters. If computing time is not a limiting
factor, however, then Markov chain Monte Carlo techniques
may be preferable because of their greater exibility.
Fang, K.-T., Wang, Y., and Bentler, P. M. (1994), “Some Applica-
tions of Number-Theoretic Methods in Statistics,” Statistical Science, 9,
416–428.
Finkel, A. M., and Evans, J. S. (1987), “Evaluating the Bene ts of Uncer-
tainty Reduction in Environmental Health Risk Management,” Journal Air
Pollution Control Association, 37, 1164–1172.
Freeze, R. A., Massmann, J., Smith, L., Sperling, T., and James, B. (1990),
“Hydrogeological Decision Analysis: 1. A Framework,” Ground Water, 28,
738–766.
Gamerman, D. (1997), Markov Chain Monte Carlo—Statistical Simulation for
Bayesian Inference, London: Chapman and Hall.
Gelman, A., Carlin, J. B., Stern, H. S., and Rubin, D. B. (1995), Bayesian
Data Analysis, London: Chapman and Hall.
Gujer, W., Henze, M., Mino, T., Matsuo, T., Wentzel, M. C., and Marais,
G. v. R. (1995), “The Activated Sludge Model No. 2: Biological Phosphorus
Removal,” Water Science and Technology, 31, 1–11.
Gujer, W., Henze, M., Mino, T., and van Loosdrecht, M. (1999), “Activated
Sludge Model No. 3,” Water Science and Technology, 39, 183–193.
Halton, J. H. (1960), “On the Ef ciency of Certain Quasi-Random Sequences
of Points in Evaluating Multi-Dimensional Integrals,” Numerische Mathe-
matik, 2, 84–90.
Hammersley, J. M., (1960), “Monte Carlo Methods for Solving Multivariate
Problems,” Annals of the New York Academy of Science, 86, 844–874.
Hammersley, J. M., and Handscomb, D. C. (1964), Monte Carlo Methods,
London: Spottiswoode, Ballantyne & Co.
Henze, M., Grady, C. P. L., Gujer, W., Marais, G. v. R., and Matsuo, T. (1986),
“Activated Sludge Model No. 1,” Scienti c and Technical Report 1, IAW-
PRC Task Group on Mathematical Modelling for Design and Operation of
Biological Wastewater Treatment Processes.
Hornberger, G. M., and Spear, R. C. (1981), “An Approach to the Preliminary
Analysis of Environmental Systems,” Journal of Environmental Manage-
ment, 12, 7–18.
Hua, L. K., and Wang, Y. (1981), Applications of Number Theory to Numer-
ical Analysis, Berlin: Springer.
Ihaka, R., and Gentleman, R. (1996), “R: A Language for Data Analy-
sis and Graphics,” Journal of Computational and Graphical Statistics, 5,
299–314.
ACKNOWLEDGMENTS
Iman, R. L., and Conover, W. J. (1982), “A Distribution-Free Approach to
Inducing Rank Correlation Among Input Variables,” Communication in
Statistics—Simulation and Computation, 11, 311–334.
Johnson, V. E. (1995), “A Technique for Estimating Marginal Posterior Den-
sities in Hierarchical Models Using Mixtures of Conditional Densities,”
Journal of the American Statistical Association, 87, 852–860.
Kadane, J. B., and Wolfson, L. J. (1998), “Experiences in Elicitation” (with
discussion), Journal of the Royal Statistical Society, Ser. D 47, 3–19.
Kalagnanam, J. R., and Diwekar, U. M. (1997), “An Ef cient Sampling Tech-
nique for Off-Line quality control,” Technometrics, 39, 308–319.
The authors thank Urmila Diwekar for stimulating discus-
sions and Mark Borsuk, two anonymous reviewers, and the
editor for suggestions for improving the manuscript. The rst
author also thanks the Swiss Federal Institute for Environmen-
tal Science and Technology (EAWAG) and the Department
of Engineering and Public Policy (EPP) at Carnegie Mellon
University for the opportunity to work temporarily at EPP.
Kruskal, W. H. (1958), “Ordinal Measures of Association,” Journal of the
American Statistical Association, 53, 814–861.
[Received May 2001. Revised March 2002.]
TECHNOMETRICS, NOVEMBER 2002, VOL. 44, NO. 4