The article "Mathematical and computational challenges in population biology and ecosystems science" by Simon A. Levin et al. (17 Jan., p. 334) discusses exciting developments and challenges for studies of complex ecological systems, with special emphasis on simulation and analytical approaches. The authors highlight how increased computation capability is affecting our ability to tackle tough questions about complex system behavior. Another related area with similar challenges and advances, however, is the statistical analysis of data from complex systems. Extracting a clear understanding of how complex ecological systems operate will depend not only on our ability to simulate component processes but will also require us to go beyond traditional experimental approaches, which are limited in their scope, duration, and realism, for practical reasons.

Relative to their rapid adoption of simulation and analytic approaches, ecologists have been slow to exploit the most recent advances in multivariate analyses. Traditional statistical approaches have often distinguished between ANOVA/MANOVA (analysis of variance/multivariate analyses of variance) approacheswhich have a limited capability to deal with multivariate, interacting factorsand descriptive multivariate methods such as principal component analysis, factor analysis, and multiple regression. With the development of new programs that use the ready availability of increased computing power, data analysts are examining more of the whole covariance structure that occurs in complex systems. Expanded capacity for covariance analysis within ANOVA as well as capabilities in structural equation modeling are leading toward more general and less restricted methods of analysis. In the forefront of this enterprise should be the difficult task of interpreting the results from these analyses. No doubt, mistakes will be made and major refinements required. Nonetheless, in order to make useful predictions about the behavior of ecological systems, the challenge of conducting and interpreting multivariate analyses of interacting components must be tackled with vigor.

James B. Grace
National Wetlands Research Center,
700 Cajundome Boulevard,
Lafayette, LA 70506, USA
E-mail: gracej{at}nwrc.gov

How to Submit a Letter to the Editor

Volume 275, Number 5308, Issue of 28 March 1997, pp. 1861-1865.
©1997 by The American Association for the Advancement of Science.