|HOME SCHEDULE AUTHOR INDEX SUBJECT INDEX|
The utility of multivariate hypothesis testing as illustrated through the study of plant diversity.
Grace, James1, 1 USGS, Lafayette, LA
ABSTRACT- The thesis of this talk is that taking a multivariate approach to developing and testing hypotheses can provide unique insights into the nature and behavior of ecological systems. To illustrate this point, I present the results of a sustained application of multivariate analyses to understanding the regulation of plant diversity. These studies have used a variety of techniques, including structural equation modeling, iterative partial least squares modeling, nonlinear signal strength modeling, and system simulation, often in combination. Examples of the discoveries made in these studies are organized into (1) those that involved partitioning effects, (2) those that involved detection of residual effects, (3) those that involved the study of composite effects, (4) those that involved cross-site comparisons, (5) those that involved comparing experimental results to multivariate expectations, and (6) those that involved partitioning reciprocal influences. Specific discoveries of special note include (1) the value of light penetration as a predictor of competitive effects, (2) the role that disturbance can play in altering the competitive effect of a unit of biomass, (3) how a nonsignificant correlation between grazing and richness can hide offsetting positive and negative effects, and (4) how residual analysis can reveal the continued impact of past disturbances. General discoveries of note include (1) most current models of diversity regulation fail to incorporate the effects of nonresource stresses on richness, (2) nonresource stress effects constitute the largest and most consistent cause of variations in species richness in nature, and (3) field data reveals little evidence of positive effects of richness on biomass production. I conclude that multivariate hypothesis testing can allow ecologists to develop more mature theories that possess greater reality, superior operational meaning, a higher empirical content, and a greater predictive capability than obtained with traditional approaches.
Key words: statistics, methodology, multivariate, richness