The statistical modelling of plant growth and its components using structural equations

The phenomenon of plant growth (or any other physiological or morphological attribute) is the result of interactions between various plant attributes that are imbedded in a nexus of causal relationships. Such multivariate causal relationships are often of interest to plant scientists, but the usual methods of randomised experiments and experimental control are not applicable to such systems, since both the hypothetical "causes" and their punitive "effects" are attributes of the experimental units (the plants). In such cases, the statistical technique of structural equation modelling can be used to test and potentially falsify alternative causal models of such interacting plant traits. This chapter introduces this method and then uses it to test a previously published model involving interacting traits involved in determining specific leaf area. The first test is based on data from 572 leaves from 194 individuals of 34 species of herbaceous angiosperms collected from the field, and divided into three groups of leaf age. The original model is falsified, but only for the oldest group of leaves. The second test is based on a data set involving 22 species of herbaceous angiosperms growth in hydroponic sand culture with factorial combinations of irradiance level (1100 and 200 mu mol m(-2) s(-1)) and nutrient concentration (full strength Hoagland solution and a 1/6 dilution). All leaves were the same approximate age. The original model is not falsified in its causal structure and is therefore applicable across the four environments, but it is falsified in its quantitative values unless these are allowed to vary across environments.