View article

The use of path analysis has grown rapidly in recent years, yet few studies employ recent advances in the field of structural equation modelling (SEM). Here I illustrate the capabilities of SEM for path analyses, using data from a study of hummingbird pollination. The main drawback to conventional path analysis is that the overall agreement between the path diagram (or `model') and the data is not quantified; this also means that there is no clear way to directly compare competing models of the same system. One of the major advantages of SEM is that it can test the descriptive ability of different models, thus allowing them to be compared. When unmeasured latent variables are required (e.g. to circumvent problems with multi-collinearity to include estimates of measurement error, or to represent general `factors' such as size), SEM can directly incorporate them as well. Furthermore, SEM provides `modification indices' …