A preliminary analysis suggested that several measured plant
growth variables might show differences among fungal treatments,
but only the proportion of plants with roots colonized were statistically
significant. Therefore we decided to create composite scores,
which are often helpful in this type of analysis. The composite
score can be considered a latent variable (a proxy for an unobservable
dependent variable) and is a weighted linear function of the
measured variables, first screened to remove those with little
information value (Table 1) based on canonical (linear) discriminant
analysis (Kramer et al., 2009). This method finds the optimal
weighting for each measured variable’s contribution to the composite
score. For these data, two orthogonal latent variables
(LDA1 and LDA2) appeared adequate to describe differences in fungal
treatments.