these analyses. I also used a more complex linear mixed model procedure to test for the
effects of plot-level variation (the block x plant age x fertilization combinations) and
subplot-level variation (the number of worker ants on each plant) on the amount of leaf
herbivory:
y iki = u + b + a, + P k + E jk + yx ki + E jki
In this model, y ykl indicates the measure of herbivory, b is the random block effect, a, is
the fertilization effect, /f k is the plant age effect, jk is the whole-plot error, yx ,jk is the
effect of the covariate ant number within individual plants and ,kl is the subplot (plant-
level) error. In this analysis, individual plants were nested within the plots, which were
defined as plant age x fertilization x block combinations.
Because the data were not balanced, I used maximum likelihood estimation
methods to fit the models. The significance of the block effect, b i, was determined using
a likelihood ratio test and was treated as a random effect contributing to total variance in
the final model rather than a fixed effect (Pinhiero and Bates 2000). Only the main
effects of plant age and fertilization treatment were included in the final model following
a likelihood ratio tests showing that interactive effects did not improve the model fit
(Pinhiero and Bates 2000). Due to significant differences in ant occupancy and herbivory
between 1- and 5-yr-old plants, I conducted analyses similar to the model above
separately for the two ages. All mixed model analyses were conducted with the Ime
program in R 2.2.0 following the protocol of Pinhiero and Bates (2000). All P-values
were calculated with marginal (Type III) tests for significance.
Leaf Palatability Bioassay
In July 2004, I tested the palatability of Cordia alliodora leaves for adult
Coptocycla leprosa beetles with a three-factor, fully crossed laboratory trial. The factors