1
Appendix B. Detailed mathematical procedure used to determine the ants that interact more than
expected by their natural abundances.
We constructed an interaction matrix between ants and plants, where aij is the number of
interactions between plant species i and ant species j. A theoretical matrix of abundance of
 F Fj
species was determined by bij   i .
 Fp Fa


 , where Fi is the absolute frequency of a given EFN
plant i in the plot, Fp is the total frequency of EFN-plants found in the plot, Fj is the absolute
frequency of a given ant j collected in plants without EFNs in the plot, and Fa is the total
frequency of ants collected in plants without EFNs found in the plot. Therefore, our estimates of
ant abundance are based on the assumption that there is no effect of EFN plants on the spatial
patterns of ant foraging. This assumption is supported by aspects of the natural history of the
system, such as: (1) interactions between ants and EFN plants are facultative and all ant species
in our study rely on other resources than EFNs. Therefore, our system differs from antmyrmecophyte interactions in which plants host ant colonies and therefore ants are spatially
aggregated around host plants; (2) ant foraging areas are often constrained to within very short
distances from ant colonies, limiting any potential aggregation effect; (3) EFN plants occur in
low densities, and even if individual ants were able to aggregate around them, the effect on
spatial patterns would be negligible; (4) if there were an aggregation effect, we would expect a
negative correlation between the abundance in non-EFN networks and the frequency of
interactions in EFN networks. However, we detected exactly the opposite: a positive correlation
between the two variables (see Results).
2
The probability of occurrence of any particular interaction between a plant species i and
an ant species j was determined as  aij 
aij
P
A
n
m

anm
, where aij is the number of interaction
events between i and j in the matrix describing the interactions between ants and EFN-plants, and
P
A
n
m

anm is the total number of interaction events between ants and EFN-plants in the plot, P
is the plant species richness, and A is the ant species richness. The probability of an interaction
expected based on the abundance of interacting partners as  bij 
bij
P
A
o
g

, where bij is the
bog
product of the relative abundances of plant species i and animal species j and
P
A
o
g

bog is the
sum of all products of relative abundances of all pairs of plants and ants. Finally, we computed
the differences between the actual probability of a given interaction occurring and the probability
derived from the species abundances, Cij  ( aij   bij ) . We standardized Cij to
Cij ´
Cij
| max(Cij ) |
, where | max(Cij ) | is the maximum value of Cij of the matrix, to allow across-
plot comparisons. Cij<0 are cases in which plant i and ant j interact less than expected from their
abundances; Cij>0 are cases in which plant i and ant j interact more than expected from their
abundances. For ant species collected on plants without EFNs that were not collected on EFNplants we did not compute Cij.