Results
In the 8 breeding seasons, the species of interest were recorded in 191 community-weeks, yielding data from 167 broods of Little Thornbird, 122 of Great Kiskadee and 79 of Greater Thornbird. Along the breeding season, the three species of interest occurred simultaneously from week 5 (mid October) to week 25 (early March) (Figure 1). Observations collected before week 5 or after week 25 were not used for the analyses.
None of the potential confounders assessed remained in the final models. The final first model (Table 1) showed that the burdens in bad alternative hosts (Little Thornbird) were negatively associated with the availability of optimal hosts (Great Kiskadee), and the magnitude of this association depended on the parasite abundance in the community (Figure 2a). For example, when the host demand in the bird community was low (i.e. low parasite abundance, <10 tL1Wi ), bad alternative hosts were parasitised only in the absence of optimal hosts, and at very low burdens. At higher levels of host demand, burdens in the bad alternative host were moderate to high only if optimal hosts were absent.
The same pattern was observed in the second model (Table 2). Availability of good alternative hosts (Greater Thornbird) reduced substantially the probability of infection in bad alternatives, and the strength of this association was related to host demand (Figure 2b). Regarding infection of good alternative hosts as a function of optimal host occurrence, the effect of availability of optimal hosts was less marked, when compared with the magnitude of the associations observed in the other two models, although at high host demand levels (~40 tL1Wi ), each brood of optimal host reduces the probability of parasitism in good alternatives by 20% (Table 3, Figure 2c).