Posterior mean (± standard deviation) of variance proportions by source for all models considered. M3 is the best model.

M0, M1, M3, M5 and M7 assume that the error variance is identical across tasks (0), while M2, M4, M6 and M8 allow the error variance to have a task-specific variance hierarchically based on a common error variance (J). M0 includes no random effects. M1 and M2 only include the species effect (S). M3 and M4 include species and species*domain effects (SD). M5 and M6 include species and individual effects (SI). M7 and M8 include species, species*domain, individual and individual*domain effects (SDID). Models M0–M2 and M5–M6 had significant lack of fit because of their failure to include both species and species*domain effects; this is also reflected in the sizable proportion of variance that both the species and species*domain effects account for whenever they were included in these models (i.e. M3–M4 and M7–M8). The principal conclusions drawn from M3 are similar to those from M4, M7, and M8. Because the discrepancy measures did not suggest that M3 needed individual effects (as in M7 and M8) or that the error variances needed to deviate per task (as in M4), M3 was the best model (see [60] for a formal assessment approach).