Hasik et al. here present an articulate and robust phylogenetic meta-analysis of the mean and variance in outcomes of species interactions between parasitized and non-parasitized hosts. The literature search strategy, inclusion criteria, and random effects seem appropriate for the questions at hand, and I was pleased to see the authors account for not only host phylogeny but also parasite phylogeny and even the host-parasite interaction (through tensor products of correlation matrices). The total dataset includes 582 effect sizes stemming from 154 studies, which suggests the authors have sufficient power for complex rma.mv models. That said, I did have some outstanding questions about the analyses themselves.

1. The authors detail the three phylogenetic random effects of their model (host, parasite, host-parasite interaction), but I do wonder about the issue of pseudo replication by host and parasite species. In a simulation study, Cinar et al. 2021 (Methods Ecology & Evolution) show that decomposing species-level variance into phylogenetic random effects (as done here) as well as non-phylogenetic random effects (simply a 1|species term) is necessary to provide unbiased estimates of fixed effects in hierarchical meta-analysis (their model 9), assuming that phylogenetic relationships among the species in question are moderately strong (defined as a mean correlation of the phylogenetic correlation matrix, excluding the diagonal, of r>=0.2). Thus assuming decent phylogenetic signal, models with a phylogenetic effect but not an additional species-level effect can produce biased estimates. Arguably, including both phylogenetic and species-level random effects for both hosts and parasites would likely be a more robust model to fit. Have the authors considered this alternative random effects structure? A formal REML comparison among full models with alternative random effects (host phylo, parasite phylo, interaction vs host phylo, host species, parasite phylo, parasite species) could be informative, as would be assessing the mean correlation (diagonal excluded) for the host and parasite phylogenies. I assume that a model with both phylogenetic and species terms for hosts and parasites, plus the tensor interaction, would likely be impossible to fit.

2. Given the complex random effects structures used here, it would also be helpful to check whether the variance components for the random effects are identifiable. This can be done with likelihood profile plots for each random effect (as a sanity check on possible overfitting). As a side note, it was helpful to see variance partitioned through I2 in Table 1, but the authors should also briefly describe I2 in the Methods.

3. On the topic of comparing among random effects above, the details of what factors were included as fixed effects in each model were somewhat unclear. L271-272 notes the 5 fixed effects (species interaction, fitness component, parasite evolutionary strategy, agent parasitized, habitat); it seems as though these were treated as univariate models, but can this be explicitly clarified? If not, how were more complex model structures decided/built? For example, L419 details an interaction between habitat and species interaction type. Were other interactions considered? Why or why not? Lastly, did all rma.mv fitting include weighting by sampling variance (and was this derived from escalc as well)?

4. One missing detail in the presentation of model results (alongside the QM statistics that are reported) is the degree to which each model explains variation in both effect size measures (mean and variance). Several different options for deriving pseudo-R2 values exist for rma.mv models (e.g., a ratio of summed variance components for a full and intercept-only model). I do think this is important for readers to assess the degree to which these moderators are informative in context of the large variation in effect sizes here.