Model description

[trvrb]

The comparison in Table 1 is crucial, but the model descriptions are very hard to understand until after reading the Methods carefully. The difficulty is because the models are described in technical terms (dimensionality, location prior, serum effects, virus effects). The most confusing part is the location prior. My understanding (and if this is wrong then things really need to be clarified) is that uninformed refers to making the predictions just from HI data, drift refers to making the predictions from HI data + date of isolation, and diffusion / drift refers to making the predictions from HI data + date of isolation + phylogenetic relationship. If the models were named in terms of the data used to make the predictions rather than in terms of the prior, it would be easier to understand intuitively. For instance, a reader could then simply see, "OK, they included the isolation date in the model and the predictions improved by this much." I recognize that the prior plays a role in addition to the data, but I feel that a more intuitive naming scheme should be considered with detailed descriptions in the Methods. There also might be a more intuitive way to denote serum effects and virus effects, although this is less confusing than the location prior.

[trvrb]

I agree that "effects" is vague. Derek uses "offsets" to describe the same term, though this may be a bit vague as well. What about "baseline"? The serum effect combined with the virus effect gives the baseline expectation for HI titre when there is no antigenic divergence.

[trvrb]

This is a very helpful suggestion. We've included a new column in Table 1 labeled 'Data' that specifies the combination of data sources that go into the model (HI vs HI/year vs HI/year/seq). We've also clarified that serum and virus effects act to give a baseline expectation for titer when virus and serum have identical antigenic locations.