Question about bioticStruct() and generating Single Species SDMs from sjSDM about s-jsdm

[MaximilianPi]

Hi @dansmi-hub,

1. bioticStruct(diag = TRUE)
   Yes, it completely turns off the JSDM characteristics. And as long as you don't use a DNN instead of a linear model it should be equal to a stacked SDM based on individual SDMs.

But I can you give already an answer to your question. The JSDM component, the variance-covariance matrix doesn't affect the coefficients of your species, the linear predictor is not affected by it, so JSDMs cannot improve/affect the predictions (see 10.1016/j.tree.2021.01.002). JSDM can only improve predictions if you have focal species for which you can condition your predictions on.

[dansmi-hub]

Thanks, very informative as always.

It sounds like JSDMs still do what I'm looking for. I'm interested in how including native mosquito species changes predictions for invasive mosquito species in my dataset.

My intuition from reading the Poggiato 2021 and Wilkinson 2020 is that this can still be done in the JSDM framework by holding out the native or invasive species portion of the model during the prediction stage and comparing these against each other and a full model?

It appears you can do this in HMSC [52] with some inbuilt code in the predict function.

Is something similar currently possible in sjSDM?

Apologies If I haven't been completely clear, I'm masquerading as a statistician here.

[MaximilianPi]

Yes, you want to condition your predictions on the native or on the invasive species, so to make only predictions for one species while the other one is used by the model.

Yes, Hmsc supports conditional predictions but sjSDM not yet.

[MaximilianPi]

...as a workaround, you could use one of the two species as a predictor/feature in your model.