Does emmeans work with delta models?

[sutjo]

Good morning. I would like to use emmeans with delta models, but it does not seem to be supported. Would modelling the delta components separately and then using emmeans be an acceptable workaround? I.e. Modeling the binomial and the positive catch components separately, then applying emmeans to each of the two models.

Here is an example of my code:
emmeans::emmeans(fit.delta, ~ predictor)
#result = Error in vcov.sdmTMB(object) : nrow(b) == nrow(vc) is not TRUE

emmeans::emmeans(fit.binomial, ~ predictor)
#result = estimates marginal means

emmeans::emmeans(fit.pos.catch ~ predictor)
#result = estimates marginal means; note that the pos.catch data set does not include zero catches

[seananderson]

Yes, you could usually fit the components separately. I'll add an issue and hope to get to this soon.

Fitting them separately should work in all cases with the following exceptions:

• if you are using the 'Poisson-link' families
• if you are sharing parameters between the models; this won't happen without a custom 'map' argument you've written with the exception of anisotropy, which by default is shared between the components.

In the meantime, I added a more helpful error message.