Specifics regarding newdata in forecasting

[aeiche01]

I've got a quick question regarding using forecasting (or the newdata section of the mvgam function). The help files say that we need to have the same variables that were included in the original data used to fit the model. Does this include every single plot, or can it just include the series?

Meaning, let's say I've got 3 plots, 2 of which were measured for 10 years and one of which was measured for 8 years. Is it possible for me to use all three plots where I fit a model on the first 8 years, and then in the newdata only forecast the 2 plots that I have data for? Or would I have to toss the third plot and only include the two plots with the full 10 years of data in the model fitting phase?

[nicholasjclark]

Hi @aeiche01, yes this would be possible. But the best way would be to include all three plots and all 10 years in the data, but put NAs for the missing observations for that single plot. This would ensure that you would get $conditional$ predictions for the third plot, where the information that you gain from the observed data for the other two plots would help to improve your predictions for the third plot