The way predict.mboost handles this use case is that you must add the new identifiers as levels to the factor variable used in constructing the brandom base learner for the random effect. Then, you specify the new data, assigning the grouping variable to the new data data.frame as a factor variable with the same levels as the grouping variable used in constructing the model fit.

e.g.,

gammod <- gamboost(resp ~ bbs(x1) + bols(x2) + brandom(x3), data = data)
newdat <- data.frame(x1 = sample(x1min:x1max, n), x2 = sample(x2min:x2max, n),
                     x3 = as.factor(sample(x3min:m3max),
                     levels = levels(data$x3))
predictions <- predict(gammod, type = "response", newdata = 

The way predict.mboost handles this use case is that you must add the new identifiers as levels to the factor variable used in constructing the brandom base learner for the random effect. Then, you specify the new data, assigning the grouping variable to the new data data.frame as a factor variable with the same levels as the grouping variable used in constructing the model fit.

e.g.,

gammod <- gamboost(resp ~ bbs(x1) + bols(x2) + brandom(x3), data = data)
newdat <- data.frame(x1 = sample(x1min:x1max, n), x2 = sample(x2min:x2max, n),
                     x3 = as.factor(sample(x3min:m3max),
                     levels = levels(data$x3))
predictions <- predict(gammod, type = "response", newdata = newdat)