confint of bivariate smooths

[scottkosty]

This is a feature request for adding support for confidence intervals of bivariate smooths.

In the below example code, there is one call to confint that succeeds, but when adding a newdata argument or a type argument, it fails.

library(mgcv)
library(schoenberg)
set.seed(2)
dat <- gamSim(2, n = 4000, dist = "normal", scale = 1)
m2 <- gam(y ~ s(x, z, k = 30), data = dat$data, method = "REML")

# this works (in the sense of no error)
test1 <- confint(m2, parm = "x,z")

# however, these fail
test2 <- confint(m2, parm = "x,z", type = "simultaneous")
newd_ <- data.frame(x = dat$x, y = dat$y, z = dat$z)
test3 <- confint(m2, parm = "x,z", newdata = newd_)

# I'm guessing these are not expected to work, but I tried them just in case,
# and they do not succeed:

# since there is only one smooth, I thought I could exclude the parm
test4 <- confint(m2, newdata = newd_)
test5 <- confint(m2, type = "simultaneous")

# try specifying parm with s()
test6 <- confint(m2, parm = "s(x,z)", newdata = newd_)
test7 <- confint(m2, parm = "s(x,z)", type = "simultaneous")

# try specifying parm with number
test8 <- confint(m2, parm = 1, newdata = newd_)
test9 <- confint(m2, parm = 1, type = "simultaneous")

[gavinsimpson]

This issue was moved to gavinsimpson/gratia#1