Is there a way to get confidence intervals for time-varying effects?

[sschooler]

The documentation suggests to get confidence intervals using the tidy() function, but there doesn't seem to be a way to get confidence intervals for time-varying effects. Must they be calculated manually from standard error?
Thanks!

[seananderson]

Yes, sorry, those aren't in tidy() yet. Random intercepts are, but not time-varying random effects. I posted an issue here: #206

Here's an example of grabbing them yourself. The format of b_rw_t may vary a bit depending on your model configuration. There's also the "ar1" and "rw0" time_varying_type, which start with a mean of zero. In that case you'd likely include main effects and then the time-varying effect would represent deviations starting from the main effect.

library(sdmTMB)
fit <- sdmTMB(
  density ~ 0, 
  time_varying = ~ 1 + depth_scaled, 
  time_varying_type = "rw",
  data = pcod_2011, 
  time = "year", 
  spatial = "off", 
  spatiotemporal = "off",
  family = tweedie()
)

est <- as.list(fit$sd_report, "Estimate")
se <- as.list(fit$sd_report, "Std. Error")

est$b_rw_t
#> , , 1
#> 
#>          [,1]       [,2]
#> [1,] 1.854868 -0.4317058
#> [2,] 1.830428 -0.4317059
#> [3,] 1.917300 -0.4317059
#> [4,] 1.602897 -0.4317059

est <- est$b_rw_t[,,1]
se <- se$b_rw_t[,,1]

est
#>          [,1]       [,2]
#> [1,] 1.854868 -0.4317058
#> [2,] 1.830428 -0.4317059
#> [3,] 1.917300 -0.4317059
#> [4,] 1.602897 -0.4317059
est + qnorm(0.975) * se
#>          [,1]       [,2]
#> [1,] 1.979196 -0.2454046
#> [2,] 1.954876 -0.2454048
#> [3,] 2.050123 -0.2454047
#> [4,] 1.760604 -0.2454047
est + qnorm(0.025) * se
#>          [,1]       [,2]
#> [1,] 1.730541 -0.6180071
#> [2,] 1.705980 -0.6180070
#> [3,] 1.784476 -0.6180071
#> [4,] 1.445190 -0.6180072

^{Created on 2023-04-28 with reprex v2.0.2}

[sschooler]

Thank you!