Standard errors missing (NA) for spatial range and spatial standard deviations

[seananderson]

Question by email:

Why does sdmTMB not provide standard errors for the spatial standard deviation, range, etc.?

[seananderson]

We did that because the standard errors are more representative in log space for those and figured it would be confusing if the SEs didn't match the confidence intervals. You’ll see the confidence intervals if you set conf.int = TRUE in tidy(), which uses the log space standard error. See ?tidy.sdmTMB. You can get them from the SD report if you want:

library(sdmTMB)
fit <- sdmTMB(
  density ~ s(depth),
  data = pcod_2011, mesh = pcod_mesh_2011,
  family = tweedie(link = "log")
)

tidy(fit, "ran_pars", conf.int = TRUE)
#>        term  estimate std.error   conf.low conf.high
#> 1     range 16.839875        NA  3.4026237 83.341979
#> 3       phi 13.677114        NA 12.4377763 15.039942
#> 4   sigma_O  2.201739        NA  0.7350579  6.594929
#> 5 tweedie_p  1.580860        NA  1.5505486  1.610573

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

est$log_sigma_O
#> [1] 0.7892475
se$log_sigma_O
#> [1] 0.5597315

est$sigma_O
#> [1] 2.201739
se$sigma_O
#> [1] 1.232383

# a matrix since the ranges aren't always shared between spatial/spatiotemporal components:
est$log_range
#>         [,1]
#> [1,] 2.82375
#> [2,] 2.82375
se$log_range
#>           [,1]
#> [1,] 0.8159348
#> [2,] 0.8159348
exp(est$log_range[1,1] + c(qnorm(0.025), qnorm(0.975)) * se$log_range[1,1])
#> [1]  3.402624 83.341979

est$range
#>          [,1]
#> [1,] 16.83988
#> [2,] 16.83988
se$range
#>          [,1]
#> [1,] 13.74024
#> [2,] 13.74024

^{Created on 2022-08-23 with reprex v2.0.2}