AR(1) Time-varying Parameters

[Nathan-Hebert]

Hi,

Is there a way to pull out the AR(1) correlation estimated for time_varying_type = 'ar1'? I've found sigma_V in sdreport, but not rho_V.

Thanks,
Nathan

[seananderson]

See rho_time_unscaled. Sorry this is a bit obscure to obtain still!

library(sdmTMB)
fit <- sdmTMB(
  catch_weight ~ 1,
  data = dogfish,
  spatial = "off",
  time = "year",
  spatiotemporal = "off",
  time_varying = ~ 1,
  time_varying_type = "ar1",
  family = tweedie(link = "log")
)
#> Detected irregular time spacing with an AR(1) or random walk process.
#> Consider filling in the missing time slices with `extra_time`.
#> `extra_time = c(2005, 2007, 2009, 2011, 2013, 2015, 2017, 2019, 2020)`

fit$sd_report
#> sdreport(.) result
#>                     Estimate Std. Error
#> b_j                4.3622661 0.38181529
#> thetaf             1.3042638 0.04396550
#> ln_phi             2.3763279 0.03156677
#> ln_tau_V          -0.2012370 0.26920637
#> rho_time_unscaled  0.8839355 0.73973581
#> Maximum gradient component: 2.277666e-05

# see `rho_time_unscaled`

pl <- as.list(fit$sd_report, "Estimate")
pls <- as.list(fit$sd_report, "Std. Error")

.rho <- pl$rho_time_unscaled[1,1]
.se <- pls$rho_time_unscaled[1,1]

bound <- function(x) 2 * plogis(x) - 1

bound(.rho)
#> [1] 0.4152742
bound(.rho + 1.96 * .se)
#> [1] 0.8232787
bound(.rho - 1.96 * .se)
#> [1] -0.2756548

^{Created on 2024-02-28 with reprex v2.1.0}

[Nathan-Hebert]

Fantastic, thanks!