Continuous-Time Dynamics with ctsem in tidyILD

tidyILD authors

2026-04-17

This vignette introduces the ctsem backend in tidyILD via ild_ctsem(). Use this path when your scientific target is continuous-time latent dynamics under irregular measurement timing.

When to use `ild_ctsem()`

Choose ild_ctsem() for continuous-time latent process modeling.
Choose ild_kfas() for discrete-time state-space workflows.
Choose ild_lme() / ild_brms() for multilevel regression targets.

Minimal workflow

library(tidyILD)

d <- ild_simulate(n_id = 1, n_obs_per = 60, seed = 501)
x <- ild_prepare(d, id = "id", time = "time")
x <- ild_center(x, y)

fit_ct <- ild_ctsem(
  data = x,
  outcome = "y",
  model_type = "stanct",
  chains = 1,
  iter = 400
)

fit_ct
td <- ild_tidy(fit_ct)
ag <- ild_augment(fit_ct)
dg <- ild_diagnose(fit_ct)

Diagnostics and plots

ild_autoplot(fit_ct, type = "fitted_vs_actual")
ild_autoplot(fit_ct, type = "residual_time")
ild_autoplot(dg, section = "fit", type = "convergence")
ild_autoplot(dg, section = "residual", type = "acf")

Guardrails and reporting

ild_diagnose(fit_ct) may trigger ctsem-focused guardrails such as:

GR_CTSEM_NONCONVERGENCE
GR_CTSEM_UNSTABLE_DYNAMICS
GR_CTSEM_SHORT_SERIES_FOR_COMPLEX_DYNAMICS

These guardrails are surfaced in print(dg), ild_methods(fit_ct, bundle = dg), and ild_report(fit_ct).

Notes

The v1 ctsem wrapper is intentionally conservative.
You can pass a custom ct_model object for advanced specifications.
Keep time scaling explicit (time_col, time_scale) for interpretability.