This is the single-call user interface. Model fitting uses all available
observations at their actual measurement times. The mandatory evaluation
grid is used only after fitting to construct the causal covariance matrices.
Baseline responses must be rows in data with time = 0; they
contribute to fitting but are excluded from the ICA evaluation grid.
ICA_contcont_long(
data,
model,
formula,
evaluation_times,
id,
treatment,
time,
endpoint,
outcome,
surrogate_label,
true_label,
treatment_levels = c(0, 1),
covariates = character(0),
residual_type = "exponential",
sensitivity = list(residual = seq(-1, 1, by = 0.1), random_effects = seq(-1, 1, by =
0.1)),
extrapolation = c("warn", "allow", "error"),
start_fit = NULL,
control = list(rel.tol = 1e-10, x.tol = 1e-10, eval.max = 5000, iter.max = 3000),
return_candidates = FALSE,
progress = FALSE,
tol = 1e-12
)An object of class surrogate_longitudinal_ica.
A data.frame in long format.
One of the model names accepted by
longitudinal_model_spec().
Fixed-effects formula written in terms of the standardized
response Y.
A common numeric grid, endpoint-specific T/S grids,
or "common_observed".
Column names in data.
Values identifying the two endpoints.
Control and active treatment values, in that order.
Optional additional covariate columns.
Temporal residual-correlation structure.
A list. Element residual is the grid for the four
unidentified residual cross-world correlations. For RI models,
random_effects is the corresponding random-intercept grid. For
RI/RS models, random_completions must be a list of cross-arm
covariance matrices Cov(b0,b1).
Extrapolation action; default "warn".
Optional fitted model used for initialization.
Control list passed to nlminb().
Retain admissible completion components.
Display progress for large calculations.
Numerical tolerance.