diff --git a/NEWS.md b/NEWS.md
index 0422639..fe203f9 100644
--- a/NEWS.md
+++ b/NEWS.md
@@ -2,7 +2,7 @@
 
   * Estimation of dynamic multivariate panel models with multiple imputation is now available via the function `dynamice()` which uses the `mice` package.
   * `predict` and `fitted` functions no longer permutes the posterior samples when all samples are used i.e. when `n_draws = NULL` (default). This also corrects the standard error estimates of `loo`, which were not correct earlier due to the mixing of chains.
-  * Added an argument `thin` for the `loo`, `lfo`, `predict` and `fitted` method.
+  * Added an argument `thin` for the `loo`, `predict` and `fitted` method.
   * Print method now only prints the run time for fastest and slowest chain instead of all chains.
 
 # dynamite 1.4.11
diff --git a/R/dynamice.R b/R/dynamice.R
index fdc6424..553998f 100644
--- a/R/dynamice.R
+++ b/R/dynamice.R
@@ -444,10 +444,12 @@ parse_predictors_wide <- function(dformula, value_vars, idx_time, group_var) {
 
 #' Compute Lagging Values of an Imputed Response
 #'
+#' Function for computing the lagged values of an imputed response in `mice`.
+#'
 #' @inheritParams mice::mice.impute.norm
 #' @param group_var \[`character(1)`]\cr Name of the grouping variable.
 #' @param resp \[`character(1)`]\cr Name of the response variable.
-#' @noRd
+#' @keywords internal
 #' @export
 mice.impute.lag <- function(y, ry, x, wy = NULL,
                             group_var, resp, ...) {
@@ -464,10 +466,12 @@ mice.impute.lag <- function(y, ry, x, wy = NULL,
 
 #' Compute Leading Values of an Imputed Response
 #'
+#' Function for computing the leading values of an imputed response in `mice`.
+#'
 #' @inheritParams mice::mice.impute.norm
 #' @param group_var \[`character(1)`]\cr Name of the grouping variable.
 #' @param resp \[`character(1)`]\cr Name of the response variable.
-#' @noRd
+#' @keywords internal
 #' @export
 mice.impute.lead <- function(y, ry, x, wy = NULL,
                              group_var, resp, ...) {
diff --git a/R/lfo.R b/R/lfo.R
index a538fe3..c44d12f 100644
--- a/R/lfo.R
+++ b/R/lfo.R
@@ -19,9 +19,6 @@
 #'   the LFO computations to the console.
 #' @param k_threshold \[`numeric(1)`]\cr Threshold for the Pareto k estimate
 #'   triggering refit. Default is 0.7.
-#' @param thin \[`integer(1)`]\cr Use only every `thin` posterior sample when
-#'   computing LFO. This can be beneficial with when the model object contains
-#'   large number of samples. Default is `1` meaning that all samples are used.
 #' @param ... Additional arguments passed to [rstan::sampling()] or
 #'   [cmdstanr::sample()], such as `chains` and `cores` (`parallel_chains` in
 #'   `cmdstanr`).
@@ -54,7 +51,7 @@
 #' }
 #' }
 #'
-lfo <- function(x, L, verbose = TRUE, k_threshold = 0.7, thin = 1, ...) {
+lfo <- function(x, L, verbose = TRUE, k_threshold = 0.7, ...) {
   stopifnot_(
     is.null(x$imputed),
     "Leave-future-out cross-validation is not supported for models
@@ -76,10 +73,6 @@ lfo <- function(x, L, verbose = TRUE, k_threshold = 0.7, thin = 1, ...) {
     checkmate::test_number(x = k_threshold),
     "Argument {.arg k_threshold} must be a single {.cls numeric} value."
   )
-  stopifnot_(
-    checkmate::test_int(x = thin, lower = 1L, upper = ndraws(x)),
-    "Argument {.arg thin} must be a single positive {.cls integer}."
-  )
   time_var <- x$time_var
   group_var <- x$group_var
   tp <- sort(unique(x$data[[time_var]]))
@@ -100,7 +93,7 @@ lfo <- function(x, L, verbose = TRUE, k_threshold = 0.7, thin = 1, ...) {
   }
   fit <- update_(x, data = d, refresh = 0, ...)
 
-  idx_draws <- seq.int(1, ndraws(x), by = thin)
+  idx_draws <- seq.int(1, ndraws(fit))
   n_draws <- length(idx_draws)
   # would be faster to use only data
   # x$data[eval(time) >= tp[L] - x$stan$fixed]
diff --git a/man/lfo.Rd b/man/lfo.Rd
index d4c10f4..75f3858 100644
--- a/man/lfo.Rd
+++ b/man/lfo.Rd
@@ -4,7 +4,7 @@
 \alias{lfo}
 \title{Approximate Leave-Future-Out (LFO) Cross-validation}
 \usage{
-lfo(x, L, verbose = TRUE, k_threshold = 0.7, thin = 1, ...)
+lfo(x, L, verbose = TRUE, k_threshold = 0.7, ...)
 }
 \arguments{
 \item{x}{[\code{dynamitefit}]\cr The model fit object.}
@@ -18,10 +18,6 @@ the LFO computations to the console.}
 \item{k_threshold}{[\code{numeric(1)}]\cr Threshold for the Pareto k estimate
 triggering refit. Default is 0.7.}
 
-\item{thin}{[\code{integer(1)}]\cr Use only every \code{thin} posterior sample when
-computing LFO. This can be beneficial with when the model object contains
-large number of samples. Default is \code{1} meaning that all samples are used.}
-
 \item{...}{Additional arguments passed to \code{\link[rstan:stanmodel-method-sampling]{rstan::sampling()}} or
 \code{\link[cmdstanr:model-method-sample]{cmdstanr::sample()}}, such as \code{chains} and \code{cores} (\code{parallel_chains} in
 \code{cmdstanr}).}