Skip to contents

Impute Missing Values with a Piecewise-Linear Mixed Model

Source: R/piecewise_lmm.R
impute_piecewise_lmm.Rd

Fits `y ~ t + pmax(t - knot, 0) + (1 + t | id)` (REML) to the observed rows and fills the missing responses with the model prediction, using each subject's random effects when the subject has observations and the population curve otherwise. A single imputation: it does not propagate the imputation uncertainty.

Usage

impute_piecewise_lmm(
  data,
  response,
  time_col = "time",
  id_col = "arl_number",
  knot = NULL,
  metric = c("AIC", "BIC", "RMSE")
)

Arguments

data

A data frame in long format.

response

Name of the numeric response.

time_col, id_col

Names of the time and subject columns.

knot

Change point. `NULL` picks it with [tune_breakpoint_lmm()] using `metric`.

metric

Criterion for the automatic knot.

Value

`data` as a tibble with two added columns, `<response>_imputed` (the response with missing values filled) and `<response>_was_imputed` (`TRUE` for filled rows). Rows with a missing time or id stay `NA`. The attribute `knot` holds the change point used.

See also

[tune_breakpoint_lmm()]

Examples

if (FALSE) { # \dontrun{
d <- lme4::sleepstudy
d$Reaction[c(5, 25, 60)] <- NA
out <- impute_piecewise_lmm(d, "Reaction", "Days", "Subject", knot = 4)
} # }