Update lsmeans (#413)

- Adjustments to the weighting of the lsmeans function, in particular
- Renamed "proportional" to "counterfactual"
- Introduced "proportional_em" to mirror implementation of emmeans
- Updated documentation to clarify differences between the implementations

DESCRIPTION

@@ -18,7 +18,7 @@ LazyData: true
 Roxygen: list(markdown = TRUE)
 URL: https://insightsengineering.github.io/rbmi/, https://github.com/insightsengineering/rbmi
 BugReports: https://github.com/insightsengineering/rbmi/issues
-RoxygenNote: 7.2.3
+RoxygenNote: 7.3.1
 Suggests: 
     dplyr,
     tidyr,

NAMESPACE

@@ -1,6 +1,9 @@
 # Generated by roxygen2: do not edit by hand
 
 S3method(as.data.frame,pool)
+S3method(collapse_values,default)
+S3method(collapse_values,factor)
+S3method(collapse_values,numeric)
 S3method(draws,approxbayes)
 S3method(draws,bayes)
 S3method(draws,bmlmi)
@@ -19,6 +22,7 @@ S3method(print,imputation_list_df)
 S3method(print,imputation_list_single)
 S3method(print,imputation_single)
 S3method(print,pool)
+S3method(repr,numeric)
 S3method(validate,analysis)
 S3method(validate,bmlmi)
 S3method(validate,bootstrap)

NEWS.md

@@ -3,6 +3,11 @@
 
 * Include vignette on how to obtain frequentist and information-anchored inference with conditional mean imputation using `rbmi`
 * Added FAQ vignette
+* Updates to `lsmeans()` for better consistency with the `emmeans` package (#412)
+    * Renamed `lsmeans(..., weights = "proportional")` to `lsmeans(..., weights = "counterfactual")`to more accurately reflect the weights used in the calculation.
+    * Added `lsmeans(..., weights = "proportional_em")` which provides consistent results with `emmeans(..., weights = "proportional")`
+    * `lsmeans(..., weights = "proportional")` has been left in the package for backwards compatibility and is an alias for `lsmeans(..., weights = "counterfactual")` but now gives
+    a message prompting users to use either "proptional_em" or "counterfactual" instead.
 
 # rbmi 1.2.6
 

R/ancova.R

@@ -12,9 +12,10 @@
 #' fit the ancova model at. If `NULL`, a separate ancova model will be fit to the
 #' outcomes for each visit (as determined by `unique(data[[vars$visit]])`).
 #' See details.
-#' @param weights Character, either `"proportional"` (default) or `"equal"`. Specifies the
-#' weighting strategy to be used for categorical covariates when calculating the lsmeans.
-#' See details.
+#' @param weights Character, either `"counterfactual"` (default), `"equal"`,
+#' `"proportional_em"` or `"proportional"`.
+#' Specifies the weighting strategy to be used when calculating the lsmeans.
+#' See the weighting section for more details.
 #'
 #' @details
 #' The function works as follows:
@@ -49,29 +50,18 @@
 #' by providing them to the `covariates` argument of [set_vars()]
 #' e.g. `set_vars(covariates = c("sex*age"))`.
 #'
-#'
-#' ## Weighting
-#'
-#' `"proportional"` is the default scheme that is used. This is equivalent to standardization,
-#' i.e. the lsmeans in
-#' each group are equal to the predicted mean outcome from the ancova model for
-#' that group based on baseline characteristics of all subjects regardless of
-#' their assigned group. The alternative weighting scheme, `"equal"`, creates hypothetical
-#' patients by expanding out all combinations of the models categorical covariates. The
-#' lsmeans are then calculated as the average of
-#' the predicted mean outcome for these hypothetical patients assuming they come from each
-#' group in turn.
-#'
-#' In short:
-#'   - `"proportional"` weights categorical covariates based upon their frequency of occurrence
-#' in the data.
-#'   - `"equal"` weights categorical covariates equally across all theoretical combinations.
+#' @inheritSection lsmeans Weighting
 #'
 #' @seealso [analyse()]
 #' @seealso [stats::lm()]
 #' @seealso [set_vars()]
 #' @export
-ancova <- function(data, vars, visits = NULL, weights = c("proportional", "equal")) {
+ancova <- function(
+    data,
+    vars,
+    visits = NULL,
+    weights = c("counterfactual", "equal", "proportional_em", "proportional")
+) {
 
     outcome <- vars[["outcome"]]
     group <- vars[["group"]]
@@ -154,13 +144,13 @@ ancova <- function(data, vars, visits = NULL, weights = c("proportional", "equal
 #' @description
 #' Performance analysis of covariance. See [ancova()] for full details.
 #'
-#' @param data The `data.frame` containing all of the data required for the model.
 #' @param outcome Character, the name of the outcome variable in `data`.
 #' @param group Character, the name of the group variable in `data`.
 #' @param covariates Character vector containing the name of any additional covariates
 #' to be included in the model as well as any interaction terms.
-#' @param weights Character, specifies whether to use "proportional" or "equal" weighting for each
-#' categorical covariate combination when calculating the lsmeans.
+#'
+#' @inheritParams ancova
+#' @inheritSection lsmeans Weighting
 #'
 #' @details
 #' - `group` must be a factor variable with only 2 levels.
@@ -174,7 +164,13 @@ ancova <- function(data, vars, visits = NULL, weights = c("proportional", "equal
 #' }
 #' @seealso [ancova()]
 #' @importFrom stats lm coef vcov df.residual
-ancova_single <- function(data, outcome, group, covariates, weights = c("proportional", "equal")) {
+ancova_single <- function(
+    data,
+    outcome,
+    group,
+    covariates,
+    weights = c("counterfactual", "equal", "proportional_em", "proportional")
+) {
 
     weights <- match.arg(weights)
     assert_that(

R/dataclasses.R

@@ -285,6 +285,7 @@ repr <- function(x, ...) {
     UseMethod("repr")
 }
 
+#' @export
 repr.numeric <- function(x, ...) {
     paste0("c(", paste0(round(x, 2), collapse = ", "), ")")
 }