| 1 |
# Default methods work for mmrm objects: |
| 2 |
# model_matrix.default() — model.matrix(obj) works; weights are all positive |
| 3 |
# residuals_CS.default() — residuals(obj) works; weights are all positive |
| 4 |
# coef_CS.default() — coef(obj) works directly |
| 5 |
|
| 6 |
#------------------------------------- |
| 7 |
# vcovCR with defaults |
| 8 |
#------------------------------------- |
| 9 |
|
| 10 |
#' Cluster-robust variance-covariance matrix for an mmrm object. |
| 11 |
#' |
| 12 |
#' \code{vcovCR} returns a sandwich estimate of the variance-covariance matrix |
| 13 |
#' of a set of regression coefficient estimates from an |
| 14 |
#' \code{\link[mmrm]{mmrm}} object. |
| 15 |
#' |
| 16 |
#' @param cluster Optional expression or vector indicating which observations |
| 17 |
#' belong to the same cluster. If not specified, will be set to the subject |
| 18 |
#' variable from the \code{mmrm} model formula. |
| 19 |
#' @param target Optional matrix or vector describing the working |
| 20 |
#' variance-covariance model used to calculate the \code{CR2} and \code{CR4} |
| 21 |
#' adjustment matrices. If not specified, the target is taken to be the |
| 22 |
#' estimated variance-covariance structure of the \code{mmrm} object, with |
| 23 |
#' weights applied if present. |
| 24 |
#' @inheritParams vcovCR |
| 25 |
#' |
| 26 |
#' @return An object of class \code{c("vcovCR","clubSandwich")}, which consists |
| 27 |
#' of a matrix of the estimated variance of and covariances between the |
| 28 |
#' regression coefficient estimates. |
| 29 |
#' |
| 30 |
#' @seealso \code{\link{vcovCR}} |
| 31 |
#' |
| 32 |
#' @examples |
| 33 |
#' |
| 34 |
#' if (requireNamespace("mmrm", quietly = TRUE)) withAutoprint({ |
| 35 |
#' |
| 36 |
#' library(mmrm) |
| 37 |
#' data(fev_data, package = "mmrm") |
| 38 |
#' |
| 39 |
#' # Fit an mmrm model with unstructured covariance |
| 40 |
#' mmrm_fit <- mmrm( |
| 41 |
#' FEV1 ~ RACE + SEX + ARMCD * AVISIT + us(AVISIT | USUBJID), |
| 42 |
#' data = fev_data |
| 43 |
#' ) |
| 44 |
#' |
| 45 |
#' # CR2 cluster-robust variance estimator (cluster auto-detected) |
| 46 |
#' vcovCR(mmrm_fit, type = "CR2") |
| 47 |
#' |
| 48 |
#' # Coefficient tests with Satterthwaite degrees of freedom |
| 49 |
#' coef_test(mmrm_fit, vcov = "CR2", test = "Satterthwaite") |
| 50 |
#' |
| 51 |
#' # Fit a weighted mmrm model |
| 52 |
#' fev_data$wt <- 1 + rpois(nrow(fev_data), lambda = 3) |
| 53 |
#' mmrm_wt <- mmrm( |
| 54 |
#' FEV1 ~ RACE + SEX + ARMCD + us(AVISIT | USUBJID), |
| 55 |
#' data = fev_data, |
| 56 |
#' weights = fev_data$wt |
| 57 |
#' ) |
| 58 |
#' |
| 59 |
#' # CR2 works with weighted models |
| 60 |
#' vcovCR(mmrm_wt, type = "CR2") |
| 61 |
#' |
| 62 |
#' }) |
| 63 |
#' |
| 64 |
#' @export |
| 65 |
|
| 66 |
vcovCR.mmrm <- function(obj, cluster, type, target, inverse_var, form = "sandwich", ...) { |
| 67 |
if (missing(cluster)) { |
| 68 |
ff <- mmrm::component(obj, "full_frame") |
| 69 |
subject_var <- mmrm::component(obj, "subject_var") |
| 70 |
cluster <- droplevels(as.factor(ff[[subject_var]])) |
| 71 |
} |
| 72 |
if (missing(target)) target <- NULL |
| 73 |
if (missing(inverse_var)) inverse_var <- is.null(target) |
| 74 |
vcov_CR(obj, cluster = cluster, type = type, |
| 75 |
target = target, inverse_var = inverse_var, form = form) |
| 76 |
} |
| 77 |
|
| 78 |
#------------------------------------- |
| 79 |
# targetVariance() |
| 80 |
#------------------------------------- |
| 81 |
|
| 82 |
#' @export |
| 83 |
|
| 84 |
targetVariance.mmrm <- function(obj, cluster) { |
| 85 |
ff <- mmrm::component(obj, "full_frame") |
| 86 |
visit_var <- obj$formula_parts$visit_var |
| 87 |
group_var <- obj$formula_parts$group_var |
| 88 |
visit_levels <- levels(ff[[visit_var]]) |
| 89 |
visits <- ff[[visit_var]] |
| 90 |
vc <- mmrm::VarCorr(obj) |
| 91 |
|
| 92 |
obs_by_subject <- split(seq_along(cluster), cluster) |
| 93 |
|
| 94 |
if (is.null(group_var)) { |
| 95 |
# Non-grouped: single covariance matrix for all subjects |
| 96 |
V_list <- lapply(obs_by_subject, function(rows) { |
| 97 |
idx <- match(as.character(visits[rows]), visit_levels) |
| 98 |
unname(vc[idx, idx, drop = FALSE]) |
| 99 |
}) |
| 100 |
} else { |
| 101 |
# Grouped: pick the group-specific covariance matrix per subject |
| 102 |
groups <- ff[[group_var]] |
| 103 |
V_list <- lapply(obs_by_subject, function(rows) { |
| 104 |
subj_group <- as.character(groups[rows[1]]) |
| 105 |
idx <- match(as.character(visits[rows]), visit_levels) |
| 106 |
unname(vc[[subj_group]][idx, idx, drop = FALSE]) |
| 107 |
}) |
| 108 |
} |
| 109 |
|
| 110 |
# Multiply by weights if non-unit |
| 111 |
wts <- ff[["(weights)"]] |
| 112 |
|
| 113 |
if (all(wts == 1)) { |
| 114 |
return(V_list) |
| 115 |
} else { |
| 116 |
wt_inv_sqrt_j <- split(1 / sqrt(wts), cluster) |
| 117 |
V_weighted <- Map(function(iw, V) tcrossprod(iw) * V, iw = wt_inv_sqrt_j, V = V_list) |
| 118 |
return(V_weighted) |
| 119 |
} |
| 120 |
} |
| 121 |
|
| 122 |
#------------------------------------- |
| 123 |
# weightMatrix() |
| 124 |
#------------------------------------- |
| 125 |
|
| 126 |
#' @export |
| 127 |
|
| 128 |
weightMatrix.mmrm <- function(obj, cluster) { |
| 129 |
V_list <- targetVariance(obj, cluster) |
| 130 |
lapply(V_list, function(v) chol2inv(chol(v))) |
| 131 |
} |
| 132 |
|
| 133 |
#--------------------------------------- |
| 134 |
# bread and scaling constant |
| 135 |
#--------------------------------------- |
| 136 |
|
| 137 |
#' @export |
| 138 |
|
| 139 |
bread.mmrm <- function(x, ...) { |
| 140 |
vcov(x) * v_scale(x) |
| 141 |
} |
| 142 |
|
| 143 |
#' @export |
| 144 |
|
| 145 |
v_scale.mmrm <- function(obj) { |
| 146 |
mmrm::component(obj, "n_subjects") |
| 147 |
} |
| 148 |
|