Debian Codegraph @

R/rma-uni.R

raw
1 #-------------------------------------
2 # vcovCR with defaults
3 #-------------------------------------
4
5 #' Cluster-robust variance-covariance matrix for a rma.uni object.
6 #'
7 #' \code{vcovCR} returns a sandwich estimate of the variance-covariance matrix
8 #' of a set of regression coefficient estimates from a
9 #' \code{\link[metafor]{rma.uni}} object.
10 #'
11 #' @param cluster Expression or vector indicating which observations
12 #' belong to the same cluster. Required for \code{rma.uni} objects.
13 #' @param target Optional matrix or vector describing the working
14 #' variance-covariance model used to calculate the \code{CR2} and \code{CR4}
15 #' adjustment matrices. If not specified, the target is taken to be diagonal
16 #' with entries equal to the estimated marginal variance of the effect sizes.
17 #' @inheritParams vcovCR
18 #'
19 #' @return An object of class \code{c("vcovCR","clubSandwich")}, which consists
20 #' of a matrix of the estimated variance of and covariances between the
21 #' regression coefficient estimates.
22 #'
23 #' @seealso \code{\link{vcovCR}}
24 #'
25 #' @export
26 #'
27 #' @examples
28 #'
29 #' pkgs_available <-
30 #' requireNamespace("metafor", quietly = TRUE) &
31 #' requireNamespace("metadat", quietly = TRUE)
32 #'
33 #' if (pkgs_available) withAutoprint({
34 #'
35 #' library(metafor)
36 #' data(dat.assink2016, package = "metadat")
37 #'
38 #' mfor_fit <- rma.uni(yi ~ year + deltype, vi = vi,
39 #' data = dat.assink2016)
40 #' mfor_fit
41 #'
42 #' mfor_CR2 <- vcovCR(mfor_fit, type = "CR2", cluster = dat.assink2016$study)
43 #' mfor_CR2
44 #' coef_test(mfor_fit, vcov = mfor_CR2, test = c("Satterthwaite", "saddlepoint"))
45 #' Wald_test(mfor_fit, constraints = constrain_zero(2:4), vcov = mfor_CR2)
46 #'
47 #' })
48 #'
49
50
51 vcovCR.rma.uni <- function(obj, cluster, type, target, inverse_var, form = "sandwich", ...) {
52
53 if (missing(cluster)) stop("You must specify a clustering variable.")
54
55 # if (length(cluster) != nrow(model_matrix(obj))) cluster <- droplevels(as.factor(cluster[obj$not.na]))
56 # if (length(cluster) != nrow(model_matrix(obj))) stop("Clustering variable must have length equal to nrow(model_matrix(obj)).")
57
58 if (missing(target)) {
59 target <- NULL
60 if (missing(inverse_var)) inverse_var <- is.null(obj$weights) & obj$weighted
61 } else {
62 if (missing(inverse_var)) inverse_var <- FALSE
63 }
64
65 vcov_CR(obj, cluster = cluster, type = type,
66 target = target, inverse_var = inverse_var, form = form)
67 }
68
69 #-----------------------------------------------
70 # residuals_CS()
71 #-----------------------------------------------
72
73 #' @export
74
75 residuals_CS.rma <- function(obj) {
76 res <- residuals(obj)
77 not_na <- obj$not.na
78
79 if (length(res) == length(not_na)) res <- res[not_na]
80 if (!is.null(wts <- weights(obj)) && !all(pos_wts <- wts > 0)) {
81 res <- res[pos_wts]
82 }
83
84 return(res)
85 }
86
87 #-----------------------------------------------
88 # coefficients for location-scale model
89 #-----------------------------------------------
90 # coef_CS.default() used for rma.uni
91
92
93 #' @export
94 #'
95
96 coef_CS.rma.ls <- function(obj) {
97 metafor::coef.rma(obj)$beta
98 }
99
100 #-----------------------------------------------
101 # coefficients for location-scale model
102 #-----------------------------------------------
103 # coef_CS.default() used for rma.uni
104
105
106 #' @export
107 #'
108 coef_CS.rma.ls <- function(obj) {
109 metafor::coef.rma(obj)$beta
110 }
111
112
113 #-----------------------------------------------
114 # Model matrix for location-scale model
115 #-----------------------------------------------
116 # model_matrix.default() used for rma.uni
117
118
119 #' @export
120 #'
121 model_matrix.rma.ls <- function(obj) {
122 metafor::model.matrix.rma(obj)$location
123 }
124
125
126 # na.action
127
128 #' @export
129
130 na.action.rma <- function(object, ...) {
131
132 if (all(object$not.na)) return(NULL)
133
134 res <- which(!object$not.na)
135 class(res) <- "omit"
136 res
137
138 }
139
140 #-------------------------------------
141 # Get (model-based) working variance matrix
142 #-------------------------------------
143
144 #' @export
145
146 targetVariance.rma.uni <- function(obj, cluster) {
147
148 vi <- obj$vi
149
150 if (obj$weighted && !is.null(wts <- obj$weights) && !all(pos_wts <- wts > 0)) {
151 vi <- vi[pos_wts]
152 }
153
154 matrix_list(vi + obj$tau2, cluster, "both")
155 }
156
157 #-------------------------------------
158 # Get weighting matrix
159 #-------------------------------------
160
161 #' @export
162
163 weightMatrix.rma.uni <- function(obj, cluster) {
164 if (obj$weighted) {
165 if (is.null(obj$weights)) {
166 wi <- 1 / (obj$vi + obj$tau2)
167 } else {
168 wi <- obj$weights
169 wi <- wi[wi > 0]
170 }
171 } else {
172 wi <- rep(1, obj$k)
173 }
174 w_scale <- mean(wi)
175 wi <- wi / w_scale
176
177 W_list <- matrix_list(wi, cluster, "both")
178 attr(W_list, "w_scale") <- w_scale
179 W_list
180 }
181
182 #---------------------------------------
183 # Get bread matrix and scaling constant
184 #---------------------------------------
185
186 #' @export
187
188 bread.rma.uni <- function(x, ...) {
189 X_mat <- model_matrix(x)
190 if (x$weighted) {
191 if (is.null(x$weights)) {
192 wi <- 1 / (x$vi + x$tau2)
193 } else {
194 wi <- x$weights
195 wi <- wi[wi > 0]
196 }
197 XWX <- crossprod(X_mat, wi * X_mat)
198 } else {
199 XWX <- crossprod(X_mat)
200 }
201 B <- chol2inv(chol(XWX)) * nobs(x)
202 rownames(B) <- colnames(B) <- colnames(X_mat)
203 B
204 }
205
206 #' @export
207
208 v_scale.robu <- function(obj) {
209 nobs(obj)
210 }