Debian Codegraph @

R/robu.R

raw
1 #-------------------------------------
2 # vcovCR with defaults
3 #-------------------------------------
4
5 #' Cluster-robust variance-covariance matrix for a robu 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[robumeta]{robu}} object.
10 #'
11 #' @param cluster Optional expression or vector indicating which observations
12 #' belong to the same cluster. If not specified, will be set to the
13 #' \code{studynum} used in fitting the \code{\link[robumeta]{robu}} object.
14 #' @param target Optional matrix or vector describing the working
15 #' variance-covariance model used to calculate the \code{CR2} and \code{CR4}
16 #' adjustment matrices. If not specified, the target is taken to be the
17 #' inverse of the estimated weights used in fitting the
18 #' \code{\link[robumeta]{robu}} object.
19 #' @inheritParams vcovCR
20 #'
21 #' @return An object of class \code{c("vcovCR","clubSandwich")}, which consists
22 #' of a matrix of the estimated variance of and covariances between the
23 #' regression coefficient estimates.
24 #'
25 #' @seealso \code{\link{vcovCR}}
26 #'
27 #' @export
28 #'
29 #' @examples
30 #'
31 #' if (requireNamespace("robumeta", quietly = TRUE)) withAutoprint({
32 #' library(robumeta)
33 #' data(hierdat)
34 #'
35 #' robu_fit <- robu(effectsize ~ binge + followup + sreport + age,
36 #' data = hierdat, studynum = studyid,
37 #' var.eff.size = var, modelweights = "HIER")
38 #' robu_fit
39 #'
40 #' robu_CR2 <- vcovCR(robu_fit, type = "CR2")
41 #' robu_CR2
42 #' coef_test(robu_fit, vcov = robu_CR2, test = c("Satterthwaite", "saddlepoint"))
43 #'
44 #' Wald_test(robu_fit, constraints = constrain_zero(c(2,4)), vcov = robu_CR2)
45 #' Wald_test(robu_fit, constraints = constrain_zero(2:5), vcov = robu_CR2)
46 #'
47 #' })
48 #'
49
50
51
52 vcovCR.robu <- function(obj, cluster, type, target, inverse_var, form = "sandwich", ...) {
53 if (missing(cluster)) cluster <- obj$study_orig_id
54 if (missing(target)) target <- NULL
55 if (missing(inverse_var)) inverse_var <- is.null(target) & (!obj$user_weighting)
56 vcov_CR(obj, cluster = cluster, type = type,
57 target = target, inverse_var = inverse_var, form = form)
58 }
59
60 #-----------------------------------------------
61 # coefficients
62 #-----------------------------------------------
63
64 #' @export
65
66 coef_CS.robu <- function(obj) {
67 beta <- as.vector(obj$b.r)
68 labs <- obj$reg_table$labels
69 if (is.factor(labs)) labs <- levels(labs)[labs]
70 names(beta) <- labs
71 beta
72 }
73
74 #-----------------------------------------------
75 # residuals
76 #-----------------------------------------------
77
78 #' @export
79
80 residuals_CS.robu <- function(obj) {
81 ord <- order(order(obj$study_orig_id))
82 resid <- obj$data.full$e.r[ord]
83
84 if (obj$user_weighting) {
85 pos_wts <- obj$data.full$userweights[ord] > 0
86 if (!all(pos_wts)) resid <- resid[pos_wts]
87 }
88
89 return(resid)
90 }
91
92
93 #-----------------------------------------------
94 # Model matrix
95 #-----------------------------------------------
96
97 #' @export
98
99 model_matrix.robu <- function(obj) {
100 ord <- order(order(obj$study_orig_id))
101 model_matrix <- obj$Xreg[ord,,drop=FALSE]
102
103 if (obj$user_weighting) {
104 pos_wts <- obj$data.full$userweights[ord] > 0
105 if (!all(pos_wts)) model_matrix <- model_matrix[pos_wts,,drop=FALSE]
106 }
107
108 return(model_matrix)
109 }
110
111 #-------------------------------------
112 # Get (model-based) working variance matrix
113 #-------------------------------------
114
115 #' @export
116
117 targetVariance.robu <- function(obj, cluster) {
118 ord <- order(order(obj$study_orig_id))
119 if (obj$user_weighting) {
120 pos_wts <- obj$data.full$userweights[ord] > 0
121 V <- obj$data.full$avg.var.eff.size[ord][pos_wts]
122 } else {
123 V <- mean(obj$data.full$r.weights) / obj$data.full$r.weights[ord]
124 }
125 matrix_list(V, cluster, "both")
126 }
127
128 #-------------------------------------
129 # Get weighting matrix
130 #-------------------------------------
131
132 #' @export
133
134 weights.robu <- function(object, ...) {
135 ord <- order(order(object$study_orig_id))
136 if (object$user_weighting) {
137 object$data.full$userweights[ord]
138 } else{
139 NULL
140 }
141 }
142
143 #' @export
144
145 weightMatrix.robu <- function(obj, cluster) {
146 ord <- order(order(obj$study_orig_id))
147 if (obj$user_weighting) {
148 W <- obj$data.full$userweights[ord]
149 W <- W[W > 0]
150 } else{
151 W <- obj$data.full$r.weights[ord]
152 }
153 w_scale <- mean(W)
154 W <- W / w_scale
155 W_list <- matrix_list(W, cluster, "both")
156 attr(W_list, "w_scale") <- w_scale
157 W_list
158 }
159
160 #---------------------------------------
161 # Get bread matrix and scaling constant
162 #---------------------------------------
163
164 #' @export
165
166 bread.robu <- function(x, ...) {
167 if (x$user_weighting) {
168 W <- x$data.full$userweights
169 } else{
170 W <- x$data.full$r.weights
171 }
172 x$N * chol2inv(chol(crossprod(x$Xreg, W * x$Xreg)))
173 }
174
175 #' @export
176
177 v_scale.robu <- function(obj) {
178 obj$N
179 }