1 ## compar.gee.R (2006-10-11)
3 ## Comparative Analysis with GEEs
5 ## Copyright 2002-2006 Emmanuel Paradis
7 ## This file is part of the R-package `ape'.
8 ## See the file ../COPYING for licensing issues.
10 compar.gee <- function(formula, data = NULL, family = "gaussian", phy,
11 scale.fix = FALSE, scale.value = 1)
13 if (is.null(data)) data <- parent.frame() else {
14 if(!any(is.na(match(rownames(data), phy$tip.label))))
15 data <- data[phy$tip.label, ]
16 else warning("the rownames of the data.frame and the tip labels of the tree
17 do not match: the former were ignored in the analysis.")
19 effect.assign <- attr(model.matrix(formula, data = data), "assign")
20 for (i in all.vars(formula)) {
21 if (any(is.na(eval(parse(text = i), envir = data))))
22 stop("the present method cannot (yet) be used directly with missing data: you may consider removing the species with missing data from your tree with the function `drop.tip'.")
24 if (is.null(phy$edge.length))
25 stop("the tree has no branch lengths.")
26 R <- vcv.phylo(phy, cor = TRUE)
27 id <- rep(1, dim(R)[1])
28 geemod <- do.call("gee", list(formula, id, data = data, family = family, R = R,
29 corstr = "fixed", scale.fix = scale.fix,
30 scale.value = scale.value))
31 W <- geemod$naive.variance
32 if (family == "binomial")
33 W <- summary(glm(formula, family = quasibinomial, data = data))$cov.scaled
35 dfP <- sum(phy$edge.length)*N / sum(diag(vcv.phylo(phy)))
36 obj <- list(call = geemod$call,
37 effect.assign = effect.assign,
39 coefficients = geemod$coefficients,
40 residuals = geemod$residuals,
41 family = geemod$family$family,
42 link = geemod$family$link,
46 class(obj) <- "compar.gee"
50 print.compar.gee <- function(x, ...)
55 coef <- matrix(rep(coef, 4), ncol = 4)
56 dimnames(coef) <- list(cnames,
57 c("Estimate", "S.E.", "t", "Pr(T > |t|)"))
58 df <- x$dfP - dim(coef)[1]
59 coef[, 2] <- sqrt(diag(x$W))
60 coef[, 3] <- coef[, 1]/coef[, 2]
62 warning("not enough degrees of freedom to compute P-values.")
64 } else coef[, 4] <- 2 * (1 - pt(abs(coef[, 3]), df))
65 residu <- quantile(as.vector(x$residuals))
66 names(residu) <- c("Min", "1Q", "Median", "3Q", "Max")
70 cat("\nNumber of observations: ", x$nobs, "\n")
72 cat(" Link: ", x$link, "\n")
73 cat(" Variance to Mean Relation:", x$family, "\n")
74 cat("\nSummary of Residuals:\n")
77 cat("\n\nCoefficients: (", sum(nas), " not defined because of singularities)\n",
79 else cat("\n\nCoefficients:\n")
81 cat("\nEstimated Scale Parameter: ", x$scale)
82 cat("\n\"Phylogenetic\" df (dfP): ", x$dfP, "\n")
85 drop1.compar.gee <- function(object, scope, quiet = FALSE, ...)
87 fm <- formula(object$call)
89 z <- attr(trm, "term.labels")
90 ind <- object$effect.assign
92 ans <- matrix(NA, n, 3)
95 ans[i, 1] <- length(wh)
96 ans[i, 2] <- t(object$coefficients[wh]) %*%
97 solve(object$W[wh, wh]) %*% object$coefficients[wh]
99 df <- object$dfP - length(object$coefficients)
100 if (df < 0) warning("not enough degrees of freedom to compute P-values.")
101 else ans[, 3] <- pf(ans[, 2], ans[, 1], df, lower.tail = FALSE)
102 colnames(ans) <- c("df", "F", "Pr(>F)")
104 if (any(attr(trm, "order") > 1) && !quiet)
105 warning("there is at least one interaction term in your model:
106 you should be careful when interpreting the significance of the main effects.")
107 class(ans) <- "anova"
108 attr(ans, "heading") <- c("Single term deletions\n\nModel:\n",
109 as.character(as.expression(fm)))