DearBegg(selectMeta)
DearBegg()所属R语言包：selectMeta
Compute the nonparametric weight function from Dear and Begg (1992)
非参数的权重函数计算亲爱的贝格（1992）
译者：生物统计家园网 机器人LoveR
描述----------Description----------
In Dear and Begg (1992) it was proposed to nonparametrically estimate via maximum likelihood the weight function w in a selection model via pooling p-values in groups of 2 and assuming a piecewise constant w. The function DearBegg implements estimation of w via a coordinate-wise Newton-Raphson algorithm as described in Dear and Begg (1992). In addition, the function DearBeggMonotone enables computation of the weight function in the same model under the constraint that it is non-increasing, see Rufibach (2011). To this end we use the differential evolution algorithm described in Ardia et al (2010a, b) and implemented in Mullen et al (2009).
在尊敬和贝格（1992）提出了非参数估计，通过最大似然权重函数w的选择模型通过集中p值2组，并假设分段常数w 。的功能DearBegg实现估算w的通过一个坐标明智Newton-Raphson算法如尊敬和贝格（1992）中描述的。此外，函数DearBeggMonotone使权重函数的计算，在同一模型中的约束下，它是不增加，Rufibach（2011）。为此，我们使用差分进化算法（2010A，B）在阿尔迪亚等，在马伦等人（2009）和实施。
用法----------Usage----------
DearBegg(y, u, lam = 2, tolerance = 10^-10, maxiter = 1000,
trace = TRUE)
DearBeggMonotone(y, u, lam = 2, maxiter = 1000, CR = 0.9,
NP = NA, trace = TRUE)
Hij(theta, sigma, y, u, teststat)
DearBeggLoglik(w, theta, sigma, y, u, hij, lam)
DearBeggToMinimize(vec, y, u, lam)
参数----------Arguments----------
参数：y
Normally distributed effect sizes.
通常情况下分布的影响的大小。
参数：u
Associated standard errors.
相关的标准误差。
参数：lam
Weight of the first entry of w in the likelihood function. Dear and Begg (1992) recommend to use lam = 2.
重量w的似然函数中的第一个条目。亲爱的贝格（1992）建议，使用lam = 2。
参数：tolerance
Stopping criterion for Newton-Raphson.
牛顿 - 拉夫逊停止标准。
参数：maxiter
Maximal number of iterations for Newton-Raphson.
牛顿 - 拉夫逊迭代的最大数目。
参数：trace
If TRUE, progress of the algorithm is shown.
如果TRUE，进步的算法。
参数：CR
Parameter that is given to DEoptim. See the help file of the function DEoptim.control for details.
参数是给DEoptim。的功能DEoptim.control的详细信息，请参阅帮助文件。
参数：NP
Parameter that is given to DEoptim. See the help file of the function DEoptim.control for details.
参数是给DEoptim。的功能DEoptim.control的详细信息，请参阅帮助文件。
参数：w
Weight function, parametrized as vector of length 1 + \lfloor(n / 2)\rfloor where n is the number of studies, i.e. the length of y.
权函数，参数化向量的长度1 + \lfloor(n / 2)\rfloor其中n是大量的研究，即长度y。
参数：theta
Effect size estimate.
效果大小的估计。
参数：sigma
Random effects variance component.
随机效应方差分量。
参数：hij
Integral of density over a constant piece of w. See Rufibach (2011, Appendix) for details.
积分密度恒定的一块w。的详细信息，请参阅Rufibach（2011年，附录）。
参数：vec
Vector of parameters over which we maximize.
我们最大限度地参数向量。
参数：teststat
Vector of test statistics, equals |y| / u.
向量检验统计量，等于|y| / u。
值----------Value----------
A list consisting of the following elements:
列表包括以下内容：
参数：<code>w</code>
Vector of estimated weights.
矢量的估计重量。
参数：<code>theta</code>
Estimate of the combined effect in the Dear and Begg model.
亲爱的的综合效应和贝格模型的估计。
参数：<code>sigma</code>
Estimate of the random effects component variance.
随机效应成分方差的估计。
参数：p
p-values computed from the inputed test statistics, ordered in decreasing order.
p值计算从输入型检验统计量，递减顺序排序。
参数：y
Effect sizes, ordered in decreasing order of p-values.
影响的大小，p值递减的顺序排序。
参数：u
Standard errors, ordered in decreasing order of p-values.
标准错误，p值递减的顺序排序。
参数：loglik
Value of the log-likelihood at the maximum.
在最大对数似然的价值。
参数：DEoptim.res
Only available in DearBeggMonotone. Provides the object that is outputted by DEoptim.
仅在DearBeggMonotone。提供输出DEoptim的对象。
（作者）----------Author(s)----------
Kaspar Rufibach (maintainer), <a href="mailto:kaspar.rufibach@gmail.com">kaspar.rufibach@gmail.com</a>, <br> <a href="http://www.kasparrufibach.ch">http://www.kasparrufibach.ch</a>
参考文献----------References----------
Differential Evolution ('DEoptim') for Non-Convex Portfolio Optimization.
Differential Evolution Optimization in 'R'. Version 2.0-7.
An Approach for Assessing Publication Bias Prior to Performing a Meta-Analysis. Statist. Sci., 7(2), 237–245.
'DEoptim': An 'R' Package for Global Optimization by Differential Evolution.
Selection Models with Monotone Weight Functions in Meta-Analysis. Biom. J., to appear.
参见----------See Also----------
IyenGreen for a parametric selection model.
IyenGreen的参数选择模式。
实例----------Examples----------
## Not run: [＃不运行：]
##------------------------------------------[＃------------------------------------------]
## Analysis of Hedges & Olkin dataset[＃对冲Olkin型数据集分析]
## re-analyzed in Iyengar & Greenhouse, Dear & Begg[＃重新分析，亲爱的贝格在艾扬格温室]
##------------------------------------------[＃------------------------------------------]
data(education)
t <- education$t
q <- education$q
N <- education$N
y <- education$theta
u <- sqrt(2 / N)
n <- length(y)
k <- 1 + floor(n / 2)
lam1 <- 2
## compute p-values[＃计算p-值]
p <- 2 * pnorm(-abs(t))
##------------------------------------------[＃------------------------------------------]
## compute all weight functions available[＃计算所有可用的权重函数]
## in this package[在此包中＃]
##------------------------------------------[＃------------------------------------------]
## weight functions from Iyengar & Greenhouse (1988)[艾扬格和温室的权重函数（1988）]
res1 <- IyenGreenMLE(t, q, N, type = 1)
res2 <- IyenGreenMLE(t, q, N, type = 2)
## weight function from Dear & Begg (1992)[＃重函数从亲爱贝格（1992）]
res3 <- DearBegg(y, u, lam = lam1)
## monotone version of Dear & Begg, as introduced in Rufibach (2011)[＃亲爱的贝格，单调的版本中所介绍的Rufibach（2011）]
set.seed(1977)
res4 <- DearBeggMonotone(y, u, lam = lam1, maxiter = 1000, CR = 1)
## plot[＃图]
plot(0, 0, type = "n", xlim = c(0, 1), ylim = c(0, 1), xlab = "p-values",
ylab = "estimated weight function")
ps <- seq(0, 1, by = 0.01)
rug(p, lwd = 3)
lines(ps, IyenGreenWeight(-qnorm(ps / 2), b = res1$beta, q = 50,
type = 1, alpha = 0.05), lwd = 3, col = 2)
lines(ps, IyenGreenWeight(-qnorm(ps / 2), b = res2$beta, q = 50,
type = 2, alpha = 0.05), lwd = 3, col = 4)
weightLine(p, w = res3$w, col0 = 3, lwd0 = 3, lty0 = 2)
weightLine(p, w = res4$w, col0 = 6, lwd0 = 2, lty0 = 1)
legend("topright", c(expression("Iyengar & Greenhouse (1988) w"[1]),
expression("Iyengar & Greenhouse (1988) w"[2]), "Dear and Begg (1992)",
"Rufibach (2011)"), col = c(2, 4, 3, 6), lty = c(1, 1, 2, 1),
lwd = c(3, 3, 3, 2), bty = "n")
## compute selection bias[＃计算选择偏倚]
eta <- sqrt(res4$sigma ^ 2 + res4$u ^ 2)
bias <- effectBias(res4$y, res4$u, res4$w, res4$theta, eta)
bias
##------------------------------------------[＃------------------------------------------]
## Compute p-value to assess null hypothesis of no selection,[计算P值，以评估零假设没有选择，]
## as described in Rufibach (2011, Section 6)[＃所描述的Rufibach（2011年第6节）]
## We use the package 'meta' to compute initial estimates for[＃我们使用的包“元”计算初步估计]
## theta and sigma[＃θ和标准差]
##------------------------------------------[＃------------------------------------------]
library(meta)
## compute null parameters[＃计算空参数。]
meta.edu <- metagen(TE = y, seTE = u, sm = "MD", level = 0.95,
comb.fixed = TRUE, comb.random = TRUE)
theta0 <- meta.edu$TE.random
sigma0 <- meta.edu$tau
M <- 1000
res <- DearBeggMonotonePvalSelection(y, u, theta0, sigma0, lam = lam1,
M = M, maxiter = 1000)
## plot all the computed monotone functions[＃图所有计算的单调函数]
plot(0, 0, xlim = c(0, 1), ylim = c(0, 1), type = "n", xlab = "p-values",
ylab = expression(w(p)))
abline(v = 0.05, lty = 3)
for (i in 1:M){weightLine(p, w = res$res.mono[1:k, i], col0 = grey(0.8),
lwd0 = 1, lty0 = 1)}
rug(p, lwd = 2)
weightLine(p, w = res$mono0, col0 = 2, lwd0 = 1, lty0 = 1)
## =======================================================================[＃================================================= ======================]
##------------------------------------------[＃------------------------------------------]
## Analysis second-hand tobacco smoke dataset[＃二手烟草烟雾分析数据集]
## Rothstein et al (2005), Publication Bias in Meta-Analysis, Appendix A[＃Rothstein等人（2005），Meta分析发表偏倚，附录A]
##------------------------------------------[＃------------------------------------------]
data(passive_smoking)
u <- passive_smoking$selnRR
y <- passive_smoking$lnRR
n <- length(y)
k <- 1 + floor(n / 2)
lam1 <- 2
res2 <- DearBegg(y, u, lam = lam1)
set.seed(1)
res3 <- DearBeggMonotone(y = y, u = u, lam = lam1, maxiter = 2000, CR = 1)
plot(0, 0, type = "n", xlim = c(0, 1), ylim = c(0, 1), pch = 19, col = 1,
xlab = "p-values", ylab = "estimated weight function")
weightLine(rev(sort(res2$p)), w = res2$w, col0 = 2, lwd0 = 3, lty0 = 2)
weightLine(rev(sort(res3$p)), w = res3$w, col0 = 4, lwd0 = 2, lty0 = 1)
legend("bottomright", c("Dear and Begg (1992)", "Rufibach (2011)"), col =
c(2, 4), lty = c(2, 1), lwd = c(3, 2), bty = "n")
## compute selection bias[＃计算选择偏倚]
eta <- sqrt(res3$sigma ^ 2 + res3$u ^ 2)
bias <- effectBias(res3$y, res3$u, res3$w, res3$theta, eta)
bias
##------------------------------------------[＃------------------------------------------]
## Compute p-value to assess null hypothesis of no selection[计算P值，以评估没有选择的零假设]
##------------------------------------------[＃------------------------------------------]
## compute null parameters[＃计算空参数。]
meta.toba <- metagen(TE = y, seTE = u, sm = "MD", level = 0.95,
comb.fixed = TRUE, comb.random = TRUE)
theta0 <- meta.toba$TE.random
sigma0 <- meta.toba$tau
M <- 1000
res <- DearBeggMonotonePvalSelection(y, u, theta0, sigma0, lam = lam1,
M = M, maxiter = 2000)
## plot all the computed monotone functions[＃图所有计算的单调函数]
plot(0, 0, xlim = c(0, 1), ylim = c(0, 1), type = "n", xlab = "p-values",
ylab = expression(w(p)))
abline(v = 0.05, lty = 3)
for (i in 1:M){weightLine(p, w = res$res.mono[1:k, i], col0 = grey(0.8),
lwd0 = 1, lty0 = 1)}
rug(p, lwd = 2)
weightLine(p, w = res$mono0, col0 = 2, lwd0 = 1, lty0 = 1)
## End(Not run)[＃（不执行）]
转载请注明:出自 生物统计家园网(http://www.biostatistic.net)。
注：
注1：为了方便大家学习，本文档为生物统计家园网机器人LoveR翻译而成，仅供个人R语言学习参考使用，生物统计家园保留版权。
注2：由于是机器人自动翻译，难免有不准确之处，使用时仔细对照中、英文内容进行反复理解，可以帮助R语言的学习。
注3：如遇到不准确之处，请在本贴的后面进行回帖，我们会逐渐进行修订。
|
发表于 2012-9-30 00:28:08
