amtest {multmod} | R Documentation |
Testing multiple endpoints or outcomes measures by providing a corrected significance level to assess the uncorrected, marginal p-values against.
amtest(modelList, varName, vcov. = c("sandwich", "model-based"), sig.level = 0.05, display = TRUE)
modelList |
list of model fits. The fits may be obtained using lm , glm , or
coxph . The fits should all be based on exactly the same dataset (data frame) and missing
values should be handled using na.omit . All models should contain the explanatory variable of
interest.
|
varName |
string specifying the name of the explanatory variable for which the effect is to be assessed. |
vcov. |
string specifying the type of estimated variance-covariance matrix to use for calculating the test statistics. Two options are available model-based estimated standard errors or robust estimated sandwich standard errors, which is the default. |
sig.level |
numeric specifying the nominal significance level or type I error rate (default is 0.05). |
display |
logical indicating whether or not the results should be shown at the command line. |
The function calculates a corrected overall significance level that asymptotically reaches the desired type I error using i.i.d. decompositions of parameter estimates by means of score components. The details are provided by Pipper and Ritz (2009).
The implemented method is less conservative than the Bonferroni correction and it flexibly adapts to the correlation structure between outcomes. Different types of outcome measures such as continuous, binomial, and event times, as well as missing values in outcome measures and explanatory variables can be handled by the procedure.
The corrected significance level as well as the Bonferroni correction and the marginal
p-values are shown by default (can be switched off using the argument 'display').
These values are also returned invisibly in a list.
Christian B. Pipper and Christian Ritz
Pipper, C. B. and Ritz, C (2009) An asymptotic correction for controlling the overall type I error when testing multiple outcomes, Submitted manuscript
##Example from: McCullagh, P. and Nelder, J.A. (1989,p. 239). Generalized Linear Models. Second Edition. ##Chapman & Hall/CRC. model1 <- glm(y1~group, family=binomial,na.action=na.omit,data=mn6.9) model2 <- glm(y2~group, family=binomial,na.action=na.omit,data=mn6.9) model3 <- glm(y3~group, family=binomial,na.action=na.omit,data=mn6.9) model4 <- glm(y4~group, family=binomial,na.action=na.omit,data=mn6.9) amtest(list(model1,model2,model3,model4),"group2",vcov.="model-based") amtest(list(model1,model2,model3,model4),"group2") ##Multiple testing in case of perfectly correlated outcomes amtest(list(model2,model2),"group2",vcov.="model-based") amtest(list(model2,model2),"group2") ## Multiple testing in a subgroup analysis and overall analysis subgroup<-mn6.9 set.seed(2982) subgroup[sample(1:2982,1500),]<-NA model1<- glm(y2~group, family=binomial,na.action=na.omit,data=mn6.9) model2<- glm(y2~group, family=binomial,na.action=na.omit,data=subgroup) amtest(list(model1,model2),"group2",vcov.="model-based") amtest(list(model1,model2),"group2")