Package multcomp

March 9, 2012
Title Simultaneous Inference in General Parametric Models Version 1.2-12 Date 2012-03-09 Description Simultaneous tests and condence intervals for general linear hypotheses in parametric models, including linear,generalized linear, linear mixed effects, and survival models. The package includes demos reproducing analyzes presented in the book Multiple Comparisons Using R (Bretz, Hothorn,Westfall, 2010, CRC Press). Depends stats, graphics, mvtnorm (>= 0.8-0), survival (>= 2.35-7) Suggests lme4 (>= 0.99937516), nlme, robustbase, mboost, coin, MASS,car, foreign, xtable, sandwich, lmtest, coxme (>= 2.21) URL The publishers web page is http://www.crcpress.com/product/isbn/978158488574 LazyData yes License GPL-2 Author Torsten Hothorn [aut, cre], Frank Bretz [aut], Peter Westfal [aut] Maintainer Torsten Hothorn <[email protected]> Repository CRAN Date/Publication 2012-03-09 14:09:31

R topics documented:
adevent . cftest . . . cholesterol cld . . . . cml . . . contrMat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 3 4 5 7 8

2 detergent . . fattyacid . . . glht . . . . . glht-methods litter . . . . . modelparm . mtept . . . . parm . . . . . plot.cld . . . recovery . . . sbp . . . . . . trees513 . . . waste . . . . Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

adevent . . . . . . . . . . . . . . . . . . . . . . . . . . 10 11 12 16 19 20 21 22 23 25 26 27 28 30

adevent

Adverse Events Data

Description Indicators of 28 adverse events in a two-arm clinical trial. Usage data(adevent) Format A data frame with 160 observations on the following 29 variables. E1 a factor with levels no event event E2 a factor with levels no event event E3 a factor with levels no event event E4 a factor with levels no event event E5 a factor with levels no event event E6 a factor with levels no event event E7 a factor with levels no event event E8 a factor with levels no event event E9 a factor with levels no event event E1 a factor with levels no event event E11 a factor with levels no event event E12 a factor with levels no event event E13 a factor with levels no event event E14 a factor with levels no event event

cftest E15 a factor with levels no event event E16 a factor with levels no event event E17 a factor with levels no event event E18 a factor with levels no event event E19 a factor with levels no event event E2 a factor with levels no event event E21 a factor with levels no event event E22 a factor with levels no event event E23 a factor with levels no event event E24 a factor with levels no event event E25 a factor with levels no event event E26 a factor with levels no event event E27 a factor with levels no event event E28 a factor with levels no event event group group indicator. Details

The data is provided by Westfall et al. (1999, p. 242) and contains binary indicators of 28 adverse events (E1,..., E28) for two arms (group). Source P. H. Westfall, R. D. Tobias, D. Rom, R. D. Wolnger, Y. Hochberg (1999). Multiple Comparisons and Multiple Tests Using the SAS System. Cary, NC: SAS Institute Inc.

cftest

Testing Estimated Coefcients

Description A convenience function for univariate testing z- and t-tests of estimated model coefcients Usage cftest(model, ...) Arguments model ... a tted model. additional arguments passed to summary.glht.

4 Details The usual z- or t-tests are tested without adjusting for multiplicity Value An object of class summary.glht. See Also coeftest Examples
## The function is currently defined as function(model, ...) summary(glht(model), test = univariate(), ...) lmod <- lm(dist ~ speed, data = cars) summary(lmod) cftest(lmod)

cholesterol

cholesterol

Cholesterol Reduction Data Set

Description Cholesterol reduction for ve treatments. Usage data("cholesterol") Format This data frame contains the following variables trt treatment groups, a factor at levels 1time, 2times, 4times, drugD and drugE. response cholesterol reduction. Details A clinical study was conducted to assess the effect of three formulations of the same drug on reducing cholesterol. The formulations were 20mg at once (1time), 10mg twice a day (2times), and 5mg four times a day (4times). In addition, two competing drugs were used as control group (drugD and drugE). The purpose of the study was to nd which of the formulations, if any, is efcacious and how these formulations compare with the existing drugs.

cld Source

P. H. Westfall, R. D. Tobias, D. Rom, R. D. Wolnger, Y. Hochberg (1999). Multiple Comparisons and Multiple Tests Using the SAS System. Cary, NC: SAS Institute Inc., page 153. Examples
### adjusted p-values for all-pairwise comparisons in a one-way layout ### set up ANOVA model amod <- aov(response ~ trt, data = cholesterol) ### set up multiple comparisons object for all-pair comparisons cht <- glht(amod, linfct = mcp(trt = "Tukey")) ### cf. Westfall et summary(cht, test = summary(cht, test = summary(cht, test = al. (1999, page 171) univariate()) adjusted("Shaffer")) adjusted("Westfall"))

### use only a subset of all pairwise hypotheses K <- contrMat(table(cholesterol$trt), type="Tukey") Ksub <- rbind(K[c(1,2,5),], "D - test" = c(-1, -1, -1, 3, ), "E - test" = c(-1, -1, -1, , 3)) ### reproduce results in Westfall et al. (1999, page 172) ### note: the ordering of our estimates here is different amod <- aov(response ~ trt - 1, data = cholesterol) summary(glht(amod, linfct = mcp(trt = Ksub[,5:1])), test = adjusted("Westfall"))

cld

Set up a compact letter display of all pair-wise comparisons

Description Extract information from glht, summary.glht or confint.glht objects which is required to create and plot compact letter displays of all pair-wise comparisons. Usage ## S3 method for class cld(object, level = . ## S3 method for class cld(object, level = . ## S3 method for class cld(object, decreasing summary.glht 5, decreasing = FALSE, ...) glht 5, decreasing = FALSE, ...) confint.glht = FALSE, ...)

6 Arguments object level decreasing ... Details An object of class glht, summary.glht or confint.glht.

cld

Signicance-level to be used to term a specic pair-wise comparison signicant. logical. Should the order of the letters be increasing or decreasing? additional arguments.

This function extracts all the information from glht, summary.glht or confint.glht objects that is required to create a compact letter display of all pair-wise comparisons. In case the contrast matrix is not of type "Tukey", an error is issued. In case of confint.glht objects, a pair-wise comparison is termed signicant whenever a particular condence interval contains 0. Otherwise, p-values are compared to the value of "level". Once, this information is extracted, plotting of all pair-wise comparisons can be carried out. Value An object of class cld, a list with items: y yname x weights lp covar signif Values of the response variable of the original model. Name of the response variable. Values of the variable used to compute Tukey contrasts. Weights used in the tting process. Predictions from the tted model. A logical indicating whether the tted model contained covariates. Vector of logicals indicating signicant differences with hyphenated names that identify pair-wise comparisons.

References Hans-Peter Piepho (2004), An Algorithm for a Letter-Based Representation of All-Pairwise Comparisons, Journal of Computational and Graphical Statistics, 13(2), 456466. See Also glht plot.cld Examples
### multiple comparison procedures ### set up a one-way ANOVA data(warpbreaks) amod <- aov(breaks ~ tension, data = warpbreaks) ### specify all pair-wise comparisons among levels of variable "tension" tuk <- glht(amod, linfct = mcp(tension = "Tukey")) ### extract information tuk.cld <- cld(tuk)

cml
### use sufficiently large upper margin old.par <- par( mai=c(1,1,1.25,1)) ### plot plot(tuk.cld) par(old.par) ### now using covariates data(warpbreaks) amod2 <- aov(breaks ~ tension + wool, data = warpbreaks) ### specify all pair-wise comparisons among levels of variable "tension" tuk2 <- glht(amod2, linfct = mcp(tension = "Tukey")) ### extract information tuk.cld2 <- cld(tuk2) ### use sufficiently large upper margin old.par <- par( mai=c(1,1,1.25,1)) ### plot using different colors plot(tuk.cld2, col=c("black", "red", "blue")) par(old.par)

### set up all pair-wise comparisons for count data data(Titanic) mod <- glm(Survived ~ Class, data = as.data.frame(Titanic), weights = Freq, family = binomial()) ### specify all pair-wise comparisons among levels of variable "Class" glht.mod <- glht(mod, mcp(Class = "Tukey")) ### extract information mod.cld <- cld(glht.mod) ### use sufficiently large upper margin old.par <- par(mai=c(1,1,1.5,1)) ### plott plot(mod.cld) par(old.par)

cml

Chronic Myelogenous Leukemia survival data.

Description Survival in a randomised trial comparing three treatments for Chronic Myelogeneous Leukemia (simulated data). Usage data("cml") Format A data frame with 507 observations on the following 7 variables. center a factor with 54 levels indicating the study center. treatment a factor with levels trt1, trt2, trt3 indicating the treatment group.

8 sex sex (0 = female, 1 = male) age age in years riskgroup risk group (0 = low, 1 = medium, 2 = high) status censoring status (FALSE = censored, TRUE = dead) time survival or censoring time in days. Details

contrMat

The data are simulated according to structure of the data by the German CML Study Group used in Hehlmann (1994). Source R. Hehlmann, H. Heimpel, J. Hasford, H.J. Kolb, H. Pralle, D.K. Hossfeld, W. Queisser, H. Loefer, A. Hochhaus, B. Heinze (1994), Randomized comparison of interferon-alpha with busulfan and hydroxyurea in chronic myelogenous leukemia. The German CML study group. Blood 84(12):4064-4077. Examples
if (require("coxme")) { data("cml") ### one-sided simultaneous confidence intervals for many-to-one ### comparisons of treatment effects concerning time of survival ### modeled by a frailty Cox model with adjustment for further ### covariates and center-specific random effect. cml_coxme <- coxme(Surv(time, status) ~ treatment + sex + age + riskgroup + (1|center), data = cml) glht_coxme <- glht(model = cml_coxme, linfct = mcp(treatment = "Dunnett"), alternative = "greater") ci_coxme <- confint(glht_coxme) exp(ci_coxme$confint)[1:2,] }

contrMat

Contrast Matrices

Description Computes contrast matrices for several multiple comparison procedures. Usage contrMat(n, type = c("Dunnett", "Tukey", "Sequen", "AVE", "Changepoint", "Williams", "Marcus", "McDermott", "UmbrellaWilliams", "GrandMean"), base = 1)

contrMat Arguments n type base a (possibly named) vector of sample sizes for each group. type of contrast.

an integer specifying which group is considered the baseline group for Dunnett contrasts.

Details Computes the requested matrix of contrasts for comparisons of mean levels.

Value The matrix of contrasts with appropriate row names is returned.

References Frank Bretz, Torsten Hothorn and Peter Westfall (2010), Multiple Comparisons Using R, CRC Press, Boca Raton. Frank Bretz, Alan Genz and Ludwig A. Hothorn (2001), On the numerical availability of multiple comparison procedures. Biometrical Journal, 43(5), 645656.

Examples
n <- c(1 ,2 ,3 ,4 ) names(n) <- paste("group", 1:4, sep="") contrMat(n) # Dunnett is default contrMat(n, base = 2) # use second level as baseline contrMat(n, type = "Tukey") contrMat(n, type = "Sequen") contrMat(n, type = "AVE") contrMat(n, type = "Changepoint") contrMat(n, type = "Williams") contrMat(n, type = "Marcus") contrMat(n, type = "McDermott") ### Umbrella-protected Williams contrasts, i.e. a sequence of ### Williams-type contrasts with groups of higher order ### stepwise omitted contrMat(n, type = "UmbrellaWilliams") ### comparison of each group with grand mean of all groups contrMat(n, type = "GrandMean")

10

detergent

detergent

Detergent Durability Data Set

Description Detergent durability in an incomplete two-way design. Usage data("detergent") Format This data frame contains the following variables detergent detergent, a factor at levels A, B, C, D, and E. block block, a factor at levels B_1, ..., B_1 . plates response variable: number of plates washed before the foam disappears. Details Plates were washed with ve detergent varieties, in ten blocks. A complete design would have 50 combinations, here only three detergent varieties in each block were applied in a balanced incomplete block design. Note that there are six observations taken at each detergent level. Source H. Scheffe (1959). The Analysis of Variance. New York: John Wiley & Sons, page 189. P. H. Westfall, R. D. Tobias, D. Rom, R. D. Wolnger, Y. Hochberg (1999). Multiple Comparisons and Multiple Tests Using the SAS System. Cary, NC: SAS Institute Inc., page 189. Examples
### set up two-way ANOVA without interactions amod <- aov(plates ~ block + detergent, data = detergent) ### set up all-pair comparisons dht <- glht(amod, linfct = mcp(detergent = "Tukey")) ### see Westfall et al. (1999, p. 19 ) confint(dht) ### see Westfall et summary(dht, test = summary(dht, test = summary(dht, test = al. (1999, p. 192) univariate()) adjusted("Shaffer")) adjusted("Westfall"))

fattyacid

11

fattyacid

Fatty Acid Content of Bacillus simplex.

Description Fatty acid content of different putative ecotypes of Bacillus simplex. Usage data("fattyacid") Format A data frame with 93 observations on the following 2 variables. PE a factor with levels PE3, PE4, PE5, PE6, PE7, PE9 indicating the putative ecotype (PE). FA a numeric vector indicating the content of fatty acid (FA). Details The data give the fatty acid content for different putative ecotypes of Bacillus simplex. Variances of the values of fatty acid are heterogeneous among the putative ecotypes. Source J. Sikorski, E. Brambilla, R. M. Kroppenstedt, B. J. Tindal (2008), The temperature adaptive fatty acid content in Bacillus simplex strains from Evolution Canyon, Israel. Microbiology 154, 24162426. Examples
if (require("sandwich")) { data("fattyacid") ### all-pairwise comparisons of the means of fatty acid content ### FA between different putative ecotypes PE accounting for ### heteroscedasticity by using a heteroscedastic consistent ### covariance estimation amod <- aov(FA ~ PE, data = fattyacid) amod_glht <- glht(amod, mcp(PE = "Tukey"), vcov = vcovHC) summary(amod_glht) ### simultaneous confidence intervals for the differences of ### means of fatty acid content between the putative ecotypes confint(amod_glht) }

12

glht

glht

General Linear Hypotheses

Description General linear hypotheses and multiple comparisons for parametric models, including generalized linear models, linear mixed effects models, and survival models. Usage ## S3 method for class matrix glht(model, linfct, alternative = c("two.sided", "less", "greater"), rhs = , ...) ## S3 method for class character glht(model, linfct, ...) ## S3 method for class expression glht(model, linfct, ...) ## S3 method for class mcp glht(model, linfct, ...) mcp(..., interaction_average = FALSE, covariate_average = FALSE) Arguments model a tted model, for example an object returned by lm, glm, or aov etc. It is assumed that coef and vcov methods are available for model. For multiple comparisons of means, methods model.matrix, model.frame and terms are expected to be available for model as well. a specication of the linear hypotheses to be tested. Linear functions can be specied by either the matrix of coefcients or by symbolic descriptions of one or more linear hypotheses. Multiple comparisons in AN(C)OVA models are specied by objects returned from function mcp.. a character string specifying the alternative hypothesis, must be one of "two.sided" (default), "greater" or "less". You can specify just the initial letter.

linfct

alternative

rhs an optional numeric vector specifying the right hand side of the hypothesis. interaction_average logical indicating if comparisons are averaging over interaction terms. Experimental! covariate_average logical indicating if comparisons are averaging over additional covariates. Experimental! ... additional arguments to function modelparm in all glht methods. For function mcp, multiple comparisons are dened by matrices or symbolic descriptions specifying contrasts of factor levels where the arguments correspond to factor names.

glht Details

13

A general linear hypothesis refers to null hypotheses of the form H0 : K = m for some parametric model model with parameter estimates coef(model). The null hypothesis is specied by a linear function K, the direction of the alternative and the right hand side m. Here, alternative equal to "two.sided" refers to a null hypothesis H0 : K = m, whereas "less" corresponds to H0 : K m and "greater" refers to H0 : K m. The right hand side vector m can be dened via the rhs argument. The generic method glht dispatches on its second argument (linfct). There are three ways, and thus methods, to specify linear functions to be tested: 1) The matrix of coefcients K can be specied directly via the linfct argument. In this case, the number of columns of this matrix needs to correspond to the number of parameters estimated by model. It is assumed that appropriate coef and vcov methods are available for model (modelparm deals with some exceptions). 2) A symbolic description, either a character or expression vector passed to glht via its linfct argument, can be used to dene the null hypothesis. A symbolic description must be interpretable as a valid R expression consisting of both the left and right hand side of a linear hypothesis. Only the names of coef(model) must be used as variable names. The alternative is given by the direction under the null hypothesis (= or == refer to "two.sided", <= means "greater" and >= indicates "less"). Numeric vectors of length one are valid values for the right hand side. 3) Multiple comparisons of means are dened by objects of class mcp as returned by the mcp function. For each factor, which is included in model as independent variable, a contrast matrix or a symbolic description of the contrasts can be specied as arguments to mcp. A symbolic description may be a character or expression where the factor levels are only used as variables names. In addition, the type argument to the contrast generating function contrMat may serve as a symbolic description of contrasts as well. The mcp function must be used with care when dening parameters of interest in two-way ANOVA or ANCOVA models. Here, the denition of treatment differences (such as Tukeys all-pair comparisons or Dunnetts comparison with a control) might be problem specic. Because it is impossible to determine the parameters of interest automatically in this case, mcp in multcomp version 1.0-0 and higher generates comparisons for the main effects only, ignoring covariates and interactions (older versions automatically averaged over interaction terms). A warning is given. We refer to Hsu (1996), Chapter 7, and Searle (1971), Chapter 7.3, for further discussions and examples on this issue. glht extracts the number of degrees of freedom for models of class lm (via modelparm) and the exact multivariate t distribution is evaluated. For all other models, results rely on the normal approximation. Alternatively, the degrees of freedom to be used for the evaluation of multivariate t distributions can be given by the additional df argument to modelparm specied via .... glht methods return a specication of the null hypothesis H0 : K = m. The value of the linear function K can be extracted using the coef method and the corresponding covariance matrix is available from the vcov method. Various simultaneous and univariate tests and condence intervals are available from summary.glht and confint.glht methods, respectively. A more detailed description of the underlying methodology is available from Hothorn et al. (2008) and Bretz et al. (2010).

14 Value An object of class glht, more specically a list with elements model linfct rhs coef vcov df alternative type a tted model, used in the call to glht the matrix of linear functions the vector of right hand side values m the values of the linear functions the covariance matrix of the values of the linear functions

glht

optionally, the degrees of freedom when the exact t distribution is used for inference a character string specifying the alternative hypothesis optionally, a character string giving the name of the specic procedure

with print, summary, confint, coef and vcov methods being available. When called with linfct being an mcp object, an additional element focus is available storing the names of the factors under test. References Frank Bretz, Torsten Hothorn and Peter Westfall (2010), Multiple Comparisons Using R, CRC Press, Boca Raton. Shayle R. Searle (1971), Linear Models. John Wiley \& Sons, New York. Jason C. Hsu (1996), Multiple Comparisons. Chapman & Hall, London. Torsten Hothorn, Frank Bretz and Peter Westfall (2008), Simultaneous Inference in General Parametric Models. Biometrical Journal, 50(3), 346363; See vignette("generalsiminf", package = "multcomp"). Examples
### multiple linear model, swiss data lmod <- lm(Fertility ~ ., data = swiss) ### test of H_ : all regression coefficients are zero ### (ignore intercept) ### define coefficients of linear function directly K <- diag(length(coef(lmod)))[-1,] rownames(K) <- names(coef(lmod))[-1] K ### set up general linear hypothesis glht(lmod, linfct = K) ### alternatively, use a symbolic description ### instead of a matrix glht(lmod, linfct = c("Agriculture = ",

glht
"Examination = ", "Education = ", "Catholic = ", "Infant.Mortality =

15

"))

### multiple comparison procedures ### set up a one-way ANOVA amod <- aov(breaks ~ tension, data = warpbreaks) ### set up all-pair comparisons for factor tension ### using a symbolic description (type argument ### to contrMat()) glht(amod, linfct = mcp(tension = "Tukey")) ### alternatively, describe differences symbolically glht(amod, linfct = mcp(tension = c("M - L = ", "H - L = ", "H - M = "))) ### alternatively, define contrast matrix directly contr <- rbind("M - L" = c(-1, 1, ), "H - L" = c(-1, , 1), "H - M" = c( , -1, 1)) glht(amod, linfct = mcp(tension = contr)) ### alternatively, define linear function for coef(amod) ### instead of contrasts for tension ### (take model contrasts and intercept into account) glht(amod, linfct = cbind( , contr %*% contr.treatment(3)))

### mix of one- and two-sided alternatives warpbreaks.aov <- aov(breaks ~ wool + tension, data = warpbreaks) ### contrasts K <- rbind("L "M "L "M for tension - M" = c( 1, -1, ), - L" = c(-1, 1, ), - H" = c( 1, , -1), - H" = c( , 1, -1))

warpbreaks.mc <- glht(warpbreaks.aov, linfct = mcp(tension = K), alternative = "less") ### correlation of first two tests is -1 cov2cor(vcov(warpbreaks.mc)) ### use smallest of the two one-sided ### p-value as two-sided p-value -> . 232 summary(warpbreaks.mc)

16

glht-methods

glht-methods

Methods for General Linear Hypotheses

Description Simultaneous tests and condence intervals for general linear hypotheses. Usage ## S3 method for class glht summary(object, test = adjusted(), ...) ## S3 method for class glht confint(object, parm, level = .95, calpha = adjusted_calpha(), ...) ## S3 method for class glht coef(object, rhs = FALSE, ...) ## S3 method for class glht vcov(object, ...) ## S3 method for class confint.glht plot(x, xlim, xlab, ylim, ...) ## S3 method for class glht plot(x, ...) univariate() adjusted(type = c("single-step", "Shaffer", "Westfall", "free", p.adjust.methods), ...) Ftest() Chisqtest() adjusted_calpha(...) univariate_calpha(...) Arguments object test parm level calpha rhs type x xlim ylim xlab an object of class glht. a function for computing p values. additional parameters, currently ignored. the condence level required. either a function computing the critical value or the critical value itself. logical, indicating whether the linear function K or the right hand side m (rhs = TRUE) of the linear hypothesis should be returned. the multiplicity adjustment (adjusted) to be applied. See below and p.adjust. an object of class glht or confint.glht. the x limits (x1, x2) of the plot. the y limits of the plot. a label for the x axis.

glht-methods ...

17 additional arguments, such as maxpts, abseps or releps to pmvnorm in adjusted or qmvnorm in confint. Note that additional arguments specied to summary, confint, coef and vcov methods are currently ignored.

Details The methods for general linear hypotheses as described by objects returned by glht can be used to actually test the global null hypothesis, each of the partial hypotheses and for simultaneous condence intervals for the linear function K. The coef and vcov methods compute the linear function K and its covariance, respectively. The test argument to summary takes a function specifying the type of test to be applied. Classical Chisq (Wald test) or F statistics for testing the global hypothesis H0 are implemented in functions Chisqtest and Ftest. Several approaches to multiplicity adjusted p values for each of the linear hypotheses are implemented in function adjusted. The type argument to adjusted species the method to be applied: "single-step" implements adjusted p values based on the joint normal or t distribution of the linear function, and "Shaffer" and "Westfall" implement logically constraint multiplicity adjustments (Shaffer, 1986; Westfall, 1997). "free" implements multiple testing procedures under free combinations (Westfall et al, 1999). In addition, all adjustment methods implemented in p.adjust are available as well. Simultaneous condence intervals for linear functions can be computed using method confint. Univariate condence intervals can be computed by specifying calpha = univariate_calpha() to confint. The critical value can directly be specied as a scalar to calpha as well. Note that plot(a) for some object a of class glht is equivalent to plot(confint(a)). All simultaneous inference procedures implemented here control the family-wise error rate (FWER). Multivariate normal and t distributions, the latter one only for models of class lm, are evaluated using the procedures implemented in package mvtnorm. A more detailed description of the underlying methodology is available from Hothorn et al. (2008) and Bretz et al. (2010). Value summary computes (adjusted) p values for general linear hypotheses, confint computes (adjusted) condence intervals. coef returns estimates of the linear function K and vcov its covariance. References Frank Bretz, Torsten Hothorn and Peter Westfall (2010), Multiple Comparisons Using R, CRC Press, Boca Raton. Juliet P. Shaffer (1986), Modied sequentially rejective multiple test procedures. Journal of the American Statistical Association, 81, 826831. Peter H. Westfall (1997), Multiple testing of general contrasts using logical constraints and correlations. Journal of the American Statistical Association, 92, 299306. P. H. Westfall, R. D. Tobias, D. Rom, R. D. Wolnger, Y. Hochberg (1999). Multiple Comparisons and Multiple Tests Using the SAS System. Cary, NC: SAS Institute Inc. Torsten Hothorn, Frank Bretz and Peter Westfall (2008), Simultaneous Inference in General Parametric Models. Biometrical Journal, 50(3), 346363; See vignette("generalsiminf", package = "multcomp").

18 Examples

glht-methods

### set up a two-way ANOVA amod <- aov(breaks ~ wool + tension, data = warpbreaks) ### set up all-pair comparisons for factor tension wht <- glht(amod, linfct = mcp(tension = "Tukey")) ### 95% simultaneous confidence intervals plot(print(confint(wht))) ### the same (for balanced designs only) TukeyHSD(amod, "tension") ### corresponding adjusted p values summary(wht) ### all means for levels of tension amod <- aov(breaks ~ tension, data = warpbreaks) glht(amod, linfct = matrix(c(1, , , 1, 1, , 1, , 1), byrow = TRUE, ncol = 3)) ### confidence bands for a simple linear model, cars data plot(cars, xlab = "Speed (mph)", ylab = "Stopping distance (ft)", las = 1) ### fit linear model and add regression line to plot lmod <- lm(dist ~ speed, data = cars) abline(lmod) ### a grid of speeds speeds <- seq(from = min(cars$speed), to = max(cars$speed), length = 1 ) ### linear hypotheses: 1 K <- cbind(1, speeds) selected points on the regression line !=

### set up linear hypotheses cht <- glht(lmod, linfct = K) ### confidence intervals, i.e., confidence bands, and add them plot cci <- confint(cht) lines(speeds, cci$confint[,"lwr"], col = "blue") lines(speeds, cci$confint[,"upr"], col = "blue")

### simultaneous p values for parameters in a Cox model if (require("survival") && require("MASS")) { data("leuk", package = "MASS") leuk.cox <- coxph(Surv(time) ~ ag + log(wbc), data = leuk)

litter
### set up linear hypotheses lht <- glht(leuk.cox, linfct = diag(length(coef(leuk.cox)))) ### adjusted p values print(summary(lht)) }

19

litter

Litter Weights Data Set

Description Dose response of litter weights in rats.

Usage data("litter")

Format This data frame contains the following variables dose dosages at four levels: , 5, 5 , 5 gesttime gestation time as covariate. number number of animals in litter as covariate. weight response variable: average post-birth weights in the entire litter. .

Details Pregnant mice were divided into four groups and the compound in four different doses was administered during pregnancy. Their litters were evaluated for birth weights.

Source P. H. Westfall, R. D. Tobias, D. Rom, R. D. Wolnger, Y. Hochberg (1999). Multiple Comparisons and Multiple Tests Using the SAS System. Cary, NC: SAS Institute Inc., page 109. P. H. Westfall (1997). Multiple Testing of General Contrasts Using Logical Constraints and Correlations. Journal of the American Statistical Association, 92(437), 299306.

20 Examples
### fit ANCOVA model to data amod <- aov(weight ~ dose + gesttime + number, data = litter) ### define matrix of linear hypotheses for dose doselev <- as.integer(levels(litter$dose)) K <- rbind(contrMat(table(litter$dose), "Tukey"), otrend = c(-1.5, - .5, .5, 1.5), atrend = doselev - mean(doselev), ltrend = log(1:4) - mean(log(1:4))) ### set up multiple comparison object Kht <- glht(amod, linfct = mcp(dose = K), alternative = "less") ### cf. Westfall (1997, Table 2) summary(Kht, test = univariate()) summary(Kht, test = adjusted("bonferroni")) summary(Kht, test = adjusted("Shaffer")) summary(Kht, test = adjusted("Westfall")) summary(Kht, test = adjusted("single-step"))

modelparm

modelparm

Generic Accessor Function for Model Parameters

Description Extract model parameters and their covariance matrix as well as degrees of freedom (if available) from a tted model. Usage modelparm(model, coef., vcov., df, ...) Arguments model coef. vcov. df ... a tted model, for example an object returned by lm, glm, aov, survreg, or lmer etc. an accessor function for the model parameters. an accessor function for the covariance matrix of the model parameters. an optional specication of the degrees of freedom to be used in subsequent computations. additional arguments, currently ignored.

mtept Details

21

One cant expect coef and vcov methods for arbitrary models to return a vector of p xed effects model parameters (coef) and corresponding p p covariance matrix (vcov). The coef. and vcov. arguments can be used to dene modied coef or vcov methods for a specic model. Methods for lmer and survreg objects are available (internally). For objects inheriting from class lm the degrees of freedom are determined from model and the corresponding multivariate t distribution is used by all methods to glht objects. By default, the asymptotic multivariate normal distribution is used in all other cases unless df is specied by the user. Value An object of class modelparm with elements coef vcov df model parameters covariance matrix of model parameters degrees of freedom

mtept

Multiple Endpoints Data

Description Measurements on four endpoints in a two-arm clinical trial. Usage data(mtept) Format A data frame with 111 observations on the following 5 variables. treatment a factor with levels Drug Placebo E1 endpoint 1 E2 endpoint 2 E3 endpoint 3 E4 endpoint 4 Details The data (from Westfall et al., 1999) contain measurements of patients in treatment (Drug) and control (Placebo) groups, with four outcome variables. Source P. H. Westfall, R. D. Tobias, D. Rom, R. D. Wolnger, Y. Hochberg (1999). Multiple Comparisons and Multiple Tests Using the SAS System. Cary, NC: SAS Institute Inc.

22

parm

parm

Model Parameters

Description Directly specify estimated model parameters and their covariance matrix. Usage parm(coef, vcov, df = Arguments coef vcov df estimated coefcients. estimated covariance matrix of the coefcients. an optional specication of the degrees of freedom to be used in subsequent computations. )

Details When only estimated model parameters and the corresponding covariance matrix is available for simultaneous inference using glht (for example, when only the results but not the original data are available or, even worse, when the model has been tted outside R), function parm sets up an object glht is able to compute on (mainly by offering coef and vcov methods). Note that the linear function in glht cant be specied via mcp since the model terms are missing. Value An object of class parm with elements coef vcov df Examples
## example from ## Bretz, Hothorn, and Westfall (2 2). ## On multiple comparisons in R. R News, 2(3):14-17. beta <- c(V1 = 14.8, V2 = 12.6667, V3 = 7.3333, V4 = 13.1333) Sigma <- 6.7 99 * (diag(1 / c(2 , 3, 3, 15))) confint(glht(model = parm(beta, Sigma, 37), linfct = c("V2 - V1 >= ", "V3 - V1 >= ", "V4 - V1 >= ")), level = .9)

model parameters covariance matrix of model parameters degrees of freedom

plot.cld

23

plot.cld

Plot a cld object

Description Plot information of glht, summary.glht or confint.glht objects stored as cld objects together with a compact letter display of all pair-wise comparisons. Usage ## S3 method for class cld plot(x, type = c("response", "lp"), ...) Arguments x type An object of class cld. Should the response or the linear predictor (lp) be plotted. If there are any covariates, the lp is automatically used. To use the response variable, set type="response" and covar=FALSE of the cld object. Other optional print parameters which are passed to the plotting functions.

... Details

This function plots the information stored in glht, summary.glht or confint.glht objects. Prior to plotting, these objects have to be converted to cld objects (see cld for details). All types of plots include a compact letter display (cld) of all pair-wise comparisons. Equal letters indicate no signicant differences. Two levels are signicantly different, in case they do not have any letters in common. If the tted model contains any covariates, a boxplot of the linear predictor is generated with the cld within the upper margin. Otherwise, three different types of plots are used depending on the class of variable y of the cld object. In case of class(y) == "numeric", a boxplot is generated using the response variable, classied according to the levels of the variable used for the Tukey contrast matrix. Is class(y) == "factor", a mosaic plot is generated, and the cld is printed above. In case of class(y) == "Surv", a plot of tted survival functions is generated where the cld is plotted within the legend. The compact letter display is computed using the algorithm of Piepho (2004). Note: The user has to provide a sufciently large upper margin which can be used to depict the compact letter display (see examples). References Hans-Peter Piepho (2004), An Algorithm for a Letter-Based Representation of All-Pairwise Comparisons, Journal of Computational and Graphical Statistics, 13(2), 456466. See Also glht cld cld.summary.glht cld.confint.glht cld.glht boxplot mosaicplot plot.survfit

24 Examples

plot.cld

### multiple comparison procedures ### set up a one-way ANOVA data(warpbreaks) amod <- aov(breaks ~ tension, data = warpbreaks) ### specify all pair-wise comparisons among levels of variable "tension" tuk <- glht(amod, linfct = mcp(tension = "Tukey")) ### extract information tuk.cld <- cld(tuk) ### use sufficiently large upper margin old.par <- par( mai=c(1,1,1.25,1)) ### plot plot(tuk.cld) par(old.par) ### now using covariates amod2 <- aov(breaks ~ tension + wool, data = warpbreaks) tuk2 <- glht(amod2, linfct = mcp(tension = "Tukey")) tuk.cld2 <- cld(tuk2) old.par <- par( mai=c(1,1,1.25,1)) ### use different colors for boxes plot(tuk.cld2, col=c("green", "red", "blue")) par(old.par) ### get confidence intervals ci.glht <- confint(tuk) ### plot them plot(ci.glht) old.par <- par( mai=c(1,1,1.25,1)) ### use confint.glht object to plot all pair-wise comparisons plot(cld(ci.glht), col=c("white", "blue", "green")) par(old.par) ### set up all pair-wise comparisons for count data data(Titanic) mod <- glm(Survived ~ Class, data = as.data.frame(Titanic), weights = Freq, family = binomial()) ### specify all pair-wise comparisons among levels of variable "Class" glht.mod <- glht(mod, mcp(Class = "Tukey")) ### extract information mod.cld <- cld(glht.mod) ### use sufficiently large upper margin old.par <- par(mai=c(1,1,1.5,1)) ### plot plot(mod.cld) par(old.par) ### set up all pair-wise comparisons of a Cox-model if (require("survival") && require("MASS")) { ### construct 4 classes of age Melanoma$Cage <- factor(sapply(Melanoma$age, function(x){

recovery
if( if( if( if( )) x x x x <= 25 ) return(1) > 25 & x <= 5 ) return(2) > 5 & x <= 75 ) return(3) > 75 & x <= 1 ) return(4) }

25

### fit Cox-model cm <- coxph(Surv(time, status == 1) ~ Cage, data = Melanoma) ### specify all pair-wise comparisons among levels of "Cage" cm.glht <- glht(cm, mcp(Cage = "Tukey")) # extract information & plot old.par <- par() ### use mono font family if (dev.interactive()) old.par <- par(family = "mono") plot(cld(cm.glht), col=c("black", "red", "blue", "green")) par(old.par) } if (require("nlme") && require("lme4")) { data("ergoStool", package = "nlme") stool.lmer <- lmer(effort ~ Type + (1 | Subject), data = ergoStool) glme41 <- glht(stool.lmer, mcp(Type = "Tukey")) old.par <- par(mai=c(1,1,1.5,1)) plot(cld(glme41)) par(old.par) }

recovery

Recovery Time Data Set

Description Recovery time after surgery. Usage data("recovery") Format This data frame contains the following variables blanket blanket type, a factor at four levels: b , b1, b2, and b3. minutes response variable: recovery time after a surgical procedure.

26 Details

sbp

A company developed specialized heating blankets designed to help the body heat following a surgical procedure. Four types of blankets were tried on surgical patients with the aim of comparing the recovery time of patients. One of the blanket was a standard blanket that had been in use already in various hospitals. Source P. H. Westfall, R. D. Tobias, D. Rom, R. D. Wolnger, Y. Hochberg (1999). Multiple Comparisons and Multiple Tests Using the SAS System. Cary, NC: SAS Institute Inc., page 66. Examples
### set up one-way ANOVA amod <- aov(minutes ~ blanket, data = recovery) ### set up multiple comparisons: one-sided Dunnett contrasts rht <- glht(amod, linfct = mcp(blanket = "Dunnett"), alternative = "less") ### cf. Westfall et al. (1999, p. 8 ) confint(rht, level = .9) ### the same rht <- glht(amod, linfct = mcp(blanket = c("b1 - b >= ", "b2 - b >= ", "b3 - b >= "))) confint(rht, level = .9)

sbp

Systolic Blood Pressure Data

Description Systolic blood pressure, age and gender of 69 people. Usage data("sbp") Format A data frame with 69 observations on the following 3 variables. gender a factor with levels male female sbp systolic blood pressure in mmHg age age in years

trees513 Source

27

D. G. Kleinbaum, L. L. Kupper, K. E. Muller, A. Nizam, A. (1998), Applied Regression Analysis and Other Multivariable Methods, Duxbury Press, North Scituate, MA.

trees513

Frankonian Tree Damage Data

Description Damages on young trees caused by deer browsing. Usage data("trees513") Format A data frame with 2700 observations on the following 4 variables. damage a factor with levels yes and no indicating whether or not the trees has been damaged by game animals, mostly roe deer. species a factor with levels spruce, fir, pine, softwood (other), beech, oak, ash/maple/elm/lime, and hardwood (other). lattice a factor with levels 1, ..., 53, essentially a number indicating the position of the sampled area. plot a factor with levels x_1, ..., x_5 where x is the lattice. plot is nested within lattice and is a replication for each lattice point. Details In most parts of Germany, the natural or articial regeneration of forests is difcult due to a high browsing intensity. Young trees suffer from browsing damage, mostly by roe and red deer. In order to estimate the browsing intensity for several tree species, the Bavarian State Ministry of Agriculture and Foresty conducts a survey every three years. Based on the estimated percentage of damaged trees, suggestions for the implementation or modication of deer management plans are made. The survey takes place in all 756 game management districts (Hegegemeinschaften) in Bavaria. The data given here are from the game management district number 513 Unterer Aischgrund (located in Frankonia between Erlangen and H\"ochstadt) in 2006. The data of 2700 trees include the species and a binary variable indicating whether or not the tree suffers from damage caused by deer browsing. Source Bayerisches Staatsministerium fuer Landwirtschaft und Forsten (2006), Forstliche Gutachten zur Situation der Waldverjuengung 2006. www.forst.bayern.de Torsten Hothorn, Frank Bretz and Peter Westfall (2008), Simultaneous Inference in General Parametric Models. Biometrical Journal, 50(3), 346363; See vignette("generalsiminf", package = "multcomp").

28 Examples
summary(trees513)

waste

waste

Industrial Waste Data Set

Description Industrial waste output in a manufactoring plant. Usage data("waste") Format This data frame contains the following variables temp temperature, a factor at three levels: low, medium, high. envir environment, a factor at ve levels: env1 . . . env5. waste response variable: waste output in a manufacturing plant. Details The data are from an experiment designed to study the effect of temperature (temp) and environment (envir) on waste output in a manufactoring plant. Two replicate measurements were taken at each temperature / environment combination. Source P. H. Westfall, R. D. Tobias, D. Rom, R. D. Wolnger, Y. Hochberg (1999). Multiple Comparisons and Multiple Tests Using the SAS System. Cary, NC: SAS Institute Inc., page 177. Examples
### set up two-way ANOVA with interactions amod <- aov(waste ~ temp * envir, data=waste) ### comparisons of main effects only K <- glht(amod, linfct = mcp(temp = "Tukey"))$linfct K glht(amod, K) ### comparisons of means (by averaging interaction effects) low <- grep("low:envi", colnames(K))

waste
med <- grep("medium:envi", colnames(K)) K[1, low] <- 1 / (length(low) + 1) K[2, med] <- 1 / (length(low) + 1) K[3, med] <- 1 / (length(low) + 1) K[3, low] <- - 1 / (length(low) + 1) K confint(glht(amod, K)) ### same as TukeyHSD TukeyHSD(amod, "temp") ### set up linear hypotheses for all-pairs of both factors wht <- glht(amod, linfct = mcp(temp = "Tukey", envir = "Tukey")) ### cf. Westfall et al. (1999, page 181) summary(wht, test = adjusted("Shaffer"))

29

Index
Topic datasets adevent, 2 cholesterol, 4 cml, 7 detergent, 10 fattyacid, 11 litter, 19 mtept, 21 recovery, 25 sbp, 26 trees513, 27 waste, 28 Topic hplot plot.cld, 23 Topic htest cftest, 3 glht, 12 glht-methods, 16 Topic misc contrMat, 8 modelparm, 20 parm, 22 adevent, 2 adjusted (glht-methods), 16 adjusted_calpha (glht-methods), 16 aov, 12, 20 boxplot, 23 cftest, 3 Chisqtest (glht-methods), 16 cholesterol, 4 cld, 5, 23 cld.confint.glht, 23 cld.glht, 23 cld.summary.glht, 23 cml, 7 coef, 1214, 17, 21 coef.glht (glht-methods), 16 30 coeftest, 4 confint, 14, 17 confint.glht, 13 confint.glht (glht-methods), 16 contrMat, 8, 13 detergent, 10 fattyacid, 11 Ftest (glht-methods), 16 glht, 6, 12, 16, 17, 2123 glht-methods, 16 glm, 12, 20 litter, 19 lm, 12, 17, 20, 21 mcp, 22 mcp (glht), 12 model.frame, 12 model.matrix, 12 modelparm, 12, 13, 20 mosaicplot, 23 mtept, 21 p.adjust, 16, 17 parm, 22 plot.cld, 6, 23 plot.confint.glht (glht-methods), 16 plot.glht (glht-methods), 16 plot.survfit, 23 pmvnorm, 17 qmvnorm, 17 recovery, 25 sbp, 26 summary, 14 summary.glht, 3, 13

INDEX summary.glht (glht-methods), 16 survreg, 20, 21 terms, 12 trees513, 27 univariate (glht-methods), 16 univariate_calpha (glht-methods), 16 vcov, 1214, 17, 21 vcov.glht (glht-methods), 16 waste, 28

31