BSagri (version 0.1-6)

simplesimint: Simultaneous confidence intervals from raw estimates

Description

Calculates simultaneous confidence intervals for multiple contrasts based on a parameter vector, its variance-covariance matrix and (optionally) the degrees of freedom, using quantiles of the multivar

Usage

simplesimint(coef, vcov, cmat, df = NULL, conf.level = 0.95,
 alternative = c("two.sided", "less", "greater"))

Arguments

coef

a single numeric vector, specifying the point estimates of the parameters of interest

vcov

the variance-covariance matrix corresponding to coef, should be of dimension P-times-P, when coef is of P

cmat

the contrasts matrix specifying the comparisons of interest with respect to coef, should have P columns, when coef is of length p

df

optional, the degree of freedom for the multivariate t-distribution; if specified, quantiles from the multivariate t-distribution are used for confidence interval estimation, if not specified (default), quantiles of the multivariate normal distribution ar

conf.level

a single numeric value between 0.5 and 1.0; the simultaneous confidence level

alternative

a single character string, "two.sided" for intervals, "less" for upper limits, and "greater" for lower limits

Value

• An object of class "simplesimint"
• estimate
  the estimates of the contrasts
• lower
  the lower confidence limits
• upper
  the upper confidence limits
• cmat
  the contrast matrix, as input
• alternative
  a character string, as input
• conf.level
  a numeric value, as input
• quantile
  a numeric value, the quantile used for confidence interval estimation
• df
  a numeric value or NULL, as input
• stderr
  the standard error of the contrasts
• vcovC
  the variance covariance matrix of the contrasts

concept

• confidence interval
• multiple contrasts

Details

Implements the methods formerly available in package multcomp, function csimint. Input values are a vector of parameter estimates $\mu$ of length $P$, a corresponding estimate for its variance-covariance matrix $\Sigma$ (P times P), and a contrast matrix $C$ of dimension $M \times P$. The contrasts $L = C \mu$ are computed, the variance-covariance matrix (being a function of $C$ and $\Sigma$) and the corresponding correlation matrix $R$ are computed. Finally, confidence intervals for $L$ are computed: if df is given, quantiles of an M-dimensional t distribution with correlation matrix R are used, otherwise quantiles of an M-dimensional standard normal distribution with correlation matrix R are used.

See Also

See ?coef and ?vcov for extracting of parameter vectors and corresponding variance covariance matrices from variou model fits.

Examples

# For the simple case of Gaussian response
# variables with homoscedastic variance,
# see the following example


library(mratios)
data(angina)

boxplot(response ~ dose, data=angina)

# Fit a cell means model,

fit<-lm(response ~ 0+dose, data=angina)

# extract cell means, the corresponding
# variance-covariance matrix and the
# residual degree of freedom,

cofi<-coef(fit)
vcofi<-vcov(fit)
dofi<-fit$df.residual

# define an appropriate contrast matrix,
# here, comparisons to control

n<-unlist(lapply(split(angina$response, f=angina$dose), length))
names(n)<-names(cofi)

cmat<-contrMat(n=n, type="Dunnett")
cmat

#

test<-simplesimint(coef=cofi, vcov=vcofi, df=dofi, cmat=cmat, alternative="greater" )

test

summary(test)

plotCI(test)

### Note, that the same result can be achieved much more conveniently
### using confint.glht in package multcomp