VCA (version 1.2.1)

test.fixef: Perform t-Tests for Linear Contrasts on Fixed Effects.

Description

This function performs t-Tests for one or multiple linear combinations (contrasts) of estimated fixed effects.

Usage

test.fixef(obj, L, ddfm = c("contain", "residual", "satterthwaite"),
  method.grad = "simple", tol = 1e-12, quiet = FALSE, opt = TRUE,
  onlyDF = FALSE, ...)

Arguments

obj

(VCA) object

L

(numeric) vector or matrix, specifying linear combinations of the fixed effects, in the latter case, each line represents a disctinct linear contrast

ddfm

(character) string specifying the method used for computing the denominator degrees of freedom for tests of fixed effects or LS Means. Available methods are "contain", "residual", and "satterthwaite".

method.grad

(character) string specifying the method to be used for approximating the gradient of the variance-covariance matrix of fixed effects at the estimated covariance parameter estimates (see function 'grad' (numDeriv) for details)

tol

(numeric) value specifying the numeric tolerance for testing equality to zero

quiet

(logical) TRUE = suppress warning messages, e.g. for non-estimable contrasts

opt

(logical) TRUE = tries to optimize computation time by avoiding unnecessary computations for balanced datasets (see details).

onlyDF

(logical) TRUE = only the specified type of degrees of freedom are determined without carrying out the actual hypothesis test(s)

...

further parameters (for internal use actually)

Value

• (numeric) vector or matrix with 4 elements/columns corresponding to "Estimate", "t Value", "DF", and "Pr > |t|".

Details

Here, the same procedure as in SAS PROC MIXED ddfm=satterthwaite (sat) is implemented. This implementation was inspired by the code of function 'calcSatterth' of R-package 'lmerTest'. Thanks to the authors for this nice implementation. Note, that approximated Satterthwaite degrees of freedom might differ from 'lmerTest' and SAS PROC MIXED. Both use the inverse Fisher-information matrix as approximation of the variance-covariance matrix of variance components (covariance parameters). Here, either the exact algorithm for ANOVA-estimators of variance components, described in Searle et. al (1992) p. 176, or the approximation presented in Giesbrecht and Burns (19985) are used. For balanced designs their will be no differences, usually. In case of balanced designs, the Satterthwaite approximation is equal to the degrees of freedom of the highest order random term in the model (see examples).

References

Searle, S.R, Casella, G., McCulloch, C.E. (1992), Variance Components, Wiley New York Giesbrecht, F.G. and Burns, J.C. (1985), Two-Stage Analysis Based on a Mixed Model: Large-Sample Asymptotic Theory and Small-Sample Simulation Results, Biometrics 41, p. 477-486 SAS Help and Documentation PROC MIXED (MODEL-statement, Option 'ddfm'), SAS Institute Inc., Cary, NC, USA

See Also

test.lsmeans, getL

Examples

data(dataEP05A2_2)
ub.dat <- dataEP05A2_2[-c(11,12,23,32,40,41,42),]
fit <- anovaMM(y~day/(run), ub.dat)
fe <- fixef(fit)
fe
lc.mat <- getL( fit, c("day1-day2", "day3-day6"))
lc.mat
test.fixef(fit, lc.mat)

# some inferential statistics about fixed effects estimates
L <- diag(nrow(fe))
rownames(L) <- colnames(L) <- rownames(fe)
test.fixef(fit, L)

# using different "residual" method determining DFs
test.fixef(fit, L, ddfm="res")

# having 'opt=TRUE' is a good idea to save time
# (in case of balanced designs)
data(VCAdata1)
datS3 <- VCAdata1[VCAdata1$sample==3,]
fit <- anovaMM(y~(lot+device)/(day)/run, datS3)
fit$VarCov <- vcovVC(fit)
fe <- fixef(fit)
L <- diag(nrow(fe))
rownames(L) <- colnames(L) <- rownames(fe)
system.time(tst1 <- test.fixef(fit, L))
system.time(tst2 <- test.fixef(fit, L, opt=FALSE))
tst1
tst2