modelMeans: Compute adjusted or unadjusted cell means in ANOVA designs

In tdeenes/eegR: Analyze EEG Signals

Description

modelMeans is a generic function for computing adjusted or unadjusted cell means in ANOVA designs.

Usage

modelMeans(...)

## Default S3 method:
modelMeans(
  .arraydat,
  factordef,
  bwdat = NULL,
  term = NULL,
  adjusted = FALSE,
  ...
)

## S3 method for class 'arrayAnova'
modelMeans(model = NULL, term = NULL, adjusted = FALSE, ...)

## S3 method for class 'tanova'
modelMeans(model = NULL, term = NULL, adjusted = FALSE, ...)

Arguments

...	not used yet
.arraydat	a numeric array with named dimnames containing the EEG (or other) data. Missing values are not allowed.
factordef	a named list of factor definitions, containing the following elements: between: character vector of between-subject factors (default: NULL) within: character vector of within-subject factors (default: NULL) w_id: name of the dimension which identifies the subjects (default: "id")
bwdat	a data.frame which contains the identification codes (factordef$w_id) and all subject-level variables (usually factors) listed in 'factordef$between'. Missing values are not allowed.
term	a character vector of model terms; a separate array of cell means is returned for each term. The default is NULL, in which case all terms (that is, the means corresponding to all main effects and interactions) are returned. For custom terms, note that interaction terms must be given as "factorA*factorB" and not as "factorA:factorB".
adjusted	a logical value if the cell means should be adjusted for unbalanced data (default: FALSE). See Details.
model	an object returned by arrayAnova or tanova

Details

If 'adjusted' is set to FALSE (the default), modelMeans fits separate models for each model term and computes the model-based predictions for a reference grid which contains the mean and the 1SD value for continuous covariates, and all levels of the factor variables in the design. If 'adjusted' is set to TRUE, the means are simply the marginal means of the predicted means of the highest-order interaction term. This is equivalent to the so-called LS means (least-squares means) in the SAS terminology. This distinction is only relevant if the design is unbalanced, that is, the sample sizes in a between-subject design are unequal (note that the input array may not contain missing values, therefore the within subject part of the design is always balanced). For such datasets the lower order interaction or main effect means can substantially differ from the averages of the corresponding higher-order means if 'adjusted' is FALSE.

Value

The function returns a named list of arrays; the names are the model terms.

Methods (by class)

• default: Default method

• arrayAnova: Method for arrayAnova objects

• tanova: Method for tanova objects

Examples

# example data
data(erps)
dat_id <- attr(erps, "id") # to get group memberships

# make the dataset unbalanced to illustrate the difference between adjusted
# and unadjusted means
erps <- subsetArray(erps, list(id = 2:20))
dat_id <- dat_id[2:20, ]

# compute the means for the group*stimclass*pairtype interaction and all
# other lower-order terms
fdef <- list(between = "group",
             within = c("stimclass", "pairtype"))
means <- modelMeans(erps, fdef, bwdat = dat_id)

# the same for the least-squares means
means_adj <- modelMeans(erps, fdef, bwdat = dat_id, adjusted = TRUE)

# this is an unbalanced dataset, so the means for the highest order
# interaction are identical, but the marginal means are not
stopifnot(identical(
    TRUE,
    all.equal(means$`group*stimclass*pairtype`,
              means_adj$`group*stimclass*pairtype`))
)
stopifnot(!identical(
    TRUE,
    all.equal(means$`stimclass*pairtype`,
              means_adj$`stimclass*pairtype`))
)

# simple means for the Fz channel at time 200, for identical pairs in the
# "A" stimulus class, separately for the two reading groups
simple_means <- sapply(c("control", "dl"), function(g)
    mean(subsetArray(erps,
                     list(chan = "Fz", time = "200",
                          stimclass = "A", pairtype = "ident",
                          id = dat_id$group == g)))
)
stopifnot(identical(
    TRUE,
    all.equal(means$`group*stimclass*pairtype`["Fz", "200", , "A", "ident"],
              simple_means))
)

# the same ignoring the groups
simple_mean_nogroup <- mean(subsetArray(erps,
                                        list(chan = "Fz", time = "200",
                                        stimclass = "A", pairtype = "ident")))
stopifnot(identical(
    TRUE,
    all.equal(means$`stimclass*pairtype`["Fz", "200", "A", "ident"],
              simple_mean_nogroup))
)
stopifnot(identical(
    TRUE,
    all.equal(means_adj$`stimclass*pairtype`["Fz", "200", "A", "ident"],
              mean(simple_means)))
)

tdeenes/eegR documentation built on April 19, 2021, 4:17 p.m.