hdanova: High-Dimensional Analysis of Variance

Description

This function provides a high-dimensional analysis of variance (HDANOVA) method which can be used alone or as part of a larger analysis, e.g., ASCA, APCA, LiMM-PCA, MSCA or PC-ANOVA. It can be called directly or through the convenince functions asca, apca, limmpca, msca and pcanova.

Usage

hdanova(
  formula,
  data,
  subset,
  weights,
  na.action,
  family,
  unrestricted = FALSE,
  add_error = FALSE,
  aug_error = "denominator",
  use_ED = FALSE,
  pca.in = FALSE,
  contrasts = "contr.sum",
  coding,
  equal_baseline = FALSE,
  SStype = "II",
  REML = NULL,
  scale = FALSE
)

Value

An hdanova object containing loadings, scores, explained variances, etc. The object has associated plotting (asca_plots) and result (asca_results) functions.

Arguments

formula: Model formula accepting a single response (block) and predictors. See Details for more information.
data: The data set to analyse.
subset: Expression for subsetting the data before modelling.
weights: Optional object weights.
na.action: How to handle NAs (no action implemented).
family: Error distributions and link function for Generalized Linear Models.
unrestricted: Use unrestricted ANOVA decomposition (default = FALSE).
add_error: Add error to LS means, e.g., for APCA.
aug_error: Augment score matrices in backprojection. Default = "denominator" (of F test), "residual" (force error term), nueric value (alpha-value in LiMM-PCA).
use_ED: Use "effective dimensions" for score rescaling in LiMM-PCA.
pca.in: Compress response before ASCA (number of components).
contrasts: Effect coding: "contr.sum" (default = sum-coding), "contr.weighted" (not for lme4 models), "contr.reference", "contr.treatment".
coding: Defunct. Use 'contrasts' instead.
equal_baseline: Experimental: Set to TRUE to let interactions, where a main effect is missing, e.g., a nested model, be handled with the same baseline as a cross effect model. If TRUE the corresponding interactions will be put in quotation marks and included in the model.frame.
SStype: Type of sum-of-squares: "I" = sequential, "II" (default) = last term, obeying marginality, "III" = last term, not obeying marginality.
REML: Parameter to mixlm: NULL (default) = sum-of-squares, TRUE = REML, FALSE = ML.
scale: Scaling of the response matrix. Defaults to FALSE (no scaling). For alternatives, see Details.

Details

Scaling of the response matrix can be done by setting the scale parameter. If scale=TRUE, each column is scaled by its standard deviation (autoscaling). A numeric value can be provided to scale the columns by specific quantities. If scale is a character string, the first element is interpreted as a factor name and the second element is interpreted as a factor level, whose samples the standard deviations are calculated from (reference group scaling).

Examples

Run this code

# Load candies data
data(candies)

# Basic HDANOVA model with two factors
mod <- hdanova(assessment ~ candy + assessor, data=candies)
summary(mod)

Run the code above in your browser using DataLab