mwlsr: mwlsr

In PfaffLab/mwlsr: Multiple Weighted Least Squares Regression

Description

Multiple Weighted Least Squares Regression (mwlsr). Used to fit gaussian glm against multiple responses simultaneously.

Usage

mwlsr(data, design, weights = NULL, scale.weights = TRUE, data.err = NULL,
  coef.method = c("chol", "ginv", "svd", "qr"), coef.tol = 1e-07,
  coefs.only = FALSE)

Arguments

data	Input response matrix with responses in columns
design	Design matrix. See model.matrix
weights	Weights matrix
scale.weights	If TRUE then weights are scaled (default behavior)
data.err	Additional per-response-value uncertainty that should be considered in the final sum of squared residual. Useful if your response values have some knowm measurement uncertainty that you'd like to have considered in the models.
coef.method	Method used to compute coefficients. This setting is passed to mols.coefs or wls.coefs
coef.tol	Tolerance setting for svd based coefficient calculation. Passed to mols.coefs or wls.coefs
coefs.only	Stop at the coefficient calculation and return only the coefficients of the models.

Value

List with the following elements:

coefficients	Model coefficients
residuals	Residuals of the fit
fitted.values	Fitted values. Same dimension as the input response matrix.
deviance	Sum of squared residuals
dispersion	deviance / df.residual
null.deviance	Sum of squared residuals for the NULL model (intercept only)
weights	Weights matrix
prior.weights	Weights matrix pre-scaling
weighted	TRUE if fit was a weighted fit
df.residual	Degrees of freedom of the model. nrows(data) - ncol(design)
df.null	Degrees of freedom of the null model. nrows(data) - 1
y	Input data matrix
y.err	Input data.err matrix
X	Design matrix
x	If design matrix was based on factor levels then this will be a factor vector that matches the original grouping vector
intercept	TRUE if the fit has an Intercept
coef.hat	If the fit has an Intercept then this is a matrix of modified coefficients that represent the per-group averages. This is calculated by adding the Intercept coefficients to each of the other coefficients. This only makes sense if your design was based on a single multi-level factor

Examples

# Using the iris data.
design <- model.matrix(~Species, data=iris)
fit <- mwlsr(iris[, 1:4], design)
# test data association with the Species factor
result <- mwlsr.Ftest(fit)
print(table(result$F.padj < 0.05))

PfaffLab/mwlsr documentation built on May 12, 2019, 5:22 p.m.