anacor (version 1.0-5)

anacor: Simple and Canonical Correspondence Analysis

Description

This function performs simple and canonical CA for incomplete tables based on SVD. Different scaling methods for row and column scores are provided.

Usage

anacor(tab, ndim = 2, row.covariates, col.covariates, 
scaling = c("Benzecri","Benzecri"), ellipse = TRUE, eps = 1e-06)
## S3 method for class 'anacor':
print(x,...)
## S3 method for class 'anacor':
summary(object,...)

Arguments

tab

Data frame of dimension n times m with frequencies. Missings are coded as NA.

ndim

Number of dimensions.

row.covariates

Matrix with n rows containing covariates for the row scores.

col.covariates

Matrix with m rows containing covariates for the column scores.

scaling

A vector with two elements. The first one corresponds to the method for row scaling, the second one for column scaling. Available scaling methods are "standard", "centroid", "Benzecri", "Goodman".

ellipse

If TRUE, confidence ellipses are computed.

eps

Convergence criterion for reconstitution algorithm.

x

Object of class "anacor" in print.anacor.

object

Object of class "anacor" in summary.anacor.

...

Additional arguments ignored.

Value

• row.scores
  Scaled row scores.
• col.scores
  Scaled column scores.
• ndim
  Number of dimensions extracted.
• chisq
  Total chi-square value.
• chisq.decomp
  Chi-square decomposition across dimensions with p-values.
• singular.values
  Singular values without trivial solution.
• se.singular.values
  Standard errors for the singular values.
• left.singvec
  Left singular vectors without trivial solution.
• right.singvec
  Right singular vectors without trivial solution.
• eigen.values
  Eigenvalues without trivial solution.
• datname
  Name of the dataset.
• tab
  Table with imputed frequencies in case of missings.
• row.covariates
  Matrix with row covariates.
• col.covariates
  Matrix with column covariates.
• scaling
  Scaling Method.
• bdmat
  List of matrices with observed and fitted Benzecri distances for rows and columns.
• rmse
  Root mean squared error of Bezencri distances (rows and columns).
• row.acov
  Covariance matrix for row scores.
• col.acov
  Covariance matrix for column scores.
• cancoef
  List containing canonical coefficients (CCA only).
• sitescores
  List containing the site scores (CCA only).
• isetcor
  List containing the intraset correlations (CCA only).

Details

Missing values in tab are imputed using the reconstitution algorithm. Setting scaling to "standard" leads to standard coordinates. Principal coordinates can be computed by means of Benzecri decomposition. Furthermore, scores can be scaled around their centroid. Goodman scaling is based on Fisher-Maung decomposition. For large datasets it is suggested to set ellipse = FALSE. If conf = TRUE, make sure that there are no rows and columns that have full 0 entries.

References

De Leeuw, J. and Mair, P. (2009). Simple and Canonical Correspondence Analysis Using the R Package anacor. Journal of Statistical Software, 31(5), 1-18. http://www.jstatsoft.org/v31/i05/

See Also

plot.anacor

Examples

## simple CA on Tocher data, asymmetric coordinates
data(tocher)
res <- anacor(tocher, scaling = c("standard", "centroid"))
res
summary(res)

## 2- and 5-dimensional solutions for bitterling data, Benzecri scaling
data(bitterling)
res1 <- anacor(bitterling, ndim = 2, scaling = c("Benzecri", "Benzecri"))
res2 <- anacor(bitterling, ndim = 5, scaling = c("Benzecri", "Benzecri"))
res1
res2

## Canonical CA on Maxwell data, Goodman scaling
data(maxwell)
res <- anacor(maxwell$table, row.covariates = maxwell$row.covariates, 
scaling = c("Goodman", "Goodman"))
res
summary(res)