NORMALIZATION Subcommand (CORRESPONDENCE command)

The NORMALIZATION subcommand specifies one of five methods for normalizing the row and column scores. Only the scores and confidence statistics are affected; contributions and profiles are not changed.

The following keywords are available:

SYMMETRICAL. For each dimension, rows are the weighted average of columns divided by the matching singular value, and columns are the weighted average of rows divided by the matching singular value. This is the default if the NORMALIZATION subcommand is not specified. Use this normalization method if you are primarily interested in differences or similarities between rows and columns.

PRINCIPAL. Distances between row points and distances between column points are approximations of chi-square distances or of Euclidean distances (depending on MEASURE). The distances represent the distance between the row or column and its corresponding average row or column profile. Use this normalization method if you want to examine both differences between categories of the row variable and differences between categories of the column variable (but not differences between variables).

RPRINCIPAL. Distances between row points are approximations of chi-square distances or of Euclidean distances (depending on MEASURE). This method maximizes distances between row points, resulting in row points that are weighted averages of the column points. This is useful when you are primarily interested in differences or similarities between categories of the row variable.

CPRINCIPAL. Distances between column points are approximations of chi-square distances or of Euclidean distances (depending on MEASURE). This method maximizes distances between column points, resulting in column points that are weighted averages of the row points. This is useful when you are primarily interested in differences or similarities between categories of the column variable.

The fifth method allows the user to specify any value in the range –1 to +1, inclusive. A value of 1 is equal to the RPRINCIPAL method, a value of 0 is equal to the SYMMETRICAL method, and a value of –1 is equal to the CPRINCIPAL method. By specifying a value between –1 and 1, the user can spread the inertia over both row and column scores to varying degrees. This method is useful for making tailor-made biplots.