rcormat

Generate a correlation matrix as R = LL' where the rows of L are of length
1, L is of rank <code>r</code> and the matrix L is sparse (depending on
<code>sparse.prop</code>. The loadings in L are sampled from a standard normal
distribution, after which <code>sparse.prop</code> is used to set a randomly
chosen number of loadings in each row equal to zero. To ensure that a
correlation matrix results, the rows are normalized.

This is an implementation of constrained dual scaling for
detecting response styles in categorical data, including utility functions. The
procedure involves adding additional columns to the data matrix representing the
boundaries between the rating categories. The resulting matrix is then doubled
and analyzed by dual scaling. One-dimensional solutions are sought which provide
optimal scores for the rating categories. These optimal scores are constrained
to follow monotone quadratic splines. Clusters are introduced within which the
response styles can vary. The type of response style present in a cluster can
be diagnosed from the optimal scores for said cluster, and this can be used to
construct an imputed version of the data set which adjusts for response styles.

Pieter Schoonees

Constrained Dual Scaling for Detecting Response Styles

rcormat function

<dl><dt>m</dt>
<dd>integer; the number of variables.</dd>
<dt>r</dt>
<dd>integer; the required rank.</dd>
<dt>sparse.prop</dt>
<dd>the proportion of zeros in the rows of the matrix.</dd></dl>

Arguments

Randomly Generate Low-Rank Correlation Matrix — rcormat

<dl>

<dt>m</dt>
<dd>integer; the number of variables.</dd>


<dt>r</dt>
<dd>integer; the required rank.</dd>


<dt>sparse.prop</dt>
<dd>the proportion of zeros in the rows of the matrix.</dd>

</dl>

rcormat: Randomly Generate Low-Rank Correlation Matrix

Description

Usage

Value

Arguments

Examples