This writes out the Table of inter-item correlations which is the basis of the principal components analysis of residuals. The basis for the correlations is controlled by PRCOMP=.
PRCOMP=R correlates the raw residuals
PRCOMP=S correlates the standardized residuals
PRCOMP=O correlates the observations
CIMPUTE= controls imputing missing values
ICORFILE=? opens a Browse window
Extreme scores: minimum possible (0) and maximum possible (perfect) person scores are omitted from the computation of the correlations. Items with extreme scores are not reported. Their correlation is 0.
Missing data: for these Winsteps substitutes their expectations when possible. For residuals and standardized residuals, these are 0. Persons with extreme scores (minimum possible or maximum possible): Winsteps drops these from the correlation computation. The reason for these choices is to make the principal components analysis of residuals as meaningful as possible.
Expected correlations: the expected inter-item correlation between any two items is -1/(L-1). Anscombe & Tukey, 1963, p. 144, courtesy of Larry Ludlow.
ICORFILE= file name
Example 1: Write out the Table of inter-item residual correlations. PRCOMP=R. ICORFILE=file.txt - Then file.txt contains, for SF.txt, fixed-field columns 12 characters wide.
ITEM ITEM CORRELATION
1 2 -.171665
1 3 -.144027
1 4 -.009290
1 5 -.103947
Example 2: When ICORFILE= is selected on the Output Files menu or MATRIX=YES, the Data Format: Matrix option can be selected:
This produces 10-character wide columns in fixed-field format
1 2 3 4 ...
1.000000 -.171665 -.144027 -.009290 ...
-.171665 1.000000 .175921 .039622 ...
-.144027 .175921 1.000000 -.101622 ...
-.009290 .039622 -.101622 1.000000 ...
-.103947 .099335 -.021021 .145339 ...
.107936 -.009881 -.039799 -.109941 ...
.108457 -.061624 .015850 -.001168 ...
Example 3: Is there a way to include extreme scores?
1. All the correlations relating to extreme scores are zero.
2. If you select ICORFILE= from the Output Files menu, and click on "matrix". You will get all pairs of non-extreme correlations.
3. You can obtain a matrix of raw or standardized response-residuals including extreme scores, by using IPMATRIX= from the Output Files menu (check the extreme score boxes). Then use this to compute the correlations with a standard statistical package etc. You will notice that the residuals for extreme scores may not be zero. This is because they are computed based on the measures shown for extreme scores in the measure Tables.
