Printing the covariance matrix through the Reliability procedure, the Correlation procedure, or the Regression procedure
The Reliability procedure is somewhat simpler in that the covariance matrix is automatically printed as a separate table. The Correlation procedure combines the correlation, significance, cross-product deviations, covariance, and N for a variable pair into a single table cell. There is a way to reorganize this table so that the covariance matrix is printed as a separate table or subtable. It should be noted that the Reliability procedure uses listwise deletion of cases with missing values, ie., the covariances will be computed with only those cases that have no missing values on the variables in the analysis. The correlation procedure uses pairwise deletion by default, i.e. each covariance is calculated with all cases that have valid values
on that variable pair and each variance is calculated with all cases that have valid values on that variable. Listwise deletion is available as an option in the Correlation procedure. (See Technote 1475199, which addresses this distinction between the options.) The Regression procedure must be run from syntax for the covariance matrix option to be included. If you want listwise deletion and want the covariance matrix to be printed in a separate table, then the Reliability procedure will be the simplest solution. If you want pairwise deletion, you will need to use the Correlation or Regression procedure. The details for each procedure are provided below.
-------
The Reliability Procedure (Analyze->Scale->Reliability Analysis):
In the Reliability Analysis dialog, paste the variables of interest into the Items box and click the Statistics button. In the Statistics dialog that opens, click the checkbox for Covariances in the Inter-Item area. (There is also a checkbox for correlations. To request the correlation and covariance matrices in command syntax, add the keywords 'Corr' and 'Cov' respectively, to the /STATISTICS subcommand, as in:
RELIABILITY
/VARIABLES=time age salnow edlevel work jobcat minority
/FORMAT=NOLABELS
/SCALE(ALPHA)=ALL/MODEL=ALPHA
/STATISTICS=CORR COV .
The correlation and covariance matrices will print as separate tables.
_______
The Correlation Procedure (Analyze->Correlate->Bivariate):
In the "Bivariate Correlations" dialog, paste the variable of interest into the Variables box on the right of the dialog box. Then click the Options button. In the Options dialog, click the checkbox for
"Cross-product deviations and covariances". Click Continue and then click OK. To make this choice in a syntax command, add the keyword 'XPROD' to the /STATISTICS subcommand, as in:
CORRELATIONS
/VARIABLES=time age salnow edlevel work jobcat minority
/STATISTICS XPROD .
You may want to reorganize the Correlation table that is printed. In that table, the rows will be organized by variable, i.e. for each variable, the Pearson correlation, Sig. value, Sum of Squares and Cross-Products, Covariance, and N will be printed in one cell of the table.If you want separate tables for the correlation,..covariance, etc., you can use the pivot trays to accomplish this. Right-click anywhere in the Correlation table. In the pop-up menu that appears, choose
'SPSS Pivot Table Object->Open'. A pivot table window will open with the Correlation table. In that window: Open the Pivot menu and choose 'Pivoting trays'. In the Pivoting tray dialog, there will be 2 icons in the Row area. The first is for Variables and the second is for Statistics. Click the Variables (leftmost) icon and drag it to the right of the Statistics icon. If you close the Pivoting Trays dialog by clicking the x in the upper right corner, you will see in the Pivot Table window that the Pearson Correlation matrix is a separate subtable, as are each of the Sig., Sum of Squares and Cross-Products, Covariance and N matrices. If you like this organization, just close the Pivot Table window and your changes will be reflected in the Output window. If you want to hide all of the matrices in the table except the covariance matrix, then reopen the Pivoting trays dialog and drag the Statistics icon into the Layers area. When you close the Pivoting Trays dialog, only the Pearson Correlations matrix will be visible. However, there will be a Statistics tab with a scroll arrow at the top of the table. Scroll down to Covariance and click that choice. The Covariance matrix will become the visible matrix in this table. Close the Pivot Table window to return to the Output window.
_______________________________
The Regression Procedure:
As stated above, you can use the Regression procedure to print the covariance matrix but you will need to run it from a syntax command. You can build the command in the Regression dialog boxes. You will need to paste one of the variables into the "Dependent:" box and the other variables into the "Independent(s):" box. (The dependent variable will be in the first row and column of the covariance matrix.) Click the
Statistics button in the Regression dialog and check the box beside 'Descriptives' in the Statistics dialog and click Continue. Click Paste rather than OK. In the Syntax window, add the keyword "Cov" to the /DESCRIPTIVES subcommand, as in:
REGRESSION
/DESCRIPTIVES MEAN STDDEV CORR SIG N cov
/MISSING LISTWISE
/STATISTICS COEFF OUTS R ANOVA
/CRITERIA=PIN(.05) POUT(.10)
/NOORIGIN
/DEPENDENT sex
/METHOD=ENTER time age salnow edlevel work jobcat minority .
The Covariance matrix will print as a subtable in the Correlations table. Note that regression uses listwise deletion by default but pairwise deletion can be requested from the "Regression: Options" dialog or by replacing "LISTWISE" with "PAIRWISE" in the REGRESSION command .