Troubleshooting
Problem
I have a set of variables in an SPSS data set that comprise responses to a set of test variables. A person's total score would be the sum of their item scores. Is there an SPSS procedure that will compute the correlation between each item variable and the total score, i.e. the item-total correlation.
Resolving The Problem
The corrected item-total correlation is provided in the RELIABILITY procedure. (The correlation is corrected in the sense that the value of item i is subtracted from the total for the correlation between the total and item 1. The rationale is that including item i in the total will result in an inflated correlation between item i and that total, as a component of the correlation will be the correlation of item i with itself. This inflation will be more severe if the number of items is low.) If you are using the menus and dialog boxes, click the Statistics button in the main Reliability dialog box. In the Statistic dialog box that opens, check the box beside Total if item deleted. Check the boxes for other desired statistics and click Continue. If you are using syntax commands, then the subcommand
/SUMMARY=TOTAL
requests the corrected item-total correlations, as well as the item-corrected scale means, standard deviations, and alpha coefficients. An example command containing this subcommand is:
RELIABILITY
/VARIABLES=aitem1 aitem2 aitem3 aitem4 aitem5 aitem6 aitem7 aitem8 aitem9
aitem10
/FORMAT=NOLABELS
/SCALE(ALPHA)=ALL/MODEL=ALPHA
/STATISTICS=DESCRIPTIVE SCALE
/SUMMARY=TOTAL .
If you want the uncorrected item-total correlation, with the total score including the item with which it is correlated, you would need to compute the total score and then use the CORRELATIONS procedure to print the item-total correlations. Suppose that your item scores were AITEM1 to AITEM10, as above, and your total score was to be called ATOT. Commands to produce the uncorrected item-total correlations would be:
COMPUTE atot = SUM(aitem1, aitem2, aitem3, aitem4, aitem5, aitem6,
aitem7, aitem8, aitem9, aitem10).
CORRELATIONS aitem1 aitem2 aitem3 aitem4 aitem5 aitem6 aitem7 aitem8 aitem9
aitem10 WITH atot .
The WITH keyword in the CORRELATIONS command allows you to produce a correlation matrix with the items in the rows and the total as the column, without printing the correlations among the items. If you're using the menus, you can compute ATOT from the Transform->Compute menu from the Data Editor. Having computed ATOT, you can build most of the CORRELATIONS command from the Analyze->Correlate->Bivariate menu. To insert the WITH keyword and avoid printing the full symmetric correlation matrix, you will need to paste the command to a syntax window (click Paste rather than OK) and then type the keyword into the variable list on the CORRELATIONS command.
If your item scores are binary variables, you should note that a Pearson correlation between a binary variable and a continuous variable is equal to the point-biserial correlation coefficient. The latter formula is a shortcut method to computing the correlation when you have the continuous variable means and variances and sample sizes for each value of the binary variable.
If the item score variables are all adjacent and ordered in the data file, you can shorten the COMPUTE command above with the TO keyword, as in:
COMPUTE atot = SUM(aitem1 TO aitem10).
Related Information
Historical Number
22227
Was this topic helpful?
Document Information
Modified date:
16 April 2020
UID
swg21477000