B3. In the situation such as the one we have with missing CES-D items, we would recommend case mean substitution for cases with five or fewer missing values, but achieving this through SPSS commands...

B3. In the situation such as the one we have with missing CES-D items, we would recommend case mean substitution for cases with five or fewer missing values, but achieving this through SPSS commands is tedious, particularly for scales with 20 items such as the CES-D. With small samples, it is likely to be simpler to manually fill in case-mean imputed values. We will guide you through the imputation of a single missing item to illustrate the process. We will do this for item 9, the one with the highest percentage of missing values. Select Transform ➜ Compute Variable, which will create a new variable. (We could have imputed the missing values into the original cesd9 variable, but it is often wise to create a new variable and to preserve the original in case you make a mistake in the imputation process.) Enter a name for the Target Variable, such as newcesd9. To set the new variable equal to old CES-D values, enter cesd9 in the field for Numeric Expression. Then click OK. Next, go back to the Compute Variable dialog box, leaving newcesd9 as the Target Variable. In the Numeric Expression field, you need to tell the computer to add the values of the other 19 items and divide by 19, to set newcesd9 equal to the person’s mean for all other items. For the CES-D scale, we need to use the four reversecoded items to get the appropriate value. Here is the command to insert in the Numeric Expression field:

Next, click the If (optional case selection) button, which brings up a new dialog box. Select “Include if case satisfies condition.” Type in the following in the blank field: MISSING (cesd9), then click Continue and OK. This will impute a value for newcesd9 only for those cases with a missing value—all others will have the original data. Now, run frequencies and descriptive statistics for both cesd9 and newcesd9 and answer the following questions. (a) How many cases were missing for the newcesd9 variable? (b) For how many cases were imputations performed? (c) Why do you think imputations were not done for all missing cases? (d) Did the mean value change for the new variable, compared to the original? How about the SD? (e) What was the range of values for just the new imputed values? (You will need to glean this information by comparing the frequency distributions for cesd9 and newcesd9.)