Effects of Nesting and Stacking on Column Means Tests

The rule for column means tests is as follows: a separate set of pairwise tests is performed for each innermost subtable. To see how nesting affects the tests, consider the previous example, but with Hours per day watching TV nested within levels of Labor force status.

1. Open the table builder again (Analyze menu, Tables, Custom Tables).
2. Drag and drop Labor force status from the variable list into the Rows area of the canvas pane.
3. Click OK to create the table.

   

   With Hours per day watching TV nested within levels of Labor force status, seven sets of column means tests are performed: one for each level of Labor force status. The same letter keys are assigned to the categories of Get news from newspapers. For respondents working full time, the B key appears in the A column. This means that for full-time employees, the mean value of Hours per day watching TV is lower for people who get their news from newspapers. No other keys appear in the columns, so you can conclude that there are no other statistically significant differences in the column means.

   Bonferroni adjustments. When multiple tests are performed, the Bonferroni adjustment is applied to column means tests to ensure that the alpha level (or false positive rate) specified on the Test Statistics tab applies to each set of tests. Thus, in this table, no Bonferroni adjustments were applied because although seven sets of tests are performed, within each set only one pair of columns is compared.

   To see how stacking affects the tests:

4. Open the table builder again (Analyze menu, Tables, Custom Tables).
5. Drag and drop Get news from internet from the variable list into the Columns area to the left of Get news from newspapers.
6. Click OK to create the table.

With Get news from internet stacked with Get news from newspapers, 14 sets of column means tests are performed--one for each level of Labor force status for Get news from internet and Get news from newspapers. Again, no Bonferroni adjustments are applied because within each set, only one pair of columns is compared. The tests for Get news from newspapers are the same as before. For Get news from internet, the category No is assigned the letter A and Yes is assigned the letter B. The B key is reported in the A column for each set of column means tests except for those respondents temporarily not working. This means that the mean value of Hours per day watching TV is lower for people who get their news from the Internet than for people who do not get their news from newspapers. No keys are reported for the Temporarily not working set; thus, the column means are not statistically different for these respondents.