Effects of Nesting and Stacking on Tests of Independence

The rule for tests of independence is as follows: a separate test is performed for each innermost subtable. To see how nesting affects the tests, consider the previous example, but with Marital status nested within levels of Gender.

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

   

   With Marital status nested within levels of Gender, two tests are performed--one for each level of Gender. The significance value for each test indicates that you can reject the hypothesis of independence between Marital status and Labor force status for both males and females. However, the table notes that more than 20% of each table's cells have expected counts of less than 5, and the minimum expected cell count is less than 1. These notes indicate that the assumptions of the chi-square test may not be met by these tables, so the results of the tests are suspect.

   Note: The footnotes may be cut off from view by the cell boundaries. You can make them visible by changing the alignment of these cells in the Cell Properties dialog box.

   To see how stacking affects the tests:

4. Open the table builder again (Analyze menu, Tables, Custom Tables).
5. Drag and drop Highest degree from the variable list into the Rows area below Labor force status.
6. Click OK to create the table.

With Highest degree stacked with Labor force status, four tests are performed--a test of the independence of Marital status and Labor force status, and a test of Marital status and Highest degree for each level of Gender. The test results for Marital status and Labor force status are the same as before. The test results for Marital status and Highest degree indicate these variables are not independent.