How To Test Differences in % Column Shares

From Q
Jump to navigation Jump to search

When statistical tests are conducted on tables containing share statistics (e.g., % Column Share), the tests are conducted using the averages, not the share statistics. This can lead to counter-intuitive results when reporting the shares. This page describes how to get Q to instead test differences between the shares.

Example of the problem

In the table below on the left, for example, the automatic statistical testing shows that Pepsi is significantly higher amongst the heavy consumers than the light consumers, but the actual numbers on the table show the opposite conclusion. The table on the right shows the averages, from which we can see that the average is higher for Pepsi amongst Heavy consumers.


The solution

The solution to the problem is to transform the data, so that the Average on the table is proportional to the % Column Share:

  • Compute a new question which contains the share computed for each case: Automate > Browse Online Library > Create New Variables > Case-Level Shares.
  • Select the new variables in the Blue Drop-down menu.
  • Select the original variables in the Variables and Questions tab, right-click and select Insert Ready-Made Formula(s) > Mathematical Function(s) by Case > Sum and No.
  • Tag the newly-created variable as a Weight.
  • On the Outputs Tab, apply the weight to the table.

The Average and % Column Share on the tables will now coincide, and the statistical tests now, as a consequence, can be interpreted as testing differences in % Column Share.