When Column Comparisons are specified on a table a number of the Multiple comparisons correction methods that can be selected for Column comparisons in Statistical Assumptions make stringent assumptions about the relationship between the columns being tested. In particular, the ANOVA-Based Column Comparison corrections all assume that the columns are not overlapping (i.e., reflect independent samples). Where this assumption is not met, Q will return an error message.
Method 1: Specifying comparisons for most ANOVA-Based Column Comparisons
With the exception of Dunnett’s Pairwise Multiple Comparison , the ANOVA-Based Column Comparison corrections require non-overlapping columns. In practice, this generally means there is a need to removed NETs from a comparison.
This can be done by either right-clicking on a NET category and selecting Hide, or, by creating a Span which excludes the NET (as when a Span is created, Q conducts the comparison using only within the spans). The following two images demonstrate the alternative ways of setting up the data appropriately.
Method 2: Specifying comparisons for Dunnett’s Pairwise Multiple Comparison
Dunnett’s Pairwise Multiple Comparison tests each category against one specific category of interest (e.g., new products are compared to a current product). It is set up by:
- Removing the NET category (as in the second example above).
- Removing and/or re-ordering categories so that the category of interest appears in the first column.
- Right-clicking on the columns and selecting Comparisons and entering appropriate comparisons. For example: A/B, A/C, A/D, A/E, A/F, A/G, A/H, A/I.