How To Create Banner Tables Using Pick One - Multi Or Grid Questions
Crosstabs involving Pick One - Multi and grid questions can be difficult to read, and have limited options (e.g., often no Statistics - Below or Statistics - Right). Usually the easiest solution is to restructure the data, by changing Question Type. However, in some situations it can be useful to construct an output that looks more like a banner has been used (e.g., when conducting sensory tests or concept tests, where there is a need to compare ratings by demographics). This is the focus of this page.
General filtering method
- Automate > Browse Online Library > Create New Variables > Filtering - Filter One Question by Another Question.
- If you desire, right-click on a column or row and select Swap Rows and Columns.
- If you wish to modify the spans, often the easiest approach is to right-click and select Revert, rearrange the rows or columns as you wish, select the rows or columns you wish to group, and select Create Span, using Find/Replace to tidy things up.
- By default, Q will only perform comparisons between columns with overlapping samples in Pick One - Multi in questions such as the one above, and also in grid questions. Thus, in the example above, Q will not, by default, compare between the data for males and females, even if they are within the same span (as they have no overlapping data). However, you can perform such comparisons by selecting the relevant tables and using Edit > Table Options > Statistical Assumptions and:
- Changing Proportions to Survey Reporter Proportions.
- Changing Overlaps to Dependent.
- Pressing OK.
- If you need to compare across spans, you can either:
- In Statistical Assumptions uncheck Recycle column letters. Next select Comparisons... from the Table Context Menu and check Compare all columns and follow the prompts that appear to change further statistical assuptions. (If you are using a version of Q earlier than 22.214.171.124, the Compare all columns option does not exist and instead you must remove all of the spans instead.)
- Override all the testing using Significance Testing - Independent Samples Column Means and Proportions Tests. However, this approach will generally have lower statistical power, by assuming paired data is independent.
Other filtering methods
In situations where there is a need to use different combinations of variables for different columns, some options are:
- First create variable with all relevant filtering variables, use the method described above, and delete any columns that are not required.
- Filtering - Apply Filters to Columns or Rows, which will allow you to create manually create filters, one-by-one.
- How to Filter Rows of a Pick One - Multi Question with Different Filters.
- Stack the data, so that each product is on top of each other product (see Stacking Data Files).
- Modify the Value Attributes of the the variable called observation at the end of the stacked data file (e.g., if you have stacked the data to put products on top of each other, then you will give the name of the first product as the Label of the first product, etc.).
- Create a Banner question, using observation and any other required variables.
- If many of your analyses do not require the data to be stacked, set up using Multiple Data Files.
Note that while this method is most flexible in terms of quickly creating and modifying tables, any statistical testing will assume that the samples are independent, even when they are not.