How to Merge or NET Categories on a Nominal - Multi

In a table of a Nominal - Multi question type, it is only possible to create a NET and perform merges in either the row or the column, but not both. For example, in the table below, the columns can be merged because they are the codes for the question, but the rows cannot be merged because they are individual questions/variables and essentially separate Nominal: Mutually exclusive categories in the data set.

This is a logical constraint rather than a limitation of the program. If such merges were possible, their meaning would be ambiguous. There isn't a standard or default way to aggregate multiple responses across questions for the same base of respondents.

Consider the consequence of merging the Coca-Cola and Diet Coke rows. When you merge or NET categories, Displayr computes the result as the proportion of people who choose at least one of the underlying categories. Thus, the Hate score would need to be the proportion of people who selected Hate for either of the brands, the Dislike score would need to be the proportion that selected Dislike for either brand, etc. The consequence would be that the percentages would add up to more than 100% (since the base for the percentages would still be the number of respondents who answered both questions). Even if a respondent answered 'Hate' for both brands, that would count as 1 response (n) in the NET/merge, leading to ambiguous results. This would be confusing, as it is inconsistent with the definition of a Nominal - Multi; when the concept of missing data is introduced, it would be even more confusing.

Nonetheless, there are workarounds to this depending on how you'd like to analyze your data.

• Method 1 - Convert the question to a binary - multi
• Method 2 - Convert to a pick any grid (binary grid)
• Method 3 - Count responses instead of respondents

Requirements

A dataset containing a Nominal: Multi-variable set.

Method 1 - Convert the question to a binary - multi

Use this method if you are only interested in aggregating one aspect of the question, such as a Top 2 Box, a specific code, or a combination of codes. This example uses the table in the screenshot above to merge and create NET Top 2 Box scores across brands.

From a table:

1. OPTIONAL: It's recommended that you duplicate your question before making edits to preserve the original structure. Select the variable set from the Data Sources tree, right-click, and select Duplicate. Right-click on the new duplicated variable set and select Rename to give it an appropriate name. For example, if your plan is to merge brands based on whether people love or like them, the name might be "Cola brands loved or liked." 
2. Create a summary table using the variable by dragging it onto the page.
3. Merge any of the codes that you wish to have merged, such as Like and Love. This can be done by simply dragging and dropping the icon of one column header onto another. 
4. Select all the other codes that you did not merge, right-click, and select Hide until you only have a single column left, as shown below.
   
5. Select the variable in the Data Sources tree, and from the object inspector , change the Structure to Binary - Multi.
6. Now you can merge the rows.
   

From the Data Sources tree:

1. Select the variable set from the Data Sources tree, right-click, and select Duplicate.
2. In the object inspector , change the Structure to Binary - Multi.
3. Click the Fix button from the warning that appears above the object inspector.
4. Check Count This Value for all of the codes you'd like to count in the final results and click OK.
   
5. You can now merge the rows in a table for this question.

Method 2 - Convert to a pick any grid (binary grid)

To retain the original structure of the question, you can convert the question to a Binary - Grid. This method is not recommended due to the ambiguous nature of the results. However, it is outlined below in case your specific data setup calculates unambiguous results. Proceed carefully! This example uses the table in the screenshot above. 

1. In the Data Sources tree, hover and click + > Convert To > Binary - Multi  > Flatten Nominal/Ordinal - Multi.
2. Right-click on the new variable set and select Rename to give it a new name. 
3. Change the Structure in the flattened question to a Binary - Grid.
4. Click the Fix button in the warning above the object inspector and then click OK on the next screen.
5. Drag the binary - grid variable onto the page to create a summary table.
6. OPTIONAL: You can select the table and then click Switch rows and columns in the object inspector to get the desired format.
   
7. Now you can merge and NET on both the rows and the columns.

Method 3 - Count responses instead of respondents 

Please note that when using this method, Displayr's statistical testing system will only ever test the Average statistic, which is not used in the table. As a result, it can be good practice to disable the significance testing by setting Object Inspector > Appearance > Significance to No in the table.  

1. Follow the directions in Method 2 to convert your question to a Binary - Grid.
2. Change the Structure to Numeric - Grid. This will make a Not selected response count as 0 and a Selected response count as 1.
3. Create a table from the new Numeric - Grid question by dragging it onto the page, then from the object inspector , go to Data > Statistics > Cells and select Sum and % Row Share and uncheck Average. The % Row Share will be the % of all responses for the categories included.
   
4. Combine your rows as desired by selecting them on the table, right-clicking, and selecting Combine.
   

Next

How to Merge/Combine Categories in Tables and Charts