Multi-punch/Multiple Response Questions Displaying as Grids

The problem

Categorical questions that permit multiple responses (i.e., multi-punch, multi, multiple response questions) are displaying in a table so that there is one row for each response containing the variable's label, and each code is appearing as a column. Or, to use different terminology, the multiple response data has been created in the data file with a max-multi structure.

The easy solution

1. The Question Type for such questions needs to be set as Pick Any - Compact (this is the correct question type in data for multiple response data that has been set up with max-multi coding).
2. If using Q 4.8.2 or earlier, where questions are not asked of the entire sample it may be necessary to construct a filter remove respondents who were not asked the question. This is done by right clicking on the number next to the NET, selecting Create Filter, and making sure the filter is selected in the Filter drop-down below the table. Since Q 4.8.3, Pick Any - Compact questions automatically apply this filtering step.

The better solution

The reason that Q fails to automatically detect the structure of such data is because it is ambiguous. It is not possible for the software to determine with complete confidence whether it is a grid (or, to use the Q terminology, Pick One - Multi), or a Pick Any - Compact, so it defaults to Pick One - Multi as this is much more common.

In most situations, the best solution is to have the data file exported so that each code is represented by a separate binary variable. This is superior to using the max-multi format in a number of ways:

• Q will be able to automatically determine the correct Question Type.
• Where there are missing values, the percentages will be correct. That is, the main reason why max-multi coding is, in general, a bad idea is that if if different respondents were shown different options, there is no way that that this can be deduced from the data (i.e., where the data is setup as binary variables, a missing value means that an option was not shown, whereas with max-multi it can either mean the option was not shown or that it was not selected).
• Most multivariate tools conduct their analysis using variables rather than questions (i.e., although you select questions when running the analysis, the maths is done using variables). For this reason, some of the multivariate methods in Q do not let you select Pick Any - Compact data, because of the risk that the user will not appreciate that the way that the data is stored will change their outputs. If you need to do multivariate analyses but have data in the max-multi format, select the question in the Variables and Questions tab, right-clicking and select Insert Ready-Made Formula(s) > Pick One Multi => Pick Any (by Category), which will create a new Pick Any question in the binary format.

The special case of large code frames

Max-multi can be used in any situation where it is known with certainty that the same code frame applies to all respondents (e.g., coding of most open-ended questions). Where this is the case and there is a very large code frame (e.g., 6,000 models of car), it will generally be superior, as the resulting data file will be smaller, the analyses will compute faster and there will be less chance of out-of-memory problems on older computers.