Setting Up a MaxDiff Experiment as a Ranking
As discussed in MaxDiff, there are some more 'exotic' types of analysis of MaxDiff data where it is useful to have the data set up as a Ranking Structure (or, to use the Q terminology, Question Type).
There are two ways to perform the setup. If you are using Q, an automated processes exists for performing his setup. Please see the Q Wiki for more information.
Alternatively, you can 'hand code' the setup, manually creating variables and setting the Structure/Question Type, as described on this page.
'Hand coding' by creating lots of variables
MaxDiff experiments are questions that present respondents with a series of options and ask them which is best and which is worst. For example, a very simple max-diff experiment may consist of the following two tasks:
Q5a. Which of these companies do you like the most and which do you like the least? Like Like most least ○ Apple ○ ○ Microsoft ○ ○ IBM ○ ○ Google ○ Q5b. Which of these companies do you like the most and which do you like the least? Like Like most least ○ IBM ○ ○ Toshiba ○ ○ Dell ○ ○ Apple ○
MaxDiff data is set up as a ranking as follows:
- The number of variables required is AT where A is the total number of alternatives (e.g., the total number of brands, or, the total number of attributes being evaluated) and T is the number of tasks. In the example shown above, T = 2 and A = 6 (i.e. even though only four brands appear in each task, the total number of brands appearing in the two tasks is six), so 12 variables are required in total.
- The first A variables relate to the first task, variables A+1 to 2A relate to the second task, etc.
- The Variable Name for each variable in the data file should be unique. It is useful to create informative variable names, such as: Task1Alt1, Task1Alt2, Task1Alt3, Task1Alt4, Task1Alt5, Task1Alt6, Task2Alt1, Task2Alt2, Task2Alt3, Task2Alt4, Task2Alt5 and Task2Alt6.
- The Variable Label for each variable should be the description of the option, even if the option was not presented. For example, for the two tasks, labels would be: Apple, Microsoft, IBM, Google, Dell, Toshiba, Apple, Microsoft, IBM, Google, Dell and Toshiba. Note that:
- The variable labels are repeated the same number of times as there are tasks. For example, with four alternatives, called A, B, C and D, and six tasks, the labels would be: A B C D A B C D A B C D A B C D A B C D A B C D. You can enter the labels for just the first task and then use Copy and Paste in Q to replicate the labels for the remaining tasks.
- The variable labels are always in the same order.
- No additional information is contained in the variable labels (e.g., no question name, nothing indicating the task).
- For example, the variable names and corresponding labels for the questions shown previously could be laid out like this (a different ordering of the brands could be used):
Task1Alt1 Task1Alt2 Task1Alt3 Task1Alt4 Task1Alt5 Task1Alt6 Task2Alt1 Task2Alt2 Task2Alt3 Task2Alt4 Task2Alt5 Task2Alt6 Apple Microsoft IBM Google Dell Toshiba Apple Microsoft IBM Google Dell Toshiba
- Each variable should contain one of four possible values for each respondent:
- A NaN if the option was not shown.
- A 1 if the option was chosen as best.
- A -1 if the option was chosen as worst.
- A 0 if the option was shown but not selected.
- With the example, if the options are laid out in the following order: Apple, Microsoft, IBM, Google, Dell and Toshiba, then the data should look something like this:
Task1Alt1 Task1Alt2 Task1Alt3 Task1Alt4 Task1Alt5 Task1Alt6 Task2Alt1 Task2Alt2 Task2Alt3 Task2Alt4 Task2Alt5 Task2Alt6 0 1 -1 0 NaN NaN 1 NaN -1 NaN 0 0 0 1 -1 0 NaN NaN 0 NaN -1 NaN 1 0 1 0 -1 0 NaN NaN 1 NaN -1 NaN 0 0 0 0 1 -1 NaN NaN 1 NaN 0 NaN 0 1
- In this example, in the first task the respondent chose Microsoft as Best and IBM as worst, while in the second task they chose Apple as Best and IBM as worst again.
- The final step is to select all the variables and:
See MaxDiff for links to key resources and concepts relating to MaxDiff.