Articles in this section

See more

How to Setup a MaxDiff Experiment as a Ranking

  • Updated
Follow

This article describes how to setup 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.

Requirements

  • A list of alternatives that you want to test in your MaxDiff experiment.

Method

There are two ways to perform the setup. If you are using Q, an automated process 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:

mceclip0.png

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):

mceclip2.png

  • 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:

mceclip3.png

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:
    • If using Displayr, select Combine and then set the Structure in the Object Inspector to Ranking.
    • If using Q, right-click and select Set Question and set the Question Type to Ranking.

While the exact method will depend on the layout of your data, these kinds of variables can be set up using JavaScript variables or R Variables.

Next

How to Create a MaxDiff Experimental Design

 

Related articles