A MaxDiff study involves presenting a sample of respondents with a series of questions, in which each question contains a list of alternatives (brands, products, etc) and the respondent is asked which alternative they like the most (best) and which the least (worst). Typically, MaxDiff studies are used to estimate the relative utility that respondents get from each alternative. There are different algorithms that can be used: Hierarchical Bayes, Latent Class, and Multinomial Logit. The average utilities can be used to calculate preference shares of each brand for the market or different market segments. In order to estimate the relative coefficients (utilities), the algorithms need to know:
- the alternatives shown for each question (called an Experimental Design)
- a variable for each question for which alternative was chosen as best or worst (these choice variables are usually shown alongside the rest of the respondent survey data).
This article outlines the format for both of those bits of information. Note that some software has different formats of MaxDiff data than shown below. For certain ones, such as Qualtrics and Alchemer, Displayr has automations on how to reformat the data.
1. Experimental Design
The algorithm needs to know the alternatives shown for each question in order to see which alternatives were not picked as best/worst. This lets the algorithm know a general ranking of the preference of the alternatives for each respondent, i.e. they like A best and C worst and B, D, E somewhere in between. The outline of what alternatives were asked for each question, how many questions there are, and how many versions of questions is known as the Experiment Design. Displayr can generate Experimental Designs for new studies, but you will need to get the Experimental Design file from the survey platform where you fielded the study. The format of the Experimental Design should have a column for the Version, Task, and one for each of the number of Alternatives shown for each question. Each row is for a specific Version and Task and lists the number of the Alternative shown in each option.
You can import the experimental data as a data set or summary table in the Data Sources pane or use Ctrl+V to paste it directly on a Page or in your Report. Or you can link to the table using a public URL.
2. Best/Worst Choices
The algorithm also needs to know which alternatives were picked as the "Best" and "Worst" for each respondent. These choices are usually stored alongside the rest of the questions from the survey. The data should be formatted where each row is a respondent and there is a column for the Version of the survey the respondent got and a variable for each Best and Worst choice for each Question. The values for the best or worst can be the number of the alternative chosen, such as the data below with 10 alternatives:
or the number of the option chosen, such as the data below where 5 options were shown for each question:
If the data is stored as the option number chosen, the Structure of the variables needs to be Numeric or Numeric-Multi. If the Best and Worst variables are Structured as numeric, you will need to supply a list of Alternative Labels in the MaxDiff analyses.