Maximum Difference Scaling (MaxDiff)
Establishing the importance of various attributes
Maximum Difference Scaling (MaxDiff) is a highly efficient process for establishing preference differences between a relatively high number of attributes of a similar nature. In surveys, the process functions in a similar manner to straightforward paired comparisons.
Respondents are given a series of similar tasks, each of which for example contains four attributes (chosen beforehand from a complete list of predefined attributes). From these 4 listed attributes they must choose the one they consider (1) most important, and (2) least important. This question is put to the respondent several times, but each time with varying items. Whereas the response items vary, the structure of the question remains unchanged.
In survey, a MaxDiff task will be presented as follows:
Using multinomial logistic regression, utility values can be established from the respondents’ answers. These are presented in standard format. The higher the value, the more important the attribute.
The evaluation shows the utility values:
The benefits of the method:
Compared to surveys using rating scales, it is easier for respondents to give clearly defined answers
By limiting questioning to four attributes per „task“, the scope of the task remains controllable (most respondents could not cope with ranking or allocation tasks containing 10+ items)
Rating scales have a tendency to generate homogenous results as respondents consider all, or nearly all, attributes to be „important“. MaxDiff results are considerably more differentiated because respondents must choose between attributes each time
Metric data result, which are then available for application in further analysis, e.g. using multivariate procedures
The absolute peak of the utility values is directly comparable (item A is x-times more important than item B)