MaxDiff technique

The MaxDiff, technique is used for preferences researching. MaxDiff may be used in many research areas, which includes surveying preferences regarding:

  • prizes in loyalty programmers
  • tastes
  • product brands
  • importance of factors:
    • affecting selection of an operator's offering, e.g.: telecommunication, web portals, gas stations, professional software
    • affecting physicians drug prescription decisions
    • likely to improve employee satisfaction levels

In social research, MaxDiff can be used to identify, e.g.:

  • social groups perceived as the most prone to social exclusion
  • self-governmental tasks key to local community

The most important feature of the MaxDiff technique is the simplicity of the task, which respondents are facing: their task is to pick from 45 objects/values, one which in the given set is the most important/attractive to them and one which is relatively least important/attractive.

Thanks to advanced analytical models (utilizing hierarchic Bayesian estimation), even a dozen of such questions permits estimation of respondent preferences with respect to several tens of attributes. What is important, is that these are individual and permanent preferences, which additionally have strong discriminating power. With large number of objects (more than 50), this task may become for respondent tiring, but if the surveyed sample is sufficiently large and the research is performed using the CAPI technique, some optimization strategies for evaluation object selection can be used, which allows the number of choices the respondent has to make to be limited, without loosing data.

Obtaining equally accurate results is only possible using the CSP (Constant Sum of Preferences technique), which involves distribution of preference points between object pairs. This task however is more difficult and less intuitive for a respondent and practically impossible to accomplish, when the number of objects exceeds five. In that case the number of compared pairs exceeds 14, which far exceeds perception skills of an average respondent.

Ranging seems to be a simple and at the same time quite an accurate technique. However when the number of attributes to be ordered exceeds ten, respondents can not perform the task reliably. Apart from that in result only an ordinal scale is obtained, e.g. it is only known what is better than what but it is not known by how much, which limits the range of possible further analyses.

Evaluation assumes that respondents are able to translate their actual assessments of researched attributes to a numeric scale. Each respondent however makes it differently and therefore results are incomparable. In particular, some respondents prefer to place their assessments on extremes of the scale, while others like to remain relatively neutral. Often very low differentiation of assessments are obtained: all attributes are equally important and the product preferred by the respondent is evaluated very high in all respects.

Maxdiff copes well with all of the following limitations:

  • it is possible to test preferences with respect to a very large group of objects
  • the respondent's task is simple and eliminates the impact of human perfection flaws
  • obtained preferences are comparable, continuous and highly differentiating both objects as well as respondents