MaxDiff analysis

What is MaxDiff analysis

MaxDiff or Maximum Difference Scaling (also called BWS for Best Worst Scaling) is a method introduced by Jordan Louvière that allows you to obtain the importance of attributes. Attributes are presented to a respondent, and they must choose the best and worst attributes (most important / least important). 

Two steps are needed to apply this method. First, a design must be generated so that each attribute is presented with other attributes an equal number of times. Then, once the respondent has selected the best and worst attribute for each choice, a model is applied to obtain the importance of each attribute. A hierarchical Bayes model is applied to obtain individual values of importance. 

In the case of MaxDiff models, individuals choose between selections of attributes. Thus, multiple choices are given to all individuals (we will select an attribute from several attributes). 

Analysis of these choices is made using a hierarchical Bayes algorithm which gives individual results. Parameters are estimated at the individual level using an iterative method (Gibbs sampling), considering each individual’s choice but also the global distribution of the choices. The obtained individual importance will be more precise. 

The MaxDiff analysis allows you to obtain individual MaxDiff scores for each respondent and each attribute. 

Thanks to our internal research, an original index ErrVarNorm to detect “bad” assessors is provided.