Switch preference analysis by the drift vector method

Abstract

The matrix of switch preference data (e.g. one’s preference for brand j, given that one has already defined his/her first choice for brand i) is not symmetric. The averaging of appropriate off-diagonal proximity entries for such data leads to the loss of information, hence the necessity to use appropriate methods for asymmetric data. Among the chosen methods of asymmetric multidimensional scaling, particular attention was paid to the drift vectors method. This method enables to present simultaneously on the perceptual map both the configuration of points representing the analyzed objects and the vectors indicating the direction and the strength of changes in the respondents preferences.