Presented at the 2010 Society of Sensory Professionals Conference in Napa, CA
Data that are obtained for many forced choice procedures such as the 2 and 3 alternative forced choice (2-AFC and 3-AFC), the duo-trio and the triangular methods are often tested using the binomial distribution or its normal approximation. In some applications however it is desirable to allow for the possibility of a no difference or no preference option which complicates the analysis of the data generated by these methods. The proper treatment of these no difference counts has been the subject of much debate in the product testing community for many years with some practitioners preferring to distribute them proportional to the item choices, others preferring to distribute them equally and others preferring to drop them entirely and restrict the results to those who expressed a preference. In addition, models that explicitly account for the no difference counts have been developed and in particular there is a Thurstonian model that estimates both the size of the difference between the items and the likelihood of generating no difference responses. In this talk we first review the motivations for allowing no preference votes in discrimination testing and we then offer definitive recommendations based on substantial simulation based power analysis for how to best analyze data that contain no preference votes.
Colleagues can download the full presentation here:
Not a Colleague? Click here to join for free!