Abstract:

Appropriate treatment of “no difference” responses in paired preference and difference testing is a persistent issue in product testing. Different approaches involving their inclusion or omission from subsequent testing using the binomial distribution have been proposed. In 1980, paired testing of four blind labeled brands of a major consumer products company was conducted to establish difference and preference testing norms for identical products. Products were manufactured at the company’s main manufacturing plant and product from the same production run for each brand was divided into two samples for paired testing. Products and a ballot were mailed to 600 category users of each brand (a total of 2400 consumers) and the return rates were 69% to 81% (a total of 1787 completed ballots.) The results of that research on identical products showed a narrow range of expected preference results (%) from 39.7:39.7:20.6 (prefer A: prefer B: no preference) to 40.8:40.8:18.5, depending on the brand, with a mean of 40:40:20. Females showed a slightly higher likelihood of choosing the “no preference” response. Although norms can be established by testing identical products in this manner, they can also be predicted from modeling routine paired product testing data as will be discussed in this report. The “no difference” version of the paired test is a special case of a relative rating method and the same model can be used to analyze data from relative-to-reference scales and just-about right scales. It may seem surprising that three methods differing so much in instructions and objectives share a common underlying model. It will be shown how the model interprets data from these methods and how the results of the analysis can be used to provide guidance for the development of norms for future testing.