Generating Optimal Sample Presentation Orders
Rousseau, B. and Ennis, D. M. (2021). IFPress, 24(1) 3-4.
Sample presentation orders arise in sensory and consumer research because sequential monadic evaluations of multiple samples are often used. Due to the successive nature of the evaluations, it is necessary to address experimental biases related to the sample presentation orders. Failing to do so will lead to biased and more variable estimates leading to reduced sample discriminability. Figure 1 illustrates typical position and sequence effects on 9-point hedonic scale averages (based on actual client-related consumer work involving the evaluation of nine samples over three days.)
In a previous technical report, we describe how various approaches differ in their ability to control for sample position, sample sequences and their spread across the design. Randomization by row (respondent) will never provide a balanced design since this would require an infinite number of respondents. A replicated Williams Square design will provide partial balancing, with a balance of sample position and sequences (under certain conditions), but there will be no control over where these sequences occur in the design (sequence spread). We showed that an approach using column randomization and search across millions of options delivers the most balanced design based on these three indices. The column randomization and search method (CR&S) is available in the Tools version of the IFPrograms® software.
We will focus on a situation that often occurs where testing is conducted over multiple days. In this situation, sequence effects within a day are important but sequence effects are not expected to occur from the last sample on one day to the first sample on the next day. Taking this into account would allow more attention to be paid to sequences that matter and deprioritize those that do not.