Order and session size effects on treatment discrimination: case study liking for dulce de leche

Abstract

This study aimed to evaluate the implications of the number of samples per panelist in experimental precision and relative efficiency of treatment comparisons when modelling order effects for Dulce de Leche samples using a Sudoku design. Precision is defined as small variance and relative efficiency is defined as smaller variance than other alternative. Both are related to increasing discrimination power. A series of 8 Sudoku 16 × 16 squares being 4 randomized independently and 4 others in the reverse order, was designed to allow a linear model fitting to experiments of different sizes (orders 1 to 4, 1 to 8, 1 to 12 and 1 to 16). Data from 112 panelists were kept. Responses were recorded using a nine-point hedonic scale. Data was analyzed using a mixed linear model panelist, order and treatment (sample) effects. Analysis has shown a pronounced order effect, showing a first sample effect overestimation. Moreover, it was noted that order effect was rather evidenced in the first four orders, making for a relatively more precise experiment than the ones using remaining samples. Therefore, the use of Sudoku can be a convenient strategy to organize smaller testing sessions that could result in efficient experiments, with small variance of the estimated averages or effects and needing a lot less panelists to make a design as efficient as the large one used. Moreover, Sudoku can also inspire an efficient design that allows different test sessions.