Proportion Tests Applied in Assessors’ Selection Under Assumption Break: The Cases of Sensory Fatigue and Learning Process

Abstract

Triangle trials are commonly employed in panelist selection, describing a Binomial distribution. However, assessors can fatigue or develop sensitivity after many trials, which violate constant proportion assumption. This violation could affect some tests obtained from normal distribution interval estimators (TN1, TN2 and TN3), F distribution (TF), Sequential Test (TS) and Poisson approximation with quantile (TP). To evaluate this tests performance, Monte Carlo simulations were performed for 1000 assessors through 5, 10, 15, ..., 100 triangle trials, fatiguing and learning in linear and nonlinear models. Violating this assumption can affect type I error and power. Power curves for linear and nonlinear fatigue models showed large type I error and low power with increasing p and small values of n. Power curves for linear and nonlinear learning models, increased simultaneous to p and n. Thus, n value has more influence on power curves than p. TS presented less power in many trials of learning functions with better behavior in few trials, while TP presented the smallest type I error rate in all situations. TN1, TN2 and TF presented small type I error and big power for linear and nonlinear learning functions, when n increases. However, it is not recommended using tests for sensory fatigue.

.