On nearly balanced designs for sensory trials

In sensory experiments, often neighbour balanced designs are used. The idea behind these designs is to control for possible carryover effects, like, e.g. a lingering taste of the products. These designs do not allow to randomize the order of presentation of the products to the assessors. We therefore consider the randomization of assessors and product labels. For some designs, this randomization validates the use of the simple row-column model, as long as there are no carryover effects present. If there are carryover effects, the estimate of the variance becomes conservative for special neighbour balanced designs, so-called totally balanced designs.Usually, for given numbers of products, of assessors and of assessments per assessor, these special designs do not exist. Several authors [e.g. Périnel, E., & Pagès, J. (2004). Optimal nested cross-over designs in sensory analysis. Food Quality and Preference, 15, 439–446] recommend algorithms to construct designs that come near to the properties of totally balanced designs. The main purpose of the paper is to demonstrate with the help of a simulation study that these designs indeed perform similarly.It has to be pointed out, however, that these results only hold if the analysis takes period effects into account. Analysis with the simple block model can lead to grossly wrong estimation of the variance.