Sensory evaluation of dehydrated tomatoes using the proportional odds mixed model: XXXX

In order to identify products acceptance and commercialization, researchers performed sensory evaluation. The responses to these studies are ordinal categorical variables, thus, they are normally based on a hedonic scale. Generally, in sensory experiments, each judge evaluate several products and the answers given by the same judge can be classed as repeated measures. In this paper, we propose an alternative statistical analysis to consider the natural ordering and repeated measures simultaneously using proportional odds mixed models. Such models were proposed to compare the taste acceptance of two varieties of tomato, Italian and Sweet Grape, after the osmotic dehydration process. The experiment was designed in the way that each judge, randomly selected and evaluated each variety once. The results showed that the Sweet Grape presented better acceptance than Italian variety. This model allowed the predictions and estimations probabilities to be calculated for each of the response categories according to each variety. PRACTICAL APPLICATIONS: Tomate is a product highly perishable and new dehydration techniques are being developed and needs to have good accepted by consumers. In this paper, we proposed an alternative statistical model to compare the taste acceptance of two varieties of dehydrated tomato, Italian and Sweet Grape. The model presented can be used in different sensorial evaluations and can help researchers to analyze ordinal categorical data, focusing on how to identify and account for specific issues arising from experimental design (such as correlated observations). The results showed that the Italian tomato was 6.89 times more likely to be classified with lower scores at hedonic scale compared to Sweet Grape tomato regardless of sex. The Sweet Grape variety has a probability of 44% (32%; 57%) of receiving score 9 and for Italian variety this happens with a probability equal of 10% (5%; 19%). In addition, the results showed no significant correlation between responses from the same judge.