Main content area

Application of Maximum Likelihood Estimation for First Serving Order Bias Adjustment

Kwak, Han Sub, Meullenet, Jean‐François, Lee, Youngseung
Journal of sensory studies 2015 v.30 no.1 pp. 68-76
beans, breads, consumer acceptance, consumer preferences, data collection, principal component analysis, statistical models, variance, wheat
Maximum likelihood estimation (MLE) is one of the statistical models to estimate parameters or missing values. MLE approach in minimizing first serving order bias (FSOB) was investigated. Two different consumer studies with three wheat breads and nine bean products were performed. To test MLE, consumer ratings of the first product were eliminated from the data, and imputed by MLE. Regression imputation was used to estimate means and intercepts, after the model parameters were set equal to their MLE. Significantly higher overall liking (OL) ratings were observed for products presented in the first order, whereas no significant serving order effects on OL were observed after MLE adjustment. Uneven consumer distribution in the consumer segmentations for unadjusted data sets revealed that actual segments of consumers might overlap with the existence of the FSOB, while the MLE approach showed different consumer segmentation patterns. The MLE approach did not distort the overall interpretation of the consumer preference, in comparison with the original data sets, by the principal component analysis. The MLE approach for eliminating FSOB was shown to be effective, especially in cases where it is necessary to adjust first‐order products at the individual consumer level. PRACTICAL APPLICATIONS: The first serving order bias (FSOB) is often seen in consumer testing, which hedonically evaluates samples that are presented in a sequential monadic design. It could inflate the experimental error variance, resulting in the delivery of inaccurate consumer acceptances. In the present study, we introduce a maximum likelihood estimation (MLE)‐based approach to minimize the FSOB. This method was proven to be quite robust and easy to perform to treat the FSOB in an effective fashion, and it is not detrimental to consumers' preference patterns, compared with the original data set. In particular, this method would be useful when specific products showing the FSOB exist in multiple products assessment, and only those products need to be selectively adjusted.