Estimation of Correlation Structures of Four Basic Tastes Data with Missing Measurements

Abstract

The means and correlation structures of four basic tastes data for Japanese and Chinese men and women, are compared. Tasting data were measured by a filter paper disk method, and recorded as missing values when subjects were not able to recognize a highest concentration of a taste substance solution. The missing values are recorded as counts, so the tasting data is expressed as censored data. A few statistics, i.e. mean vectors, variance-covariance matirices, and correlation matirices, are estimated by a multivariate statistical model for censored data to avoid biases on the estimation. Moreover, a correlative map among four tastes for men and women of Japan and China is drawn by using a principal components analysis. These results suggest that: (i) An appropriate statistical model is needed for analysing data with missing values. (ii) Positive correlations among the four basic tastes were estimated. (iii) The differences in experience and custom of eating appear as taste sensitivities and mean and correlation structures, such that the correlation between men and women is positive, while between Japanese and Chinese is negative. (iv) The correlation structures suggest a kind of an ordering of the tastes for Japanese and Chinese, and for men and women.