Comparison of the results of Multivariate Finite Mixture Model with Factor and Cluster Analysis Methods in Dietary Pattern Identification

Today nutritionists use dietary pattern to find out the effect of food in health. Most common statistical methods to determine dietary pattern are factor analysis and cluster analysis.Mixture models are a combination of k probability distribution with different probability and provide a parametric model for unknown distributional shapes. Recently mixture model as the third method is used to determine dietary pattern. Then we compared this new method with other two methods available for this purpose. We analyzed data from 25 food groups of 400 high school girls in Ahar (Rashidi’s data), and compared the results of factor analysis, cluster analysis and multivariate normal mixture model for dietary pattern. Selection of the best mixture model was done by AIC and BIC criteria. Three, two and five dietary pattern were obtained from factor analysis, cluster analysis and normal mixture model, respectively. Prevalence of these dietary patterns in normal mixture model was 6%,12%,34%,28%, and 20%, respectively. It is concluded that mixture model has two advantages over the two other methods. First, the proportion of each pattern in population is known and secondly, the average of consumption of each food group gets clear; so more information can be obtained compared to the usual methods.