Quality of food products in Bangladesh: analysis in the framework of quality control and econometric techniques

Abstract

An abstract of the thesis of Md. Anwar Hossain for the Ph.D Degree in the Department of Statistics, Biostatistics and Informatics, University of Dhaka. Recent global increases in food contamination and adulteration have further exacerbated vulnerabilities and made it imperative to examine food products in the country. Food is identified as one of the sustainable approaches to farming and offers insights towards a paradigm shift in food and nutritional quality. The binary logit or probit regression model is one of the popular choices to study effects on dichotomous responses. This study is designed at monitoring a definite response i.e., food acceptability result in terms of some predictors, determines the goodness of fit as well as strength of the assumptions and selecting an appropriate and more parsimonious technique there by proffer helpful suggestions and recommendations. The technique was used as a tool to representation the major factors viz. food quality characteristic (moisture, protein, fat, carbohydrate and iron etc.) and microbial (aerobic plate count, coliform etc.) that affected the acceptability of food. The fit, of the technique was illustrated with 8 (eight) types of food, 678 food samples and several physiochemical and microbial test parameters records obtained from Institute of Food Science and Technology (IFST), BCSIR, Dhaka by the method of Single Stage Cluster Sampling. This study contents some visible and numeric statistical methods for checking the level of significance of technique. The technique spotlight on assessing acceptability of statistical methods for specific food analysis variate effects. The tested technique showed best fit and performed differently depending on classification of result, sufficiency in relation to assumptions and goodness of fit. The outcome specify that different methods for analyzing binary data are likely to vary in their presentation under the subsequent conditions: 1. the distribution of the interpreter variables follows significantly from normality, 2. the two group must have equal variances. While these conditions are by no means inclusive, the being of conditions should concern the study that the choice of a particular method should be made carefully. Reliable with precedent studies, the presentation of logit and probit was similar under the different conditions. Hence, a selection between these two may not be significant (except in computational cost). However, the selection among logit, probit model, Linear Discriminant Analysis and OLS is still not uncomplicated. Therefore, the study be supposed to first carry out different preliminary data study to decide the statistical properties of the forecaster variables. Possibly part of the data might be analyzed by these techniques to decide which one is most suitable. Otherwise, the study could change the data to comply with the assumptions of a particular technique. Comparing the consequences, we observed that the Logit and Probit model outperforms other models and techniques as because they fulfill necessary assumptions as required. If in the case of normality assumptions fullfill, Discriminant analysis also yields better results. The results of one sample t-test and wilcoxon signed-rank/ sign test indicate that, in almost every case when the null hypothesis was false, the t-test performed same results with the Wilcoxon signed-rank/ sign test though the normality assumptions violated. But when H0 was true, the Wilcoxon signed-rank test perform efficient or more power predicted probability (p-value) than the t-test as fullfill the assumptions.