期刊名称：Pakistan Journal of Statistics and Operation Research
印刷版ISSN：2220-5810
出版年度：2012
卷号：8
期号：4
页码：821-837
DOI：10.1234/pjsor.v8i4.254
语种：English
出版社：College of Statistical and Actuarial Sciences
摘要：Abstract The main objective of this study is to pinpoint the main factors that affect the percentage who suffers of malnutrition in developing countries. Three locations are randomly chosen: Asia, Africa, and Middle east and North Africa ( MENA); A total of 96 countries were chosen randomly from 137 developing countries of the three locations; and were cross classified by " Location" and " Human Development Index (HDI) as high, middle, and low (UNDP, 2005). Data for the study was compiled from FAO (2005). The analysis started with seven explanatory variables and the dependent variable; however, stepwise regression reveals that the average Protein intake and Infant mortality rate were the only two significant variables. "Location and "HDI" are dummy coded and OLS regression is performed using the two significant variables, but the only significant variable was the "average protein intake". OLS multiple regression Model is re-applied to the data using dummy variables technique with interaction with the "average Protein intake", nine regression equations were reached. The Linear Mixed effect Models are also applied, using "location" as the random factor and "HDI" as the fixed factor. Five models were applied: (1) a null model (baseline model)where no predictors are introduced to the model; (2) the fixed model: where predictors used are the covariate and the HDI; (3) the random model: where predictors used are the covariate and Location ; (4) the mixed model: where predictors used are the covariate and the HDI I ( fixed) and the location( random); and (5) the random coefficient model: where predictors used are the covariate , the HDI Index and the location but produces different prediction equations that differ in slopes and intercepts. Models are compared based on information criterions. The random coefficient model produces the least criterion values and thus fits better than all previous ones. A comparison between the Random Coefficient model results and GLM model is made, and conclusions are reached.
