Although child malnutrition is improving over the world in the last couple of decades, still now it is concerning issue among the developing countries including Bangladesh. In general, malnutrition is a dichotomous response variable fitted with logistic regression model. But in this study, counting number of malnourished children in each household is defined as response variable. UNICEF with co-operating Bangladesh Bureau of Statistics (BBS) conducted Multiple Indicator Cluster Survey (MICS) covering 64000 households in Bangladesh by using two stage stratified sampling technique, where 21000 households have children age 0-5 years. We use bivariate analysis figuring out significant association between target and socio-demographic predictor variables. Then Negative binomial regression model is used over poisson regression model due to arising over-dispersion problem ($variance > mean$). Zero inflated negative binomial model also is applied for the excess of zeros in the target variable. Considering standard error and significant level of individual factors NB model provides better result as compare to ZINB.

翻译：尽管过去几十年全球儿童营养不良状况有所改善，但其在包括孟加拉国在内的发展中国家仍是值得关注的问题。通常，营养不良作为二分类响应变量适用于逻辑回归模型。但本研究中，将每户家庭中营养不良儿童的数量定义为响应变量。联合国儿童基金会与孟加拉国统计局合作，采用两阶段分层抽样技术，在孟加拉国开展了覆盖64000户家庭的多指标类集调查，其中21000户家庭有0-5岁儿童。我们通过双变量分析揭示了目标变量与社会人口统计学预测变量之间的显著关联。由于存在过度离散问题（$方差 > 均值$），本研究采用负二项回归模型替代泊松回归模型。针对目标变量中过多的零值，同时应用了零膨胀负二项模型。就标准误差与各因素的显著性水平而言，NB模型相较于ZINB模型提供了更优的结果。