A multi-level statistical modelling approach to multidimensional poverty alleviation in Namibia

ABSTRACT

This thesis examined household demographic factors and how they influence poverty levels in Namibia. While most previous studies have used income and expenditure to define household socio-economic status levels, this study used a three poverty dimension approach namely health, education and living standard. This is because poverty is multidimensional. The data used came from the Namibia Household Income and Expenditure Survey (NHIES) of 2009/10. Initially, the Alkire Foster method was used to select variables for modelling. To identify key determinants of poverty classification, a binary logistic regression was used. In this case, the aim was to determine whether the predictors, age of household head, gender/sex of head of house, household size, household head educational level, physical location of the household (rural or urban), main language spoken in the household and ethnicity/region were associated with poverty. To measure the structural relationship among endogenous and exogenous variables, the study used structural equation modelling (SEM).To understand the relationship between causes of poverty, the study used multilevel structural modelling which is recommended for hierarchical data.

Results show that the variables: gender, education, age, language, household size, region and location are statistically associated with poverty.

The structural equation modelling standardised solutions indicate that location (urban/rural) defines poverty significantly with a load factor of 0.54 and a residue value of 0.70. Religion and age of head of household define poverty significantly with a load factor of 0.30 each and error value of 0.91 for all the two variables. The size of the household influenced poverty significantly with a load factor of 0.22 and an error value of 0.95; while the household’s main language and gender of the head of household influenced poverty insignificantly with loads of 0.01 and -0.02 respectively.

Using the multilevel structural equation modelling, the results revealed that within level 1 and level 2 hierarch, the household head was the most influential factor of poverty.

It can be deduced that the variables do significantly define poverty even though the error values are very high. High error values indicate that all the observed variables were difficult to measure. The latent endogenous variable, health, is influenced by poverty with a load factor of 0.44; while the latent endogenous variable education is influenced by poverty indirectly with a load factor of -0.72.