Vulnerability to Multidimensional Poverty in Algeria and Tunisia Using the Counting Based Approach and Bayesian Networks Classifiers

This paper assesses levels and trends in vulnerability to multidimensional poverty for two different years in Algeria (2012/13 and 2019) and Tunisia (2012 and 2018). Using as benchmark the M-gamma multidimensional poverty measures as developed by Alkire and Foster (2019), it follows the approach suggested by Gallardo (2022). To preserve the multidimensional nature of poverty, the joint probability of being poor and deprived in each dimension is modelled using multidimensional Bayesian networks classifiers and the vulnerability by mean risk approach (VMR) to vulnerability measurement. Despite similar levels of multidimensional poverty, vulnerability measures are higher in Tunisia than in Algeria. In addition, the achievements in poverty reduction are more fragile in Tunisia than in Algeria. The results show that moderate vulnerability prevails over severe vulnerability both in Algeria and Tunisia. Trends over time indicate that in Algeria, vulnerability seems to be shifting more towards moderate vulnerability while the opposite is observed in Tunisia. The indicators that differentiate severe from moderate vulnerability are mainly related to health and education dimensions both in Algeria and Tunisia. We show that chronic poverty among the vulnerable is larger in Tunisia than in Algeria. Our results reveal also different trajectories in the evolution of the vulnerability components in these two countries.