Focusing on determinants of Tunisian middle class: a spatial approach

Many empirical studies in developing countries show that the expansion of the middle class is associated with better governance, economic growth and poverty reduction. Thus, assessing the weight of the middle class and exploring its determinants should be of primary interest to policy makers. We find several papers analyzing the determinants of middle class in developing countries, among others, we can cite the manuscripts of Martinez and Parent (Middle class determinants in Latin America (2000–2010): a gender perspective, 2012), Filali and Bouabid (Région et développement 44: 79–101, 2016). These studies were conducted in a non-spatial framework. However, the distribution of household income classes in one region can be expected to influence those of neighboring regions. This analysis uses a spatial ordered probit model which allows interregional spatial interactions and heteroscedasticity to define the determinants of Tunisian middle class. Based on the economic approach in defining this class, I use the total annual consumption expenditure as an approximation of the permanent household income. I identified three subcategories: a lower middle class, an intermediate class and an upper class.Model parameters are estimated in a Bayesian structure using MCMC simulation, by sampling sequentially from the complete set on conditional distributions for all parameters. The results indicate that the spatial model outperforms the standard model and that the age of the head of the household, a high level of education, the number of employees in the household, living in an urban area and home ownership have a positive effect on the probability of being in the upper income classes. Furthermore, the spatial coefficient is significantly positive, showing that spatial relationships should be reflected in model specification. This observation is confirmed by the mean values of regional specific errors $$\left({\uptheta }_{\mathrm{i}}\right)$$ θ i estimates. Indeed, a clustering pattern of similar values is visible.