An inquiry into the regional unemployment disparities in Tunisia: a spatial Durbin model

ABSTRACT

This study uses a spatial Durbin model in identifying possible causes of the overdispersion into regional unemployment in Tunisia. Data properties were identified using exploratory spatial data analysis which indicates significant neighbouring effects for several variables. Differences in socio-economic structure between regions explain in part these phenomena. Education is a key factor and also some constraints prevent married women from taking up jobs, thus exacerbating regional unemployment. Regions with important tourism activity do well than others. Tourism sector exhibit important spillover effects on regional unemployment while the impact of agricultural activity is confined to local labour market. The diversification of the industrial fabric at a regional level is not a sine qua non for differences reduction in regional unemployment rates. Developing road infrastructure helps to reduce unemployment disparities.

KEYWORDS:

• Unemployment disparities
• spatial autocorrelation
• spatial Durbin model
• Tunisia

Acknowledgments

I would like to thank an anonymous referee for useful comments. I am grateful to Mrs. Olfa Ben Amor for substantially improving the English and proofreading the text.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Tunisia is divided into 24 administrative areas (‘governorates’). The ‘governorates’ are divided into territorial ‘administrative delegations’ whose number varies with area size and population importance. These subregions correspond approximately to level 3 of the Nomenclature of Territorial Units for statistics (NUTS 3) in Europe.

2 An additional justification for the use of this kind of spatial weight matrix is provided in footnote 5.

3 In all calculations and in the empirical investigation, the variable road density is expressed in logarithm to reduce dispersion of this variable characterized also by a wide range due to the presence of outliers. This is also the only variable that is not expressed in percentage.

4 This local index associated to an observation i of a variable x, is defined as $I_i=x_i\sum _{j\ne i}{w}_{ij}{x}_j$. The sum of all local Moran's gives the Moran's I statistic.

5 Partitioning is only possible with a sparsely specified W (as used in this study) because such matrix gives rise to higher neighbour effect (i.e. second-order neighbour effects, third-order neighbour effects, etc.). With a densely specified W, such as an inverse distance matrix without cut-off (i.e. inverse distance between all regions) there are no higher order neighbour effects. This is due to the fact that all regions are neighbours of one another so there are only first-order neighbour effects.

6 All calculations and empirical results are computed using R (R Core Team, 2016). In particular, spatial regression results were obtained using the spdep package (Bivand & Piras, 2015; Bivand et al., 2013).

7 Exception is the variable 'youth' with a negative indirect effect and a positive but not, however, statistically significant direct effect

8 Another alternative interpretation is that a global rise in the percentage of people with tertiary education by one percentage point would decrease each region's unemployment rate by $-1.26$ percentage points. These percentage points collapse to a direct effect of $-0.36$, which is due to the rise in particular region, and to an indirect effect of $-0.9$ percentage point, which is due to the rise in all other regions.