Multidimensional Child Poverty: How different approaches compare for measurement and applied poverty reduction interventions under the Sustainable Development Goals

Calderon, Maria Cecilia (1); Evans, Martin (2) (2016). 'Multidimensional Child Poverty: How different approaches compare for measurement and applied poverty reduction interventions under the Sustainable Development Goals' Paper presented at the annual conference of the HDCA, Tokyo 2016.

Many developing countries are adopting multidimensional poverty analysis to capture the sustainable development goal (SDG) 1 and to measure progress against the goal. The SDG poverty goal additionally stipulates disaggregation of poverty targets for children. This paper empirically considers two approaches to measuring multidimensional poverty for children to fulfil SDG purposes. Our analysis outlines two methodologies: (i) the United Nations Development Programme (UNDP)/Human Development Report Office (HDRO) Multidimensional Poverty Index (MPI), and (ii) the UNICEF’s Office of Research Multiple Overlapping Deprivation Analysis (MODA). In many countries, national governments and statistical offices are adopting the MPI methodology for multidimensional poverty measurement of the whole population, often sponsored through national UNDP offices. UNICEF’s own methodology, MODA, is different in many ways and this leads to a real applied policy problem in an increasing number of developing countries, “how do we approach child poverty when MPI and child-level multidimensional poverty measurement are so different?”
This paper begins by empirically comparing the poverty results from these two different approaches for two separate age groups of children: aged 0-4 and 5-17 year olds. Furthermore, their treatment of the selected indicators and their reporting of child deprivations are analysed. Our paper proceeds as follows. First, we outline the methodologies in summary form for both MPI and MODA and provide MODA and MPI results for two developing countries, Cambodia and Tanzania, using the 2010 Demographic and Health Survey microdata.
Second, we explore two key questions on the resulting profiles. The first question addresses a commonly held preconception of the difference between MPI and direct child level multidimensional poverty measures – that the household level MPI will collect less or worse information on children. Given that MPI measures at the household level and MODA measures at the individual child level, what differences in multidimensional child poverty profiles result and how are they attributable to individual level (child and/or other household members) or household level factors?  
Our second question addresses the less transparent differences in methodology between the two approaches. MPI uses a score calculated on the aggregation of deprivation indicators based on equal dimensional weighting and then uses a dual cut-off approach to set thresholds and to allow assessment of severity and decomposition by sub-groups, indicators and dimensions. MODA uses a counting of dimensions with each dimension being assessed using the union approach for its component indicators. Similarly to the MPI, it then assesses severity and decomposition by dimension and by population sub-groups. This latter use of a shared Alkire-Foster (2010) approach to poverty decomposition and consistent comparison of sub-group poverty scores means that MODA is sometimes considered as a child MPI. However, the differences in aggregation and weighting assumptions that precede the common approach to decomposition and comparison of sub-groups need to be more rigorously assessed. We consider a set of robustness characteristics and sensitivities that are apparent in consistent scaling and distributions created by the indices and in the treatment of common deprivation components by MPI and MODA. We only consider children and child poverty/multiple deprivation and, thus, are able to more accurately assess the nature of a child MPI and MODA and see their similarities and differences.
The findings from this analysis are then used to discuss optimal approaches to disaggregation of MPI for children verses MODA, a child specific measure, at least a priori. Finally, the paper uses a range of simulated sensitivity tests to show how each methodology performs to changes in the underlying deprivation profile. We simulate different MODA and MPI poverty headcounts based on ‘all children’ and ‘no children’ being deprived in sanitation and compare these extremes to the original baseline poverty headcount. Additional exercises test for the sensitivities of both indices to school attendance and for ‘item level’ sensitivity to households possessing a cellular phone.
The overall analysis included in this paper allows for a further discussion of the properties of each methodology to capture poverty reduction in the SDG context. The paper concludes with an applied discussion of findings and their relevance for the SDG multidimensional poverty measurement for children. The key conclusion is that it is probably not safe to assume that MODA captures more poverty based purely on child level indicators than MPI. Even when considered on their own, the MODA decomposition clearly shows domination by household level factors, even if the size of the exact difference between MODA and MPI is uncertain in precise terms.

scroll to top