Since the late nineties, the discussion on the concept and measurement of poverty has centred on
the Oxford Poverty and Human Development Initiative (OPHI) and its proposal known as
Multidimensional Poverty Index (MPI). The MPI displays some advantages over other indices,
like being more reliable than the unidimensional income/consumption poverty or more applicable
for poverty alleviation programs. However, the MPI also displays some contradictions such as
the approaches and techniques used to weight and to aggregate the multivariate indicators. In
order to obtain the vector of weights and aggregate MPI components, there had been analysed
some multivariate approaches that fluctuate between two extremes: the complete arbitrariness and
the data-driven principle.
On one side, the complete arbitrariness means that all calculations must be taken as given.
This approach has used by OPHI to produce annual and comparative reports across more than
one hundred
1 countries. However, it can be said that the expert criteria reduces some of the
arbitrariness in weighting the indicators because those weights are based on recommendations of
“poverty-experts”. The standard MPI consists of 3 components (M0, H and A), calculated based
on 4 set of parameters: dimensions (3) which are equally weighted with 1/3 each one; indicators
1 The last report was in 2014 over 108 countries.2
(10) that are equally weighted within each dimension; weights and cut-offs (Alkire & Santos,
2010).
On the opposite side, the data-driven approach means leaving data to speak for themselves, as
they reveal the preferences of poor people. This approach could be used to select not only the
weights of indicators, like here we do, but also to choose the set of variables and their cut-offs
previous to the aggregation. The focus of this research is to contribute to the data driven approach
through the analysis of different choices of weights to aggregate alternative MPI measurements.
Those vectors of weights are based on the multivariate technique called Principal Components
Analysis (PCA) and applied on Timor-Leste data.
Although, the technical advantages in the use of PCA, and other multivariate techniques as
well, there are relatively few studies focused on this discussion and applications in economics.
From an econometric perspective, this study encourages a review of measurement problems,
survey issues, multivariate methods and specialized software.
Next sections are developed as follows. Section 2 presents the literature review consisting in
3 sub-sections, starting with a discussion on poverty index, and then describing the
multidimensional poverty components of MPI, lastly the description of PCA technique with an
alternative variation called polychoric PCA. Section 3 briefly describes the data used in this study
which come from “The 2009-10 Timor-Leste Demographic and Health Survey (TLDHS)”.
Section 4 analyses the main results in two parts; first the outcomes replication of the standard MPI
exercise, (fixing those as the initial values); second the resolution of both PCA (regular and
polychoric), with the calculation of alternative MPI and comparative analyses with the initial
values plus the disaggregation by districts. Section 5 rehearses some possible extensions and
limitations of this paper. Finally, section 6 summarizes the main results and findings.