Construction of a composite indicator on multidimensional poverty in Kenya using multiple correspondence analyses (MCA) : An application on the KDHS 2008 Data
Abstract
This study is on the construction of a composite indicator, which can be used to measure
poverty more effectively and precisely on the basis of Demographic and Health survey data. This paper looks at the history of measuring poverty in Kenya, some basic indicators and recommends the use of Multiple Correspondence Analysis in computing a poverty scores using the KDHS 2008 data.
Poverty is a phenomenon which cannot be described simply using a causal relationship,
basically poverty is seen as a state of lack of adequate resources which in turn reduce or hinder the social welfare of a household. Due to the fact that poverty manifests itself differently in different areas, the notion that poverty should be measured on the basis of a large number of variables has enjoyed an increasing support in the recent years.
As observed in the work of Achia et el, (Achia, 2010), PCA can be used in composing a
composite index for determining asset index, In the PCA technique, the distance between
population units in the population space are based on usual Euclidean Metric, this
representation is build by a step-by-step process, on a successive orthoganal projection on subspace (otherwise know as rotations) on subspace with increasing dimension
Multiple correspondence analysis (also Homogeneity Analysis) can be said to be a "special case of generalized canonical analysis" (Asselin, 2009), the beauty with this technique is that it can be applied to categorical or nominal data, where each of the variables Zj is assumed to have Kjdistinct categories.
The main technical difference between PCA and MCA is the use of the X2 metric (chi-square) in MCA, instead of the usual Euclidean metric used in PCA, to measure distances between two lines or two columns of the data matrix being analyzed.
The main strength of Multiple Correspondence Analysis (MCA) is that it can deals with
categorical data, unlike other techniques, it makes it easy for a wide range of variables to be included in the analysis, for variables which are quantitative, they should be transformed to qualitative through recoding. The CPI constructed from Multiple Correspondence Analysis is a better measure, once
compared to other commonly used indices; wealth index and the principle component analysis, results from this study show that it gives a better indication. Another important advantage of the CPI is its ability to use categorical variables which are the majority in asset based information in households
Publisher
School of Mathematics