Measuring Poverty as a Fuzzy and Multidimensional Concept: Theory and Evidence from the United Kingdom

摘要

Previous research shows that poor people define poverty not only in material terms, but also in psychological and social terms, though it has been consistently characterized by economic resources in social sciences. Using a method based on `fuzzy-set' theory can be uniquely placed to answer the question as it allows us not only to tackle the problem of arbitrary poverty line, but also to integrate multiple dimensions into one index in an intuitive way. It can avoid the problem of poverty line entirely by introducing the concept of `membership function' which represents a degree of inclusion in a fuzzy subgroup poor.ududI therefore argue that the fuzzy measures of poverty can be a strong multidimensional alternative for the measures centered around income. To support the argument, two crucialudpoints are clarfied. Firstly, the difference between traditional measures and the fuzzy measures needs to be discussed further since the discussions on the new measures so far lean more toward the fresh insights from the measures, so that the distinction in policy-relevant information has not been emphasized enough. From the comparison, I present that the fuzzy measures can provide a richer description of the social phenomenon, enabling audmore acceptable distinction between different sub-populations. Secondly, how the measures behave statistically should be considered in depth because one of the most frequent critiques for poverty measurements is that present methods depend too much on arbitrary decisionsudlike setting a poverty line. Utilizing a Monte Carlo simulation, I find that the measures (Totally Fuzzy, Totally Fuzzy and Relative, and Integrated Fuzzy and Relative) acknowledge two points quite well: (i) poverty is a multidimensional concept, and (ii) the `poor' and `non-poor' are not two mutually exclusive sets and the distinction can be `fuzzy'. It also turns out that the sampling distribution of the fuzzy measures is well-behaved, and they are robust to arbitrary choice in the estimation as well as reliable with relatively small sample size. Besides, I show that they are robust to measurement errors. Finally, I investigate the identification performance of each measure and show that the measures have a strong consistency.