Some Unpleasant Poverty Determination Procedures: Different Researchers, Different Poverty Rates

There is a lot of interest in measuring poverty, comparing it across regions and countries and analyzing its evolution over time. This interest has spurred a large literature on poverty, especially in developing countries where poverty is considered as a primary indicator of development. A “cookbook” methodology for poverty determination was progressively developed by the World Bank. It is widely assumed that it produces poverty measures which depend only on the survey data, so that different researchers working on the same household survey and following the methodology reach identical poverty rates. The paper shows that this understanding is incorrect. It shows that different practitioners can derive different poverty rates while working on the exact same survey. The paper formalizes the existing methodology and illustrates its shortcomings. It proposes a number of improvements on the methodology. These improvements which include the use of more efficient kernels with bandwidths derived from the data seek to make it fully automatic. More efficient and less ad-hoc estimates of food and non-food expenditures are then obtained. In the end, the researcher is left to smooth the data “by eye” by setting one single parameter. These improvements produce a fully automatic procedure which ensures that different researchers working on the same household survey derive identical poverty rates. The paper applies the methodologies to the 1995/96 Nepal Living Standard Survey.