Developing & testing poverty assessment tools : results from accuracy tests in Uganda
November 2005
USAID
USAID commissioned the IRIS Center to develop, test, and disseminate poverty assessment tools which meet U.S. Congressional requirements for accuracy and cost of implementation. Accuracy tests of poverty indicators have been implemented by IRIS in Bangladesh, Peru, Uganda, and Kazakhstan. Comprehensive information on the project is available at www.povertytools.org, and will not be summarized in this report.
This report presents the results of poverty assessment tool accuracy tests conducted in Uganda in 2004. 1 Chapter 1 (this Introduction) provides an overview of the design of the field research for the accuracy test, and the computation of the applicable poverty line. Chapter 2 provides an overview of the analysis. In Chapter 3, we present the results on selected poverty indicators from eight regression models (known as “Ordinary Least Squares,” or OLS, regression models). Each of these models potentially represents a newly designed poverty assessment tool, calibrated for Uganda. The regression models are run in the econometric package SAS, using the maximum R2 improvement technique (MAXR) procedure that seeks to maximize the explained variance of the dependent variable (per capita daily expenditure) by a set of the best 5, 10, and 15 regressors (referred to in this report as BEST5, BEST10, and BEST15). Any set of five,
ten, or fifteen poverty indicators can be considered a poverty assessment tool for purposes of identifying the poverty status of a household.
The first five regression models differ with respect to the set of poverty indicators allowed in the model, starting from a model with a full set of potential regressors and gradually restricting the set of regressors on the basis of implementation practicality. A sixth model is run as an example of a tool that considers only those poverty indicators that were rated as “highly verifiable” by Nkoola Institutional Development Associates (NIDA), the survey firm in Uganda.2 A subsequent model compiles these indicators with five powerful subjective and two monetary indicators. Finally, the last model makes use of poverty indicators usually available in the World Bank’s Living Standards Measurement (LSMS) surveys. Thus, the first seven models can be considered alternative best combinations of poverty indicators which were mainly derived from existing practitioner tools for poverty assessment, while the last model is a tool derived from
poverty indicators usually available in LSMS surveys.
Chapter 4 presents results from an alternative estimation approach, the so-called “two-step” models. In addition to Ordinary Least Squares (OLS), we also test Quantile, Probit, and the so-called Linear
Probability regression technique. Compared to the models presented in Chapter 3, the performance of models presented in Chapter 4 is overall much better. Finally, Chapter 5 summarizes the results.
Footnotes:
-
This report consists of original work and data analysis. Citations of entire paragraphs or tables in published material by other authors is only permitted after prior consent with the authors and the IRIS Center. The cleaning and processing of data, as well as the entire analysis presented in this report, was carried out at the Institute of Rural Development, Georg-August-University of Göttingen, Germany. We gratefully acknowledge the valuable comments and support given by the IRIS project members Thierry van Bastelaer, Tresja Denysenko, Kate Druschel, and Anthony Leegwater; by Advisory Panel members Lauren Hendricks (CARE), Jonathan Murdoch (Princeton University), and Laura Foose (SEEP, PAWG); and by Stefan Schwarze, Isabelle Jaisli, Marinella Fader and Norbert Binternagel of the Institute of Rural Development at the University of Göttingen. The input by the SEEP Network and its Poverty Assessment Working Group (PAWG), the Advisory Panel for the Developing Poverty Assessment Tools project, and USAID is gratefully acknowledged. In particular, Christian Grootaert provided valuable comments and advice during all phases of the field research and data analysis, especially with regard to the choice of regression technique and the definition of accuracy measures. We gratefully acknowledge the excellent cooperation with Charity Irungu,
the IRIS consultant for the field survey in Uganda, and the staff of the survey firm NIDA in Kampala, Uganda. All remaining errors are ours.
-
Contact Information: Nkoola Institutional Development Associates (NIDA), 209 Upper Mawanda Rd., Old Mulango Hill, P. O. Box 22130, Kampala, Uganda. Tel./ Fax: 256 (0)41-530696, Email: .
|
