Note: I would like to thank the Norwegian Agency for Development Cooperation (NORAD) for financial support and the Uganda Bureau of Statistics (UBoS) for providing the data. I am in debt to John K. Dagsvik for advice and suggestions. I thank Bjørn Wold and Geir Øvensen for useful discussions and comments. I would also like to thank two referees of this journal for very helpful reviews.
Testing Prediction Performance of Poverty Models: Empirical Evidence from Uganda
Article first published online: 11 DEC 2012
© 2012 International Association for Research in Income and Wealth
Review of Income and Wealth
Volume 59, Issue 1, pages 91–112, March 2013
How to Cite
Mathiassen, A. (2013), Testing Prediction Performance of Poverty Models: Empirical Evidence from Uganda. Review of Income and Wealth, 59: 91–112. doi: 10.1111/roiw.12007
- Issue published online: 3 FEB 2013
- Article first published online: 11 DEC 2012
- Norwegian Agency for Development Cooperation (NORAD)
- household survey;
- money metric poverty;
- poverty model;
- poverty prediction;
This paper examines the performance of a method of predicting poverty rates. Because most developing countries cannot justify the expense of frequent household budget surveys, additional low-cost methods have been developed and used. The prediction method is based on a model linking the proportion of poor households to suitable explanatory variables (consumption proxies). These consumption proxies are variables that can be collected at much lower cost through smaller annual surveys. Several applications have shown that such models can produce poverty estimates with confidence intervals of a similar magnitude to the poverty estimates from the household budget surveys. There is, however, limited evidence of how well the methods perform out-of-sample. A series of seven household budget surveys conducted in Uganda in the period 1993–2005 allows us to test the prediction performance of the model. We test the poverty models by using data from one survey to predict the proportion of poor households in other surveys, and vice versa. The results are encouraging, as all models predict similar poverty trends. Although in most cases the predictions are precise, sometimes they differ significantly from the poverty level estimated from the survey directly.