Developing and Testing Poverty Assessment Tools: Results From Accuracy Tests in Kazakhstan

This paper presents the results of the accuracy tests of poverty assessment tools in Kazakhstan. The paper discusses:

• The design for the field research of the accuracy test and the computation of the applicable poverty line.
• The results on selected poverty indicators from 9 regression models. Facts about the models:
  • Each of them potentially represents a newly designed poverty assessment tool;
  • The first 6 models differ with respect to the set of poverty indicators;
  • The seventh model is an example of a tool that considers only those poverty indicators that a survey firm in Kazakhstan rated highly verifiable;
  • The eighth model compiles these indicators with subjective as well as monetary indicators;
  • The last model uses poverty indicators usually available in Living Standards Measurement surveys.
• The results from an alternative estimation approach, the two-step model.

The paper finds that:

• All nine models show total accuracy in predicting the percentage of households that actually fall into a given category. However, they show low accuracy among the very-poor and underestimate the poverty headcount;
• These single-step models are ill-suited for use in Kazakhstan;
• Model 7 is the most practical model;
• The inaccuracies in prediction are higher for the very-poor than for the not-very-poor. This problem of unbalanced accuracies can be solved by the use of the two-step models, which is more accurate.