Poverty Targeting in Public Programs - A Comparison of Some Nonparametric Tests and their Application to Indian Microfinance

This paper explores statistical methods that are appropriate to test hypotheses about the relative exclusion of households in the bottom tail of an income or asset distribution. The paper states that:

• If a program is designed to cater to the poor, but not to the destitute, the population distribution of participants would cross that of non-participants from below;
• Standard tests of first order stochastic dominance would not be applicable in such cases.

The authors perform numerical simulations with alternative pairs of distributions and find that:

• Non-parametric tests that explicitly test for distribution crossing are powerful, even when the two distributions under consideration are fairly similar;
• A combination of tests for distribution crossing and quantile tests can be useful in examining the pattern of household participation in a social program.

The paper:

• Applies its methods to test for poverty targeting in a rapidly growing microfinance program in India;
• Uses data that contains annual household expenditures under several heads such as food, clothing, education, health and entertainment;
• Obtains data from a household survey administered to a random sample of members of newly-formed microcredit groups and a randomly selected sample of non-members.

The authors find that:

• The distribution of members cross that of non-members from below;
• There is a preponderance of non-members among the poorest quarter of the population;
• In this range, expenditure quantiles for the member distribution appears to be higher than those for the population as a whole.