This paper presents new evidence about the association between gender and poverty. Using stochastic dominance analysis to examine data from 10 developing countries in Africa, Asia, and Central America, the study attempts to more accurately understand information that has been misrepresented by other types of analysis.

The Study
The study used household survey data collected by the IFPRI, its affiliates, and the World Bank. The IFPRI studies investigated patterns and determinants of food security in Botswana, Ethiopia, Madagascar, Rwanda, Bangladesh, Indonesia, Nepal, and Honduras from 1985 to 1993. Data sets from the Côte d'Ivoire and Ghana, both urban and rural, were from the Living Standards Measurement Survey conducted by the World Bank from 1986 to 1988. Most of the IFPRI sets included at least two rounds of data collection.

The study used these data to calculate poverty indices for male and female-headed households. Since the study focused on the very poorest, it specified an endogenous bound of the bottom third of the income distribution for each area as the poverty line. It applied first- and second-order stochastic dominance criteria to the poverty curves for male and female-headed households, testing for varia- bles including (1) per capita expenditure, (2) per adult equivalent income, (3) the poverty gap index, (4) depth of poverty, and (5) relative poverty of individuals in each household. The use of stochastic dominance methods enabled the au- thors to rank distributions in a way that would not have been possible simply from the comparison of mean and variances, and it allowed poverty comparisons to be made without prior specification of a poverty line.

The study found weak evidence that female-headed households are overrepresented among the poor.

Findings
The study found weak evidence that female-headed households are overrepresented among the poor. In most of the data sets it was actually difficult to determine whether poverty is greater among males or females. Only in two countries, Bangladesh and Ghana, did the data consistently indicate higher levels of poverty among women. In these countries the data suggested that cultural and institutional factors led to this disparity.

Before applying stochastic dominance analysis, per capita expenditure measures revealed more women than men below the 33 percent poverty line in 7 out of the 11 data sets. Again in 7 out the 11 data sets, the poverty gap was larger for female-headed households than for male-headed households. Per adult equivalent measures showed that in 8 out of 11 sets, more female-headed households were poorer than male-headed households. In less than half of these comparisons, however, were the data statistically significant.

Yet stochastic dominance analysis revealed a different picture. In most of the cases, the differences between male and female poverty levels were not statistically significant enough to conclude that one group was better-off than the other. In first degree stochastic dominance analysis of per capita measures and adult equivalent measures, 10 out of 11 data sets revealed no dominance by either male- or female-headed households. Second degree stochastic dominance analysis showed that while male-headed households dominated in most cases, the curves were not sufficiently different to conclude that there was significant difference between male and female poverty levels.

Policy Implications
The general lack of robust evidence indicating greater poverty among women in developing countries suggests several implications for future research and policy intervention.

1. The diversity among male and female groups in this study was not considered. Future studies should use multivariate analysis to examine multiple determinants of household income and expenditure, factors that will permit a better understanding of poverty levels.
2. Other variables in addition to income should be studied and incorporated into policy models directed at alleviating poverty. A problem too complex to attribute solely to gender, poverty must be examined through factors such as nutrition, health, and time and workload allocation.
3. Self-reporting of who heads the household is misleading, causing inaccurate definitions of male- and female-headed households. In order to better define which households are truly female-headed, future studies need to analyze the endogenous processes that lead to female headship.
4. Women in developing countries generally have lower levels of education, resources, and social support than men. Despite an apparent lack of significant gender poverty differences in this and other studies, therefore, policy intervention should continue to consider gender when formulating models.