Analyzing Vulnerability

Comparisons of vulnerability across groups

With panel data, poverty profiles can also prove a powerful tool to reveal differences in poverty dynamics between various household groups. For example, one may analyze the movements in and out of poverty of population groups defined according to various characteristics such as demographics and place of residence. This approach answers such questions as 'are female-headed households more likely to remain poor?'. Are households in specific regions more likely to escape poverty?

In the case of China for instance, most of those who experienced poverty in Guangdong province were transiently poor while a larger share were persistently poor in Guizhou. Such differences suggest different underlying characteristics of poverty, and thereby different policy responses [Jalan and Ravallion 1999].

In the same way that a static poverty profile can be presented in two different ways, as seen earlier, when long observation periods are available, one may compare the characteristics of the “vulnerable”, “very vulnerable”, and “non vulnerable”, and how these change over time.

Determinants of vulnerability

In the same way that regressions can be used to assess the determinants of poverty at any given point in time, regressions can also be used to assess the determinants of changes in income or poverty over time. Again, the advantage of panel data are that they go beyond finding the static correlates of poverty to identify the determinants of income or spending changes over time. Some of the problems of mutual causality with cross-sectional data do not arise in this case, since the initial conditions of households cannot be caused by the changes in household welfare. There are different ways to address the issue:

• First, when data is observed for two periods, one can run regression of income or consumption in the second period on household and individual characteristics in the first period. Doing so allows to estimate households’ ex-ante distribution of future consumption or income, and therefore to estimate the probability of each household to fall in poverty in the future.
• An alternative would be to relate change in household welfare over time to exogenous variables and to ‘initial starting conditions’ of the household. This approach does not capture vulnerability in the sense used above (that of probability of falling into poverty) but focuses on explaining absolute changes in consumption (In order to focus on vulnerability, one could carry out the regression only with those households who fell into poverty in the second period of observation).
• Regressions could also be run to explain entry and exit rates and the duration of poverty. This usually requires long panels which are not as common in low-income countries. These regressions can help explain the triggers that cause households to fall into poverty, such as death of a family member, illness, or unemployment, and the triggers that pull them out of poverty. They also allow the analyst to test the impact of potential alternative policies, for example, social protection interventions, on the probability of exit from, and entry into, poverty.
• Finally, the analyst can carry out regressions of low-vulnerability (in the sense of transient poverty) and high vulnerability (in the sense of chronic poverty). For instance, using data for rural China, Jalan and Ravallion (1998, 1999) suggests that both ‘’acute vulnerability” (or “chronic poverty”) and “vulnerability” (or “transient poverty”) are reduced by greater command over physical capital, such as wealth and land, and certain demographic characteristics. These are, however, the only similarities. Smaller and better-educated households, and those who live in areas with better attainments in health and education, have lower chronic poverty, but these factors have little influence on transient poverty. Thus, interventions aimed at reducing chronic poverty may have little impact on transient poverty.