In This Module We Want to Introduce the Concept of Vulnerability to Poverty, and to Distinguish

Slide #1

In this module we want to introduce the concept of vulnerability to poverty, and to distinguish it from measuring poverty itself.

Slide #2

More concretely, there are seven distinct objectives for this module. First we need to explain why current measures of poverty are sometimes inadequate for our purposes. But we have an approach that may work better, and that is the concept of vulnerability to poverty. It turns out that it is possible to measure vulnerability to poverty if we have a measure of a household’s expected consumption, the variance of its expected consumption, a poverty line, and if we assume some type of distribution, for example an Normal distribution, that applies to expected consumption.

We then need to outline the steps that are actually required to measure vulnerability in practice, given that we have data from a household living standards survey. Then there are some methodological issues that require our attention. After discussing them, we itemize the main sources of risk that are faced by households, and finally we describe the nature and purpose and contents of Risk and Vulnerability Assessments, which the World Bank favors.

Slide #3

Logically, the first question we need to address is why we need a measure of vulnerability at all. And the starting point for this discussion is the observation that poverty rates may not always be as helpful as we think. When we measure a poverty rate we are looking into the past, so it is an ex post measure. But when we are designing policy, we need to be thinking about the future. We are interested in identifying who will be poor tomorrow, so that we may help them. The practical problem is this: just because you were poor in the past does not necessarily mean that you will be poor in the future.

A good way to see this is in the table that we show here. Take, for example, the case of Ethiopia. Based on surveys that were done in 1994 and 1997, we see that 25% of the population was poor in both years, but 30% of the population was poor in one of the two years, but not in the other. So, many poor people must have moved out of poverty during this period, and others must have gone into poverty. That means that if you were making policy in the mid-1990s, just knowing who was poor in 1994 would not be a good guide to who was poor in 1997. These transitions into and out of poverty make it difficult to target the future poor. Most broadly, a household is vulnerable to poverty if it is likely to be poor in the future.

Slide #4

Actually, we need to be a bit more precise about how we define vulnerability. A serviceable definition is “the propensity to suffer a significant welfare shock, bringing the household below a socially-defined minimum level.” Let’s parse this more carefully. What sort of shock are we talking about? Most usually it would be a shock to consumption per capita, although this does generally suppose that households are able to smooth shocks to their income, at least to some extent. The “socially-defined minimum level” is usually taken to be the poverty line. It is less obvious how best to measure the propensity to fall into poverty. The commonest approach is to ask, “What is the probability that you will be poor next year?” We have written this out algebraically: the vulnerability of household h in year t is the probability that its consumption in year t+1 will be at or below the poverty line. An alternative would be to measure the probability that you will fall below the poverty line within the next few years.

Slide #5

Everybody has some probability that they may be poor next year – meteors could strike, there could be war – but we are of course most interested in identifying those for whom there is a high probability of poverty. We consider a household to be highly vulnerable to poverty if there is at least a 50% chance that it will be poor next year. The group that is not considered to be vulnerable consists of those folks for whom the probability of being poor next year is less than the poverty rate. For example, if 10% of the population is poor now, anybody who has a less than 10% chance of being poor next year is considered to be “not vulnerable”. This does not mean that you certainly won’t be in poverty – you might be unlucky – but it certainly means that the likelihood that you will be poor is less than that of the average person. In between, there is a group that has a higher-than-average chance of being poor, but it is not as high as 50%, and that group is generally referred to as “vulnerable”.

Slide #6

The central challenge we face is one of measuring vulnerability in a way that is actually useful. Theoretically, vulnerability, which is a probability, is based on an enormous amount of information. We would want to know, for each household, what are its assets, its educational level, its skills, what risks it faces – such as drought, highly food prices, illness – and what is its ability to handle those risks: Does it have access to loans? Could family members work harder? Does the family have savings that it could dip into? And ideally we would have all this information before we could determine whether a household is vulnerable to poverty.

In practice we rarely have enough information, so we have to simplify. There are the are in fact four bits of information where that we need in order to come up with a serviceable measure of vulnerability. We need to know what a household’s expected consumption is for next year; then we need to know the variance of that expected consumption; third, all we need a poverty line so that we can judge the probability that it would be poor; and finally we need to make an assumption about the way in which expected consumption next year is distributed. We generally assume that it follows a Normal distribution.

Slide #7

Once we have these basic bits of information, the actual computation of vulnerability is relatively straightforward. So let’s take a couple of examples. Suppose we expect that your consumption per capita next year will be 55, and the standard deviation of expected consumption will be 13. Assume a poverty line of 40, and a normal distribution. The graph at the bottom here shows the situation. You can see that expected consumption is 55, because that’s where the distribution is centered. We are assuming a bell-shaped distribution, and have marked the poverty line at 40. What we are now trying to measure is the probability that you will find yourself below the poverty line, and that is given by the shaded area relative to the total area under the curve. We can actually calculate this in Excel, using the norm.dist function as shown here. In this example there is a 12.4% probability that you will be poor next year; this is a measure of your vulnerability.

The second example is similar, but here are your expected consumption is only 47.5, and the standard deviation of that is 6.5. In this case the probability that you will be poor is also 12.4%! We can see that vulnerability is fundamentally driven by two things: first, where your expected consumption is relative to the poverty line; and second, the variability of your expected consumption.

Slide #8

We are not quite out of the woods, because we have not yet measured expected consumption in a real situation, nor its variance. But there is a reasonably robust approach that Chaudhuri and others have developed that allows us to measure vulnerability relatively easily. First we need to estimate an econometric model of the determinants of consumption; in module 4 we discuss estimation issues in more detail. In the example shown here, the log of consumption might depend on a number of household and other characteristics, represented here by X. Of course the X variables do not perfectly predict consumption, so we are left with an error term. This is the eh term, and this error will be different from household to household. It is a larger number if your income is very variable, and small if your income is stable, and we need to estimate it as well.

So in the second step we take the residuals from the estimated consumption equation, square them so they look like variances, and then regress them on the same set of X variables. When we do this we get an estimate of theta hat, and this allows us to predict the variance for each household. So the regression estimates from steps 1 and 2 give us both expected consumption, and the expected variance (and hence standard deviation) of that consumption, for each household. We plug these numbers into Excel, or use the formula shown here, to get a measure of the vulnerability of poverty for the each household.

Slide #9

The study by Chaudhuri and his co-workers used information from Indonesia to illustrate this procedure, and we present the key results in this slide. In 1998-99, shortly after the onset of the Asian Financial Crisis that began in 1997, the estimated poverty rate in Indonesia stood at 22%. But the interesting finding here is that 45% of the population was vulnerable, in the sense of having a greater than average probability of being poor in the year ahead. That’s an important finding, because if a lot of people are vulnerable, they may rightly be worried about the possibility of falling into poverty, and this may drive them to turn to the government for social protection, and for help in the event that things go wrong.

There is another implication, which is that it makes targeting the future poor much harder, because more than half of the vulnerable population in Indonesia in 1998-99 were not actually poor at that time. We also note that some of the poor were not considered vulnerable! This may sound a little strange, but these were the folks who have fundamentally good prospects, but just temporarily fell on very hard times. The numbers in the table here, which come from the study by Chaudhuri, Jalan, and Suryahadi, show that 8% of the Indonesian population was highly vulnerable, meaning they had at least a 50% chance of being poor, and a further 37% had what might be termed low vulnerability; the remaining 55% were not vulnerable.

Slide #10

As with any technique in the social sciences, there are some methodological issues, and these need to be mentioned. One of the more serious problems is that the measurement of income or consumption, based on household survey data, is subject to error. The difficulty with this is that it adds noise to our measures of vulnerability. Lant Pritchett and others argue that perhaps as much as a third of the variation in consumption that we observe is not real, and is just noise based on measurement error. That would imply that people’s incomes are not as variable as they look, and that fewer people move into and out of poverty from one year to the next than we may think. This matters. The more serious the problem of measurement error, the less useful our measures of vulnerability will be, because we can fall back on our measure of poverty in the past as a good predictor of poverty in the future.

A second issue, and not a trivial one, is that as soon as a household has a new baby, consumption per capita immediately falls sharply. But it is hard to argue that this means that the family has suddenly become worse off. Kamkanou and Morduch claim, based on their work in the Côte d’Ivoire, that a quarter of the variability in per capita consumption is due to demographic changes such as this. The third difficulty is the challenge of measuring the standard deviation of consumption; Chaudhuri uses cross-sectional data, but that may not be very satisfactory. We’d really like to have longitudinal data, meaning data collected from the same households over a number of years, but such data are extremely rare in developing countries. Finally, we have assumed that the log of future consumption is Normally distributed; this is an approximation, and perhaps we should be using the empirical distribution instead.

Slide #11

This slide simply presents a slightly simplified version of a table from a review article by Stefan Dercon, who has done a lot of thinking about issues related to vulnerability. As we have seen, it is quite difficult to model vulnerability in practice. We can make a list – as in this table – of the factors that influence vulnerability, such as your assets, your income, and your capabilities. There are also some nice examples of the sort of risks that you face – the risk of losing a job, or facing lower output prices, or facing higher food prices, for instance. But actually quantifying these into a model that can be measured is not at all easy, which is of course why we had to take some short-cuts in a simplified approach to measuring vulnerability, as we saw in the earlier slides.

Slide #12

The World Bank favors the creation and use of risk and vulnerability assessments, seeing these as diagnostic tools that help one understand the sources of vulnerability to poverty. These assessments identify the commonest and most severe shocks, and seek to determine which groups are most at risk from these shocks. A good assessment will also catalog the public interventions that are related to risk – for instance, the provision of food for work, or cash grants for the poor – and try to identify what else might be tried. In the jargon of economic development, this is sometimes referred to as looking for the “policy gap”.

Vulnerable households can try to reduce the risks they face, for instance by having some family members migrate to the cities to work, or building irrigation schemes. They may also mitigate their risk, by diversifying their sources of income, or building up savings and other assets. This then allows them to cope better with shocks, by selling assets. The point here is that public policy may sometimes be able to strengthen these mechanisms, although as we will see in module 8, it is not always easy to find interventions that work well.