Suppose you have come up with a way of identifying the poor people in a poor country (where income is hard to verify) in transparent way (one reason for doing this would be for social assistance targeting). How would you go about doing this? Here are a couple of options, as I see them:
1. Suppose you have household survey data that contains variables for easily observable characteristics, such as the household’s house, its assets, characteristics of the household (such as education of the household head, location), etc. as well as actual consumption (adjusted for regional price differences). You could try Proxy Means Testing: that is you regress actual consumption on a set of variables (e.g. gender of household head; age of household head; age of household head squared; presence of car, air conditioner, etc.) to arrive at a set of betas, or weights, which you would then use to predict consumption for out-of-sample households. Those with predicted consumption below a certain cutoff or threshold can be defined as poor. For this example let us define as poor those in the bottom quintile of predicted consumption.
One problem with this approach is that if the regression does not explain a lot of the variation in the data then many errors of inclusion (noon-poor being deemed poor) and errors of exclusion (poor being deemed non-poor) will result. Ideally, your PMT model would be limited to 10-14 variables, and this would limit the explanatory power of the model; the consequence is that the inclusion and exclusion errors would be significant. One mitigating fact, though, is that most of the errors of inclusion consist of those in the second lowest quintile of consumption. In a very poor country, these people are still quite poor.
2. One idea suggested to me was to use open public meetings to determine who the poorest people are. The idea is that you’d gather some people in the village, and ask them: who are the poorest 20 percent in this village? The extent to which this would work might depend on the power dynamics in the village. The advantage is that it could allow local knowledge to inform the determination of who is in the poorest quintile.
Some modifications to this approach are worth considering. For example, one could approach several people in the village and ask them to pick out those they think are in the poorest quintile. If there is significant overlap in the selection, it would be reasonable to conclude that the such overlap constitutes the poorest people in the village.
I went to a small farming village in a dirt-poor Central Asian country a few days ago. I spoke with a couple of farmers, and asked them about how in general to identify the poorest people in the village. It seems that the answer is that ‘you just know.’ One farmer took me to see a lady who, in his opinion, is the poorest person in his village.
He was right – you just know. This lady’s husband left her to become a non-remitting labor migrant. She has a disabled daughter, and at least one young son. She has a house (everyone has property) with a decent sized yard, but the kitchen and other rooms are tell the story about the well-being of this family. To get by, this lady gathers stalks, makes brooms out of them, and sells them for $0.75 each.
While there I saw that others in her village grow crops in their yards right outside their houses (Unlike the farmers, they are not forced to allocate 60 percent of their land to growing cotton). I was wondering if there is potential for this lady to do so as well. I could provide a small loan to pay for inputs (labor, seeds, fertilizer, etc), the farmer can provide technical advice. I can see how small social enterprises like this get started. My only concern is that the funds might be diverted other needs (such buying medicine and food, which is understandable) rather than being spent solely for the purpose of farming, which would not bring in revenue for some time. I’ll have to make sure I have a plan for this for my next trip out there.