Executive summary

Annex 3.2: Identifying the extreme poor population: Constructing a Proxy Means test

Once the target group is established, a methodology must be found for identifying individuals or households that are in that group and for excluding those who are not. For instance, if the extreme poor population is identified as the target group for the program, one must be able to make a precise judgment about the level of welfare of the means of the recipient. In practice, it is difficult, time consuming, and costly to collect that information for each household of the country.

An alternative method used to measure household welfare is to administer a Proxy Means Test (PMT). This approach relies on indicators that are highly correlated with total consumption expenditure, yet are easy to collect, observe, and verify. With statistical analysis, weights can be assigned to the selected indicators. Then the eligibility for program benefits can be determined on the basis of a total score, as a proxy for household consumption.

To measure welfare, we use per capita household consumption expenditure. In development literature, consumption expenditure is generally considered a more accurate measure of welfare than income because consumption expenditures tend to be less variable than income over seasons. Additionally, in practice consumption is generally measured with far greater accuracy than income in household surveys.

Estimation strategy

We first identify variables that exist in the surveys that are highly correlated with household consumption, that can be easily observed and that cannot be easily manipulated by the households in an attempt to get into the program. We use four groups of independent variables given by: quality of the household, ownership of durable goods, family characteristics, and location.

A two step-strategy is used for the estimation. In the first one, we use a stepwise function to eliminate from the regression those variables that are not statistically significant. In the second one, we estimate an Ordinary Least Squares (OLS) regression using the selected variables.

Given that the explanatory variables have a different impact in the household consumption depending on the region where the household is located, we propose the estimation of two separate models. The first one takes into account the urban population while the second one considers the rural area. It is important to note that we do not estimate a regression for the indigenous area since it is implicitly supposed that the entire household population resident in that area is eligible to participate in the program. As Figure A.3.3.1 shows, this design does not provoke larger targeting error given that most of the population in this area is living in extreme poverty. The extreme poverty ratios in the indigenous ‘corregimientos’ reach levels superior to 0.8 points.




Table A.3.2.1: Estimation of the Total Household Consumption - Proxy Means Test

Source: Own estimation based on ENV 2003 data. Note 1: Robust standard errors in brackets. * significant at 10%; ** significant at 5%; *** significant at 1% Note 2: We used two steps to estimate the household’s welfare. In the fist one, we used a stepwise function to eliminate from the regression those variables that are not statistically significant. In the second one, we estimated an Ordinary Least Squares (OLS) regression using the selected variables. Note 3: Household population


The estimations of the household per capita consumption are shown in Table A.3.2.1. Considering the inherent characteristics of each household, we constructed its level of welfare. Using this information we can estimate the probability of being extreme poor. This probability is the score attributable to each household. In the following section we develop a briefly analytical explanation of the `household score estimation’.

Estimation of the household score

Using the regression results shows in Table A.3.2.1, it is possible to construct for each household a score based on its characteristics.
As it is known, the OLS estimation can be resumed in the following equation:
(1)
where means per capita household consumption, represents the vector of observe characteristics that affects the consumption, and is a random error with normal distribution,
The main idea of the score computation is the estimation of the probability of being extreme poor for each household given its characteristics. This idea can be resumed in the following expression:
(2)
where is the extreme poverty line.
Using the parameters obtained in the OLS estimation, and , and the vector of observed characteristics , it is possible to estimate the equation (2) in the following way:



(3)
where is the cumulative normal standard distribution.
Then, the beneficiary population is selected comparing the estimated score with a determinate cutoff value. If the value of the household score is higher than the value of the cutoff, the household result to be eligible to participate in the program.