The Union cabinet last week paved the way for the new round of the below poverty line (BPL) census to be conducted later this year. The new census assumes importance given the dissatisfaction with the previous three attempts to identify the poor. It is also important because BPL census remains the primary delivery mechanism for the majority of our social protection schemes, including the much-maligned public distribution system and most social pension programmes.
But can the new BPL census provide answers to the targeting problems that have ailed our subsidy delivery system? Not really, though the proposed methodology will reduce, albeit to a limited extent, the earlier inclusion and exclusion errors. Thus, it will definitely be better than previous such attempts.
The basic framework of the new methodology involves a three-step procedure. The first step is the exclusion of rich households. This is followed by automatic inclusion of extremely vulnerable social and community groups, and ranking of the rest of the households on various deprivation parameters. Though the basic framework remains similar to the proposals of the Saxena committee appointed by the ministry of rural development, the indicators have been refined based on data from a pilot survey that the ministry conducted last year. The pilot data are useful in understanding the strength of the various indicators used as possible measures of deprivation.
The results of the pilot survey are revealing though not very different from similar results from other secondary surveys. First, it is easy to identify the top 20-30% using verifiable and observable exclusion criteria. Second, only a few social and community groups can be automatically included without any error. Third, there are very few indicators that work as suitable measures of deprivation. The pilot suggested seven such measures, in dimensions such as housing, land, social vulnerability, family composition, education, health and disability, and occupation. These will then be used to generate a ranking of households.
Analysis of the pilot data suggests that this method is simpler to understand, and works better in ranking households, compared with alternative specifications such as scoring that were suggested by the Saxena committee. But like any method of ranking based on a limited set of indicators, this method does not yield an exact number of deprived households. What happens is a large bunching of households at every deprivation score, particularly at the lower scores of 1, 2 and 3.
This also implies that the results of this method will not be entirely consistent with the Planning Commission’s poverty estimates. However, such an outcome is neither possible nor desirable theoretically and empirically. These two measures rank households on completely different dimensions, and there is no reason to expect them to be synchronous with each other except through coincidence.
Though the final methodology does not explicitly set any poverty caps (a fixed number of households specified by the Planning Commission based on poverty estimates), it does provide for a method of adhering to a cap if the government desires to impose one. This is where the problem lies, because any attempt to segregate the people who are essentially similar, with the same amount of deprivation, means choosing arbitrarily among the poor.
Not only does this create perverse incentives among households that are not too different from each other to use all means to sneak on to the BPL list, it also means that the genuinely poor are at risk of being pushed out. Past experience of targeting has confirmed the inclusion of non-poor in the BPL list at the cost of the poor. And even one wrong household getting on to the list means a genuine household squeezed out.
The pilot data suggests that even with the best of designs, the error rate would remain at around 20%. That number is bound to go up when the survey goes and hits the field. This would be partly due to errors in implementation, but also due to vested interests trying to corrupt the system by wilfully concealing information. In the end, the method may reduce the errors in identifying the poor, but may not eliminate them altogether.
The more important message emerging from the pilot is the fact that it is almost impossible to have a method of perfect targeting. This should not be surprising to those who are aware of previous experiences of targeting in the Indian context or elsewhere. For the record, inclusion errors in Brazil are close to 49% (that is, half of the beneficiaries are non-poor) and in Mexico, 36%. The corresponding exclusion errors (percentage of non-beneficiary poor to total poor) are 59% in Brazil and 70% in Mexico. Incidentally, both these countries have been highlighted as success stories of cash transfers. The message is clear; the problem is targeting, not the form in which subsidies are delivered. It is high time we had a debate on the method of BPL identification. The issue of cash transfers is secondary.
Comments are welcome at firstname.lastname@example.org
Himanshu is assistant professor at Jawaharlal Nehru University and visiting fellow at the Centre de Sciences Humaines, New Delhi