Counting Approaches in Targeting

The implementations of the counting approach observed in Europe, the US, Latin America, and elsewhere were originally developed mostly within universities, and later became a tool for policy design and even targeting, although usually complemented by some other methodology.

One good illustration is the case of India, where a series of different methodologies have been used to identify rural households as ‘below the poverty line' (BPL). BPL households are eligible for certain benefits, such as subsidized food or electricity, and programmes to construct housing and encourage self-employment. Poverty measurement in India has largely been based on consumption and expenditure poverty. Since 1992, the Indian government's census-based targeting methods have gradually evolved towards a counting approach (GOI 2009; Alkire and Seth 2013c). For example, in 2002, the BPL census collected information on thirteen dimensions covering topics such as food, housing, work, land ownership, assets, and education, and an aggregate achievement approach was implemented. This methodology was criticized on a number of grounds, including the cardinalization of ordinal variables and the substitutability of achievements among others.^[153] Alkire and Seth (2008) compare the 2002 BPL method with a method based on a counting approach and show the possible mismatches that may occur between the two methods.

In 2008, the Indian government appointed an Expert Group Committee, under the chairmanship of N. C. Saxena, to provide a critical review of the 2002 BPL methodology and data contents, and to propose a new method for identification.^[154] Their three-stage proposal implicitly used a counting method with a union approach in the first two stages leading to a counting-based identification in the third.

Other subnational initiatives in South Asia use counting approaches for targeting. Two cases might illustrate this. The first concerns the Indian state of Kerala, an emblematic case of development and poverty reduction, whose government has been using a counting approach for targeting poor households since the late 1990s. The method was originally developed by non-governmental organizations (NGOs) and subsequently used for a women-based participatory poverty eradication programme named 'Kudumbashree' (Thomas et al. 2009). Kudumbashree uses nine equally weighted indicators related to housing, water, sanitation, literacy, income sources, food, presence of infants, presence of mentally or physically challenged or chronically ill persons, and caste/tribe. If the household presents deprivations in four or more indicators, it is considered poor; if it presents eight or nine, it is destitute. The identification of poor households is verified by neighbourhood groups comprising households that live in proximity. The identified households are eligible for a number of programmes, including microcredit.^[155]

In our second case, a counting approach to celebrating 'graduation from poverty' is used by two acclaimed Bangladeshi NGOs, the Grameen Bank and the Bangladesh Rural Advancement Committee (BRAC). The Grameen Bank, the 'bank for the poor', was founded by Muhammad Yunus in 1976 in Bangladesh, originally as a local microcredit project. The project evolved into a nationwide bank with over eight million borrowers, of whom 96% were women, and has spread elsewhere. Grameen uses a set of ten indicators to identify participants. When a household has zero deprivations (intersection approach), it is considered to have 'graduated' from poverty. A counting approach to identifying the poor is also implicitly used by BRAC, another prominent microfinance NGO, initiated by Fazle Hasan Abed in 1972 in Bangladesh, which has spread widely.

Moving further east, in Indonesia poverty is primarily measured using the indirect income approach. However, multidimensional perspectives using counting approaches are emerging (CBS 2008). A 'family welfare approach' was initially proposed by the Family Planning Coordination Board in 1999 (CBS 2008: 10). This approach identified a family as poor if it was deprived in one of five indicators (a union approach): religious freedom, meals per day, clothing, size of house, and access to modern medicine. The approach was not implemented, however, because the five indicators were not relevant to all families. The Central Bureau of Statistics (CBS) then proposed a ‘poverty criteria approach', which identified people as poor if they were deprived in five out of eight indicators. The eight indicators were floor area; type of floor; water access; type of water; asset ownership; income per month; expenditure spent on food; and consumption of meat, fish, eggs, and chicken. A census instrument conducted in three provinces—South Kalimantan, DKI Jakarta, and East Java—in the years 1999, 2000, and 2011, respectively (CBS 2008: 18), used this method to determine whether households had the right to receive basic necessity subsidies (CBS 2008: 19).^[156]

A distinct yet related methodology for identifying the poor is the poverty scorecard developed by Mark Schreiner (Schreiner 2002, 2006, 2010). Schreiner proposed the method both for measuring poverty as well as for targeting beneficiaries. The poverty scorecard uses an individual or household card, and grades five to ten achievements to produce a score. Indicators are sought that are strongly correlated with income poverty and have the following characteristics: ease of acceptance, inexpensive to observe and verify, already commonly collected, objective, liable to change over time as poverty status changes, variety vis-a-vis other selected indicators, and applicable across countries and across regions within a country.

The indicators are fielded in nationally representative household surveys that also collect information on income or expenditure. Indicator weights are set through a logit regression as follows. The individuals are categorized into two groups: income poor and income non-poor, and this categorization is used as the dependent variable in the regression, with the selected indicators as explanatory variables. The logit weights are transformed such that all weights are non-negative integers and the minimum score is 0 and the maximum is 100.^[157] For example, in his proposed scorecard for Pakistan (Schreiner 2010), if the household does not have a flush toilet (most deprived), it receives a score of 0; if it has a flush toilet to pit (less deprived), it receives fourteen points; and if it has a flush toilet to public sewer (not deprived at all), it receives nineteen points. The total poverty score for each household is obtained as the sum of the household's scores obtained in all indicators. A person is identified as poor if that person's poverty score lies below a poverty threshold which, as Schreiner indicates, can be determined according to the aim and scope of the particular programme.

As in the case of Boltviniks method, Schreiner's poverty scorecard method departs from a counting approach. Furthermore, it cardinalizes ordinal data, based on logit regressions, which does not seem legitimate, as section 2.3 argued. Scores are then standardized and aggregated to obtain an overall score, which is compared to an overall threshold. This step, like the aggregate achievement approach (section 2.2.2), allows substitutability between non-deprived and deprived achievements. If all the variables had been cardinal, the score would be the (weighted) sum of achievements.

With the method described above, Schreiner developed poverty scorecards for various microfinance institutions and also developed adaptations such as the Progress out of Poverty Index (PPI).^[159] A related method was used by the Benazir Income Support Programme in Pakistan, which targets benefit recipients using a scorecard of twelve observable indicators, each of which receives a weight based on an Ordinary Least Squares (OLS) model of household expenditure per adult equivalent regressed on various sets of predictors (proxy-means test) (Khan and Qutub 20 1 0).^[160] The approach has thus spread widely yet without clarifying fundamental methodological concerns.

In the area of targeting, the AF methodology is also spreading both via academic studies and in policy programmes. For example, Robano and Smith (2014) examine the TUP programme of BRAC, developing M₀ measures for the existing targeting methods as well as for a proposed alternative, and present and implement an impact evaluation methodology using M0 rather than any single outcome as the dependent variable. Azevedo and Robles (2013) propose an M0 multidimensional targeting approach to identifying beneficiaries that explicitly takes into consideration the multiple objectives of conditional cash transfer programmes and the multiple deprivations of the poor household. Using data from Mexico's prominent Oportunidades programme, they find M₀ multidimensional targeting to be significantly better than either the current targeting method or an alternative income proxy-means test at identifying households with deprivations that matter for the programme objectives.

In sum, the necessity of defining a target population for poverty reduction programmes has motivated the use of counting methods with a variety of specificities and prompted the development of related new identification methods. However, the measurement properties and features of the alternative targeting instruments are rarely discussed (or, one suspects, clearly communicated), which makes it difficult for policymakers to make an informed decision.

4.5