Sen (1976) states that ‘in the measurement of poverty, two distinct challenges must be faced, viz., (1) identifying the poor among the total population, and (2) constructing an index of poverty using the available information on the poor’ (p. 219). This paper aims to address Sen (1976)’s second challenge by exploring the steps taken in constructing a composite index of poverty alongside illustrating the role composite measures may play in supporting sound policy. More importantly, it aims to contribute to the improvement of existing composite measures of poverty by examining one crucial source of their technical weaknesses; i.e. the arbitrary nature of the weighting process by which multiple indicators are combined. A Delphi survey was first conducted with policy makers in Indonesia, in order to gather relevant information on the dimensions and indicators alongside a range of respective weights to be included within the composite measure. The use of the Delphi method enabled the process of selecting weights to be explicit, clear and subject to public scrutiny, allowed weights to take into account trade-offs between dimensions and indicators and entitled selected weights to reflect people’s preferences. The method however presented two challenges with regard to determining the final weights for the composite measure; i.e. (1) mitigating the existence of a degree of disagreement with regard to the exact value of the weights per dimension and indicator and (2) achieving a set of indicator weights of which the total sum equals 1. A linear programming method was suggested to mitigate these challenges. Resulting weights were utilised to calculate the composite measure using the Indonesian Susenas 2013 dataset. A comparative exercise was then conducted to contrast the use of the weights resulting from the Delphi survey vs. the commonly used equal nested weights. Significantly high correlation coefficients were found between household-rank according to their total weighted deprivation when both weighting schemes were applied. This indicates that the use of both weighting schemes may identify similar households as poor. This conclusion was further supported when an examination of composition of the poor using both weighting schemes, was conducted. Policy wise, these findings suggest that the use of the more parsimonious equal nested weighting scheme, may lead to similar households being deemed eligible to receive social assistance. Despite this, the use of the Delphi in determining weights still adds value to the creation of a measurement of poverty that aims to truly reflect contextual priorities and trade-offs. The Delphi survey initiated discussion for the crucial need of a multidimensional measure, whilst establishing a sense of ownership to the resulting measure. As Nardo et al. (2005) states, ‘however good the scientific basis for a given composite measure, its acceptance relies on negotiation’ (p. 8).
Short Bio of Putu Natih
Putu is a Jardine-Oxford graduate scholar currently in the final stages of a DPhil in Social Policy at the University of Oxford.