Multidimensional poverty in the Philippines, 2004–2013: How much do choices for weighting, identification and aggregation matter?

Sen’s writings on this subject are many; referenced here are only a few examples (including one of the early ones).

Even from the welfarist perspective of consumption expenditure as a money metric of utility, consumption may be inadequate as there are other arguments in individual utility functions for which markets (and hence prices) may not exist or, if they exist, may be distorted.

Even in the absence of market failures, total consumption or income as a linear combination of prices and quantities of goods and services implies perfect substitutability between these goods and services—an assumption that is questionable from a non-welfarist or human rights perspective that insists on the essentiality of minimum levels of achievement across a range of dimensions.

These official estimates are based on per capita incomes from the first-semester Family Income and Expenditure Surveys (FIES) for 2003, 2006, 2009 and 2012, and the Annual Poverty Indicators Surveys (APIS) for 2013 and 2014. The APIS surveys are only conducted for the first semester of the survey year. Since 2013, the APIS has used the same income module as the FIES.

Note that the standard errors for full-year official poverty incidence (percentage of population below the poverty line) for 2006, 2009, 2012 and 2015 were 0.49, 0.52, 0.54 and 0.76, respectively, while those for the first semester incidence were 0.39, 0.46, 0.56 and 0.55, respectively (PSA 2016a and tables accessed online from https://​psa.​gov.​ph/​tags/​official-poverty-statistics).

Balisacan (2015) in fact presents estimates for a longer period for some of his data sources. In particular, the study uses FIES data for the years 1998, 1991, 1994, 1997, 2000, 2003, 2006, 2009 and 2012; NDHS for 1993, 1998, 2003 and 2008; and APIS data for 1998, 2002, 2004, 2007, 2008, 2010 and 2011.

Most of the literature has concentrated on the identification and aggregation issues with the choice of dimensions and weights having remained relatively under-explored.

For related work, see Alkire et al. (2017) who present an analysis of changes in multidimensional poverty over time for 34 countries following the global MPI approach. Their detailed analysis at both national and sub-national levels, however, does not explore robustness to methodological choices in measuring multidimensional poverty.

The survey is considered representative at the national and regional levels, but not at the provincial level. For further information on the APIS survey design and instruments, see Philippines Statistics Authority (2014).

The indicator list was developed in consultation with the staff from the Philippines Statistics Authority.

See Appendix in the online supplementary material for details of how the official food poverty line is determined. Alternatively, nutritional deprivation could be defined as falling below the food poverty line itself, but the more stringent criterion is used here to focus on those facing significant shortfalls in food consumption.

Information on illness/injury over the reference period is collected for all household members. APIS survey documentation indicates that the respondent is “an adult knowledgeable member of the sample family who can provide accurate answers to all or most of the questions in the survey.”

See for instance, Dooley et al. (1994), Dooley et al. (2000), Helliwell and Putnam (2004), Stutzer and Lalive (2004), Lundin and Hemmingsson (2009), Stuckler et al. (2009a, b), ILO (2010), Blanchflower and Oswald (2011).

There are many examples. For instance, for Colombia’s MPI, two of the 15 indicators relate to the absence of long-term unemployment and formal employment, and these indicators are included under “Labour” as a separate dimension. For South Africa, “Economic Activity” is included as a separate dimension, with an indicator on unemployment among 15–64-year-old adults. The Ecuador MPI has “Work and Social Security” as a separate dimension, with an indicator on unemployment or inadequate employment among those 18 years or older. Similarly, for Costa Rica, “Work” is included as a separate dimension, with indicators on long-duration unemployment, informal employment and earnings less than the minimum wage. Employment dimension or indicators are also included in MPIs for El Salvador and Chile.

See Decanq and Lugo (2013) for a review.

See Bellani (2013), Bellani et al. (2013), Decancq et al. (2013) and Mitra et al. (2013) for related applications of this approach.

This is arbitrary but yields a subjective poverty incidence of about 22% that is not very different to the official income poverty incidence of 25% (for 2003 based on FIES).

The estimation of marginal effects takes into account the fact that the independent variables in the regression are also dichotomous.

See Datt (2018) for a formal statement of this property as well as further discussion of why the union approach may be worth considering.

See Datt (2018) for the properties of this class of multidimensional poverty measures; also see Chakravarty and D'Ambrosio (2006) who proposed such a measure of social exclusion.

For the present analysis based on this select set of indicators, survey non-response is not an issue. Based on the available documentation, unit non-response is low at about 4% or less across different APIS rounds. Item non-response for the variables we are interested is virtually zero in the data set.

Note that Balisacan (2015) does not use exactly the same indicators as in this paper. The table reports his weights for the APIS application, lining them up as closely as possible against the indicators in this paper.

Using the 10 indicators, however, the decline in multidimensional poverty over 2004–2011 is of similar magnitude across the three weighting schemes. But this also omits the last year 2013 due to the comparability issue with the illness/injury-related indicator.

Also see Roche (2013) and Alkire et al. (2017) for related application of Shapley decomposition, though applied to changes in multidimensional poverty over two points in time. Dimensional decomposition by comparison has an added degree of complexity as it involves a large number of decomposition factors determined by all possible paths of elimination of deprivations in individual dimensions.

Note that the contribution of food consumption is still higher than its weight, which reflects a larger fraction of the population deprived in terms of food consumption relative to other indicators.