1 Introduction
Income, or consumption poverty measures such as the World Bank’s dollar-a-day headcount ratio (Ravallion et al.
2009), is still the most prevalent measure of poverty used across the globe. However, from Asia to Africa (Batana
2013; Klasen
2000; Santos
2013; Ranis and Stewart
2012; Yu
2013), and across Europe to Latin America (Battison et al.
2013; Brandolini and D’Alessio
1998; Whelan et al.
2004), scholars have consistently documented that the lack of money is not always an accurate proxy for deprivations that society cares about. It has been argued that money metrics do not tell the whole story of human suffering, because poverty is not only about one’s inability to spend on essential goods and services. More than that, it is about one’s inability to enjoy valuable beings and doings (Sen
1985). Indeed, what is now generally accepted is a notion of poverty (or well-being for that matter) as an intrinsically multidimensional construct that encompasses the whole range of ways in which an individual can participate effectively in society.
Since the seminal works of Townsend (
1979) and Sen (
1985), different multidimensional poverty measures have been developed. Yet, as noted by Santos and Ura (
2008: 1), ‘some of the proposed measures seem to have incorporated a multidimensional perspective at the cost of giving up the simplicity and intuition that characterise the unidimensional measures’. Statistical approaches to multidimensional poverty measurement (Filmer and Pritchett
2001; Sahn and Stifel
2003), for instance, rely on multivariate or latent-variable techniques to the extent that parameters are completely data-driven, leaving evaluators with limited control over the measure. Some axiomatic alternatives such as Bourguignon and Chakravarty (
2003), on the other hand, satisfy a number of useful measurement properties but do strictly necessitate the availability of cardinal data; in reality, vital social indicators such as literacy and completion of primary school are usually ordinal in nature.
In an attempt to address these problems, Alkire and Foster (
2011a; henceforth AF) proposed a new sort of multidimensional poverty measure: one that is not only simple to construct, but also retains many of the properties of the well-known Foster–Greer–Thorbecke (FGT) measures (Foster et al.
1984) of unidimensional poverty measurement. The AF method combines the FGT with the counting approach (Atkinson
2003), which is easy to understand and has a long history in sociology. The method deals with ordinal data in a straightforward manner by dichotomising individuals’ achievement into deprived and non-deprived states. Aggregation is then performed, first across deprivations experienced by each individual, and then across individuals, yielding a measure that is intuitively interpretable as the share of deprivations that poor individuals experience out of the total deprivations that the society could possibly experience.
As a generalisation of the classical FGT, the AF family of multidimensional poverty measures satisfies an array of desirable axioms (Alkire and Foster
2011a). Foremost among them are their ‘subgroup decomposition’ and ‘dimensional breakdown’ properties, which allow the overall poverty measure to be broken down into its social, geographical or dimensional constituents in a way that is both conceptually and technically defensible. A thorough characterisation of joint deprivations is further made possible by the availability of partial indices that capture the incidence as well as the intensity of poverty. The methodology is also transparent in the sense that all parameters are under the control of the evaluator, allowing normative decisions with regard to the selection of indicators, dimensional and poverty cut-offs, and weighting schemes to be easily incorporated into the analysis. In fact, acknowledging these novelties, in 2010 the United Nations Development Programme (UNDP) replaced its Human Poverty Index (HPI; first published in 1997) with the new multidimensional poverty index (MPI), based on the AF family of multidimensional measure (UNDP
2014,
2010: 95).
Applying the AF method to the National Socio-economic Survey data from Indonesia, this paper seeks to estimate the extent and to investigate the regional as well as the temporal patterns of multidimensional poverty in Indonesia for 11 consecutive years spanning from 2003 to 2013. The aim of this study is not to replace the official consumption poverty estimate with a new one, but rather to augment the conventional poverty measure with additional information on health and education using the same data source that has historically been used to estimate the official consumption poverty figure in Indonesia. This version of an Indonesian multidimensional poverty index (MPI) is constructed in a way that income poor individuals are ‘automatically’ multidimensionally poor, but not the converse. The present study considers the following questions: taking into account income, health and education dimensions, how many Indonesians are poor overall? Are urban areas always better off? Which island of the archipelago is the most deprived? Did recent progress, if any, benefit the poorest of the poor? And what happened to gender and spatial inequities during the last decade?
Indonesia, the world’s largest archipelagic state and the third-most populous developing country, is known for its exemplary achievement in terms of income poverty reduction and overall human development (Ranis and Stewart
2012). However, there is little research attempting to understand the nature of
simultaneous deprivations experienced by its people. The majority of recent poverty evaluations have been conducted exclusively within the monetary space (Ilmma and Wai-Poi
2014; Strauss et al.
2004; Sumarto et al.
2014); even when multidimensionality is sought, it has always been computed using variants of marginal method (BPS
2015b; BPS et al.
2004) that are blind to joint deprivation (Alkire
2011: 503–504).
To date, only two studies attempted to measure the extent of simultaneous deprivations in Indonesia. Alkire and Foster (
2011a), in the earliest showcase of their methodology, provided a national poverty estimate for the year 2007 using the Indonesia Family Life Survey data (IFLS; Thomas et al.
2012). But it is known that the IFLS sampling frame is not entirely representative of the population (RAND
2007); it neglects individuals living in the eastern islands of the archipelago (RAND
2014), yielding a sample that favours the relatively well-developed areas in western Indonesia. Alkire and Santos (
2014) carried out further study on Indonesia using the Demographic Health Survey (ICF International
2012) data as a part of a grand endeavour to construct a globally comparable MPI (UNDP
2010). While they are completely representative of the population, the DHS data do not, however, provide household consumption expenditure information, preventing a useful comparison with the official measure of consumption poverty.
The contribution of the present study to the existing literature is threefold. Firstly, in estimating the extent of multidimensional poverty in Indonesia, this study uses large and nationally representative data that have been regarded as the primary source of information among Indonesian policy makers as well as international observers. While concurring with Alkire and Santos (
2014: 266) who stress that data availability has been the major bottleneck in the development of an internationally comparable MPI, we would like to demonstrate that even when using an existing data source, the construction of an Indonesian MPI is not only technically feasible but also substantively meaningful. The collection of better well-being data is of course desirable, but Indonesians do not have to wait until the ‘perfect’ data becomes available to have their progress assessed. Secondly, the inclusion of consumption expenditure information makes this version of Indonesian MPI not only comparable to the official poverty measure, but also sensitive to economic fluctuations (Ravallion
2010: 11). Lastly, by providing an annual analysis of the trend of multidimensional poverty in the last 11 years, this study presents a richer picture compared to one that analyses only selected points in time over the same period.
The remainder of the paper is structured as follows. The next section describes the AF method, the data and the dimensions. It then investigates the degree to which income poverty correlates with non-income deprivations. Section
3 presents the results. Initially, unidimensional deprivations are investigated using the marginal dashboard approach. Then, MPI estimates at national and sub-national levels are presented along with robustness checks. Finally, changes in the distribution of deprivations among the poor are studied. Section
4 concludes.
4 Conclusion
Applying the Alkire–Foster method of multidimensional poverty measurement to the National Socio-economic Survey (Susenas) data of Indonesia, this study estimates the extent and investigates the regional as well as the temporal patterns of multidimensional poverty in Indonesia from 2003 to 2013. An Indonesian version of multidimensional poverty index (MPI) is developed through an augmentation of the existing consumption poverty measure with information on health and education that are represented by indicators of illness episode, morbidity, completion of primary school, and literacy.
It is found that, irrespective of the poverty cut-offs or weights specified, there was an unambiguous multidimensional poverty reduction over the last decade at both national and sub-national levels. About half (48 %) of Indonesian adults were multidimensionally poor in 2003 and, collectively, they experienced about one-fifth (0.19) of the total possible deprivations that the society could experience. In 2013, the situation was unmistakably better: only one in ten adults (11 %) was identified as multidimensionally poor, while the overall poverty figure fell to 0.04 (78 % reduction). The data suggest that the rate of poverty reduction was faster in the 2003–2008 period (60 %) than in 2008–2013 (44 %).
With the exceptions of rural areas and the Nusa Tenggara islands, there was minimal improvement with regard to the average deprivations experienced by the poor (intensity); overall poverty reduction was driven mainly by the decline in poverty incidence. It is further found that, when the overall measure is broken down into its dimensional constituents, income deprivation remains the main contributor to multidimensional poverty (60–70 %), albeit with a 2 % rate of decrease annually. Also estimated in the national-level analysis is the mismatch between income and multidimensional poverty identification. Results show that approximately 3 % of adult Indonesians (4.5 million individuals in 2013) would be classified as non-poor if poverty identification did not take into account deprivations in health and education. This figure could be as high as 7–17 % (11–26 million), depending on how much importance is assigned to schooling and/or illness episode indicators.
In an attempt to gain a more complete understanding of joint deprivation, the overall poverty measure is broken down by relevant population sub-groups. The data show that for each year from 2003 to 2013, multidimensional poverty was unambiguously higher in rural than in urban areas, but the gap between them has been progressively narrowing thanks to substantial improvement in both the incidence and the intensity of poverty in rural areas (rural-to-urban poverty ratio was 1.53 in 2003 vs. 1.25 in 2013). The data further reveal that Indonesian women are not unambiguously more deprived than men, although they appeared to have slightly more deprivations on average in the 2003–2007 period. Nevertheless, we cannot ascertain whether Indonesia has fared well in terms of gender equality because we cannot disentangle fully the information of women’s income using household expenditure data available at present.
In contrast to the clear trend seen in urban/rural and gender decompositions, we found only faint dominance in between-island comparisons over the 11-year period. It is only from 2010 onwards that it can be asserted with statistical confidence that poverty is unambiguously higher in Papua, Maluku and Nusa Tenggara (or lower in Kalimantan) than anywhere else in the archipelago. Even so, it is still important to note that such between-island comparisons mask a substantial amount of within-island and between-district variations, echoing both Ilmma and Wai-Poi (
2014) and Sumarto et al. (
2014). While five out of the ten poorest districts in 2013 are indeed located in Papua, three of them are in Sumatra and the other two are in Java, neither of which are thought of as places with extreme poverty. Analysis at the district level further reveals that, departing from the pattern observed in a cross-national study (UNDP
2010: 98), the intensity of poverty among districts in Indonesia does not seem to be related in a linear way to its incidence.
When the distribution of deprivations among the poor is studied, it is found that between 2003 and 2013, there were statistically significant improvements in terms of inequality among the poor and disparity across subgroups. The data show that poorer subgroups progress faster than the less poor, irrespective of the social or geographical groupings considered (converging subgroup poverty level). This finding indicates that the progress achieved within the last 11 years is relatively inclusive, although it should be noted that the between-district inequality within the Indonesian archipelago remains striking.
Overall, these trends are comparable to those obtained from recent consumption poverty evaluations conducted by Ilmma and Wai-Poi (
2014) and Sumarto et al. (
2014), highlighting the fact that, even a decade after a ‘big-bang’ decentralisation (Hill
2014) was initiated, spatial inequity remains a serious challenge for Indonesia. It has been argued that the immense variation in poverty levels across districts reflects heterogeneity in the ‘capacity and resources of local governments to develop and implement poverty reduction strategies, and to quickly provide good public services’ (Sumarto et al.
2014: 310). Only competent local government can formulate sound development plans, allocate budgets efficiently, and deliver public services effectively. Therefore, there is plenty of room for local administrators to learn lessons from the top-performing districts (Maharani and Tampubolon
2014).
While this study has presented a thorough investigation into the state of multidimensional poverty in Indonesia over the last decade, it is inevitably bound by several limitations. Firstly, the present study is unable to include children and adolescents younger than 18 years old in the analysis because information on the relevant dimensions of their well-being (Trani et al.
2013) are not available in Susenas survey. Secondly, the health indicators used in this study (illness episode and morbidity) are weak and by no means comparable to the indicators stipulated in the Millennium Development Goals (malnutrition). Thirdly, with the absence of preference data obtained from large-scale participatory study, the trade-offs between social indicators used in this study are entirely normative. In addition, although the measurement of chronic multidimensional poverty under the Alkire–Foster methodology has recently become feasible (Alkire et al.
2014), this study was unable to make use of it due to the cross-sectional nature of Susenas survey. It is indeed indisputable that future poverty evaluations would benefit from the availability of more comprehensive micro-data.
Even with these limitations, the study still contributes to the literature in at least three ways. First, using nationally representative survey data from Indonesia, the present study shows that the conventional measure of income poverty is not comprehensive. The Indonesian data reveal that income poverty only weakly correlates with deprivations in the domains of health and education, confirming the findings documented in other Asian (Ranis and Stewart
2012; Santos
2013; Yu
2013), African (Batana
2013; Klasen
2000), European (Brandolini and D’Alessio
1998; Whelan et al.
2004) and Latin American (Battison et al.
2013) countries. This may motivate future assessment of multidimensional poverty in other parts of the world.
Second, in using consumption expenditure data as the indicator of income, this study allows the poverty measure to become more sensitive to economic fluctuations than the current version of the international MPI (UNDP
2010), which uses asset ownership as a proxy for deprivation in living standards. This not only addresses one of the criticisms of the MPI (Ravallion
2010: 11), but also makes the MPI more comprehensible to Indonesians, who have for decades been accustomed to the conceptualisation of poverty as a consumption shortfall in essential goods and services.
Finally and most importantly, the present study demonstrates the feasibility of adapting the Alkire–Foster methodology to the Indonesian context using an existing official data source that has been in production since the 1960s (Surbakti
1995). Because the data are readily available, and the proposed multidimensional poverty measure presented here makes identification of multiply-deprived Indonesians possible, the MPI could nicely complement the existing indices that are routinely reported by the Indonesian Statistical Bureau (BPS). The new measure is suitable as a tool for monitoring the progress of national development, and could also be used as a device for prioritising investment projects or other forms of intervention that are funded by transfers from central to local governments (see Salazar et al.
2013 for a recent proposal in Colombia). With the demonstrated novelty, feasibility and utility of the Alkire–Foster method, policy makers should now more than ever want to incorporate the idea of poverty as an experience of multiple deprivations into the discourse of national development.