Official Poverty Measurement in Indonesia since 1984: A Methodological Review

Abstract

This article describes how the measurement of the official Indonesian poverty figures has evolved since 1984, when Badan Pusat Statistik (BPS), Indonesia's central statistics agency, published its first poverty report. Since then, BPS has on several occasions revised the underlying methodology for how it calculates poverty. These changes have, in general, improved the way that poverty in Indonesia is measured, but they make it difficult to compare poverty figures over time. In fact, only poverty estimates (at the national and provincial level) since 2007 are based on the same methodological approach. This article presents the first detailed description of official poverty measurement in Indonesia since Booth's (1993) study, in English, and Sutanto and Avenzora's (1999) study, in Indonesian. It constitutes a unique repository for anybody who wants to understand the technical details of official poverty measurement in Indonesia.

Tulisan ini menjelaskan evolusi pengukuran angka kemiskinan Indonesia yang resmi, sejak pertama kali laporan kemiskinan dipublikasikan oleh Badan Pusat Statistik pada tahun 1984. Sejak saat itu, BPS telah beberapa kali merevisi metodologi dasar yang mereka gunakan dalam menghitung kemiskinan. Secara umum, perubahan ini telah memperbaiki cara pengukuran tingkat kemiskinan di Indonesia. Di sisi lain, perubahan-perubahan tersebut juga mengakibatkan sulitnya melakukan perbandingan angka-angka kemiskinan antar waktu. Kenyataannya, baru sejak tahun 2007 angka-angka estimasi kemiskinan didasarkan pada metodologi yang sama (baik pada tingkat nasional maupun tingkat provinsi). Tulisan ini merupakan tulisan pertama yang menjelaskan secara rinci tentang pengukuran resmi kemiskinan di Indonesia sejak kajian Booth (1993), dalam bahasa Inggris, dan Sutanto dan Avenzora (1999), dalam bahasa Indonesia. Tulisan ini juga bisa menjadi acuan baru bagi mereka yang ingin memahami rincian teknis tentang pengukuran resmi kemiskinan di Indonesia.

Keywords:

• poverty measurement
• poverty line
• prices
• methodology
• Susenas

JEL classification:

• I31
• I32

TNP2K = Tim Nasional Percepatan Penanggulangan Kemiskinan (National Team for the Acceleration of Poverty Reduction), Office of the Vice-President. I would like to thank Anne Booth, Isis Gaddis, Fiona Howell, Stephan Klasen, Riyana Miranti, Suahasil Nazara, Ririn Purnamasari, Elan Satriawan, Sudarno Sumarto, Matthew Wai-Poi, and all three anonymous referees for valuable input and comments, and Mercoledi Nikman Nasiir, Wisnu Harto Adi Wijoya, and Novat Pugo Sambodo for their outstanding research assistance. Special thanks goes to BPS, and in particular Ahmad Avenzora, for providing me with clarifications on the way that it has calculated official poverty rates. The views expressed in this article are mine alone and do not represent the views of TNP2K.

Notes

¹ Other BPS documents that address poverty measurement in Indonesia include those of Ritonga and Avenzora (2002), BPS (2002, 2010), and Maksum (2004). However, these documents lack important information and therefore do not help in obtaining a more thorough overview of poverty measurement in Indonesia.

² Confusion still arises on occasion about the World Bank's poverty figures ($1.00 or $1.25 per day [purchasing power parity]). While these figures make sense only in an international comparison, they are often inappropriately used in policy forums and the media by those interpreting them as national poverty rates and trends. See Sumner and Edward's (2014) study, in this issue.

³ The survey was based on 348 households of public servants in urban areas. The methodology required them to record expenditure over several weeks, which caused the sample to be biased to middle-income households able to read, write, and calculate.

⁴ Recent instances of the BPS publication Analisis dan penghitungan tingkat kemiskinan (Analysis and calculation of the poverty rate) list further unofficial poverty lines and poverty estimates that emerged between 1975 and 1985.

⁵ A province-specific rural and urban spatial price deflator was applied to the consumption aggregate whereby the reference category was Jakarta. Indonesia defines districts as municipalities (kota) or predominantly rural areas (kabupaten). Each district can then be divided into urban precincts (kelurahan) and rural villages (desa). BPS's definition of rural and urban areas in its poverty measurement uses the desa and kelurahan classifications.

⁶ In addition, the World Bank (1990) draws on an internal analysis by V. V. Rao from 1983, in which poverty estimates are based on valuing the price of 16 kilograms of rice (step one), multiplying the obtained value by 1.25 to allow for other food expenditures (step two), and then dividing this figure by the share of food expenditures in total expenditures of the expenditure group whose total food expenditures were closest to those in step two. Rao's estimates were calculated separately for each region (provinces and rural and urban areas).

⁷ The selection of households to be interviewed for Susenas is based on a stratified sampling procedure. Consequently, BPS provides individual and household weights with the Susenas dataset that allow BPS and researchers to derive representative welfare statistics. It is not yet well understood to what degree BPS is able to avoid the undercounting of the very poor and the very rich during the data collection process as well as in its construction of individual and household weights. See the World Bank (2007) study for a more detailed discussion on this issue.

⁸ BPS first included questions on consumption expenditures in the Susenas core in 1992, and these questions now appear in every Susenas round. Since 1993 BPS has greatly increased the number of households it samples for each annual Susenas round, from about 65,000 in 1992 to 202,000 in 1993. Since 1993 BPS has therefore been able to calculate annual district-level welfare statistics. However, only since 2002 has BPS published official district poverty figures. Moreover, while all selected households are asked to answer the core section, only about 25% of households had to answer the module section during 1993–2010. Since 2011 all selected households have been asked to answer both the core questions and the module questions. During 1993–2014, the consumption expenditure module was included in the Susenas rounds of 1993, 1996, 1998, 1999, 2002, 2005, 2008, 2011, 2012, 2013, and 2014. Although not all Susenas modules contain data on consumption expenditure, in this article the term ‘module’ will refer to the consumption expenditure module.

⁹ Susenas collects income data too, but these are not used by BPS for poverty measurement. Please see Leigh and Van der Eng's (2010) and Nugraha and Lewis's (2013) studies on welfare (inequality) measurement using income data from Susenas.

¹⁰ The years of the 1997–98 Asian financial crisis are exceptions. BPS conducted additional Susenas consumption surveys, Susenas ‘Mini’, in December 1998 and August 1999, to monitor the effects of the crisis. However, only the December 1998 figure from these two surveys entered official poverty estimates.

¹¹ The selection of new panel households occurs only once the panel is completed: up to the end of 2010, this was three years; from 2011, five years.

¹² Sutanto and Avenzora (1999) provide more details on these expenditure categories.

¹³ Nyberg (1976) and Van der Walle (1988) give good overviews of the composition of the welfare consumption aggregate derived from earlier Susenas rounds.

¹⁴ The poverty lines are absolute not relative.

¹⁵ The 2,100-kilocalorie norm stems from the national food and nutrition workshop (Widyakarya nasional pangan dan gizi) in Bogor in 1978 (LIPI 1978). BPS has not changed its calorific base of poverty measurement since its 1984 report. Likewise, despite internal discussions on the introduction of adult equivalence scales into poverty measurement—see the BPS publications Analisis dan penghitungan tingkat kemiskinan (Analysis and calculation of the poverty rate) from 2005 and Penyempurnaan metode pengukuran kemiskinan rumah tangga di Indonesia (Refinement of the method of measuring household poverty in Indonesia) from 2010—BPS has kept its per capita approach. BPS does not use any sort of protein anchor in measuring poverty.

¹⁶ Since 1984 BPS has called its method of measuring poverty a ‘basic-needs approach’ (pendekatan kebutuhan dasar). Yet only since 1993 has it measured poverty in a way that is in line with the usage of this term in the literature (Ravallion 1998, 2008).

¹⁷ In order to convert the recorded food expenditure information into calories, BPS relies on nutrition composition tables published and regularly updated by Direktorat Bina Gizi (the Directorate of Nutrition) in the Ministry of Health (previously Depkes). These tables were first published in 1964, with revisions in 1990, 1995, 2001, 2005, and 2009. Although BPS relies on these tables, it does not document which poverty estimates are based on which table.

¹⁸ Most developing countries began to determine the non-food poverty line by specifying a detailed, item-specific food basket without determining an explicit non-food basket. In contrast, Indonesia began the opposite way, specifying a detailed non-food basket while letting the food poverty line be determined by the average calorie price of the reference population.

¹⁹ As documented by Sutanto and Avenzora (1999), only in 1993 did BPS start to calculate its poverty figures by considering the entire per capita consumption distribution. Before 1993, BPS interpolated aggregated consumption statistics to estimate how many individuals were poor.

²⁰ BPS publishes an overall national poverty line for illustrative purposes—it has never been used to derive poverty figures in Indonesia. During 1976–87, only the rural and urban poverty lines (at the national level) entered poverty calculations (Sutanto and Avenzora 1999). The definition of rural and urban areas in Indonesia is updated approximately every 10 years, after each population census; McCulloch, Weisbrod, and Timmer (2007) show that such revisions in certain areas can have significant effects on Indonesia's poverty profile. In 2014 BPS recalculated its 2013 poverty estimates using newly available population projections calculated using the Population Census 2010. According to BPS the switch to the updated population numbers leads to only minor changes in the poverty figures (Table 1).

²¹ The food poverty lines from 1976 until 1990 are based on the food energy intake method. These lines are derived by multiplying an average kilocalorie price by 2,100. As explained in more detail by Sutanto and Avenzora (1999), BPS applied several intermediary steps before this average calorie price was obtained. One of these intermediary steps involved the adjustment of the entire expenditure per capita distribution using the correction factors for calories consumed outside of the home.

²² The nine remaining provinces were Bengkulu, Central Kalimantan, Central Sulawesi, East Kalimantan, East Timor (now the nation of East Timor), Irian Jaya (now Papua and West Papua), Jambi, Maluku, and Southeast Sulawesi. Only since 1993 has BPS derived rural and urban poverty lines for each province.

²³ In its 1976–1990 estimates, BPS used the reference population only to find the average calorie price (food poverty line). Since 1993 it has used the reference population to determine the prices for the food poverty line, derive the non-food poverty line, and create sampling frames for the SPKKD. Some circularity is involved in selecting the reference population: the rupiah value of the poverty line depends on a reference group whose being defined as such depends on the rupiah value of the poverty line. Pradhan et al. (2000, 2001) discuss iterative methods of addressing this problem. BPS has taken a very practical approach by simply inflating by the consumer price index the poverty lines of the preceding round. It then uses this inflation-adjusted poverty lines to select the reference population. Since its 1984 poverty report, BPS has always selected a reference population with wealth levels above the poverty line, which differs from the recommendations of many academics—such as Ravallion (1998)—who propose selecting a group of households whose wealth levels are around or just below the poverty line.

²⁴ To better understand how BPS handles the SPKKD in combination with Susenas, let us assume that BPS has decided (based on SPKKD data) that three pieces of soap should be included in the non-food basket, and that soap belongs to the toiletries subgroup. Let us also assume that the expenditure on three pieces of soap constitutes 20% of all toiletries expenditures of the reference group, as recorded in the SPKKD. In the next step, BPS multiplies this 20% value by the recorded expenditure amount in rupiah of the toiletries subgroup, as recorded in Susenas. Let us then assume that data from Susenas showed that the reference population spends about Rp 100,000 on toiletries each month. Accordingly, 0.2 × Rp 100,000 = Rp 20,000 is the amount that enters the non-food poverty line. BPS applies this method to all selected non-food items.

²⁵ It is reasonable to assume that the determination of the upper limit(s) of the rupiah value that determines the reference population (the lower value is given by the poverty line[s] for the 1976–99 estimates) was based on BPS having a certain share or number of the population in mind. Therefore, the change in the definition of the reference population might have had only limited consequences in practice.

²⁶ Since Indonesia's CPI is based on urban areas only, BPS had to assume that prices between rural and urban areas increased by the same amount. Likewise, since the CPI captures price developments that do not necessarily represent the consumption basket of the reference population, BPS had to assume that these differences were negligible.

²⁷ As shown by Pradhan (2009), the under-reporting of consumption in the Susenas core section compared with that in the modules is more prevalent among the poor than the non-poor. It is therefore questionable whether poverty estimates based on the modules are comparable with those based on the core.

²⁸ In this article, all information on changes during 2002–13 is based on various BPS sources, including the annual rounds of Statistik Indonesia: Statistical Yearbook of Indonesia and Data dan informasi kemiskinan (Poverty information and data); Analisis dan penghitungan tingkat kemiskinan (Analysis and calculation of the poverty rate) for 2005, 2008, 2009, 2010, and 2011; the (bi-)annual rounds of Tingkat kemiskinan di Indonesia (The level of poverty in Indonesia) since 2006; and BPS's (2010) report. Certain information described in this article is not tabulated in these publications but draws on discussions with BPS representatives.

²⁹ Although not directly comparable with the BPS approach, Ravallion and Bidani's (1994) study investigates whether food energy intake and the cost of basic needs would produce similar spatial distributions of poverty in Indonesia. The authors find that both measures produce very different findings in classifying who is poor in Indonesia.

³⁰ Even BPS's poverty estimates since 2007 lack comparability and consistency if we define consistency as applying the same welfare standard over time. BPS still applies the subgroup expenditure share values from the 2004 SPKKD. Although these shares remain constant, it does not mean that the interpretation of these shares remains constant. If, for instance, three pieces of soap compose 20% of expenditures in the toiletries subgroup in 2005, this does not mean that 20% of expenditures on toiletries in 2014 can be interpreted as corresponding to three pieces of soap. One could, however, say that BPS applies what Esmara called the ‘dynamic poverty line’, by changing from round to round the reference living standard for poverty measurement.