Individual Income Inequality and Its Drivers in Indonesia: A Theil Decomposition Reassessment

See, for instance, Ahluwaria (1976a, b), Robinson (1976), Blejer and Guerrero (1990), Anand and Kanbur (1993), and Jha (1996), among many others.

The literature on this subject is large, and the review of past studies in this paper is less than exhaustive. More recent studies on inequality in Indonesia include, but not limited to, Nugraha and Lewis (2013) and Akita et al. (2011). The former re-assesses inequality measures in Indonesia by examining market and non-market income. The latter examines the determinants of interprovincial income inequality associated with structural change.

Several other studies that produce disaggregate evidence in developing countries, apart from Indonesia, include Van Cao and Akita (2008) for Vietnam, Baliscan and Fuwa (2004) and Estudillo (1997) for the Philippines, Akita (2003) for China, Adams (2002) for Egypt, Adams (1994) for rural Pakistan, Mishra and Parikh (1992) for India, and Glewwe (1986) for Sri Lanka.

A rise in income inequality together with a decrease in poverty incidences has been observed in many developing Southeast Asian countries such as Cambodia, Lao PDR, the Philippines, and Vietnam, in addition to Indonesia (ADB 2012).

Figure 1 underlines that structural factor such as economic policies and financial crises may be a key driver of income inequality in Indonesia. However, assessing the structural change requires different empirical approaches from the Theil decomposition analysis which captures contributions of individual attributes to overall inequality at a point in time.

The IFLS dataset is downloadable at: http://​www.​rand.​org/​labor/​FLS/​IFLS.​html.

The first wave of IFLS was fielded in 1993–1994 by RAND Corporation and Lembaga Demografi, University of Indonesia. The second wave was implemented in 1997–1998 by RAND Corporation in collaboration with University of California, Los Angeles and Lembaga Demografi, University of Indonesia, covering about 25 % of the IFLS samples. The third wave covered the full sample and was fielded in 2000 by RAND Corporation and the Population Research Center, University of Gadjah Mada.

Although the IFLS samples do not cover the entire Indonesian population, it can satisfactorily serve as a nationally representative dataset. For instance, in the 2007 IFLS, urban samples account for 54 %, while the national figure for urban population in 2010 is 44 %. For gender composition, males make up around 65 % in the 2007 IFLS while the national figure in 2010 is around 50 %. Likewise, in comparison with the Population Census 2010, the interprovincial components are satisfactorily close to the national figures.

Like the SUSENAS dataset, the IFLS dataset employs a stratified random sampling technique whereby the selection of households (and thus individuals) is made by classifying enumeration areas into strata, choosing several areas from each stratum and surveying households from each of the selected enumeration areas.

The examples of income inequality measures that are decomposable are Atkinson, Theil and Log-Variance. See Bourguignon (1979) and Shorrocks (1980) for the detailed discussions.

Some words of caution should be highlighted. First, Theil measures of inequality based on information theory is rather normative and thus in contrast to other alternative concepts like the Lorenz Curve. Additionally, the levels of inequality under Theil’s measurement are influenced by a member with higher income.

One limitation of the Theil decomposition technique is that it measures only the individual effects of each characteristic on overall income inequality and is thus unable to capture the interplay across different factors. For instance, education investment can reduce inter-provincial inequality, thereby ameliorating overall inequality. Although it is possible to confine samples by province and capture the effects of education investment on inequality in each province, the Theil decomposition does not yield conclusion about the second-round effects on overall inequality.

Another vital spatial dimension is inequality within municipalities/districts. As discussed in Sect. 5.1, Indonesia’s decentralization reform in 2001 empowered local governments and redirected resources toward municipal areas. However, limited scope and data availability of this paper necessitates future research to come back and address this issue more rigorously.

The sample size across provinces is unbalanced ranging from 22 respondents in Kalimantan Selatan to 3,645 respondents in Jawa Barat. Sample size variation reasonably represents differences in population size across provinces. However, the relatively limited samples in some provinces may not be sufficiently large to represent the entire population, resulting in biased estimates. One way to mitigate this problem is to exclude provinces with a small number of samples from the estimation; however, this comes at a cost of losing information on some provinces.

The island of Java accommodated 60 % of the total population and contributed 56 % to total gross domestic product (GDP). In the island of Java, Jawa Barat is the most densely populated province in Indonesia.

Recent studies such as Yusuf et al. (2014) posited that interregional inequality in Indonesia is more pronounced across districts than that across provinces. To inspect this, we perform the Theil decomposition across 196 districts in the 2007 IFLS and find that inequality across districts contributes approximately 13.7–15.5 %—significantly higher than the contributions of interprovincial inequality and, as shown in Sect. 5.4, as high as income inequality across education levels.

Like inequality in income, inequality in education can also be measured by the Gini coefficient that captures unevenness of the educational attainment distribution among the population.

The detailed calculation and results are downloadable at: https://​drive.​google.​com/​folderview?​id=​0B44A5V96WsRWUmZ​LSFdMbmNtanM&​usp=​drive_​web.

As in Sect. 5, decomposition by districts, in lieu of provinces, yields the result that inequality across districts explains 20.6–23.0 % of total expenditure inequality. This is consistent with Yusuf et al. (2014) and implies that in the context of Indonesia, interregional inequality emerges more significantly along the district level.

Expenditure inequality across districts also appears to be much more significant in the eastern provinces. It constitutes as much as 31–46 % of total expenditure inequality.

The school enrolment ratio should be interpreted with caution. It is possible that low school enrolment at the secondary and tertiary levels are attributable to the fact that the effects of education investment will take long time for the preprimary and primary cohorts to pass and enroll in the secondary and tertiary levels. However, this extent should be relatively modest in the context of Indonesia because school enrolment at the secondary and tertiary levels are considerably lower than that at the preprimary and primary levels, especially in comparison with the developed nations’ figures. That said, the low school enrolment ratios at the secondary and tertiary levels are more likely to be explained by inadequate access to education (e.g. inadequate number of schools and teachers and lack of education facilities).