Intragenerational Economic Mobility in Indonesia: A Transition from Poverty to the Middle Class in 1993–2014

Abstract

Economic mobility, especially through expansion of the middle class, will dominate the future of Indonesia’s development agenda. Based on data from five waves of the Indonesia Family Life Survey (IFLS), we found that (1) poverty decreased significantly between 1993 and 2014, from 86.1% to 20.2%, while the middle class grew by almost nine times; (2) 34.4% of the poor moved into the middle class, but 11.9% were still categorised as chronically poor; (3) 42.3% of the middle class did not move into the upper class; (4) the middle and upper classes are vulnerable and easily fall into the lower classes. Our econometric estimations confirm that the drivers of economic mobility are educational attainment, formal employment, water and electricity supply, land ownership, and health investment. These findings suggest that investment in human and physical capital are the two main strategies to expand the middle class.

Mobilitas ekonomi, khususnya melalui ekspansi kelas menengah, akan mendominasi masa depan agenda pembangunan ekonomi Indonesia. Berdasarkan data dari lima gelombang Indonesia Family Life Survey (IFLS), penulis menemukan bahwa (1) angka kemiskinan menurun drastis pada periode 1993-2014, dari 86,1% ke 20,2%, sedangkan kelas menengah bertumbuh hampir sembilan kali lipat, (2) sebanyak 34,4% masyarakat miskin naik menjadi kelas menengah, meskipun 11,9% masih masuk kategori golongan miskin kronis, (3) sebanyak 42,3% kelas menengah tidak pernah naik menjadi kelas pendapatan tinggi; (4) kelas menengah dan kelas pendapatan tinggi rentan jatuh ke kelas ekonomi di bawahnya. Estimasi ekonometrika yang dilakukan penulis mengkonfirmasi bahwa penentu mobilitas ekonomi adalah pencapaian pendidikan, status pekerjaan formal, penawaran air dan listrik, kepemilikan lahan, serta investasi kesehatan. Temuan ini mengusulkan perlunya investasi sumber daya manusia dan modal fisik sebagai dua strategi utama untuk menambah kelas menengah.

Keywords:

• poverty
• poverty dynamics
• middle class
• upper class
• economic mobility
• intragenerational
• transient poverty

JEL classifications:

• I32
• R20
• Z13
• A14

ACKNOWLEDGMENTS

The authors thank the two referees who provided valuable and insightful suggestions for improving this article; and the JICA Research Institute, and the Universitas Indonesia for providing financial support for this article (NKB-0190/UN2.R3.1/ HKP.05.00/2019). We also extend our thanks to Dr Jossy P. Moeis for helping us calculate household expenditure using 2014 IFLS data, and the participants of the Indonesian Bureau of Economic Research’s special session at the Indonesian Regional Science Association’s 2018 conference, for their insightful comments. All errors are ours.

Notes

1 The headcount index of Statistic Indonesia (BPS) indicates that the proportion of the Indonesian population lving below the national poverty line fell from 21.60% in 1984 to 10.12% in 2017. The progress of poverty reduction seems slow, since the rate of poverty appears to have fallen by only half in the past 30 years. However, it is inappropriate to compare these two poverty figures, because of revisions to the methodology used to calculate them (see Priebe 2014). For instance, applying the 2017 poverty line to the 1996 Susenas results in a poverty rate of 17.47%, whereas applying the 1984 poverty line results in an 11.3% poverty rate. Therefore, we use an international standard poverty measure in this paper to achieve consistency in welfare measurement.

2 The later samples may not represent the same share of the population, because of attrition, split-off households and migration.

3 Applying a two-sample t-test, we found significant differences between the means of some household characteristics in the attrited and non-attrited samples. However, our later econometric estimation shows that the attrition does not significantly change the result. The online appendices can be found at http://dx.doi.org/10.1080/00074918.2019.1657795.

4 The individual attrition rate may be higher than the household attrition rate.

5 Information on the weighting methodology is available from the corresponding author upon request.

6 Although this study applies consistent thresholds over time ($1.9, $3.2, $5.5 and $15.3), the real values of thresholds seem inconsistent and decrease over time (see table 1). This is because the consumer price index (CPI) increased faster than the PPP index: in 1993–2011, the CPI increased 7.6 times, while the 2011 PPP increased only 4.65 times. Comparing the real value of thresholds should be done with the same base year between PPP dollars and CPI. However, the adjustment of the base year should be conducted carefully, since, without any correction, change will also create an unnecessarily biased calculation. Kakwani and Son (2016) discuss in detail how to construct a consistent poverty line in PPP dollars in both monetary and real value.

7 The steps to construct the new poverty line are (1) applying the BPS poverty lines for 2000, 2007 and 2014 to the IFLS data set to calculate the incidence of poverty; (2) scaling the BPS poverty lines to achieve the same poverty rate as the one published by BPS—the BPS poverty lines for 2000, 2007 and 2014 had to be adjusted by 106.17%, 120.86% and 114.75%, respectively; (3) choosing a reference year (either 2000 or 2014), as we use the absolute approach to construct the new poverty line; (4) adjusting the new 2000 or 2014 poverty line with the consumer price index to obtain the constructed poverty lines for 1993, 1997, 2000, 2007 and 2014; and (5) calculating the extent of poverty, and the size of the emerging, middle and upper classes, based on the 2000 and 2014 reference years, in order to check for robustness and consistency.

8 See online appendices 3A, 3B and 3C for the proportion of households in each welfare category, based on different thresholds (http://dx.doi.org/10.1080/00074918.2019.1657795). The proportions vary depending on the thresholds used, though the trends are similar. The results of welfare classification are also sensitive to the choice of reference year. For example, using a reference year of 2000 leads to the underestimation of the size of the poor and vulnerable groups; using 2014 leads to underestimation of the size of the middle class. This study recommends using consistent thresholds over time to minimise biased measurement of welfare classification.

9 For example, a middle-class household that falls into poverty five times is considered worse off than a middle-class household that falls into poverty four times. According to the spell approach to determining economic status, households that experience poverty in all periods of observation (five times) can be categorised as chronically poor; those that never experience poverty can be categorised as never poor. Because these households can be ordered in such a way, an ordered logit model is appropriate.

10 To distinguish between classes such as the poor and middle class, this study applies an ordered (multinomial) logit model instead of an ordered probit model. When selecting a discrete choice model, the choice between a logit and a probit model is generally considered unimportant.

11 As we believe that there is no certain order of groups in terms of the transition from being poor in 1993 to being in a higher class in 2014, the multinomial logit is the most appropriate approach.

12 The independent variables are drawn from only four waves of the IFLS in order to avoid problems of endogeneity; the dependent variable (welfare condition) represents the welfare conditions of households surveyed in IFLS 1–5. It seems that the welfare conditions from 1993 to 2014 were influenced by the conditions during 1993–2007. This approach could not resolve all endogeneity problems, because unobserved heterogeneity could not be completely identified. We therefore need to be cautious in interpreting causal effects from the results.

13 Online appendix 4 shows other results estimated using unified thresholds without considering regional variation (see http://dx.doi.org/10.1080/00074918.2019.1657795). These results are consistent with the estimation results in table 4. Unlike the proportion of household welfare classification that is sensitive to applied thresholds, the determinants of poverty and middle-class dynamics are robust, stable, consistent and insensitive to thresholds.