The determinants of poverty dynamics in Indonesia: evidence from panel data

Abstract

We use the ‘spell’ approach to identifying poverty and apply an ordered logit model to examine the determinants of poverty dynamics in Indonesia, categorising households as poor, transient poor (–), transient poor (+) or non-poor. Observing the National Socio-Economic Survey (Susenas) balanced-panel data sets of 2005 and 2007, we found that 28% of poor households are classified as chronically poor (that is, remaining poor in two periods) while 7% of non-poor households are vulnerable to being transient poor (–). Our estimations confirmed that the determinants of poverty dynamics in Indonesia are educational attainment, the number of household members, physical assets, employment status, health shocks, the microcredit program, access to electricity, and changes in employment sector, employment status and the number of household members. We also found that households in Java–Bali are more vulnerable to negative shocks than those outside Java–Bali.

Keywords:

• poverty dynamics
• transient poverty
• vulnerability
• shocks
• government assistance

Acknowledgements

We would like to thank the University of Indonesia and the Directorate General of Higher Education, Ministry of National Education, the Republic of Indonesia, for financing this research through the National Research Strategic Fund 2010 (DRPM/Hibah Strategis Nasional/ 2010/I/4024). We would also like to thank Ms Lily Yunita and Mr Usman from the Institute for Economic and Social Research, University of Indonesia, for their assistance. We would like to thank Professor Mohamad Ikhsan (University of Indonesia) for providing Susenas panel data sets. Professor Shigeru Otsubo (Nagoya University) and his seminar participants provided valuable comments, as did Professor Hal Hill, and other participants of the 2011 Singapore Economic Review Conference, and Assistant Professor Mark Rebuck. We would also like to thank the three anonymous referees for their constructive and valuable comments and suggestions, which helped improve the quality of this paper. Any remaining errors are our responsibility.

Notes

¹For example, Contreras et al. (2004) found that health problems were correlated with households falling into poverty in Chile. Dercon and Krishnan (2000) showed that risk contributes to poverty fluctuations in Ethiopia.

²Sparrow, Suryahadi and Widyanti (2012), using the National Socio-Economic Survey (Susenas) panels of 2005 and 2006, showed that Askeskin increases the use of outpatient health care among the poor. This policy may therefore protect households from falling into transient poverty because of health shocks.

³Dercon and Shapiro (2007) contend that the impact of shocks and risks on poverty mobility has received relatively limited attention in the literature of poverty dynamics. Analysing poverty dynamics can provide insights into the effects of socio-economic and antipoverty policies on household poverty status. It can also help policy makers to identify effective ways to help households escape poverty.

⁴We intended to use a longer sequence of Susenas data sets (for instance, 2002–07), to capture greater changes in poverty status. The 2002 and 2007 databases do not match, however, because Susenas modules collect information from different categories every three years. We also found many inconsistencies in the 2006 data, so we could not include them. Analysing poverty dynamics using panel data covering a short period (three years) may not reveal all of the long-run changes in poverty. Given the limitations of available data, however, analysing a short period of poverty dynamics using a Susenas data set that provides rich information about household socio-economic conditions and covers all provinces will nevertheless contribute to a deeper understanding of the recent situation of poverty. It will also provide useful insights into why some households remain poor and why others escape poverty.

⁵In merging the 2005 and 2007 sample identifiers of Susenas core and module data sets, we found 9,491 balanced-panel samples. Around 1,120 samples were lost during the merger, which might be due to a split of provinces. We not only merged the sample identifiers but also included household information such as educational attainment, physical assets, shocks and the poverty line.

⁶ The FGT class of poverty measures follows: where π is the poverty index, n is the total population size, z is the poverty line, y_i is the income of the i ^th individual (or household), q represents the number of individuals just below or on the poverty line, and α is a parameter for the FGT class.

⁷According to BPS, the 2005 and 2007 Susenas panel data sets should be presented at the national, rural and urban levels but not at the provincial level. However, there is still the possibility and validity of analysing at the regional level both Java–Bali and outside Java–Bali, because the 2005 and 2007 Susenas balanced panel had been distributed proportionally between Java–Bali (4,626 households) and outside Java–Bali (4,100 households). A regional analysis would follow Suryadarma et al. (2006), who used the 2002 and 2004 Susenas panel data sets to analyse the level of access to basic services at regional level.

⁸Negative shocks and risks include economic risks and health shocks. Positive shocks include improvements to public facilities surrounding living areas, more new jobs, and microcredit. Economic risks include crop loss, job loss, falling crop prices and increased production costs. This vector also includes interaction variables between savings and socio-economic shocks, and the policy variables of Raskin and Askeskin. These variables are intended to examine the effectiveness of saving and government policies in coping with negative shocks.