Long-Term Effects of Temporary Income Shocks on Food Consumption and Subjective Well-Being

Abstract

A national level unconditional cash transfer programme, Bantuan Langsung Tunai (BLT), in Indonesia allows an empirical assessment on whether there are long-term benefits in terms of food consumption and overall well-being. The results show that a positive albeit temporary income shock increases the quantity of food consumed by the poorest households and the overall subjective well-being among the poorest recipients. It is also found that poor households are more likely to invest in farm and non-farm businesses, which in turn helps them sustain a higher level of food consumption and overall satisfaction months after the end of the programme.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1. This is true only if transfers are regular and predictable (Tiwari et al., 2016).

2. One being the rollout of the programme to the control group as well, leaving no possibility of a rigorous impact evaluation. Studies that do look at longer-term impacts focus on the comparison of the outcomes for early and later treatment groups (Gertler, Martinez, & Rubio-Codina, 2012).

3. The post-treatment data comes from the fourth wave of IFLS (discussed in detail in the Data and Context section). The data was collected between November 2007 and March 2008. This means that there is around 24 months between the first transfer and the last month of the survey.

4. Bazzi et al. (2015) uses the Susenas from 2005 as baseline and that from 2007 as post-treatment data. The post-treatment data, collected in February/March 2007, provides seven/eight months after the end of BLT in July 2006. In contrast, the post-treatment data in this paper provides 16–20 months after the end of BLT.

5. The analysis on attrition is done using a probit model where the dependent variable identifies households that were present in the pre-treatment survey but were absent in the post-treatment survey. The main independent variable is the treatment variable. The purpose of the analysis is to check whether receiving the BLT programme affected whether or not they are in the post-treatment survey. If attrition from the survey is statistically significantly determined by receipt of the programme, any analysis on the impact of the programme would be biased.

6. Yang (2006) uses this threshold and it seems a reasonable estimate given the programme and data.

7. These criteria are, namely, the size of the house in square metres; the flooring material of the house; the material used for house walls; sanitary facilities; source of drinking water; source of main lighting; fuel used for cooking; meat consumption; number of meals; clothing consumption; job of household head; financial ability to go to hospital; and possession of specific assets (Hirose, 2008). Although the data used in this study is not the same as the one used to identify the 19.5 million poor households in Indonesia, the authors have done their best to follow these criteria as closely as possible. Nevertheless, some discrepancies in identification are expected.

8. The estimation of propensity scores relies on household characteristics five years prior to the targeting of the programme since it comes from the 2000 IFLS wave as baseline data. Many of these variables are bound to fluctuate over this time period. This is potentially problematic if they systematically change across the treatment and control, which is unlikely. Further, since only around 2 per cent (117 out of 6011) of households were trimmed out after the propensity score estimation, having more recent baseline data is only likely to have a minimal effect.

9. The IFLS survey asks about 34 food items.

10. Extremely poor households are identified as households whose per day, per capita consumption is below $1.91 PPP. Marginally poor households are identified as households whose per day, per capita consumption is between $1.91 and $2.80.

11. Inverse propensity weights equal $\frac{1}{propensityscore}$ for households in the treatment group and $\frac{1}{1-propensityscore}$ for households in the control group.

12. Bazzi et al. (2015) report that the fuel price shock affected everyone uniformly. Nevertheless, fuel products are a small share of overall household expenditures among both rich and poor (less than 4%). In our sample, fuel expenditure makes up less than 1 per cent for both treatment and control households and there is no statistically significant difference between the two groups.

13. The first wave of IFLS collected in 1993 is used as the pre-treatment data and the third wave collected in 2000 is used as the post-treatment data. This allows around the same time frame between baseline and follow-up in both scenarios.

14. All regressions control for household, household head characteristics, shocks at the household level (such as death and drought), and prices of staples and fuel.

15. This is done to use the space more efficiently. Results on other coefficients are available upon request.

16. For more on the choice of cultivation techniques to secure property rights, see Bandiera (2007).