Economic shocks prompt parents to work more in order to reduce the fallout, and children may spend more time helping in the household or on the family farm, thereby interrupting their schooling. Analysing data from rural India, this article shows that a 100% increase in risk – in the form of more volatile agricultural production and consumption – reduces the probability that a child attends school by 4-5%.
In economics, we have long considered the costs of risk – for instance, forward-looking farmers may forego profitable investment opportunities if they are uncertain about the possibility of realising the return to such an investment in the future (Rosenzweig and Binswanger 1993, Karlan et al. 2014). Likewise, individuals are less willing to engage in entrepreneurial activities if exposed to more risk (Bianchi and Bobba 2013).
When we think about schooling – which is also a form of investment as parents have to pay for school materials, uniforms, tuition, and while children are in school they cannot help out in the household or on the family farm – we tend not to attribute too much importance to uncertainties in the process. After all, children go to school every day, and eventually they will have accumulated sufficient knowledge to complete schooling. Right?
Unfortunately, this is not quite true for millions of school-aged children. Even without a global pandemic, schooling can be disrupted for a multitude of reasons – due to illness, for example, or because children need to support their families in the aftermath of economic shocks. When wages are low, parents try to work more to reduce the economic fallout of the shock (Jayachandran 2006). For children, that often means that they have to spend more time helping in the household or on the family farm. And every time schooling is disrupted, it becomes more difficult for children to catch up with the course work they missed.
Is it possible that this risk of having to interrupt schooling potentially keeps parents from sending their child to school in the first place?
Risk and schooling decisions in rural India
In recent research, we investigate if children who are exposed to higher income risk and as a consequence, are more likely to interrupt schooling at some point during the school year, are less likely to be enrolled in school at the start of the year (Foster and Gehrke 2020). We first approach the question theoretically, and model household investment decisions in child human capital over two periods, allowing for parental uncertainty about the ability to follow up with schooling investments in the second period1. To account for the cost of schooling interruptions in the dynamics of learning, we let consecutive investments in child human capital be complementary to each other, that is, an investment in child human capital in the beginning of the school year only pays off, if the parent can invest similar amounts continuously throughout the school year. The model predicts that an increase in uncertainty (a higher probability that the child might have to interrupt schooling) discourages human capital investments in the first period, for fairly standard assumptions about parental preferences.
To understand the importance of this phenomenon in rural India, we then combine data from three rounds of the Rural Economic and Demographic Survey (REDS) with high-resolution rainfall data obtained from the European Centre for Medium-Range Weather Forecasts (ECMWF). These data are unique in that they include time allocation of children and mothers for three seasons during the year in each round2. The REDS data gives us schooling information for approximately 7,500 children between the years 1981-82 and 2007-08. These children are aged 6-15 at the time of the interview, and almost all of them have some duties in terms of domestic work, on the family farm, or in paid work.
The REDS data are broadly representative of rural India and span multiple decades, allowing us to use long-term changes in the risk structure across villages as a source of variation. Our concept of ‘risk’ exploits three key facts from rural India. First, agriculture is the dominant source of income for the majority of households over the time period of our study – either through own production or through casual work on other farms. Second, agriculture is inherently risky – agricultural yields, wages, and employment levels in rural India are strongly influenced by rainfall conditions (see, for example, Jayachandran 2006, Shah and Steinberg 2017, Kaur 2019). High rainfall leads to good harvests, high demand for workers, and high wages. In contrast, low rainfall levels lead to poor harvests, low demand for agricultural workers, and low wages. Third, the use of irrigation in agricultural production expanded rapidly in rural India over the last decades, which reduced the reliance of agricultural production on good weather conditions in some places, but not in others (Duflo and Pande 2007).
To construct the risk measure, we estimate the relationship between household consumption expenditures, rainfall, and village-level irrigation, allowing for the effect of rainfall to vary with the extent of irrigated area. We find that a 10% decline in rainfall decreases consumption by 2.3% in villages without any irrigation. In contrast, in fully-irrigated villages, a decline in rainfall has an almost zero effect on household consumption. The ‘interaction’ between rainfall and irrigation also influences child time use – less rainfall means less time in school in the study sample, and the magnitude of the shock is mediated by a higher share of irrigated area in a particular village4. With these estimates, we can construct the probability distribution of consumption expenditures in a specific village at a certain point in time – the village’s historical rainfall distribution gives us the probability of rainfall outcomes, and the current availability of irrigation determines how strongly a given level of irrigation would determine consumption. The spread in this distribution4 is our time-varying village-level measure of risk.
We explore changes in village-level risk5 within villages over time, to estimate the effect of risk on study time, and find that risk significantly reduces the probability that children attend school. Our findings imply that a 100% increase in risk would reduce the probability that a child attends school by 4-5%. These results are very robust to various specifications, including controls for household wealth, three lags of rainfall (that is, rainfall in the previous three years) to account for the effect of past shocks on the current stock of child human capital), and state-specific shocks. We also investigate a range of alternative explanations, for example, parents may be required to hold a higher stock of savings when exposed to more risk and may be less willing to spend their limited income on schooling, or returns to schooling might be differentially affected by village-level risk. However, we cannot find any evidence for these to play a role.
What can public policy do?
In order to assess the scope for policy, we also simulate the effects of an income-smoothing policy, specifically the Mahatma Gandhi National Rural Employment Guarantee Act (MNREGA)6, on schooling investments. We estimate the extent to which MNREGA reduces variability in consumption by augmenting the REDS data with the 2014-2016 follow-up of the REDS survey, and exploiting differences in programme intensity (the number of person-days of MNREGA employment generated per capita) across villages as source of variation. We find that MNREGA mediates the effect of rainfall on consumption expenditures, just as irrigation does. We can use these estimates to simulate the programme’s risk-reducing and attendance-enhancing effect, and find that a similar programme, that held the level of wages fixed, would increase school attendance by 1 percentage point.
However, the wage effects of the programme should not be ignored. Shah and Steinberg (2019) and Li and Sekhri (2020) find negative effects of MNREGA on schooling using the rolled phase-in of the programme as a source of variation, and argue that MNREGA raised wages and led adolescents to substitute away from school and to join the labour force at younger ages. Arguably the risk-reducing effect of MNREGA could not be fully internalised by households at the early stages, when its implementation was spotty and long-term viability unknown. It is unclear how the direct effect of rising wages and the indirect effect of less variable incomes balance out in the longer term. However, our results suggest that the negative wage consequences of MNREGA on schooling might be mitigated to the extent that the programme is recognised as a reliable source of support during periods of adverse shocks.
This simulation highlights the benefits of programmes that support low-income households in developing countries, in dealing with various kinds of risk. Income transfers, either in the form of cash transfers or public works programmes, could increase investments in human capital by reducing the probability that parents (or the student) cannot follow up with early levels of schooling investment in later time-periods, as long as these are carefully designed. Welfare effects can be maximised by making access to work transparent and reliable, and by differentially targeting such programmes to villages that are more regularly affected by adverse shocks.
I4I is now on Telegram. Please click here (@Ideas4India) to subscribe to our channel for quick updates on our content
- For example, the parent has to enrol the child in May-June but only knows if the child needs to help with the harvest in September.
- The three seasons refer to planting, harvesting, and slack season. In the questionnaire, the reference months are October/November (season one; harvesting season for rice cultivators), February (season two; slack season), and April/May (season three; planting season for rice cultivators).
- See Foster and Gehrke (2020) for more details.
- We use the standard deviation and the interquartile range as two distinct measures. Standard deviation is a measure that is used to quantify the amount of variation or dispersion of a set of values from the average of that set. Interquartile range is the difference between the 75th and 25th percentiles of the data.
- For example, say villages with higher risk are more likely to see a diversification out of agriculture and the expansion of industries. In such a case, the wage returns to education would be different between villages with high risk and villages with low risk. However, we find no evidence that risk is systematically related to the wage returns to education, the presence of factories in a village, or the share of households engaged in agricultural production.
- MNREGA guarantees 100 days of wage-employment in a year to a rural household whose adult members are willing to do unskilled manual work at the prescribed minimum wage. An objective of MNREGA was to provide work in rural areas outside of the agricultural seasons.
- Bianchi, Milo and Matteo Bobba (2013), “Liquidity, risk, and occupational choices”, The Review of Economic Studies, 80(2): 491-511. Available here.
- Duflo, Esther and Rohini Pande (2007), “Dams”, Quarterly Journal of Economics, 122(2): 601-646. Available here.
- Foster, A and E Gehrke (2020), ‘Start What You Finish! Ex Ante Risk and Schooling Investments in the Presence of Dynamic Complementarities’, NBER Working Paper No 24041.
- Jayachandran, Seema (2006), “Selling Labor Low: Wage Responses to Productivity Shocks in Developing Countries”, Journal of Political Economy: 114(3): 538-575. Available here.
- Karlan, Dean, Robert Osei, Issac Osei-Akoto and Christopher Udry (2014), “Agricultural decisions after relaxing credit and risk constraints”, Quarterly Journal of Economics, 129(2): 597-652. Available here.
- Kaur, Surpreet (2019), “Nominal wage rigidity in village labor markets”, American Economic Review, 109(10): 3585-3616.
- Li, Tianshu and Sheetal Sekhri (2020), “The Spillovers of Employment-based Safety Net Program on Child Labor and Education”, The World Bank Economic Review, 34(1): 164-178.
- Rosenzweig, Mark. R and Hans P Binswanger (1993), “Wealth, weather risk and the composition and profitability of agricultural investments”, Economic Journal, 103(416): 56-78.
- Shah, M and BM Steinberg (2017), ‘Good monsoon, bad test scores? Substituting away from schooling’, Ideas for India, 27 June.
- Shah, Manisha and Bryce Millett Steinberg (2017), “Drought of opportunities: Contemporaneous and long-term impacts of rainfall shocks on human capital”, Journal of Political Economy, 125(2): 527-561. Available here.
- Shah, Manisha and Bryce Millett Steinberg (2021), “Workfare and Human Capital Investment: Evidence from India”, Journal of Human Resources, 56(2): 380-405. Available here.