سیستم توزیع عمومی هند به عنوان ارائه دهنده امنیت غذایی : مدارک و شواهد از تغذیه کودک در آندرا پرادش

We study whether a sudden increase of the price of rice supplied by the Indian Public Distribution System in Andhra Pradesh, a large Indian state, had a negative impact on child nutrition. A few months after the price increase, a health survey started to record weight for a large sample of children. The data collection continued for several months, so that children measured later lived for a longer period of time in a less favorable price regime. Using different estimation techniques we find that longer exposure to high prices are not accompanied by worse nutritional status, as measured by weight-for-age.

Food subsidization has a very long tradition in India. For most of the last three decades, it has accounted for more than 2 percent of total government expenditure, and its cost peaked in 1993–1994 at 55 billion Rupees (roughly 1.8 billion 1993 US$), almost 50 percent of total expenditure allocated to poverty alleviation programs, and approximately 0.8 percent of Gross Domestic Product (Radhakrishna et al., 1997). The bulk of these sums sustains the Public Distribution System (PDS hereafter), which mainly supplies rice, wheat, edible oils, sugar, and kerosene at subsidized prices, through a network of retail outlets known as Fair Price Shops. The main proclaimed task of the PDS is to provide food security to poor households, but corruption, inefficiencies, and limited scope are widely believed to prevent the system from fulfilling its goal. The existing literature does include several insightful descriptive studies of the program that analyze its functioning and ability to reach the poor, and provide estimates of the implicit subsidy it offers.1 However, notwithstanding the size of the program and its importance in policy debates in India, an evaluation of the efficacy of the PDS in improving the nutritional status of the beneficiaries is still missing. The main goal of this paper is to contribute to fill this gap, by studying the relation between PDS and children's nutritional status, as assessed through anthropometric measurements. The evaluation is complicated by the fact that take-off rates and program placement are not randomly assigned. As a consequence, a database recording both purchases from the PDS and child nutritional status would not easily allow to identify any causal link between the two variables. If Fair Price Shops are concentrated in poorer areas, one might observe a negative relation between purchases from the PDS and nutritional status, even if the former has a positive impact on the latter.2 Similarly, a negative bias might arise if—as it is the case—purchases from the PDS are concentrated among poorer households, which in India frequently face poor nutrition, as well as health and epidemiological hazards. The picture is further complicated by the presence of other factors—such as preferences—that may cause unobserved systematic differences between eligible households that purchase from the system and those that do not. The literature on program evaluation sometimes eludes the selection problem making use of natural experiments, whereby policy changes translate into arguably exogenous variations in the level of the examined treatment.3 This is the path we follow here, and the natural experiment is provided by a sudden drop in the level of the subsidy offered by the system that took place in the state of Andhra Pradesh. At the beginning of 1992—with effect starting January 27th—the State Government of Andhra Pradesh almost doubled the price of subsidized rice, which had been kept for years at levels much below the market price (Krishna Rao, 1993). In April of the same year, the National Family Health Survey (NFHS hereafter) began collecting data on fertility, family planning and health from a sample of ever married women in fertile age, recording also weight and age of all children up to age four. Since the collection of data continued for several months, children measured later in the survey lived for a longer period of time in a less favorable PDS regime. We will show that most children in the sample show poor anthropometric performances, and that there is ample evidence that weight typically reacts quickly to changes in the nutritional status. So, if the subsidy offered by the system really helped providing food security, one should expect lower weight in children measured later in the survey, given their age and sex. Another, methodological, goal of this paper is to show how the very length of the process of data collection in a household survey can be exploited to identify the effect of a treatment on an outcome measured in the survey itself. Other papers have used the timing of the survey with respect to a policy change to evaluate the impact of a program (see, for example, Duflo 2001 and Duflo 2003). Here, we show that the point in time within the survey in which the outcome has been measured can be used to study the causal relationship between the policy change and the outcome of interest. This is possible if the outcome responds quickly to a change in the economic environment, and if the data collection continues for a long enough period of time. The appropriateness of such an identification strategy can be undermined if the outcome is affected by seasonality, or if the subsamples examined in different time periods are systematically different. In our context both issues are potentially relevant: to deal with the former we introduce difference-in-differences estimators, while we control for a series of child and household specific variables to solve the latter. Since the NFHS contains information on asset ownership, but not on income or expenditure, we control for household wealth using an asset index constructed by applying principal component analysis. Admittedly, confining the analysis to the rice scheme in Andhra Pradesh limits the generality of our evaluation. However, there are three factors that mitigate such limitations. First—as we will show—the sale of subsidized rice is by far the most important component of the PDS in Andhra Pradesh. Second, this is one of the very few states in which the implicit subsidy offered by the PDS is thought to be relatively large. There are important interstate differences in the scope and functioning of the PDS, and in most states its relevance is so modest that the task of identifying an effect on nutrition would be practically hopeless.4 Third, Andhra Pradesh is one of the few states where targeting towards the poor was already present during the period considered in our evaluation. This makes our evaluation more interesting from a current perspective, because in most states targeting was only introduced in June 1997, when a new set of guidelines for the PDS were issued by the Government of India. Our results show that for subsidy withdrawals that we estimate to be equivalent, for the poorest households, to about 5 percent of total budget, we cannot detect any effect on child nutrition, as measured by weight-for-age. We find that the effect of a longer exposure to a less favorable PDS regime is negative only for females, and even in this case it is small and statistically insignificant. These results are robust to the use of different specifications. This casts further doubts on the ability of the Public Distribution System to provide food security to the most vulnerable sections of the Indian society. The paper is organized as follows. The next section presents a brief description of the PDS in India, and in Andhra Pradesh in particular. Section 3 describes the data used in the paper, briefly describes the construction of the asset index we use as a proxy for household wealth, and lays out in detail our identification strategy. Section 4 provides evidence of the poor nutritional status of children in Andhra Pradesh, while Section 5 evaluates the size of the support provided by the PDS to poor households in this state. In Section 6 we present a preliminary exploration of our identification strategy, using nonparametric estimation. Parametric and semiparametric estimators are introduced in Section 7, and Section 8 concludes.

In this paper, we investigate whether a large and sudden reduction of the implicit subsidy offered by the Public Distribution System in Andhra Pradesh had negative effects on children's weight-for-age. Our identification strategy uses the fact that the NFHS recorded children's weight during a 4-month period, form April to July 1992, while the increase in PDS rice happened at the end of January in the same year. Given their age, children measured later in the survey have been exposed to a longer period of high prices. Using a range of different estimators, we find that a subsidy reduction estimated to be equivalent to about 5 percent of the poorest households’ total budget, did not affect child nutrition, as measured by weight-for-age. We find a negative impact only on girl weight, and even in this case the effect is small and not statistically significant. This suggests the important possibility that relatively large changes in food subsidies can have little or no effect on child nutrition. We can think of different—possibly concurring—explanations for this negative finding. First, even if the changes in the rice scheme reduced considerably the subsidy, the size of the subsidy itself might have been too small for its drop to be detected by our estimates, given the relatively small size of our sample. In Section 5 we showed that, even among the poorest households, a return to the more generous former rice scheme might have been associated with a relative increase in calorie availability per child as small as 2–4 percent, possibly lower, if one accepts the smallest calorie–income elasticity estimated in the literature.35 In other words, even if weight responds quickly to changes in nutrition, it is possible that nutrients consumption itself did not change significantly as a consequence of the changes introduced in the PDS regime. However, the lower elasticities are not widely accepted in the field (e.g. Deaton, 1997, Chapter 4), and other studies (e.g. Rand et al., 1985) suggest that even small discrepancies between calorie intake and physical activity can lead to significant changes in weight over a short period of time. Moreover, anthropometric measurements are the results both of nutrient intakes and of the epidemiological environment in which the child lives, so one might expect weight-for-age to respond to a drop in income even through channels different from nutrition. For example, the household might have to reduce health-related expenditure, or have to resort to less safe sources of drinking water. So, our results suggest a low income elasticity of nutritional status, as measured by weight-for-age, and do more than merely corroborate the low income elasticity of calorie intake sometimes suggested in the literature. Second, because the changes took place only 3 months before the starting of the NFHS, we cannot exclude that even the poorest households have been able to cope with the reduction in subsidy, making use of accumulated savings, or reducing the budget share devoted to non-food items. Third, it is possible that child nutrition was buffered at the expense of adult consumption of food. In particular, given the very young age of the children in our sample, it is possible that breast-feeding protected them, at the expense of the mother's nutritional status. This hypothesis cannot be tested directly using our data, since anthropometric measurements were not taken for the mothers. However, our conclusions do not change significantly if one performs a separate analysis for children aged up to 24 months, and for older children, for which breast-feeding is much less likely to play an important role. Fourth, one might be worried about attenuation bias due to error in measuring exposure to the high price regime. In fact, our methodology does not distinguish between, for example, the exposure of a child born at the end of December 1991, and measured at the end of June of the following year, and that of another for which birth and measurement took place at the beginning of the same 2 months. Both children can have the same age—in months—and both have been measured in June, but there is approximately a 1-month difference in actual exposure to the high-price regime, which started at the end of January 1992. Therefore, as a robustness check, we have repeated the estimations calculating the period of exposure to the high-price regime in days (by making use of the day of measurement reported in the survey), but this does not change any conclusion.36 With all these caveats, our results cast further doubts on the ability of the Public Distribution System to offer food security to poor households, which is its primary proclaimed objective. Even in a state like Andhra Pradesh, where the system is working relatively well, we do not find evidence that this task is being achieved.