Does Food Price Subsidy Affect Dietary Diversity? Evidence from South India

Abstract

The present study has tried to address the impact of subsidised rice distribution through the public distribution system on dietary diversity and nutrition intake in the state of Tamil Nadu in India as the state is considered a pioneer in introducing a number of food security programmes in India. We used National Sample Survey Organisation’s data for the years 2004-05 and 2011-12, and the propensity score matching technique to estimate the actual impact of the subsidy programme on food consumption patterns and nutrient intake, as the data-set used for analysis was subjected to non-randomisation and selection bias. The estimated results reveal that the subsidy on rice has significantly and positively impacted food consumption and nutritional intake across households, irrespective of income groups. The increased purchasing power of the poor due to the subsidy is limited to the staple food commodities—rice, millets, pulses and vegetables—whereas middle- and high-income households are more likely to consume high-value commodities such as fruits, processed food and livestock products, with a resultant higher gain in fat and calcium. Our study indicates that extending the price subsidy to nutritious foods, besides rice can help the poor diversify their diets towards healthy and nutrient-rich foods.

JEL Codes: C5, C54, D01, D11, D12, Q11, Q18

Keywords

Food security Nutrient intake Dietary diversity Propensity score matching

1. Introduction

The first three sustainable development goals (SDGs)—no poverty, zero hunger, and improved health and well-being—are directly or indirectly associated with households’ food consumption, dietary pattern and nutrient intake. These goals call for a substantial reduction in the magnitude of undernourishment and malnutrition in many of the developing and underdeveloped countries of Africa and Asia by 2030 (Buse & Hawkes, 2015; Griggs et al., 2013; Weber, 2017). The Indian government has undertaken several measures towards these goals, such as increasing food production, intervening in the procurement and distribution of food to the end consumers, subsidising food prices and creating employment and income generating opportunities to enable the households with reasonable purchasing power to access healthy and nutritious food products. In fact, India has managed to achieve self-sufficiency in food grain production in addition to increasing its production of vegetables, fruits, animal products, etc., considerably through green revolution technologies. To boost food distribution, the country has introduced several food security programmes, including mid-day meals for school children, food for work, the integrated child development programme and the distribution of food grains at subsidised prices through the public distribution system (PDS). The supply of food grains such as rice and wheat at subsidised prices has had a significant impact on addressing food insecurity and malnutrition problems over the decades.

However, the burden of undernutrition, micronutrient deficiency and overnutrition remains high in the country, especially among children and women (Aurino, 2017; Kehoe et al., 2014; Misra et al., 2011); for instance, about 15 per cent of the population suffer from undernourishment and 39 per cent and 20 per cent of children under 5 years are stunted and wasted, respectively. Children from low-income groups and rural areas are often more affected than their counterparts from high-income groups and urban households. Overall, one-fifth of India’s population suffers from chronic hunger, even though the country ranks 100th among 119 countries in the 2017 Global Hunger Index (von Grebmer et al., 2017). On the other hand, problems such as obesity, diabetes and overnutrition are on the rise across all types of households (Food and Agriculture Organisation [FAO], 2017), especially among the higher income and urbanised households.

Most malnutrition-related problems are associated with the quantity of food consumed and the variety of foods featuring in the diet. The literature indicates that diversity in food consumption patterns can help to overcome problems of overnutrition and undernutrition (Arimond & Ruel, 2002; Ekesa et al., 2008; Hatløy et al., 2000; Taren & Chen, 1993). A couple of studies have identified the lack of diversity in regular diets as the primary cause of malnutrition (Chandrasekhar et al., 2017). An inadequately diversified diet in terms of quantity and composition of the food basket is associated with below optimal growth, development and long-term health outcomes (Charmarbagwala et al., 2004). Moreover, the consumption of dairy products, vegetables, fruits and legumes is important for a balanced diet and micronutrient intake. Children, in particular, whose diet consists of more than four food groups are less likely to be underweight or stunted (Desai & Vanneman, 2015). In this context, evaluating the effect of price subsidies on food consumption patterns and nutritional improvement is important, especially in the Indian context where around 1 per cent of the total gross domestic product (GDP) is spent on food security programmes.

Few studies attempting to address the impact of price subsidy programmes on nutritional improvement in India have yielded mixed results. Pingali et al. (2017) observe that although PDS has helped to tackle the problem of hunger, to what extent it has nutritional benefits remains unclear. Some studies in literature have found that the expanded coverage of PDS has not only increased calorie intake but also improved dietary diversity through the income effect and declining poverty rates (Kaul, 2018; Kishore & Chakrabarti, 2015; Krishnamurthy et al., 2017; Rahman, 2016). Some studies suggest that PDS may have led to a shift from more nutritionally superior coarse cereals and millets to the subsidised wheat the PDS system supplies (Khera, 2014), while Shrinivas et al. (2018) suggest that in-kind staple food subsidies can lead to large improvements in nutritional outcomes of poor households. Conversely, Desai and Vanneman (2015) observe that the PDS system has had no effect on the consumption of micronutrient-rich foods.

With this background, this study aims at estimating the effect of rice price subsidy on dietary diversity and nutrient intake in Tamil Nadu, as the state is considered a pioneer in introducing food security programmes from mid-day meals for school children to supplying free rice through PDS to all households. The specific objectives of this study are as follows: to estimate the determinants of choice in PDS rice consumption and to estimate the impact of PDS rice consumption at subsidised prices on food consumption pattern and nutrient intake in Tamil Nadu. Keeping these objectives in view, we have set up two hypotheses to be tested: (a) there is no relationship between socio-economic household characteristics and choice of PDS rice consumption and (b) the price subsidy on PDS rice does not have a significant impact on food consumption patterns, dietary diversity and nutrient intake.

The article is divided into four different sections. Section 2 outlines the data and conceptual framework of the study. Section 3 presents the results and discusses the cause and effect of the variables in the study and the Section 4 provides concluding remarks.

2. Methodology

2.1 Data

The study used National Sample Survey Organisation’s consumer expenditure surveys on food and non-food commodities for 2004–2005 and 2011–2012 to capture spatial and temporal variations. Both surveys covered over 100,000 sample households across India, with Tamil Nadu accounting over 7,000 sample households in each period. Data on the price response of demand were obtained on the basis of unit values. We used the state-wise poverty line to classify the sample into low-, middle- and high-income groups (the Planning Commission’s poverty estimates for 2004–2005 and 2011–2012 were used for each individual state). Accordingly, the ‘low-income’ class comprised households with an income level below poverty line (BPL); those that fell between BPL and up to 150 per cent of BPL were grouped as ‘middle-income’ households; and households with a per capita income above 150 per cent of BPL were categorised as the ‘high-income’ group (Kumar et al., 2011).

2.2 Propensity Score Matching Technique

In order to assess the effect of any government intervention or development activity, we need data on the respective individuals in the period before and after the intervention. Randomised Control Trial (RCT) is the most scientifically rigorous study design in experimental studies to investigate the effect of a treatment with minimal bias, where samples or participants are randomly assigned to treatment or control groups (Pocock & Elbourne, 2000). However, such trials are difficult to conduct as they are labour-intensive and expensive; further, it may be ethically unacceptable to establish a causal effect of a development activity on welfare outcomes (Mendola, 2007), that is, the experimental programme or project may benefit only a group of people and not the others, thus making it difficult to establish both treatment and comparison groups at the baseline.

Research in economics at the household-unit level relies on observational or survey data, especially for the study of the impact of large-scale development interventions. In such data, the respondents in the treatment and control groups are likely to vary due to confounding variables,¹ and variations in the relevant outcomes could be due to variations in baseline conditions, rather than an actual treatment effect. Matching each respondent in the treatment group with respondents in the control group with comparable baseline confounding variables is an intuitive way to minimise the bias due to confounding variables in observational studies. However, matching simultaneously one or more than one confounding variables is a complex process and often results in very few similar matches (Benedetto et al., 2018). Further, estimating the casual effect by simply comparing a treatment group with a non-experimental group could be biased, when a group of respondents or households is exposed to a well-defined treatment with no systematic method of experimental design to maintain control and treatment groups. This is because of the problem of self-selection or systematic judgement by the researcher in selecting the samples to be assigned to treatment (Caliendo & Kopeinig, 2008; Dehejia & Wahba, 2002).

The propensity score matching (PSM) technique is used as an alternative to evaluate the effect of a treatment by using observational or survey data. The key idea behind estimating the difference between the treated and untreated samples is to match similar individuals representing both samples based on the confounding characteristics of the individuals. The propensity score (PS) is the probability of a subject receiving a treatment conditional on a set of confounding variables. Logistic regression function is commonly used to estimate PS (Austin, 2011; Brookhart et al., 2006; Staffa & Zurakowski, 2018; Zhang et al., 2014). PS obtained from logit regression helps to identify the treatment group and its counterfactual (control) group and to simplify the matching process by collapsing all the confounding variables into a single indicator. Matching respondents with a similar estimated PS generates an approximate balance for all the confounding variables, and difference in outcomes within groups with a similar PS gives unbiased estimates of the treatment effect (Austin, 2011; Stuart, 2010). We, therefore, employed the PSM technique to estimate the true effect of the PDS rice price subsidy on dietary diversity and nutrient intake, by using household-level consumer expenditure survey data in our study.

According to Caliendo and Kopeinig (2008), PSM follows the following steps: first, we estimated PS by using a logit model where the choice of PDS rice consumption is regressed on various socio-economic household characteristics. Second, we matched the PDS beneficiaries and non-beneficiaries by using the estimated PS from the logit model. Various procedures have been followed for matching the treatment and control groups—nearest neighbour matching (NNM), kernel matching, radius matching, Mahalanobis matching, Spline matching, etc. From these, we chose the NNM technique because it is the most straightforward method where an individual from the control group is selected as a counterpart of the treated respondent, that is, nearest to the estimated propensity score. In NNM method, we followed matching with replacement technique (i.e., a subject in the control group can be matched with more than one treated unit), as our data set contained several treated individuals (beneficiaries of rice price subsidy) with high propensity scores, and only a few control respondents with high propensity scores to avoid bad matches, as some of high-score beneficiaries are expected to get matched to low-score nonparticipants. Third, the quality of matching was checked by using the propensity score distribution balance test and the mean absolute standardised bias (MASB) test (Rosenbaum & Rubin, 1985). Fourth, we estimated the average treatment effect (ATE) of rice price subsidy on dietary diversity and nutrient consumption by taking the average difference between the treatment and control groups (Abebaw et al., 2010).

Thus, we used the following equation to compute the ATE:

A T E = E_{x} (E (Y_{1} | X, Z = 1) - E (Y_{0} | X, Z = 0))

where Y indicates the output variable on which we want to estimate the food subsidy effect, X indicates a vector of covariate determining the choice of PDS rice consumption (Z), which includes the price of rice (PRICE), monthly per capita consumer expenditure (MPCE) as a proxy for income, non-food expenditure (NONFOCE), household size (HHS), age of the household head (AGE), gender of the household head (SEX), social groups (Scheduled Tribe—ST, Scheduled Caste—SC, other backward classes—OBC and forward caste—FC), owned land (OWL), presence of regular salary earners (RSE), own dwelling unit (DWU), educational levels (illiterate—ILL, non-institutional education—NONINSEDU, primary school—PRIMARY, high school—HIGH, higher secondary schooling—HIGHSEC, college level—COLLEGE), food away from home (FAFH), home production of rice (HPP), presence of a refrigerator (REFRAVL), presence of a liquid petroleum gas facility (LPG), location of the household (REGION) and time period (YEAR).

3. Estimation Results and Discussion

3.1 Food Consumption Pattern and Nutrient Intake in Tamil Nadu

Details of the food consumption patterns and nutrient intake across income groups, PDS rice consumers and non-PDS consumers are given in Table 1. In general, PDS rice beneficiaries consume more rice per month across all income categories. Further, PDS rice beneficiaries in high- and medium-income households consume more rice than those in low-income households. Non-PDS beneficiaries belonging to the low-income category consume less rice than the other categories. Low-income households consume fewer processed food items derived from rice and wheat compared to high-income households, while non-PDS beneficiaries of all income categories have a higher consumption of these processed food items.

Table 1.
Income-wise Food Consumption Patterns of PDS and Non-PDS Rice Households

Low-income Households
Middle-income Households
High-income Households

Non-PDS PDS Non-PDS PDS Non-PDS PDS

Per month food consumption (in kg)

Rice 6.375 9.192 7.923 9.800 7.205 9.969

Processed food items 0.316 0.297 0.730 0.696 1.447 1.158

Millets 0.215 0.341 0.118 0.149 0.122 0.116

Roots and tubers 0.224 0.250 0.357 0.421 0.522 0.599

Vegetables 2.059 2.291 3.118 3.563 4.492 5.240

Fruits 0.322 0.296 0.779 0.894 2.302 2.050

Meat 0.131 0.171 0.295 0.378 0.444 0.725

Eggs 0.054 0.068 0.143 0.163 0.221 0.316

Fish 0.067 0.105 0.132 0.229 0.216 0.365

Pulses 0.513 0.596 0.918 0.957 1.242 1.325

Milk products 1.557 1.604 4.685 3.664 7.997 6.176

Oil and fat foods 0.325 0.377 0.550 0.575 0.784 0.864

Nuts 0.005 0.004 0.021 0.033 0.088 0.084

Per day nutrition intake

Energy (in kcal) 1,332.585 1,453.825 1,749.711 1,850.119 2,301.723 2,363.445

Protein (in g) 31.493 34.747 44.440 47.358 59.897 63.277

Fat (in mg) 19.398 19.814 35.362 33.780 57.326 53.919

Calcium (in mg) 6,260.531 7,336.659 1,3219.890 1,2465.710 2,1425.150 1,9473.610

Iron (in mg) 173.497 201.014 258.190 281.629 375.002 396.402

Vitamin A (in µg) 69.824 77.605 105.975 117.880 169.996 179.201

	Low-income Households	Middle-income Households	High-income Households
Per month food consumption (in kg)
Rice	6.375	9.192	7.923	9.800	7.205	9.969
Processed food items	0.316	0.297	0.730	0.696	1.447	1.158
Millets	0.215	0.341	0.118	0.149	0.122	0.116
Roots and tubers	0.224	0.250	0.357	0.421	0.522	0.599
Vegetables	2.059	2.291	3.118	3.563	4.492	5.240
Fruits	0.322	0.296	0.779	0.894	2.302	2.050
Meat	0.131	0.171	0.295	0.378	0.444	0.725
Eggs	0.054	0.068	0.143	0.163	0.221	0.316
Fish	0.067	0.105	0.132	0.229	0.216	0.365
Pulses	0.513	0.596	0.918	0.957	1.242	1.325
Milk products	1.557	1.604	4.685	3.664	7.997	6.176
Oil and fat foods	0.325	0.377	0.550	0.575	0.784	0.864
Nuts	0.005	0.004	0.021	0.033	0.088	0.084
Per day nutrition intake
Energy (in kcal)	1,332.585	1,453.825	1,749.711	1,850.119	2,301.723	2,363.445
Protein (in g)	31.493	34.747	44.440	47.358	59.897	63.277
Fat (in mg)	19.398	19.814	35.362	33.780	57.326	53.919
Calcium (in mg)	6,260.531	7,336.659	1,3219.890	1,2465.710	2,1425.150	1,9473.610
Iron (in mg)	173.497	201.014	258.190	281.629	375.002	396.402
Vitamin A (in µg)	69.824	77.605	105.975	117.880	169.996	179.201

Source: The authors.

Note: Units are kg, unless otherwise indicated.

PDS rice beneficiaries consume more rice, roots and tubers, vegetables, meats, eggs, fish, pulses and oil than non-PDS beneficiaries, across all income categories. This indicates that PDS rice beneficiaries are able to allocate their disposable income to all kinds of food groups. A mixed consumption pattern can be observed for wheat, millets, fruits and nuts among all the categories of households. As regards nutrition, generally the per day intake of all the nutrients is low for low-income households relative to middle- and high-income households. The intake of energy, proteins, iron and Vitamin A is found to be high among PDS rice consumers, irrespective of the income category. The intake of fat and calcium is higher among PDS rice consumers from low-income households, whereas it is lower for middle- and high-income households.

3.2 Determinants of Public Distribution System Rice Consumption

Estimated logit model (marginal effects) of covariates determining PDS-rice consumption across income groups are presented in Table 2. The results reveal that a high market price of rice increases the probability of consuming rice at subsidised price provided under PDS among low-income households, whereas an inverse relationship exists for middle- and high-income households; for instance, if the price of rice in open market increases by ₹10 per kg, the probability of consuming PDS rice would increase by 6.7 per cent, indicating that increased price of rice in the open market forces poor households to rely more on PDS for their rice consumption, that is, open market rice and PDS rice are treated as substitutes among poor households. But it is not the case when it comes to middle- and high-income households, with both PDS and open market rice being treated as complementary to each other. There is no significant impact observed on the PDS rice consumption among poor households with respect to income changes. Nevertheless, increased income has a significant and negative impact on the probability of middle- and high-income households consuming PDS rice. One unit increase in the income level would result in a substantial reduction in the probability of consuming PDS rice by 50 per cent and 44 per cent for middle- and high-income households, respectively. There is a higher probability of consuming PDS rice with an increase in household size and LPG facilities. In contrast, households in urban areas with higher levels of education are less likely to consume PDS rice, irrespective of income groups. Also, the results infer a positive tendency towards PDS rice consumption for 2011–2012 as against 2004–2005.

Table 2.
Income Group-wise Estimated Logit Model (Marginal Effects) for Determinants of PDS Rice Consumption

Covariates Determining PDS Rice Consumption Poor Households Middle-Income Households High-Income Households All Households

Ln price of rice 0.067*
(0.027) −0.136
(0.022) −0.237
(0.032) −0.153
(0.016)

Ln monthly per capita
Consumer expenditure −0.008 NS
(0.029) −0.501
(0.029) −0.445
(0.042) −0.305
(0.019)

Ln non-food expenditure 0.037 NS
(0.022) 0.193
(0.017) 0.253
(0.029) 0.136
(0.012)

Household size 0.037
(0.004) 0.015
(0.003) 0.03
(0.005) 0.027
(0.002)

Age of household head 0.003
(0) 0.002
(0) 0 NS
(0.001) 0.002
(0)

Gender of household head 0.016 NS
(0.016) 0.026*
(0.013) −0.022 NS
(0.019) 0.01 NS
(0.009)

Social groups (Scheduled Tribes)

Scheduled Caste −0.004 NS
(0.07) 0.077 NS
(0.056) 0.06 NS
(0.088) 0.056 NS
(0.041)

Other backward classes −0.03 NS
(0.07) 0.026 NS
(0.056) −0.023 NS
(0.086) 0.004 NS
(0.04)

Other classes −0.027 NS
(0.078) −0.035 NS
(0.062) −0.17*
(0.089) −0.089 NS
(0.044)

Owned land 0.091**
(0.014) 0.03*
(0.013) 0.006NS
(0.021) 0.053
(0.01)

Regular salary earners 0.127 NS
(0.09) 0.005 NS
(0.018) 0.021 NS
(0.016) 0.017 NS
(0.011)

Owning dwelling unit 0.059 NS
(0.05) 0.088*
(0.019) 0.053
(0.025) 0.027
(0.013)

Educational level (illiterate)

Non-intuitional education −0.105 NS
(0.068) 0.033 NS
(0.064) −0.058 NS
(0.139) −0.012 NS
(0.052)

Primary −0.01 NS
(0.015) −0.015 NS
(0.012) −0.121
(0.029) −0.034
(0.009)

High school −0.047
(0.016) −0.059
(0.013) −0.168
(0.028) −0.085
(0.01)

Higher secondary −0.05 NS
(0.039) −0.155
(0.024) −0.25
(0.033) −0.168
(0.017)

Collegiate −0.135
(0.039) −0.206
(0.024) −0.355
(0.03) −0.262
(0.016)

Food away from home −0.036 NS
(0.02) 0.036
(0.011) 0.001 NS
(0.015) 0.005 NS
(0.008)

Home production of rice −0.091 NS
(0.058) −0.058*
(0.031) −0.078 NS
(0.042) −0.074**
(0.024)

Presence of refrigerator 0
(omitted) −0.071 NS
(0.064) 0.027 NS
(0.034) 0.004 NS
(0.026)

Presence of LPG 0.228*
(0.106) 0.116
(0.018) 0.081
(0.02) 0.089**
(0.012)

Year 2011–2012 0.229*
(0.108) 0.413
(0.019) 0.426
(0.023) 0.338
(0.013)

Urbanisation −0.06
(0.015) −0.052
(0.011) −0.122
(0.015) −0.082
(0.007)

Income class (poor)

Middle income −0.021 NS
(0.011)

High income −0.077
(0.02)

Covariates Determining PDS Rice Consumption	Poor Households	Middle-Income Households	High-Income Households	All Households
Ln price of rice	0.067* (0.027)	−0.136** (0.022)	−0.237** (0.032)	−0.153** (0.016)
Ln monthly per capita Consumer expenditure	−0.008 NS (0.029)	−0.501** (0.029)	−0.445** (0.042)	−0.305** (0.019)
Ln non-food expenditure	0.037 NS (0.022)	0.193** (0.017)	0.253** (0.029)	0.136** (0.012)
Household size	0.037** (0.004)	0.015** (0.003)	0.03** (0.005)	0.027** (0.002)
Age of household head	0.003** (0)	0.002** (0)	0 NS (0.001)	0.002** (0)
Gender of household head	0.016 NS (0.016)	0.026* (0.013)	−0.022 NS (0.019)	0.01 NS (0.009)
Social groups (Scheduled Tribes)
Scheduled Caste	−0.004 NS (0.07)	0.077 NS (0.056)	0.06 NS (0.088)	0.056 NS (0.041)
Other backward classes	−0.03 NS (0.07)	0.026 NS (0.056)	−0.023 NS (0.086)	0.004 NS (0.04)
Other classes	−0.027 NS (0.078)	−0.035 NS (0.062)	−0.17* (0.089)	−0.089 NS (0.044)
Owned land	0.091** (0.014)	0.03* (0.013)	0.006NS (0.021)	0.053** (0.01)
Regular salary earners	0.127 NS (0.09)	0.005 NS (0.018)	0.021 NS (0.016)	0.017 NS (0.011)
Owning dwelling unit	0.059 NS (0.05)	0.088*** (0.019)	0.053** (0.025)	0.027** (0.013)
Educational level (illiterate)
Non-intuitional education	−0.105 NS (0.068)	0.033 NS (0.064)	−0.058 NS (0.139)	−0.012 NS (0.052)
Primary	−0.01 NS (0.015)	−0.015 NS (0.012)	−0.121** (0.029)	−0.034** (0.009)
High school	−0.047** (0.016)	−0.059** (0.013)	−0.168** (0.028)	−0.085** (0.01)
Higher secondary	−0.05 NS (0.039)	−0.155** (0.024)	−0.25** (0.033)	−0.168** (0.017)
Collegiate	−0.135** (0.039)	−0.206** (0.024)	−0.355** (0.03)	−0.262** (0.016)
Food away from home	−0.036 NS (0.02)	0.036** (0.011)	0.001 NS (0.015)	0.005 NS (0.008)
Home production of rice	−0.091 NS (0.058)	−0.058* (0.031)	−0.078 NS (0.042)	−0.074** (0.024)
Presence of refrigerator	0 (omitted)	−0.071 NS (0.064)	0.027 NS (0.034)	0.004 NS (0.026)
Presence of LPG	0.228* (0.106)	0.116** (0.018)	0.081** (0.02)	0.089** (0.012)
Year 2011–2012	0.229* (0.108)	0.413** (0.019)	0.426** (0.023)	0.338** (0.013)
Urbanisation	−0.06** (0.015)	−0.052** (0.011)	−0.122** (0.015)	−0.082** (0.007)
Income class (poor)
Middle income				−0.021 NS (0.011)
High income				−0.077** (0.02)

Source: The authors.

Note: ** Indicates significance at 1% level; * indicates significance at 5% level; and NS indicates non-significance; figures in parentheses are standard errors.

3.3 Effect of Public Distribution System Rice Consumption on Food Consumption and Nutrient Intake in Tamil Nadu

Before estimating the impact of PDS rice consumption, we tested the quality of the matching process between the treatment and control categories, as explained in the methodology section. After estimating PS for the PDS rice beneficiaries and non-PDS beneficiaries, we checked the common support condition. Figures 1–6 show the distribution of PS estimated from the previous step of the logit model determining the choice of PDS rice consumption with respect to the covariates of household socio-economic characteristics. As shown, matching is balanced only for poor households, whereas the rest show an unbalanced distribution, emphasising the need for an appropriate matching procedure to take care of the underlying bias. Moreover, we conducted an MASB test for covariate balancing. MASB was estimated for each variable before and after matching and the average for all the variables was calculated. Figures 7-12 show that the MASB scores between the treated and control groups after matching are less than 20 per cent for most of the covariates, indicating that the matching process was successful and valid.

Figure 1.

Source: The authors.

Figure 2.

Source: The authors.

Figure 3.

Source: The authors.

Figure 4.

Source: The authors.

Figure 5.

Source: The authors.

Figure 6.

Source: The authors.

Figure 7.

Source: The authors.

Figure 8.

Source: The authors.

Figure 9.

Source: The authors.

Figure 10.

Source: The authors.

Figure 11.

Source: The authors.

Figure 12.

Source: The authors.

After estimating the propensity scores and checking their matching quality, we estimated ATE using the nearest neighbourhood algorithm. Table 3 reports the average effect of subsidised rice through PDS on food consumption patterns, which include rice, processed food, millets, pulses, roots and tubers, vegetables, fruits, meats, eggs, fish, milk products, oil items and nuts and the intake of nutrients such as energy, protein, fat, Ca, Fe and vitamin ‘A’. The results reveal that, overall, the consumption of food items such as pulses, roots and tubers, vegetables, meat, eggs, fish and oils shows an increase for PDS beneficiaries vis-à-vis non-beneficiaries; for instance, after matching, the difference between PDS beneficiaries and non-PDS beneficiaries with respect to pulses consumption was 124 g per month.

Table 3.

Effect of Subsidised Rice Through PDS on Food Consumption Patterns Across Income Groups

Output Variables	Low Income	Middle Income	High Income	Overall
1. Staple foods (in kg/month)
Rice	1.95**	2.131**	2.146**	2.167**
Millets	0.108*	−0.008 NS	−0.004 NS	0.007 NS
Vegetables	0.326**	0.571**	0.566**	0.506**
Pulses	0.055**	0.152**	0.15**	0.124**
Roots and tubers	0.032*	0.071**	0.078**	0.068**
2. High-value food commodities (in kg/month)
Processed food items	0.021 NS	0.081*	−0.155 NS	0.008 NS
Fruits	0.000 NS	0.105*	0.111 NS	0.019 NS
Meats	−0.014 NS	0.085**	0.182**	0.067**
Egg	0.014*	0.037**	0.064**	0.022**
Fish	0.036 NS	0.065**	0.104*	0.045*
Milk products	−0.252 NS	−0.084*	−0.409 NS	−0.197 NS
Oil and fat foods	0.032**	0.053**	0.105**	0.066**
Nuts	0.000 NS	−0.002 NS	0.006 NS	0.002 NS
3. Nutrient intake (per day)
Energy (kcal)	236.692**	291.812**	272.66**	285.545**
Protein (g)	5.257**	7.683**	7.46**	6.919**
Fat (g)	0.696 NS	2.072**	3.339**	2.208**
Calcium (mg)	532.894 NS	939.074**	422.212 NS	602.209 NS
Iron (mg)	33.624**	45.186**	38.881**	37.37**
Vitamin A (µg)	10.349*	22.228**	24.342**	18.416**

Source: The authors.

Note: ** Indicates significance at the 1% level; * indicates significance at the 5% level; and NS indicates non-significance.

Similarly, after matching, the consumption of roots and tubers, vegetables, meat, eggs, fish and oil has increased by 68, 506, 67, 22, 45 and 66 g, respectively, for PDS beneficiary households. The results also suggest that the subsidy on rice might have increased the real disposable income of PDS beneficiary households and shifted the food consumption pattern towards the increased consumption of rice as well several other food commodities significantly. Specifically, a considerable increment in the consumption of high-value commodities such as vegetables, meat, eggs, fish and oil appears to be beneficial for nutritional improvement in the dietary pattern under PDS. These results also confirm the existence of both substitution and income effects due to the policy intervention of rice subsidies under PDS as the quantity of rice consumed and other food commodities has increased.

In terms of nutritional gains, the intake of energy, protein, fat, iron and vitamin A has increased significantly among PDS rice beneficiaries vis-à-vis non-beneficiaries; for instance, after matching, the energy intake has increased by 285.54 kcal for PDS rice-consuming households. It is no surprise that there is an income effect on the quantity of rice consumed, given the price subsidy, while the energy calorie intake has not declined across income groups. Similarly, the intake of protein and fat among PDS beneficiaries shows an increase of 6.92 and 2.21 g, respectively, while iron has increased by 37.37 mcg and vitamin A by 18.42 mcg vis-à-vis non-beneficiaries, clearly indicating that the subsidy on rice has led to an increase in nutrient intake.

It can also be observed that changes in food consumption patterns after the subsidy intervention are not the same across income groups. The average effect for PDS beneficiaries from the middle- and high-income households is significant for most commodities, while for poor households, it is significant only for commodities such as rice, millets, vegetables, pulses and roots and tubers. It is interesting to note the significant gain in the consumption of millet only among poor households, indicating that they continue to be inclined towards staple foods even after the rice subsidy, while the diets of middle- and high-income households have become more diversified. These results reveal a gain in purchasing power for poor households, which results in a rise in the quantity of consumption of staple foods such as rice, millets, vegetables and pulses, whereas for middle- and high-income households, the gain is in their higher consumption of horticulture and animal-based food items.

Table 3 also reveals a gain in the intake of energy, protein, iron and vitamin A across all income groups. Since there is a substantial increase in the quantity of pulses and vegetables consumed, all income groups show a significant improvement in their protein intake. However, while the subsidy on PDS rice has had a significant positive effect on the quantity of fat intake for middle- and high-income households, the gain in fat intake is insignificant for poor households, indicating clearly that PDS beneficiary households from the middle- and high-income categories are simultaneously interested in adding taste to their diets. Thus, middle- and high-income households progress towards dietary diversity, with their increased consumption of high-value commodities such as fruits, vegetables, milk, meat, fish and processed food items, when rice is distributed at subsidised price.

Table 4 presents the effect of the PDS rice subsidy on food consumption patterns and dietary diversity across rural and urban regions. The PDS rice subsidy has a significant and positive effect on the consumption of staple foods in both the rural and urban regions, with the exception of a negative effect on millets in urban regions. The PDS rice subsidy has a significant positive effect on the consumption of livestock products in urban regions, but no significant effect is observed on the consumption of most high-value food items in rural regions. In particular, the quantity of milk consumed decreases by 0.516 kg per month in rural areas, while the PDS rice subsidy was found to increase energy and nutrient consumption in urban rather than rural regions.

Table 4.

Effect of Subsidised PDS Rice on Food Consumption Patterns Across Rural and Urban Regions

Output Variables	Rural	Urban
1. Staple foods (kg)
Rice	2.050**	2.215**
Millets	0.104**	−0.039**
Vegetables	0.604**	0.583**
Pulses	0.122**	0.143**
Roots and tubers	0.046*	0.089**
2. High-value food commodities (kg)
Processed foods	−0.010 NS	0.034 NS
Fruits	0.005 NS	0.068 NS
Meats	0.052 NS	0.080**
Egg	0.014 NS	0.055**
Fish	0.004 NS	0.091**
Milk products	−0.516*	0.137 NS
Oil and fat foods	0.065**	0.101**
Nuts	−0.002 NS	0.005 NS
3. Nutrient intake
Energy (kcal)	237.884**	299.818**
Protein (gm)	5.410**	7.781**
Fat (gm)	−0.948 NS	3.993**
Calcium (mg)	350.607 NS	1282.376**
Iron (mg)	39.872**	41.182**
Vitamin A (µg)	25.849**	20.683**

Source: The authors.

Note: ** Indicates significance at 1% level; * indicates significance at 5% level; and NS indicates non-significance.

4. Concluding Remarks

Our results empirically show that price subsidy interventions on rice through PDS significantly influence food consumption patterns and the nutrient intake of households in Tamil Nadu. The subsidy on rice has had differing effects on the consumption patterns of the poor, middle- and higher-income households. The real income gains from the price subsidy are confined to staple foods such as rice, millets, pulses and vegetables in poor households, whereas middle-income households gain across a variety of food groups, especially high-value commodities such as fruits, milk, meat, fish and processed food. This indicates that PDS encourages the beneficiaries of poor households to consume larger quantities of cheaper food products rather than high-value commodities. Furthermore, among the poor, the nutrient intake of fat and calcium is insignificant compared to the intake of middle-income households. There was also considerable difference in the effects of the subsidy on the dietary patterns of rural and urban regions.

Since subsidy on rice has led to a shift away from high-value commodities by the poor, thereby reducing their dietary diversity, an extension of the subsidy to nutrition-rich food items other than rice could encourage poor people to diversify their diets towards a healthy and nutrient-dense consumption pattern. In addition, the existing food subsidy can be effectively implemented by ensuring the availability of nutritious food such as pulses, millets and animal products at affordable prices. This would be made possible by extending and supporting current measures related to crop and farm diversification at various stages, from production to marketing.

Footnotes

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.

Funding

The authors received no financial support for the research, authorship and/or publication of this article.

Notes

References

Abebaw

, Fentie

, & Kassa

(2010). The impact of a food security program on household food consumption in Northwestern Ethiopia: A matching estimator approach. Food Policy, 35(4), 286–293.

Arimond

, & Ruel

M. T.

(2002). Progress in developing an infant and a child feeding index: An example using the Ethiopia demographic and health survey 2000 (No. 583-2016-39595). IFPRI.

Aurino

(2017). Do boys eat better than girls in India? Longitudinal evidence on dietary diversity and food consumption disparities among children and adolescents. Economics & Human Biology, 25, 99–111.

Austin

P. C.

(2011). An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivariate Behavioral Research, 46(3), 399–424.

Benedetto

, Head

S. J.

, Angelini

G. D.

, & Blackstone

E. H.

(2018). Statistical primer: Propensity score matching and its alternatives. European Journal of Cardio-Thoracic Surgery, 53(6), 1112–1117.

Brookhart

M. A.

, Schneeweiss

, Rothman

K. J.

, Glynn

R. J.

, Avorn

, & Stürmer

(2006). Variable selection for propensity score models. American Journal of Epidemiology, 163(12), 1149–1156.

Buse

, & Hawkes

(2015). Health in the sustainable development goals: Ready for a paradigm shift? Globalization and Health, 11(1), 13.

Caliendo

, & Kopeinig

(2008). Some practical guidance for the implementation of propensity score matching. Journal of Economic Surveys, 22(1), 31–72.

Chandrasekhar

, Aguayo

V. M.

, Krishna

, & Nair

(2017). Household food insecurity and children’s dietary diversity and nutrition in India. Evidence from the comprehensive nutrition survey in Maharashtra. Maternal & Child Nutrition, 13, e12447.

10.

Charmarbagwala

, Ranger

, Waddington

, & White

(2004). The determinants of child health and nutrition: A meta-analysis. World Bank.

11.

Dehejia

R. H.

, & Wahba

(2002). Propensity score-matching methods for nonexperimental causal studies. Review of Economics and Statistics, 84(1), 151–161.

12.

Desai

, & Vanneman

(2015, August). Enhancing nutrition security via India’s national food security act: Using an axe instead of a scalpel? In India Policy Forum: [papers]. India Policy Forum. Conference (Vol. 11, p. 67). NIH Public Access.

13.

Ekesa

B. N.

, Walingo

M. K.

, & Abukutsa-Onyango

M. O.

(2008). Influence of agricultural biodiversity on dietary diversity of preschool children in Matungu division, Western Kenya. African Journal of Food, Agriculture, Nutrition and Development, 8(4), 390–404.

14.

Food and Agriculture, Organisation (FAO) (2017). The state of food security and nutrition in the world 2017. Building Resilience for Peace and Food Security, FAO. http://www.fao.org/3/a-i7695epdf

15.

Griggs

, Stafford-Smith

, Gaffney

, Rockström

, Öhman

M. C.

, Shyamsundar

, Steffen

, Glaser

, Kanie

, & Noble

(2013). Policy: Sustainable development goals for people and planet. Nature, 495(7441), 305.

16.

Hatløy

, Hallund

, Diarra

M. M.

, & Oshaug

(2000). Food variety, socioeconomic status and nutritional status in urban and rural areas in Koutiala (Mali). Public Health Nutrition, 3(1), 57–65.

17.

Kaul

(2018). Household responses to food subsidies: Evidence from India. Economic Development and Cultural Change, 67(1), 95–129.

18.

Kehoe

S. H.

, Krishnaveni

G. V.

, Veena

S. R.

, Guntupalli

A. M.

, Margetts

B. M.

, Fall

C. H.

, & Robinson

S. M.

(2014). Diet patterns are associated with demographic factors and nutritional status in South Indian children. Maternal & Child Nutrition, 10(1), 145–158.

19.

Khera

(2014). Cash vs. in-kind transfers: Indian data meets theory. Food Policy, 46, 116–128.

20.

Kishore

, & Chakrabarti

(2015). Is more inclusive more effective? The ‘New Style’ public distribution system in India. Food Policy, 55, 117–130.

21.

Krishnamurthy

, Pathania

, & Tandon

(2017). Food price subsidies and nutrition: Evidence from state reforms to India’s public distribution system. Economic Development and Cultural Change, 66(1), 55–90.

22.

Kumar

, Kumar

, Shinoj

, & Raju

S.S.

, (2011). Estimation of demand elasticity for food commodities in India §. Agricultural Economics Research Review, 24(1), 1–14.

23.

Mendola

(2007). Agricultural technology adoption and poverty reduction: A propensity-score matching analysis for rural Bangladesh. Food Policy, 32(3), 372–393.

24.

Misra

, Singhal

, Sivakumar

, Bhagat

, Jaiswal

, & Khurana

(2011). Nutrition transition in India: Secular trends in dietary intake and their relationship to diet-related non-communicable diseases. Journal of Diabetes, 3(4), 278–292.

25.

Pingali

, Mittra

, & Rahman

(2017). The bumpy road from food to nutrition security–Slow evolution of India’s food policy. Global Food Security, 15, 77–84.

26.

Pocock

S. J.

, & Elbourne

D. R.

(2000). Randomized trials or observational tribulations? The New England Journal of Medicine, 342, 1907–1909.

27.

Rahman

(2016). Universal food security program and nutritional intake: Evidence from the hunger prone KBK districts in Odisha. Food Policy, 63, 73–86.

28.

Rosenbaum

P. R.

, & Rubin

D. B.

(1985). Constructing a control group using multivariate matched sampling methods that incorporate the propensity score. The American Statistician, 39(1), 33–38.

29.

Shrinivas

, Baylis

, Crost

, & Pingali

(2018). Do staple food subsidies improve nutrition? (Unpublished manuscript). http://barrett.dyson.cornell.edu/NEUDC/paper_520.pdf

30.

Staffa

S. J.

, & Zurakowski

(2018). Five steps to successfully implement and evaluate propensity score matching in clinical research studies. Anesthesia & Analgesia, 127(4), 1066–1073.

31.

Stuart

E. A.

(2010). Matching methods for causal inference: A review and a look forward. Statistical Science: A Review Journal of the Institute of Mathematical Statistics, 25(1), 1.

32.

Taren

, & Chen

(1993). A positive association between extended breast-feeding and nutritional status in rural Hubei Province, People’s Republic of China. The American Journal of Clinical Nutrition, 58(6), 862–867.

33.

von Grebmer

, Bernstein

, Hossain

, Brown

, Prasai

, Yohannes

, Patterson

, Sonntag

, Zimmerman

, Towey

, & Foley

(2017). 2017 Global hunger index: The inequalities of hunger. International Food Policy Research Institute (IFPRI).

34.

Weber

(2017). Politics of ‘leaving no one behind’: Contesting the 2030 sustainable development goals agenda. Globalizations, 14(3), 399–414.

35.

Zhang

, Wang

, & Zhang

(2014). Application of propensity scores to explore the effect of public reporting of medicine use information on rational drug use in China: A quasi-experimental design. BMC Health Services Research, 14(1), 492.