Andaleeb 2016.

Study characteristics
Methods	Study design: CBA How were missing data handled? NR (only available survey data used) Randomisation ratio: N/A Recruitment method: ration cards issued to individuals or families based on their poverty status, and purchased from the PDS. In Odisha state, the KBK region, comprising 8 districts, has universal PDS access; whereas it was a targeted programme in the 22 non‐KBK districts. Sample size justification and outcome used: no sample size calculation. National‐level surveys carried out to be nationally representative of consumer expenditure; conducted by the National Sample Survey Organization. Sampling method: purposive programme placement by government as the decision to make PDS a universal programme in the KBK region based upon its history of poor nutritional outcome. Hence, the selection of districts into the programme was not random. Sample was restricted to rural areas of Odisha since the PDS revival was more effective in rural areas. KBK region – with a universal PDS – was the treatment group while the rest of Odisha was the control group. Non‐KBK districts within the same states were control group. For further robustness checks, samples restricted to the KBK districts and considered HHs without any ration card as the alternative control group, with all other HHs with a ration card (AAY/BPL/APL) as the treatment group. Study aim or objective: to determine the role of consumer food subsidies in improving nutritional intake and diet quality by evaluating the expansion of the government food assistance programme coverage in the hunger prone state of Odisha in India. Study period: about 8 years. 2004–2005 survey was baseline and the 2011–2012 survey was postintervention information Unit of allocation or exposure: cluster: districts 8 KBK districts with universal PDS (treatment) vs 22 non‐KBK districts (control). Alternative: within KBK districts: no ration card (control) vs any ration card (treatment)
Participants	Baseline characteristics Intervention or exposure Age: HH head, years: 42.5 Place of residence: 8 KBK districts in Odisha state Sex, %: female HH heads: 10 Ethnicity and language: religion, %: Hinduism 100 Occupation, %: employed in non‐agriculture: 20; other: 50 Education: HH heads with no education or incomplete primary schooling, %: 100 SES: not regularly salaried, %: 90; estimated MPCE (INR): 294.95; type of ration card, %: Antayodaya Anna Yojana 2.71, BPL 48.94, APL 7.41, no card 40.95 Social capital: scheduled caste, %: 20, other backward classes, %: 30 Nutritional status: monthly rice consumption from PDS (kg (% share of total)): Antayodaya Anna Yojana 24.8 (47.6), BPL 14.1 (30.9), APL 6.4 (11.6), no card 2.1 (4.5). Intake of calories: 1801.4.3; protein (g): 41.7; fat (g): 15.7. Sources of calories: cereals (1416.9); non‐cereals (384.5); pulses (46.3); egg, fish and meat (9.6); dairy products (25.3); vegetables and fruit (70.5); edible oil (70.1); other food (162.8) Morbidities: NR Concomitant or previous care: NR Control Age: HH head, years: 42.3 Place of residence: 22 non‐KBK districts in Odisha state Sex, %: female HH heads: 10 Ethnicity and language: religion, %: Hinduism 100 Occupation, %: employed in non‐agriculture: 20; other: 50 Education: HH heads with no education or incomplete primary schooling, %: 100 SES: not regularly salaried, %: 90; estimated MPCE (INR): 415.32; type of ration card, %: Antayodaya Anna Yojana 1.85, BPL 41.29, APL 25.42, no card 31.43 Social capital: scheduled caste, %: 20, other backward classes, %: 30 Nutritional status: monthly rice consumption from PDS (kg (% share of total)): Antayodaya Anna Yojana 26.4 (45.2), BPL 6.5 (9.6), APL 0.6 (0.9), no card 0.1 (0.2). Intake of calories: 2159.3; protein (g): 51.4; fat (g): 21.3; sources of calories: cereals (1649.4); non‐cereals (509.8); pulses (66.5); egg, fish and meat (17.7); dairy products (40.3); vegetables and fruit (145.0); edible oil (101.4); other food (139.0) Morbidities: NR Concomitant or previous care: NR Overall Age: NR Place of residence: Odisha state Sex: NR Ethnicity and language: NR Occupation: NR Education: NR SES: % based on ration card: 1.99 (poorest of the poor), 42.57 (BPL), 22.41 (APL), 33.02 (not targeted as poor wealthiest?) Social capital: NR Nutritional status: NR Morbidities: NR Concomitant or previous care: NR Inclusion criteria: HHs in the Odisha state of India. Treatment: HHs in 8 KBK districts in the state (alternative: HHs with ration card in KBK districts). Control: HHs in 22 non‐KBK districts in the state (alternative: HHs without a ration card in KBK districts) Exclusion criteria: NR Pretreatment: outcome variables related to macronutrient and calorie sources were all higher for the control group. Poverty levels: We estimate that the MPCE stood at INR 294.95 in the KBK districts as compared to INR 415.32 in the rest of Odisha at 2004–2005 constant prices. Mean HH consumption of rice from PDS: 8.9 kg per month KBK districts, 3.3 non‐KBK districts. Share of monthly rice consumption from PDS to total: 19% KBK, 5.2% non‐KBK. Attrition per relevant group: N/A as they are 2 repeated cross‐sectional HH surveys. Description of subgroups measured and reported: besides KBK vs non‐KBK districts, there are also analyses done comparing ration card vs no ration card HHs in the KBK districts. Total number completed and analysed per relevant group: at follow‐up, 2973 HHs were surveyed and contributed outcome data to the total group. Numbers NR for intervention and control separately. Total number enrolled per relevant group: at baseline, 3819 HHs were surveyed. Numbers NR for intervention and control separately. Total number randomised per relevant group: N/A
Interventions	Intervention characteristics Intervention or exposure Food access intervention category: food prices Intervention type: universal food subsidy Description: universal access to PDS, providing grains at extremely low prices to all ration card holders; whether they were AAY, BPL or APL. All HHs were eligible to 25 kg of subsidised rice Duration of intervention period: about 4 years: universal PDS in Odisha started in 2008 Frequency: ongoing Number of study contacts: 2; 2004–2005 survey was baseline while 2011–2012 survey was postintervention information. Providers: state government Delivery: to be able to access the PDS, HHs needed to possess a ration card (document issued by government that entitles an individual/family to purchase from the PDS). Ration cards are also used as an identity card to avail many of the other government schemes, since it classifies HHs based upon their poverty status. Ration cards were of 3 types: AAY card for the poorest of the poor, BPL for the poor and APL for those HHs who were not identified as poor. Co‐interventions: ration cards availed many people to other government schemes, since it classifies HHs based upon their poverty status. Resource requirements: NR Economic indicators: NR Control Food access intervention category: food prices Intervention type: targeted food subsidy Description: poorer HHs: 25 or 35 kg of subsidised rice? Duration of intervention period: 1947 to present Frequency: ongoing Number of study contacts: 2; 2004–2005 survey was baseline while 2011–2012 survey was postintervention information. Providers: state government Delivery: to be able to access the PDS, HHs needed to possess a ration card (document issued by government that entitles an individual/family to purchase from the PDS). Ration cards are also used as an identity card to avail many of the other government schemes, since it classifies HHs based upon their poverty status. Ration cards were of 3 types: AAY card for the poorest of the poor, BPL for the poor and APL for those HHs who were not identified as poor. Co‐interventions: ration cards availed many people to other government schemes, since it classifies HHs based upon their poverty status. Resource requirements: NR Economic indicators: NR
Outcomes	Dietary diversity: contribution of different types of food to caloric intake, e.g. cereals; non‐cereals; pulses; milk; eggs, fish and meat; vegetables and fruits; edible oils; other foods Dietary intake: intake of protein; intake of fat; ratio of nutrient intake to the RDA, multiplied by 100, e.g. caloric; protein; fat
Identification	Sponsorship source: System of Promoting Appropriate National Dynamism for Agriculture and Nutrition (SPANDAN) initiative; housed in IGIDR and supported by Bill & Melinda Gates Foundation. Country: India Setting: rural HHs in Odisha, India Author's name: Andaleeb Rahman Email: arahman@iihs.ac.in Declarations of interest: NR Study or programme name and acronym: Public Distribution System (PDS) Type of record: journal article
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (Selection bias)	High risk	CBA. KBK regions were by definition poorer than non‐KBK regions.
Allocation concealment (Selection bias)	High risk	CBA. Intervention was given to the poorest regions of the Odisha state, and control to less‐poor regions.
Baseline characteristics similar (Selection bias)	Low risk	Although the HHs in the treatment groups are not comparable based on the fact that the allocation of the districts in the programme is based on the poverty level, this has been taken into account in the analysis. Quote: "The present case is of purposive program placement by the government as the decision of make PDS a universal program in the KBK region was based upon its history of poor nutritional outcome. Hence, the selection of districts into the program (here, PDS) is not random. We do a slew of robustness check to ensure that we control for this later in the paper." "Results from the DID regressions are presented in Table 6. Estimates as reported in column (1) were arrived at by controlling for the district fixed effects but not for the HH characteristics. In the column (2), both district fixed effects and the HH characteristics were controlled for."
Baseline outcome measurements similar (Selection bias)	Low risk	Outcome variables related to macronutrient intake and sources of calories were all significantly higher in the control group: calories; protein; fat as well as calories from cereals; non‐cereals; pulses; egg, fish and meat; dairy products; vegetables and fruits and edible oil all P < 0.01. However, PSM was performed to account for the non‐random allocation of the intervention and control: there were no significant group differences for the covariates used to perform the PSM.
Blinding of participants and personnel (Performance bias)	Low risk	Participants were not blinded to allocation as their subsidy status was determined by their residential district; however, it is unlikely that they would have known that the national survey (from which the data were obtained) would be comparing their nutritional intake with participants receiving another intervention. Implementation of the project by local state government seems totally separate from implementation of the surveys by the respective organisation.
Blinding of outcome assessment (Detection bias)	High risk	It is NR whether outcomes were assessed blindly. Outcomes were assessed by self‐reported measures.
Protection against contamination (Performance bias)	Low risk	As intervention and control treatments were government‐assigned according to state district, it is unlikely that meaningful contamination could have occurred. Any small spillover effect at the border between the 2 areas would be diluted by the large sample sizes and geographical area. NR, but although it is controlled who could and could not buy the rice (ration card), it did not control who actually used it.
Incomplete outcome data (Attrition bias)	High risk	Very little information provided about attrition, but 3819 HHs were surveyed at baseline and 2973 at follow‐up, indicating a 87.8% 'response rate'. According to the supplementary data sheet, difference‐in‐difference estimates were performed on 6722 observations; roughly the sum of available baseline and follow‐up data. Therefore, it would appear that missing values were not adjusted for or imputed.
Selective outcome reporting (Reporting bias)	Unclear risk	No protocol available and the paper did not follow the usual journal format with a designated Methods section. The aim of the study was to assess unspecified 'nutrient intake indicators' as well as a variety of food items in the diet, according to 6 groups. The latter have all been reported on.
Other bias	Low risk	Selection bias was overcome by correction for HH variables. Since we were testing for the significance of a large number of dependent variables, it might lead to higher probability of Type I errors leading to false rejection of the null hypothesis. To control for this bias, we used the summary indices approach of Clingingsmith et al. (2009) and Kling et al. (2004).