Upscaling Participatory Action and Videos for Agriculture and Nutrition (UPAVAN) study data – User Guide

Persistent identifier

10.17037/DATA.00003642

Description

Data collected as part of a four-arm cluster randomised controlled trial to test the effects of three nutrition-sensitive agriculture interventions on maternal and child nutrition in rural Odisha, India.

The clusters (village and its surrounding hamlets) were allocated 1:1:1:1 to a control group or an intervention group of fortnightly women's group meetings and household visits over 32 months using: NSA videos (AGRI group); NSA and nutrition-specific videos (AGRI-NUT group); or NSA videos and a nutrition-specific participatory learning and action (PLA) cycle meetings and videos (AGRI-NUT+PLA group). Primary outcomes were the proportion of children aged 6–23 months consuming at least four of seven food groups the previous day and mean maternal body-mass index (BMI). Secondary outcomes were proportion of mothers consuming at least five of ten food groups and child wasting.

All outcomes were assessed using cross-sectional surveys at baseline and endline, 36 months later. The trial's impact evaluation is available at: 10.1016/S2542-5196(21)00001-2.

Data collection methods

We did a parallel group, observer-blind, four-arm cluster RCT in Keonjhar district, Odisha, India. A cluster was a village and its surrounding hamlets.

The interventions were delivered at the cluster level and all women in intervention clusters were eligible to participate. We evaluated effects on trial participants: one child aged 0–23 months per household, their mothers or female primary caregivers when the mother was absent (hereafter mothers) aged 15–49 years, and the mother's spouse (or household head, if unavailable). We aimed to sample 32 households per cluster, giving an intended sample size of 4736 mother–child dyads. The sample size was calculated to detect differences in the two primary outcomes of the trial. Data collectors sought informed consent from adult participants in writing or by thumbprint, and, for children, from their primary caregivers.

The primary outcomes were percentage of children aged 6–23 months consuming at least four of seven food groups in the previous 24 hours and the mean BMI of non-pregnant, non-postpartum (gave birth >42 days ago) mothers. Secondary outcomes were the percentage of mothers consuming at least five of ten food groups in the previous 24 hours and the percentage of children with a weight-for-height Z score of less than −2 SD. Other additional outcomes were mothers' and children's haemoglobin concentrations and mid-upper arm circumference (MUAC), and exploratory outcomes along the causal pathways including indicators of women's empowerment and agricultural production. The list of outcomes was published in the trial protocol: 10.1186/s13063-018-2521-y.

Following Food and Agriculture Organization guidelines for women and WHO guidelines for children, dietary intakes were elicited using the free recall method and following prespecified probes. Children's length and mother's height were measured using Seca 417 Infantometer (Seca, Germany) and 213 Stadiometers (Seca, Germany); their weight using MAX-CRUZER scales (Axis Electronics, India); MUAC using Médecins Sans Frontières produced standard MUAC tapes (MegaCare International, India); and haemoglobin using HemoCue Hb 301 machines (HemoCue, India)

All outcomes were assessed through cross-sectional surveys at baseline and endline, 36 months later (at the end of the 32-month intervention). To reduce respondent burden, we randomly allocated half of the spouses to answer questions on empowerment, and the other half answered questions on household consumption.

At baseline, we collected data on paper questionnaires, and a quality assurance team checked them for plausibility and logic at the field site before double entry into a database in the nearest city (Bhubaneswar). All data at endline were captured on Android tablets using Open Data Kit software (version 1.29.3).

Data analysis and preparation

Data analysis

The primary analysis of outcomes was by intention to treat and included all randomised clusters and participants. The analyses were cross-sectional, comparing outcomes in each of the intervention groups and the control group at endline. To adjust for the baseline measures of the outcomes, the analysis included all individuals at each timepoint linked by cluster. This approach uses the individual-level measures to effectively calculate cluster-level summaries of each outcome at baseline. The models therefore included a cluster-level summary of each outcome at baseline and the inclusion of a time variable additionally allowed us to estimate changes in outcomes over time in the study area. Each measure was analysed using separate generalised estimating equations (GEEs) to account for clustering. Statistical significance was taken at the 5% level (p<0.05). We carried out analyses adjusted only for baseline measures of the outcomes, and adjusted analyses that additionally included the stratification variables: distance to the nearest town and proportion of Scheduled Castes and Scheduled Tribes.

Preparation

Data is shared in the form of two datasets:

Quantitative combined mothers, child, anthropometry and indicator dataset (UPAVAN_mother_anthro_indicators)

• This dataset includes data that were collected during cross-sectional surveys at baseline and endline administered to mothers and their children (aged 0-23 months).
• Baseline and endline data have been appended and are identified by the variable ‘time’.
• Variables include mother and child anthropometry and haemoglobin concentration, infant and young child feeding practices, mothers and child dietary intake, care seeking and health practices, women’s empowerment (e.g., access to credit; decision-making in productive and health-related domains; group membership; time use; and resources/asset ownership) and exposure to the UPAVAN interventions.
• This dataset also includes all UPAVAN trial outcome indicators. For more information on indicator definitions see: 1016/S2542-5196(21)00001-2; appendix pp 15-16. Note that while all UPAVAN trial outcome indicators are available in this dataset, data used to generate several of these indicators (e.g., those related to household consumption/expenditure, agriculture production, and male empowerment) are found in the males dataset.
• Other indicators generated during data processing are also available in this dataset, alongside several key variables merged from the males dataset.

Quantitative males (mother’s spouse or household head, if unavailable) dataset (UPAVAN_males)

• This dataset includes data that were collected during cross-sectional surveys at baseline and endline administered to mother’s spouses or household head, if unavailable.
• Baseline and endline data have been appended and are identified by the variable ‘time’.
• Variables include sociodemographic information, asset ownership, men’s empowerment, household agriculture production, and household consumption.
• 50% of households were randomly allocated the questions on household consumption and 50% on expenditure; this is indicated by the variable ‘survey_type’.
• This dataset also includes trial outcome indicators related to agriculture production, and household expenditure/consumption in addition to several other indicators generated during the making of these indicators, and other data processing.
• Whilst in the generation of agriculture production and household expenditure-related indicators the authors conducted quality checks to ensure logic and remove any potential outliers/implausible values, variables shared to generate these indicators have not been changed based on these quality assurance procedures. Therefore, the authors highly recommend that appropriate quality checks are applied when using these variables. For more information on quality checks and indicator generation please contact helen.harris-fry@lshtm.ac.uk or emily.fivian@lshtm.ac.uk

Further information

• The accompanying codebooks have additional information required to understand aspects of the data.
• Mothers and males surveys can be merged on ID (unique household identifier). Some mother’s surveys are missing a male survey; these can be identified by using Stata code ‘tab survey_type, m’.

Geographic regions

Keonjhar district, Odisha, India.

Key dates

• Baseline surveys took place from Nov 24, 2016, to Jan 24, 2017
• Clusters were randomised from December, 2016, to January, 2017
• Interventions were implemented from March 20, 2017, to Oct 31, 2019
• Endline surveys were conducted from Nov 19, 2019, to Jan 12, 2020

Quality controls

• The study followed a pre-registered protocol: 10.1186/s13063-018-2521-y.
• Dietary diversity assessments were standardised by observing each interviewer administer a 24-hour recall and comparing results against a gold standard interviewer.
• Certified laboratory technicians completed anthropometry standardisation on a set of ten mothers and ten children (aged 0–23 months) before data collection began. Technical error of measurement results for the laboratory technician teams at endline were less than 0.49 cm for height, 0.25 cm for adult MUAC, 0.06 kg for adult weight, 0.19 cm for child length, 0.12 cm for child MUAC, and 0.08 kg for child weight. Supervisors retrained interviewers and technicians with unsatisfactory results.
• Data collection training took 5 weeks. A data quality assurance team did spot checks on 10% and back checks on 20% of households.
• The statistical analysis plan was approved by the trial steering committee, which included an independent statistician.

Population

Mothers with a child 0-23 months and their households in four administrative blocks (Ghatgaon, Harichandanpur, Patna, and Keonjhar sadar) of Keonjhar district, Odisha, India.

Mothers with any disability impairing participation in the surveys, children with any disability affecting weight, standing height, or recumbent length, and household members who were not residents of the household for at least half a year before data collection were not eligible trial participants.

Privacy

No names of participants or exact date of birth are included in this dataset. Village and block names have been removed, and instead, different villages and blocks are indicated by unique numerical values (the corresponding village/block names and values are only known to the research and project team). Geolocation data has been corrected to missing. To protect anonymity of participants, the variable related to the question “Which of the following tribes do you belong to?” has been removed due to low-frequency responses across several categories. Additionally, due to several unique values for household size, variables related to household size have been placed into broader categories.

Ethics

Keywords

Nutrition, randomised-controlled trial, complex intervention, agriculture, India, nutrition-sensitive agriculture

Language of written material

English

Project information

Creators

File description