Quality of anthropometric data in India's National Family Health Survey: Disentangling interviewer and area effect using a cross-classified multilevel model

Document Type

Article (peer-reviewed)

Publication Date



India has adopted a target-based approach to reduce the scourge of child malnourishment. Because the monitoring and evaluation required by this approach relies primarily on large-scale data, a data quality assessment is essential. As field teams are the primary mode of data collection in large-scale surveys, this study attempts to understand their contribution to variations in child anthropometric measures. This research can help disentangle the confounding effects of regions/districts and field teams on the quality of child anthropometric data. The anthropometric z-scores of 2,25,002 children below five years were obtained from the fourth round of India's National Family and Health Survey (NFHS-4), 2015–16. Unadjusted and adjusted standard deviations (SD) of the anthropometric measures were estimated to assess the variations in measurements. In addition, a cross-classified multilevel model (CCMM) approach was adopted to estimate the contribution of geographical regions/districts and teams to variations in anthropometric measures. The unadjusted SDs of the measures of stunting, wasting, and underweight were 1.7, 1.4, and 1.2, respectively. The SD of stunting was above the World Health Organisation threshold (0.8–1.2), as well as the Demographic and Health Survey mark. After adjusting for team-level characteristics, the SDs of all three measures reduced marginally, indicating that team-level workload had a marginal but significant role in explaining the variations in anthropometric z-scores. The CCMM showed that the maximum contribution to variations in anthropometric z-scores came from community-level (Primary Sampling Unit (PSU)) characteristics. Team-level characteristics had a higher contribution to variations in anthropometric z-scores than district-level attributes. Variations in measurement were higher for child height than weight. The present study decomposes the effects of district- and team-level factors and highlights the nuances of introducing teams as a level of analysis in multilevel modelling. Population size, density, and terrain variations between PSUs should be considered when allocating field teams in large-scale surveys.