It was a great plan for Dr Sumon Das of the School of Demography to rehearse this talk on Tuesday 4 July for the School , as he will be presenting it to the International Statistical Institute Conference known as the World Statistics Congress in Ottawa later this month. This work was partially funded by an ANU College of Arts & Social Sciences Small Grant, allowing for the employment of a research assistant to undertake the wrangling of the remote-sensed data.
I was concentrating on how long the talk went for, any typos in the slides and so on and so for the content, I don’t think I can do better than copy in Sumon’s comprehensive abstract! Here it is. Look out for the journal article, under review as I write.
“This study aims to estimate trends in chronic undernutrition (stunting) among children under five years old in Bangladesh at micro-level administrative domains, encompassing 64 districts and 544 sub-districts. We utilize remote-sensed data such as night-time light intensity, serving as a proxy for urbanization, in addition to precipitation, land surface temperature, and the Normalized Difference Vegetation Index (NDVI) as environmental factors. Our analysis incorporates data from six rounds of the Bangladesh Demographic and Health Survey (BDHS) spanning the period from 2000-2018, which provides geographic coordinates of the sampled clusters.
Bayesian multilevel time-series models are developed, assuming the outcome variable follows a binomial distribution with the mean equating to the estimated stunting prevalence derived from the micro-data. These models leverage cross-sectional, temporal, and spatial data to interpolate stunting levels in non-survey years for all small domains, conditional on the relationship of outcome variables with the remote-sensed data. The remote-sensed variables significantly contribute to the borrowing of strength across space and time. Estimates for higher aggregation levels are obtained by aggregating the small domain predictions, allowing for the examination of numerical consistency in stunting prevalence from micro to macro levels.
Results show that the national-level trend in stunting has experienced a steep decline over the period, decreasing from approximately 50% in 2000 to about 30% in 2018. The trends at the district level reveal that some districts with higher stunting levels over the last two decades exhibit consistently higher vulnerability, while others demonstrate more variability. At the sub-district level, the direct estimates, which were highly volatile and left 30-50% of domains unobserved in all surveys, are significantly improved through the use of multilevel time-series modeling. The findings of this study provide data-driven evidence for monitoring progress towards meeting nutrition goals to date within detailed administrative domains in Bangladesh. The utilization of accessible remote-sensed data enhances both the precision and reliability of these findings.”
aliceinstatisticsland 