Missingness of Height Data from the Demographic and Health Surveys in Africa between 1991 and 2016 Was Not Random but Is Unlikely to Have Major Implications for Biases in Estimating Stunting Prevalence or the Determinants of Child Height |
Authors: |
Amelia B Finaret, and Matthew Hutchinson |
Source: |
Journal of Nutrition, 148(5): 781–789; DOI: https://doi.org/10.1093/jn/nxy037 |
Topic(s): |
Child health Data quality Nutrition
|
Country: |
Africa
Multiple African Countries
|
Published: |
MAY 2018 |
Abstract: |
Background
Obtaining accurate information on child height is essential for targeting interventions to reduce stunting. Thus, large-scale nutrition surveys must ensure that samples are representative of underlying populations of interest. Without accurate representation, resources for combating child stunting may be inefficiently allocated.
Objective
This study examined differences between children with (92.7%) and without (7.3%) complete and biologically plausible height data available from the Demographic and Health Surveys.
Methods
A total of 116 Demographic and Health Surveys conducted between 1991 and 2016 from 35 countries in sub-Saharan Africa were merged. Differences between children with and without biologically plausible height data were examined with the use of chi-square tests, t tests, and bivariate and multivariate logistic regression with survey cluster-level fixed effects.
Results
Of the whole sample, 97.9% of children had complete height data and 92.7% of children had complete and biologically plausible height data. There were sociodemographic and socioeconomic differences between those with and those without complete and biologically plausible height data. Children with usable height data were more likely to have a health card seen by the survey enumerator [mean height-for-age z score (HAZ): -1.32] than not (mean HAZ: -1.44) (P < 0.001), be older (mean HAZ: -1.63) than younger (mean HAZ: -1.11) (P < 0.001), have been ill in the previous 2 wk (mean HAZ: -1.43) than not ill (mean HAZ: -1.33) (P < 0.001), live in urban areas (mean HAZ: -1.13) than in rural areas (mean HAZ: -1.44) (P < 0.001), have literate mothers (mean HAZ: -1.16) than illiterate mothers (mean HAZ: -1.53) (P < 0.001), have mothers with more education (mean HAZ: -1.23) than not (mean HAZ: -1.54) (P < 0.001), and have more household wealth (mean HAZ: -0.82) than not (mean HAZ: -1.56) (P = 0.038).
Conclusions
Missing data from the DHS anthropometry questionnaires may affect research on child height, but overall effects are likely small. Given the trends in nutritional epidemiology toward the use of large-scale national surveys, understanding the ways in which biases arise as sample sizes increase is essential. |
|