ABSTRACT Reliable anthropometric data are essential for tracking child malnutrition, yet such data are prone to measurement errors, particularly in large‐scale surveys conducted in resource‐limited settings. This study assesses the quality of anthropometric data in the Bangladesh Demographic and Health Surveys (BDHS) from 2014, 2018, and 2022, focusing on how different data filtering approaches affect malnutrition estimates. We analyzed data from children aged 6–59 months using the BDHS children's recode files. Data quality was assessed using WHO‐defined thresholds for biologically implausible values and SMART plausibility criteria, which exclude values beyond ±3.1 standard deviations from the sample mean. Additional checks included z‐score distribution properties, digit preference scores (DPS), and demographic consistency measures. Prevalence of stunting, underweight, and wasting was calculated before and after data filtering, with subgroup analyses by age, household wealth, and family size. SMART flagging excluded about 12% of records per round—substantially more than WHO thresholds. Stunting and underweight estimates remained broadly stable over time, while wasting prevalence dropped by 1.5–2.5 percentage points post‐flagging, reflecting sensitivity of prevalence to filtering thresholds. Children aged 6–23 months were more frequently flagged. A steady decline in digit preference scores suggested improved measurement consistency, though variability persisted by socioeconomic group. Overall, BDHS anthropometric data showed stable internal patterns across survey rounds rather than conclusive reliability. These findings are presented as scenario analyses comparing WHO and SMART filters, illustrating how methodological choices influence malnutrition estimates. Strengthening field protocols—particularly for weight‐for‐height among younger children—could further enhance data precision and comparability.
Ahmed et al. (2026) studied this question.