Table 2. Risk of bias assessment and potential approaches.

	Criteria	Potential biases	Proposed approach admin data	Proposed approach survey data
1.	Population representativeness of available birthweight data	Biases in newborns included in data source and biases in birthweight availability for included newborns. (Table 1 – Potential biases 1,2,3)	Exclude if the total births with a weight in the data source is <80% of UN-estimated population of live births. Consider sensitivity analyses.	Include only surveys designed to be nationally representative. Only include surveys with valid birthweights for ≥30% of births, and for those, undertake multiple imputation to impute birthweight data for included newborns with missing birthweight. Set a stricter inclusion criterion of ≥95% requiring a valid birthweight for surveys where multiple imputation is not possible.
2.	Birthweight distribution	Biases due to missing birthweight for very sick babies and those born around the threshold of viability (Table 1 – Potential biases 1,3)	Categorize data where possible into LBW subgroups % for very low birthweight, extremely low birthweight and <500g. Review distributions and identify data with evidence of under-capture of those <1,000g. Consider adjusting these data or sensitivity analysis based on excluding these data.	Multiple imputation to impute birthweight data for included newborns with missing birthweight. Very sick or small babies who die immediately after birth may not be captured in the birth history at all. Thus, consider sensitivity analysis based on excluding data points with evidence of under-capture of those <1,000g.
3.	Measurement errors due to heaping	Heaping of recorded birthweight on 2,500g. (Table 1 – Potential biases 4)	Consider use of administrative data birthweight heaping index for countries with available information to identify indicators of countries that have higher and lower prevalence of heaping. Use model terms for categories of administrative data in the Bayesian model to adjust data in countries that are expected to have high heaping.	Exclusion of surveys with extreme heaping (>55% of all birthweights falling on the three most frequent birthweights and <5% of births on the tail ends of ≤500g and ≥5,000g) Also, heaping adjustment undertaken as part of the pre- modelling data processing.
4.	Measurement errors due to misclassification of live births as stillbirths	Most likely in babies around the perceived thresholds of viability, which vary by context (Table 1 – Potential biases 4)	Methods detailed above on birthweight distribution to attempt to identify missing babies around the threshold of vulnerability.	Misclassified newborns will be missing from the survey dataset. Methods detailed above on birthweight distribution to attempt to identify missing babies around the threshold of vulnerability.
5.	Measurement unit error	Confusion in surveys where birthweights may be provided in both pounds and grams (Table 1 – Potential biases 5)	Not applicable	Exclusions based on >10% of all birthweights ≥4,500g. 5
6.	Incorrect denominator used	For example, where a large number of newborns in the data source did not have a recorded birthweight and the denominator used includes all newborns in data source, rather than all newborns with a birthweight in the data source. (Table 1 – Potential biases 6)	Re-calculate LBW prevalence estimates using the correct denominator, if available; explore other approaches to account for bias if not.	Re-calculate all LBW prevalence estimates.

Note: Any remaining error will be captured by model terms for non-sampling variability