Abstract

The maternal hemoglobin (Hb) data are highly skewed in sub-Saharan Africa, with some women showing higher and others lower levels of Hb. Therefore, it is essential to use appropriate statistical methods when analyzing such data to avoid biased conclusions. This chapter examined the performance of mean, quantile, and robust regression diagnostic methods in identifying outlier women to Hb levels in Malawi. The analysis involved both simulations and real women Hb data from the 2015–2016 Malawi Demographic and Health Survey (DHS). The calculations were done using STATA software version 17. The simulation results revealed that in large sample sizes the detection rates for outliers were similar among the conventional mean, quantile, and robust regression models. In small sample sizes, the Matthews correlation coefficient (MCC) and Cohen’s kappa \((\kappa )\) results showed perfect agreement between the actual perturbed outliers and those detected by each model’s residual for all the models and moderate agreement in large samples, except for the Q90 model residual which showed disagreement. Further, when the datasets had no outlier observations, all the models performed with similar estimates based on standard errors and bias. However, for datasets that had outlier observations, the robust and quantile regression methods with first and second quartiles yielded most accurate estimates with smaller biases compared to the linear model and 75th and 90th percentile models. The real data analysis showed that directions of estimates were similar across the models, but the linear, robust maximum (M), and multiple maximum (MM) likelihood estimator models produced estimates with smallest standard errors. The estimated average Hb level for women was 13.7 gram/deciliter (g/dl). Residing in rural area, higher body mass index, and having primary and secondary education were linked to high Hb levels, while older pregnancy, drinking from safe water sources, and living in a rich household were associated with low Hb levels. The model residuals detected considerable amount of outliers in the data; mostly, they were women with extremely low Hb levels. These findings suggest that triangulating a variety of statistical methods to analyze Hb data will help in concretizing evidence of the burden of maternal anemia in sub-Saharan Africa. This will help in determining appropriate evidence-based interventions to combating the problem in the region. Policymakers in the maternal health sector should devise interventions to boost women’s Hb in Malawi, targeting expectant mothers in the second and third trimester and other outlier groups of women in the society.

Original language	en
Pages (from-to)	313-353
Publication status	Published - 2025