Multivariable multilevel analysis

The EDHS used a multistage cluster sampling technique, whereby data were hierarchical (i.e., mothers and infants were nested within households, and households were nested within clusters). Considering the hierarchical nature of EDHS data, mothers and infants who lived within the same cluster may have had similar characteristics to other mothers and infants compared to those in other parts of the country. Considering the clustered sampling approach, a two-stage multivariable multilevel logistic regression analysis was used to estimate the effects of individual-household- and community-level determinants on infant mortality. Backward stepwise multilevel logistic regression analysis was performed to select individual-, household- and community-level variables to each model and those variables with p-value > 0 .25 were removed.

The fixed effect sizes of individual-, household- and community-level determinants on infant mortality were expressed as AORs with 95% confidence intervals. A p value of .05 was used as the cut-off for statistical significance. Additionally, the measure of variance (random effects) was reported in terms of the intraclass correlation coefficient [35] and proportional change in variance [36].